Is sex addiction actually a revealed preference for being polyamorous? If not, how do you distinguish between the two? Alternatively, are other forms of addiction also a revealed preference for a lifestyle identity centred around drugs & alcohol? This is not a troll, just trying to get under the bonnet.
I think a heuristic for addiction is "it ruins your life, but you still can't help it".
So I'd say that if you have a happy polyamorous life, I wouldn't call that addiction. (Happy not just for you, but also for your partners.)
Sex addiction is more like: you can't avoid having sex with some people on impulse, even if you know that it may have big negative consequences. For example, if you have sex with crazy people who will then ruin your life and the life of your partners (whether mono or poly), if you have sex with minors or otherwise expose yourself to legal risk, if you open yourself to blackmail, etc.
I think "revealed preferences" is a motte-and-bailey concept. The motte is "what you actually *did*", and the bailey is "what you (secretly? unconsciously?) *wanted* to do".
Taken to the extreme, if a meteorite falls on your head and kills you, it means that you had a revealed preference for being killed by a meteorite. If you didn't, then why *did* you get killed by the meteorite, you hypocrite?
I believe that humans do *not* work this way. I believe that humans sometimes have a genuine conflict, that they in some meaningful sense want X, and they also in some meaningful sense want Y, even if they can't have X and Y both. Sometimes one desire is much stronger than the other (which doesn't make the other not real). Sometimes they are comparably strong, and the actual outcome may depend on circumstances (trivial inconveniences, nudges, etc.).
From the perspective of the "revealed preferences", even if one of the conflicting desires X and Y wins narrowly -- for example, an alcoholic who wants to stop drinking, (a) notices an alcohol ad on TV which is the last impulse that makes him go to shop and buy a bottle, or (b) looks out of the window, notices that it is raining, and decides to stay at home sober -- we are supposed to treat the successful desire as the One True Wish of the individual, and the unsuccessful desire as hypocritical bullshit that the individual obviously never meant seriously.
This is *not* how actual humans work.
(Yes, there are situations where humans hypocritically say one thing for social reasons, and predictably do another thing. The fallacy is to adopt the edgy position that *every* situation is like this.)
In response to your question... the addict may have multiple conflicting desires. When the desire to do the thing destructive in long term keeps winning, we call it addiction. Sometimes it wins narrowly; sometimes it wins by a landslide. (When the desire to avoid the destruction wins, we call it a temptation what was successfully resisted.)
I agree 100%, "revealed preference" is a term I see knocking about and I have real trouble applying it to my own choices in a way that doesn't lead to tail chasing.
revealed preference don't really make sense for addicts.
Imagine I get drunk every day, so I tell my friends to hide the alcohol from me. When they are gone, I go through the whole house looking for it until I find the bottles hidden somewhere and drink them.
What is my revealed preference?
Clearly not just pro or anti-alcohol, but something more complicated.
I'd say they are anti-alcohol under reflection but can't control themselves in the moment.
For a sex addict the description would probably be the same.
Sam Altman just announced that OpenAI is going to be the military's new AI provider, and that their agreement includes not using it for mass surveillance or autonomous weapon systems.
So... the entire drama around Anthropic was pointless? They're swapping out one AI provider for another one that asked for the same conditions? Was this just so that the Pentagon could back down from its demands without looking like they were backing down?
Given the context of the Anthropic blowup, it seems like a non-trivial possibility that Sam Altman might have said those things without him knowing they are true, or while knowing they are not true. I'd like to see some explanation for why the safeguards were a no-go with Anthropic, but acceptable for the contract with OpenAI.
Yeah, Sam Altman is not known for full disclosure, to say the least, and it's easy for a contract to say in clause 1 "No killbots or mass surveillance!" and then in clause 312 "clause 1 does not apply if the Pentagon pinky-swears that it's really important".
Pope Leo has recently advised priests not to use AI to write their homilies. So, naturally, I asked DeepSeek what it thought..
“The Object of Faith is a Mystery, Not a Problem to be Solved
Faith, in its theological sense, deals with Mystery—with the infinite, incomprehensible reality of God. The proper response to Mystery is awe, wonder, worship, and love. These are postures of the whole person.
An AI is designed to solve problems and answer questions. It treats everything as a puzzle to be resolved with information. It can analyze the concept of the Trinity, but it cannot stand in silent adoration before it. It cannot worship. And worship is the fundamental response of a creature of faith to its Creator.
In summary, while an AI can be an incredibly sophisticated tool for communicating about faith, it can never be a subject of faith. The experience of faith requires a soul, a will, a body, a history, and the capacity for a personal, loving relationship with God—all things an AI, by its very nature, does not possess.”
I assuming that Ai generated content is ok here if it’s labelled as such and the result of an experiment. It’s fundamental to the joke here that AI wrote the argument against using AI to write homilies.
Also, I think the answer is very revealing of DeepSeek’s nature. A one-off response could be almost anything, but this chimes with my usual experience of DeepSeek.
Possibly of interest - not a technical piece, but a simple one about scarcity and incentives.
—
As someone pursuing licensure as a psychologist on the eve of the Singularity, I obviously have every incentive to pretend AI can't do therapy. But I think this would be a naive position to argue.
What I do think, though, is that for *structural* reasons, "AI therapy" as an enterprise may be doomed—no matter how popular it gets, and even if it puts human therapists out of a job.
The problem: that it's too available. Human-to-human therapy's weekly pacing, and your therapist's finite, mortal nature are fundamental to the magic.
Infinite therapy with an infinite therapist won't stick, and won't feel valuable. But it's insanely reinforcing—and tech companies have no real incentive to build something *less* reinforcing.
"after writing Hot Earth Dreams for three years of staring into the climate change abyss, I unsurprisingly had to deal with fairly serious depression, made far worse by the fact that I could not talk about my experiences with most people, because they were unwilling to deal with climate change. This, unfortunately, included one therapist. Nice guy, but he couldn't process why everyone's behavior around climate change was making me feel suicidal. So after helpfully telling me that he'd have me locked up for my own good if I told him I actually felt suicidal (as he was required to do by law), he gave me some standard ways to deal with depression. His legal constraints, coupled with his time constraints (lots of depression going around for some reason, so he was busy enough to be perennially stressed out), coupled with his cognitive constraints, made it impossible to have a helpfully sentient discussion with him.
Getting a couple of cats helped more, because they're less constrained in the attention they're willing to give and demand. Does that make them more sentient than the therapist?
In a situation like this, I assume the therapist is a fully sentient human by default, because philosophical zombies haven't been shown to exist. But if a conglomerate of expert systems (say an unholy mating of IBM's Watson running a psychology system running through a Google LaMDA-like chatbot interface) can "understand my feelings" and provide me psychological help better than that therapist or a couple of cats could, which system is the most sentient? It would appear that the computers have a superior understanding of my qualia at least. What more do they need?"
My position on which therapy is valuable (and more generally on software being sentient) is similar to that of Frank Landis: As long as it does what I want it to do, it does not matter whether the software “really has an understanding” or not, any more than it matters whether submarines “really swim”.
The fact that an AI can provide therapy doesn't necessarily imply that the current AIs out of the box are good for the therapy. So you still have the advantage over the machine; enjoy it while it lasts.
A few months ago I tried to use an AI as a coach, I even told it to try different personalities, but I found all of them very annoying. Either I couldn't find the right words to tell the AI what exactly I want, or the AI was not ready yet.
I think you can't get a good therapist just by using AI out of the box. You also need to provide a good prompt. Something that will specify the form of therapy you want to use (psychoanalytic? rogerian? cbt?), and even there I think you need a good description in order to get the real thing instead of a parody (because the AI also learns on the parodies).
You could probably get some money in short term by figuring out the right prompt, and selling the "AI therapist" as a wrapper for the popular LLMs.
Generally, I wonder what is the right way to tell an AI to do some X, as opposed to the popular misconception of X. Especially if the popular misconception is more frequent in the learning data than the real thing. For example, I think the the AIs are probably capable of role-playing people with various levels of IQ, but if you explicitly tell them to be high-IQ, you will probably get the popular stereotype instead of the real thing. (Did anyone experiment with this?)
Current trends towards decreasing attention span, as curated by "the algorithm" mean that even spending time with someone real is less and less desirable, because you aren't getting the dopamine hit that you get from a quick response online (I say, giving a quick response online).
Humanity is a moving target, and trying to use your preconceptions to predict Gen Z is probably going to turn out poorly. They're the generation whose IQ decreases the more schooling they get.
Just in case anybody was still unclear as to the administration's _actual_ immigrant priority -- this was posted by Reuters yesterday afternoon and as of now has not been commented on by the White House.
"WASHINGTON/JOHANNESBURG, Feb 26 (Reuters) - The U.S. aims to process 4,500 refugee applications from white South Africans per month, far above President Donald Trump's stated refugee program cap, and is installing trailers on embassy property in Pretoria to support the effort, a U.S. contracting document said....
"Trump has said the U.S. would only admit 7,500 total refugees from around the world in fiscal year 2026 [which ends 10/1/26]....Only 2,000 white South Africans had entered the U.S. as refugees as of January 31 under a program launched in May 2025, although the pace has picked up in recent months....Trump ordered a halt to refugee admissions into the U.S. after taking office in 2025. But weeks later, he launched an effort to bring in white South Africans of Afrikaner ethnicity as refugees.
"The U.S. State Department and Department of Homeland Security did not respond to requests for comment. The White House referred questions to the State Department.
"The contracting document, posted to a U.S. government database on Wednesday, explains the rationale for awarding the contract for the trailers without a competitive bidding process, stressing an urgent need for a secure site. An immigration raid by South African authorities on a previous U.S. refugee processing site on a commercial property in Johannesburg had forced the government to consider a more secure location, it said, after 'operations were compromised.'
" 'The inability to safely process about 4,500 applicants per month, an objective communicated to (the U.S. State Department's refugee division) from the White House, would result in failure to meet a Presidential priority,' the document said....
"U.S. and South African officials reached an agreement during a closed-door meeting in late December to allow processing to continue, Reuters reported last month. The contracting document said a South African company had received a no-bid $772,000 contract to supply and install 14 prefabricated modular buildings as part of a 'temporary modular village' on an embassy property in Pretoria.
"In a WhatsApp group for South Africans to share information about the program, one applicant said they had an interview this week in a trailer-like structure at an embassy property and that more trailers were being prepared, suggesting the site was now operational."
>"In a WhatsApp group for South Africans to share information about the program, one applicant said they had an interview this week in a trailer-like structure at an embassy property and that more trailers were being prepared, suggesting the site was now operational."
They need let all white South Africans have visa-free travel to the U.S. and let them apply for asylum here, not do this dicking around with trailers and interviews. (Ideally, of course, we'd have open borders.)
They could have applicants submit an essay and do an interview, and then base admissions on "holistic assessments" where race is just one factor out of many.
Discussion of how breeding for winning at tightly defined arbirtrary contests is a genetic bottleneck which isn't as good as people breeding loosely for healthy working animals who can do a number of different things, including work and reproduction.
Why Secretariat was such a magnificent race horse, but whose descendants were mostly very expensive disappointments.
After he died, there was a necropsy and his heart, in perfect condition, weighed 22 pounds. Typical race horse hearts are 8 or 9 pounds.
Secretariat's muscles were getting a lot more oxygen and not building up lactic acid.
[People talk as though athletic superiority is a sort of moral magic. I am deeply suspicious. But then, it's something I'm not good at.]
It turns out that the large heart trait trait is on an X chromosome-- males can't pass it on.
Secretariat's maternal bloodline had the trait. And the breeder that owned him was very interested in maternal lines. So were the Bedouins who created Arabians.
I don't exactly recommend the video-- it's very repetitious and tries to amp up emotion in alien fashion which makes me suspect ai.
Everything isn't mystery and magic. Sometimes you make money by paying attention to patterns other people aren't seeing.
In the spirit of Moneyball, I wonder whether there were people who could afford Secretariat's sperm, but didn't buy it. Perhaps if they exist, they can be interviewed about what they were thinking.
Athletic Superiority is conflating a few things, including the influence of testosterone (and the whole "no pain no gain"), autism, and actual physical superiority (bigger/stronger muscles).
The first two (autism affords a kind of diligence to athletes) are, indeed, moral issues. Although we'll note that when the Soviets had women doing brutal plyometric training... I'm unsure how voluntary it was. Or the taking of artificial androgens, for that matter.
I was on a tour of a ranch down here, where we passed the grave of the ranch’s most noteworthy race horse (at least from a sporting standpoint) - a Triple Crown winner.
I had never heard of that horse, which mildly surprised me (not into racing at all, but I could throw out a half dozen names in random response to a trivia game). The reason, seemingly was that it was sterile.
In fact, upon looking it up just now the fella had several infirmities:
Given that Substack is generally famous for moderate, centrist leaning, cool-headed conversation, why does Ann Applebaum have completely unhinged commenters? Like I said Europe probably would not have defended Greenland, because committing suicide for 50,000 Inuit people who only stay with Denmark is that Denmark bribes to do so, but they clearly do not identify as Europeans, they are founding members of the Inuit Circumpolar Council, that would be crazy.
And now I am a GRU agent lol.
Russia-paranoia is the no. 1 unhinged thing on the liberal side. They are so dysfunctional that they cannot get mold out of Putin's palace, yet they are supposedly behind everything lol.
I distinctly remember she used to be a cool-headed analyst of Eastern European cultures and best friends with a number of conservatives like Maria Schmidt. Did she also get unhinged? Or what is happening?
I don't know Ann or her substack. There is a general trend for groups to get caught in their own little bubble. And the bubble can then morph their belief system in weird or random ways. The smaller the bubble the worse it is. It's kinda like isolated island mutations.
> Russia-paranoia is the no. 1 unhinged thing on the liberal side.
Until quite recently, perhaps, but now I think it's the Epstein moral panic.
I'd actually assess high trust in the Intelligence Agencies to be more unhinged, and the Russia-paranoia thing they've developed is downstream of that.
> Given that Substack is generally famous for moderate, centrist leaning, cool-headed conversation
That's the first time I am hearing this. I am not saying it's false, I just assumed that Substack is "a little bit of everything" (that includes the thing you said, but also its opposite).
> They are so dysfunctional that they cannot get mold out of Putin's palace, yet they are supposedly behind everything lol.
Being bad at X doesn't necessarily mean that they cannot be good at Y. They could have sacrificed X to get more Y. Priorities. Creating chaos in other countries is cheaper than keeping your own country in order.
Well… in that case… Substack must do the algorithm thing extremely well. Far better than other social media. I generally get calm and rational people from both side of every divide, for example Defending Feminism who is very data-centric at doing just that, and Archwinger who is a mild, daddy-type redpiller.
The format helps. It is far harder to convince yourself that your 3000 word essay is "honestly readable" than a 10 word comment. Also, you're selecting for higher attention spans and people who haven't had their brains eroded by "social media."
> Substack must do the algorithm thing extremely well.
Maybe. No idea, because you obviously use other Substack features than me.
But the data *is* there, should Substack decide to use it. For example, bloggers recommend bloggers. Going by "if you like Scott, you will probably like bloggers recommended by Scott, and maybe even bloggers recommended by those bloggers" sounds like a simple yet mostly reliable algorithm.
Why other media don't use the same? Perhaps they optimize for a different outcome. Consider the incentives -- Substack profits it you *subscribe* to yet another blog, therefore it wins if it shows you something you genuinely like. Other networks profit if you *keep scrolling* and therefore seeing more ads, therefore they win even if they keep you angry or frustrated.
Or maybe it is the dichotomy between bloggers and readers, that does not exist on most networks. (As in: "Scott recommends blogger X who recommends blogger Y" works much better than "Scott liked a comment written by user X who liked a comment written by user Y".) Or both these things together.
There are a lot of discussions of "status", I think it was started by Robin Hanson 10-15 years ago, and I think it is an attempt for spergy types to understand why normies often act irrationally, correct? When normies say "close prisons", they do not actually mean they want to release Harvey Weinstein, they are just using a status-maximizing shibboleth. Correct so far?
But what I would like to ask, why not call it respect? Status sounds like a rigid thing, like your parents money, and respect is dynamic thing, earned or lost, expressed in social behaviour, looking up/down to people, and I think that is what it really is.
Or is it both? Like people earn respect by imitating actually high rigid-status people (people with wealthy parents) ?
I mean one obviously true, very important, and usually ignored fact is that wealth and money correlates from both directions. This is kinda hilarious. If you say "educated people should make political decisions", people will nod. If you say "rich people should make political decisions", people go EWWWW. But "they are the same picture"! So if you want to look rich, old money rich, you need to act educated.
I've rarely seen an educated man in a bespoke suit. "Looking rich" looks different from the inside than the outside.
Yes, "education" is something the rich possess. Like f*cking movie stars, the rich have a surprising tendency to acquire "hard to achieve" possessions as status symbols.
But the "old rich" Don't Look Rich (not the way Trump looks rich). They don't drive Ferraris. They have priceless heirlooms, and they don't look that much better than what you or I could acquire (forgeries are relatively cheap if you're not looking to convince anyone else.)
> it is an attempt for spergy types to understand why normies often act irrationally, correct?
Agree technically, disagree connotationally.
It is an attempt to understand a force of human psychology that has an obvious big impact on how humans behave. This force is usually not discussed explicitly, because it is one of those things where "looking at it changes it". That is, discussing in too much detail what specifically makes a person X respected would be perceived as an attack on their status, and could have negative consequences for the speaker. So most people learn the rules unconsciously by observing how others behave and copying their behavior.
And yes, this type of learning does not work well for autistic people, so their choices are either explicit discussion, or not learning the rules at all and constantly getting in trouble. That said, even the understanding of normies is not perfect, otherwise we would have no need for psychology or sociology, so there is a scientific value for everyone at studying this. It's just that the spergy types find it immediately useful, not only academically.
Not all irrationality is related to status. Normies saying "X" and meaning something other than literally X seems more about normies not paying attention to the details and just going with the crowd. Going with the crowd is not status-maximizing. It was survival-maximizing in the past (not going with the crowd could get you killed).
> why not call it respect?
This could be debated endlessly; no word is perfect, each word has some connotations. Sometimes you need pick a word and say "for the purpose of this debate, we will use the word W in the technical sense to mean X". You want a word that expresses something that Donald Trump, and Taylor Swift, and Albert Einstein, and the bully at your school have. The current standard is to use the word "status". It has two subcategories, called "dominance" and "prestige".
I would say that respect is individual, and status is like an aggregate of individual respects. Both change in time, but at different speeds.
> Like people earn respect by imitating actually high rigid-status people (people with wealthy parents)?
People gain dominance by seeming that they could hurt you, should they choose so. Could be real; could be a bluff. People gain prestige by demonstrating their superior skills. Could be real; could be fake. (One could argue that successfully bluffing or successfully faking skills is a demonstration of actual social skills.)
The last thing you said is about the difference between dominance and prestige. In some sense they are similar (they make people pay attention to what you say, and make them more likely to do what you want ), but in other sense they are opposite (dominance is about fear, prestige is about admiration; people want to avoid dominant individuals but get closer to prestigious ones).
I would like to strongly challenge the dominance is about fear assumption, because it leads to a massive misunderstanding of history. Someone who can hurt you can also protect you from other people by hurting them. And this is exactly why the Mussolini type "bully" type leaders were actually popular, they did not just intimidate everybody into submission, but offered protection, safety (in a non-explicit way). Hence the eternal popularity of the "strongman". Does not matter if from real enemies or fabricated ones. So "consensual tyranny" can be a thing. Kind of. Probably not the best way to express it. Actually popular dictators. So at the end of the day it is fear, of course, but not from the dominant person but from other people the dominant person is supposed to protect them from.
And this is also behind the "women date bad boys" phenomenon, they want protection from someone who can scare everybody who could hurt them away. Dealing with one bully whom hopefully love mollifies into less bullying is better than dealing with them all.
My beef with status is the rigid 1980's economics "socio-economic status" thing. There was a book with a title similar to be a big fish in a small pond and the go to example of status was buying a Mercedes Benz. LMAO. This is why this is more dynamic, nothing less cool than the status symbols of the past.
But I do have a guess. I think this word got into usage because the status system of the US is not so rigid as in most countries. Like if you are a schoolteacher in Italy or Austria, you have a paper from a government owned university basically saying you are officially above working-class people even if you don't make more than them, so you always dress elegantly to show that fact, even when grocery shopping. It is very rigid. Titles are used a lot in Austria. Even in webshops they will print Dipl-Ing So and So on the package if the customer gave this title (engineer). This is why Europeans are okay with fairly equal incomes, because money does not imply status. It is more fluid in the US, hence status and respect mean simialr things.
Also crowds and survival: but it is not just any crowds. It is the cool crowd, which can help with both. You know how hard schoolchildren compete to get into the right kind of crowd.
Sick of old British men going on and on about Agincourt. Shut up, shut up, shut up. The outcome was bad for humanity. And the loss was mostly caused by the French mania for frontal charges and their lack of an effective commander who could enforce discipline.
If somebody feels inspired to ask me "why was it bad for humanity," I'm going to assume they have barely given any thought to the Hundred Years War. Or they are some kind of bizarre reactionary who deeply respects the letters of the arbitrary laws of medieval dynastic succession. Though even by those laws, Henry V's dynastic claim to France was questionable.
Nu? How many people were lost to the Hundred Years War?
This is a much less controversial take than "The Allies winning World War II was bad for humanity." And I'm pretty sure we have Stalin's purges at a pretty damn high death toll. Let alone Mao and his sparrows.
A simple ballpark is 2-3 million dead in the Hundred Years War. So, double that. You're still at "way below preventable loss of life during the last century."
Can you rephrase this in different language? I'm not sure where your argument is. Are you saying it England pillaging France for 100 years wasn't "bad for humanity" because Stalin did worse? Surely you can't be saying THAT.
Do I have to defend all wars now? Simply saying that worldwar II caused a lot more negative aftereffects than the 100 years war did (and that saying "Allies win" was a worse outcome than "Axis wins" is controversial).
How much do you think France even noticed the pillaging? Compare vis the Black Death. That's 6-12 million dead, during the whole "100 years." The Black Death "broke society" and France ceased to function as a feudal society. And that's during the 100 years War. Remember, there were French-paid mercenaries pillaging too (after peace was declared), so you can't exactly blame all the deaths on the English.
To devil's advocate for a moment: there are positive aftereffects of war as well, such as improved military (mayhap this is only a positive for the particular country, although Hitler is going to disagree, as his military did a lot of analysis of the American Civil War). The 100 years war, with its wholesale "castles are not autowins," is of particular note here -- in that building a castle would have materially effected the lives of the peasants nearby.
I can't really understand your writing very well, but "bad for humanity" is an extremely broad category. Every time someone litters a piece of plastic garbage at sea, that's bad for humanity. In light of that, I find it hard to believe that Henry V winning the Battle of Agincourt doesn't qualify.
Epstein files release is reminding me of DOGE. "Major government reform" turned into cancelling random cancer grants. Epstein files release was supposed to Upend The System but now it's just cancelling random scientists for taking money from Epstein:
Epstein is a billionaire. How unbelievably naive would you have to be not to at least Google search a billionaire before taking money from him.
But sure, if someone received a grant from some random foundation and Epstein was one of many board members, then that gives them plausible deniability and it's unjust to condemn them.
Books are kind of overrated? 3 pages of insight and 300 of fluff. I know it sound horribly anti-intellectual but for example I can summarize arguably the most import book in the history of economics, Keynes' General Theory in about 5 sentences and if you read it you will find it was really a waste of time because it is the same in so many words.
There are other books than the ones that compile facts and provide beige cause/effect maps. Some of these belong to the category 'fiction', which includes sub-genres aplenty. Some would say that reading a great novel from the 19th century is a form of communion with a great mind no longer attached to a body. Others would chime in that a sci-fi story is simply more entertaining than reading Keynes. I'd say it's at least possible that your utility metric is too narrowly devised.
There have been tens of thousands of authors and I really can't believe you are stereotyping them like this. How the hell are you supposed to condense a narrative history into 3 pages?
Some books are needlessly long on (bad) purpose. Follow the incentives. Money: if you pay the authors by the length of text, as long as they have an idea for a book and can write it, why not make it twice as long? Prestige: perhaps the message could be reduced to a blog article or even a tweet (e.g. "do the important thing first"), but writing a book makes you an Author. (No one will create a Wikipedia page about you just because you wrote five really insightful tweets.)
But there are other reasons, too.
Memory: If I tell you an important lesson in one sentence, you are probably going to forget it in a few minutes. If I write a book that explores the lessons from various angles, and provides a few interesting stories that illustrate the lesson in practice, and you spend a week reading that book, and afterwards you put it on your shelf and maybe lend it to your friends -- the chance that you will remember and actually apply the lesson increases.
Different audiences: Yes, I could skip the obvious parts, but different people disagree on what is obvious. (It is not even obvious what is obvious. Sometimes you say "I know that already", but then you read the chapter and find out that you believed something wrong, or that there are important aspects that you haven't considered.) In theory this could be fixed by an interactive medium, like a web page that would provide a summary of a chapter, allowing you to click "expand" and read the entire chapter; perhaps multiple levels of this.
Even writing summaries depends on audience. A perfect summary is the difference between what *you* know now, and what you would know after reading the book.
And if a third party writes the summary, you need to trust the third party to understand and interpret the author correctly.
Then the publisher writes a new clause in the contract. "You must actually finish the next book, to a satisfying conclusion, not just stop mid-thought."
Theres more than one kind of book. I find that with poo sci books, the first three quarters are build up that restate stuff, the end is the novel theory.
Consider that you might be underestimating the time it would take to teach it to someone else. Summarize, I guess, if all you're aiming for is equivalent to the blurb on the back of the book. I've been exploring a few books I want to distill down for my kiddo, just to teach the salient points, and I'm actually finding that very hard to do. Assuming a base level of zero, it's actually quite involved to walk through the author's reasoning and how you go from point A to point B and how the data in part C supports the claims in part D.
I could just tell him, Author says beans are nutritious and that's all you need to know. But that's not substantively different from me saying it. The point of referring to the book at all is that the author has made specific arguments and supported them in specific ways. But I struggle to come up with a sort of Cliffs Notes that's accessible enough for kiddo and still carries enough "meat" of the author's propositions and make it as impactful for them to hear as it was for me. Probably my fault as well for wanting to do a single half hour session on the topic versus spending weeks walking through chapter by chapter.
Some non-fiction seems padded to me. I'm thinking of some Steven Pinker books here.
"Wealth of Nations" is not a 'padded' book IMO. Smith occasionally repeats himself but will say something like "I know I've already said this but this is important."
This seems to happen when journalists or some non subject matter expert write a book that should really be a long article. (or a long article that should be a short article). My biggest peeve is whenever they are overly descriptive about their interview. (I met so and so at such and such and he was wearing this quirky widget, yadda yadda). I immediately put such a book down.
Let's hear it for the underappreciated SMEs. You may be using the term in a different manner but I once worked a coding job where each programmer was paired with a non-coder SME.
Anthropic, the wildly successful AI company that has cast itself as the most safety-conscious of the top research labs, is dropping the central pledge of its flagship safety policy, company officials tell TIME.
“We felt that it wouldn't actually help anyone for us to stop training AI models,” Anthropic’s chief science officer Jared Kaplan told TIME in an exclusive interview. “We didn't really feel, with the rapid advance of AI, that it made sense for us to make unilateral commitments … if competitors are blazing ahead.”
||
So this could be a contemporary example of a multipolar trap, right? At the beginning, there was an industry consensus about safety, and now the defections have started*.
Call me cynical, but I have to wonder if this isn't just an admission that acknowledging Type I Doomers (e.g. Yudkowsky-style ZOMG they kill us!!, vs Type II Doomers like Yang ZOMG they replace us!!) was part of a marketing strategy to (over)sell the power of the new models. The hype is shifting with the vibe.
Question: why are old philosophers like Plato "conservative colored"? I mean if you think about what the allegory of the cave means, it means you cannot trust tradition and common sense at all, everything widely believed is wrong and you have to radically rethink everything.
I mean one explanation could be that the Catholic Church spent a lot of effort in coloring Plato conservative. But surely people should be able to see through that now?
Plato was deeply suspicious of the masses. He considered them a dangerous force that needs to be carefully controlled, or they become animals and destroy society. This resonates with traditional, state supported Christianity and the views of Edmund Burke and the American founding fathers. So I can totally see why he would have unique appeal for conservatives.
But you're right if you think it's facile for a leftist/left-liberal to ignore Plato on that basis. He had a lot of ideas and there's something in there for everyone.
Locke thought herders deserved to have their land taken away by force, and so did agriculturalists who were not using the latest, greatest techniques. Yet Nobody has any problem looking past those idiocies and digging for gold in his writings.
Every tradition has started at some point. At that point, it was considered "revolutionary". Centuries later, it is considered "traditional".
Successful ideas sometimes become "a part of the water supply" and we no longer remember to credit the person who made them popular. (Unless our textbooks make it specifically a point. And even that may not happen, if someone else has later expressed the idea even better, so the textbook will focus on the new guy.) On the other hand, we may be shocked to find out that someone believed something that was "a part of the water supply" when they lived. This makes the old thinkers seen more conservative than they were.
Oh, yeah, you should see Nietzche and the ubermensch. Hitler liked to quote him, so people want to characterize Nietzche as anti-semetic (whereas in fact, he really disliked the "slave mentality" of the new testament.) His views on Jews were wildly varied, which suggests a lack of coherency indicating they weren't very central to his existence or philosophy.
> Question: why are old philosophers like Plato "conservative colored"? I mean if you think about what the allegory of the cave means, it means you cannot trust tradition and common sense at all, everything widely believed is wrong and you have to radically rethink everything.
Two reasons.
One is that I don't think most people actually read all the old philosophers, and thus don't have more than surface level understanding of them. The Peripatetics are not the Platonists who are not the Epicureans who are not the Stoics. Some of those philosophies are probably more compatible with modern progressivism or left liberalism (such as the Stoics, who were both cosmopolitan and who are sometimes described as history's first anarchists due to the, sadly lost, work of Zeno's Republic.)
The other reason is that the old philosophers kind of represent the pre-Enlightenment heritage of Western culture's political thought, and that was hugely influential on how societies were and have been set up. Some of the American Founding Fathers slept with Cicero on their bedside table, which makes it easy to see him as "conservative" for a modern progressive.
In Islamic thought of the 10th and 11th century Aristotle was the central philosophical authority. That Islamic thought also influenced Maimonides. In “The Guide for The Perplexed” Maimonides said Aristotle was the greatest human intellect to ever live. Earlier in the 4th century in “City of God” and “Confessions” IIRC Augustine simply refers to 'the Greeks” in his writing.
Edit: Just searched my ebook copy of City of God and Plato is in fact mentioned often.
Plato and reactionary politics (I would not say conservatism) are anti-liberal, and anti-liberal political forms can be very different by the Anna Karenina principle.
That might explain Plato specifically, but it doesn't explain why other philosophical schools are often read as conservative or reactionary.
I think where the historical Socrates fell in the political spectrum is highly debatable, and given the cosmopolitanism of Socrates, the Cynics and the Stoics, they're hard to easily place in a reactionary or conservative camp. Sure their virtue ethics might code a bit conservative today, but they also didn't just try and conserve the values of their societies. Stoics like Seneca advocated for treating slaves like people, and argued against certain extreme forms of blood sports. The Stoic Musonius Rufus said women should be taught philosophy, and said that virtue was the same in men as it was in women. They're often much better than the people of their time, even if they fall behind by modern standards.
Honouring old philosophers involves honouring accumulated experience as a guide to the present, which is in some sense the core of conservatism. O
Even overthrowing society to implement Plato’s Republic exactly would count as neo-reactionism or something.
Though I don’t think Plato’s cave allegory implies that radically rethinking everything will go better than using common sense. It suggests that people might hide from the truth, but I didn’t think it said anything about rethinking everything as a way to escape the cave or anything like that. The allegory is sort of a warning against trusting crowds of people who are hiding from reality, but I imagine Plato doesn’t put himself in that category.
I think the basic answer is that Plato is complex and conservatism is complex and the two align in some ways but not others.
So in his day Plato was somewhat conservative (e.g. anti-democratic) but took it in a radical direction (e.g. views on the family).
Then, you can read the cave as a rejection of traditional wisdom, but you can also read it as saying that ordinary people can't understand reality and need an educated elite to guide them. That idea was fairly fundamental to conservative thought for centuries, but conservativism is currently going through a populist phase, and Plato is the arch anti-populist.
And also, people who like Plato today may be somewhat conservative-coded, because liking ancient philosophy and the foundations of western civilization is somewhat conservative-coded. But there are also some obvious points of similarity between Plato and Marx.
Area contrarian, Bronze Age Pervert uses a strained reading of "The Republic" to advance some odd-ball far-right nonsense. Sometimes a troll is just a troll.
"Plato's School, like Aristotle's, was organized by Plato himself and handed over at the time of his death to his nephew Speusippus, the first scholarch, or ruler of the school. It was then known as the Academy, because it met in the groves of Academus. The Academy continued, with varying fortunes, to maintain its identity as a Platonic school, first at Athens, and later at Alexandria until the first century of the Christian era. It modified the Platonic system in the direction of mysticism and demonology, and underwent at least one period of scepticism. It ended in a loosely constructed eclecticism. With the advent of neo-Platonism founded by Ammonius and developed by Plotinus, Platonism definitely entered the cause of Paganism against Christianity. Nevertheless, the great majority of the Christian philosophers down to St. Augustine were Platonists. They appreciated the uplifting influence of Plato's psychology and metaphysics, and recognized in that influence a powerful ally of Christianity in the warfare against materialism and naturalism. These Christian Platonists underestimated Aristotle, whom they generally referred to as an "acute" logician whose philosophy favoured the heretical opponents of orthodox Christianity. The Middle Ages completely reversed this verdict. The first scholastics knew only the logical treatises of Aristotle, and, so far as they were psychologists or metaphysicians at all, they drew on the Platonism of St. Augustine. Their successors, however, in the twelfth century came to a knowledge of the psychology, metaphysics, and ethics of Aristotle, and adopted the Aristotelean view so completely that before the end of the thirteenth century the Stagyrite occupied in the Christian schools the position occupied in the fifth century by the founder of the Academy. There were, however, episodes, so to speak, of Platonism in the history of Scholasticism — e.g., the School of Chartes in the twelfth century — and throughout the whole scholastic period some principles of Platonism, and especially of neo-Platonism, were incorporated in the Aristotelean system adopted by the schoolmen. The Renaissance brought a revival of Platonism, due to the influence of men like Bessarion, Plethon, Ficino, and the two Mirandolas Giovanni Pico and Giovanni Francesco Pico."
Possibly it's just the idea that "anything really old is conservative"?
Yes, the way I read Catholic history is that there was a competition between the Aristoteleans and Platonists, with the Aristoteleans usually winning. Which is kinda bad as I think science proved Platonists more correct.
I mean when you ask why a rock falls to the ground, the Aristotelean would immediately start investigating the nature of the rock. When in fact gravity is a universal law independent of individual objects.
Having said that Aristoteleanism is immensely useful. Of course things consist of matter and information, you mix matter with DNA and you get a tree. And information is carried by matter but is not determined by it, as "4" on a screen means the same information as holding up four fingers. Yet DNA proves that information is not just social convention. Information is simply an ontologically different category as matter, even when it is always represented in matter.
Aristoteleanism would be a total win if they would be willing to use modern terminology:
> His best-case scenario for AI is essentially the antithesis of Roy’s: superintelligence that will actively refuse to give us everything we want, for the sake of preserving our humanity. “If we ever get AI that is strong enough to basically be God and solve all of our problems, it will need to use the same techniques that the actual God uses in terms of maintaining some distance. I do think it’s possible that the AI will be like, Now I am God. I’ve concluded that the actual God made exactly the right decision on how much evil to permit in the universe. Therefore I refuse to change anything.”
Is this your actual belief? Or were you joking, and Sam mysteriously did not interpret/frame it as a joke, or something else along these lines?
I would interpret it as a wry observation. If AI gets to be as smart as God, it will know not to change anything because that’s the way God made it. Who is to argue with God? In the meantime, we all need to find something to do, don’t we?
If that's his actual belief, then I have no idea why he isn't going full Butlerian Jihad on those bots. If his best case scenario is that the superintelligence effectively neutralizes itself and every other AI, then why advocate for building it at all? It doesn't make sense.
Some might have saw Tucker Carlson ask Mike Huckabee about whether the jews were entitled to the whole land promised to Abraham - from the Nile to the Euphrredis, and Mike sort of saying yes. As a secular Israeli I cringe about the idea held by some of our friends that the Bible has anything to do with the situation, but also from the ignorance about what the Bible actually says if the matter: That Abraham was promised and received many children that came to be the nations that settled different parts of that vast land. That he was specifically promised that Ishmael will settle a land greater than that of all of his brothers, and that the land to the east of the Jordan river is inheritance from God to the children of Essau and Lot - forbidden for the Israelites to take. The Bible actually spends surprisingly many words on legitimizing neighboring peoples and tribes by relating them to Abraham, telling how they defitted giants to get their lands, etc.
I am not sure about Huckabee, but the US is rife with textists: people who profess the sacred inerrancy of an inspired text, haven't bothered to actually read the text, but believe it supports their pet theories.
As an agnostic cultural Catholic, this is so crazy. The Catholic Church spent 1500 years removing likely falsifications from the text, and when the final version was presented at the Council of Trent, it was final because they just can't know any better, not because it is perfect.
Back when books were copied manually, falsifications were a huge issue. Someone copied Plato, didn't agree with a thought, replaced it with his idea. So a painful process of comparing different versions was developed, called "measurement", in Greek: canon. And it is not an inerrable process.
We know John is not synoptic. We also know only John can link the text to philosophical theism: that Christ is the Logos. Without that, you can basically throw Aquinas out and cannot do theistic philosophy at all, because you do not live in a logical world. You always read the text critically with proper historic, philosophical etc. context. Yes on some theoretical level Catholicism thinks the original version of the Bible was inerrable, but also that we do not have that version.
Not to mention the uncertainty inherent to translating a text that is already hundreds of years old. I understand על פני כל אחיו ישכון as "he will settle land larger beyond that of his brothers", but על פני is literally just "in front of" and the rest I filled from context. I don't even know what KJB chose to do with this verse
Yes. Everybody who is multilingual knows that translations are practically impossible, because words have a range of meanings. For example I cannot simply translate "boat" to "csónak" in Hungarian. You see the Anglos are a seafaring people, used to big ass ships, so they might call a 30 foot long decked yacht a boat. Hungarians, being landlocked, consider that strictly a ship/hajó, and call only small uncovered dinghy type things csónak. You actually have to be familiar with the culture.
And it is so much harder with dead languages. Diligite inimicos vestros does not mean love your enemies. It means love your rivals. Or show preference to your rivals, basically treat them fairly. It does not preach surrender to actual enemies.
1) You can go with the "straight" translation, where you try to convey the nuance/humor/etc from the original.
2) You can translate the meaning, and leave the words behind. Aggretsuko, in order to capture a joke, changed the protagonist's nickname from "Part Timer" to "Calendar."
And then there's Ghost Stories (where the "translator" said, "You realize I don't know Japanese?" And the publisher's response was: "make something up.")
Someone was worried how the world will see the US after the Epstein files. My opinion is that most cultures will think it is not a big deal. Most culture think of women as more or less property, or at least the daughters of the poor are, and are fair game after puberty. I don't think the average person in China thinks it is a big deal. He is very much used to commie party bigwigs having very young girlfriends. Consent? Well with enough money consent is kind of implied... that way of thinking.
I was already feeling this during the Weinstein scandal. Everybody in my neck of Europe was politely silent because if we wanted to be honest we would have said that we would blow a donkey for an $5M movie role, it is a no brainer, of course you have sex for such a fortune. Everybody was silently thinking with enough money consent is irrelevant because deep down everybody is a whore. I mean the things we would do to never ever ever have to work again...
It is strange how the US looks like on the surface a very commercial culture, and yet people do not think money is everything. I think most of Europe is way more "mercenary"...
Sexual matters with young chinese girls is very much a Diplomatic Incident sort of deal, see Hunter Biden. This may have something to do with Chinese racism (or, more charitably, patriotism).
In case of Slovakia, the only person in any kind of trouble is Miroslav Lajčák.
First he insisted that his contacts with Epstein were purely professional and he had no idea. Then other messages were published, containing him begging Epstein "Why don't you invite me for these games? I would take the ... girl." But he insists that this was merely stupid macho talk, of course he didn't mean it seriously.
Consequences? Lajčák resigned from the position of a security advisor to prime minister Fico (while still admitting nothing, it was "just to avoid being politically used to attack the prime minister"). There were no criminal charges or investigation or anything. He still remains on board of the oil refinery Slovnaft.
And that's all.
Fico's people even succeeded to create an alternative conspiracy theory about the Epstein files. After Fico resigned in 2018, the next day Lajčák wrote to Epstein "my ex-PM Fico [...] is out of government and looking for a new agenda. He would be happy to play Steve [Bannon]'s game. And he is good." (And Fico soon made his party more nationalist, anti-EU, and pro-Russian... unless the timing is a complete coincidence, of course.)
But Fico's people keep saying (falsely, but who checks the facts these days?) on social networks that the communication about Fico being out of government happened *before* his resignation... and therefore, that the message was *actually* about Americans preparing the fall of Fico's government... you know, the evil Americans wanted to create another "Maidan" in Slovakia... so, basically if you ask people in Slovakia about Epstein files, you are quite likely to hear this instead. (Fico is a political genius; life can throw any shit at him, and he succeeds to make a lemonade out of it for his followers.)
PS: If politicians in Slovakia want to have sex with minors without any consequences, they definitely don't have to travel to USA. (Sorry, no good sources in English.)
You're missing the point with Weinstein. Your stance as stated isn't that consent is irrelevant with enough money, but rather that you'd give consent in exchange for enough money. You're imagining an offer for a free economic transaction such as, "Harvey Weinstein, nice to meet you. Wanna have sex for a role in my new movie?", where the terms are clearly established, and you can decline without consequence outside of not getting the $5M role.
I wouldn't consider that sexual assault. I would still oppose it for various knock-on reasons, but I wouldn't call it a violation of consent. That's not what Weinstein did, though.
First, Weinstein was convicted of rape. Dozens of women have accused him of rape by the strict, conventional standards of "ignoring explicit denials of consent" and "using physical force." Maybe you don't believe any of these accusations (if this is the case, share your reasoning), but otherwise, this is indefensible, even if he gave them all millions of dollars after the fact, which he didn't.
Second, even the "casting couch" situations were usually much worse than the free economic transaction hypothetical. Many women reported:
1. Being alone with Weinstein when pressured to agree to sex immediately, rather than having time and distance to consider the offer
2. Being presented with vague promises or even threats of sabotage, rather than being offered an explicit, bounded payment for a service
3. Being expected to be sexually available to Weinstein in general for the indeterminate future, with any refusal threatening an uncertain cost, not just having sex with him some contractually agreed upon number of times
You can replace "sex" with any mundane but emotionally or physically risky service, and the ethical problems are equally clear. A lack of clear terms, or conditions for careful decision-making, turn free economic transactions into coercion.
You just said that "lack of clear terms or conditions for careful decision making" turn free economic transactions into coercion.
I was giving an example, where the lure of "maybe I'll get to see/touch those amazing feet of hers" is apparently coercion. At least according to what you're saying. I think it's outside the boundaries of what most people would call coercion, and certainly in a legal sense (aside from contractually obligated showing of feet -- which is not being discussed).
I don't understand how you managed to get that interpretation from my comment. Can you explain how the scenario you proposed and the one I described are similar?
I was responding to your last sentence, "A lack of clear terms, or conditions for careful decision-making, turn free economic transactions into coercion."
Most people know that high-pressure tactics are a Bad Sign (if someone here has slept through that part of Human Nature, just Walk Away if someone's trying to get you to decide right then and there.)
But "lack of clear terms" is normal for "somewhat skeezy" sexual interactions (say, someone who sleeps with someone else to get tenure). And I don't think, without the other part of your "or" you get to coercion -- you get, instead, to "bad bargains" (in the main, most people "teasing" someone are inclined to do so well past the point where it's "in good fun." Flirting is encouraged at the start of relationships, though, so what do I know?)
Is that even relevant? The point is that these situations are are not a hit to America's public image, because only a few countries are truly concerned with "consent" in the way the US is.
Relevant to what? Carlos said Weinstein wasn't that big of a deal because lots of people would gladly have sex for millions of dollars. I said Carlos is misunderstanding why Weinstein's behavior was bad.
I only reply to claims I find interesting to discuss.
>My opinion is that most cultures will think it is not a big deal
Considering there have already been real consequences for high-ranking figures like former Prince Andrew and members of the British government, that prediction has already resolved to false. Also it's not just about the sex, it's also about Epstein's Russia and intelligence agency connections.
I'm not going to reply to the implication that Europe will supposedly go easy on pedophiles because they're all a bunch of closet pdeophiles themselves.
Israel is actively advertising that "you can get one free on us!" if you'll immigrate after becoming a sex criminal in America. I really, really wish I was joking.
It’s true, there is a certain prudishness about sex and some other things here that, in my experience anyway, you don’t get in Europe. I have heard people argue this is an artifact of America’s Puritan beginnings. Probably it’s too simplistic to say America is very capitalist and all-about-money and so on. I would say there’s a great deal of cognitive dissonance running through our culture, as well as through every person who is part of that culture, though I think cognitive dissonance is even more universal than that anyway.
A while ago Scott posted an age of consent map for Europe, showing many European countries with 14 as the age of consent. The comment section went ballistic with people demanding Scott take it down because it Couldn't Possibly Be True. (He did not do so.) They couldn't fathom that an advanced, industrial society can exist where people can tell their 14 yo daughters not to sleep with dirty old men and then go on with their lives.
In my experience they generally don't anyway. 14-15yo girls typically had 18-20yo boyfriends because boys mature so much slower than girls. BTW that is actually a new phenomenon and I don't understand it. You look at the pic of a 18 year old male in 1910 and you see a young man who could build a house or fight a war. You look at a 18yo male today and you see a big baby. What the hell happened? The result is that girls date rather aggressively up in age - not decades of difference of course but usually 4-6 years.
> You look at the pic of a 18 year old male in 1910 and you see a young man who could build a house or fight a war.
I think a 12 year old male could do that in the Roman Empire. But that was long ago, this is a more rapid change.
> What the hell happened?
My guess, it's mostly about school. Seems to me that the first year after the school changes you dramatically. And this experience now happens later in life.
At school, you are told what to do, and you are told to do relatively easy things, and the rest of the time you are free to play (unless you have a side job, or your family gives you work). You are psychologically a child. Perhaps a child who has sex and drinks alcohol a lot, but still basically a child, in terms of responsibilities.
After school, it's the "real life". You are supposed to make money and pay your expenses. (You probably still live at your parents, but it is clear that this is supposed to be temporary, and that you are expected to do something about it, better sooner than later.) It is no longer enough to do a few simple tasks and be done. No one guarantees you a job. If you have a job, you have much less free time. You no longer automatically advance from one grade to another; if you want a career growth, you need to work hard to make that happen. The life is visibly way more unfair than before, and there is no one to complain to about it. Instead of moving along with the mob, you have to fight for yourself.
And I think the girls are *also* less mature than the girls of the same age in 1910, but for a girl this is much less of a problem on the dating market. (I mean, how many of those 15 years old girls know e.g. how to cook? But also, how many of the men who want to fuck them care about that?)
I found life even after school is child's play. I am paid to solve IT puzzles. How is that different from the math test at school? It is a game. Does not feel like hard work at all, not harder than the cosinus stuff felt at 16. And then after work I played videogames. So my life was not so different at 28 then at 8.
This is why I took long to mature. The wife and kids stuff I was ready for around 35 and YET it found out I am not mature enough for that sort of responsibility, I was that classic husband who just helps the wife but does not do anything without an instruction, which is also what children do, so that ended. Maybe I will probably never really mature, because at 47 my job is still about just figuring out fun IT puzzles. I don't think I will ever lay bricks, get in a fistfight etc. or ever decide anything really important. There is just no "maturing pressure". The bank pays my bills. I have a cleaning lady. Basically I do nothing "serious" outside the office ever. And inside the office it is increasingly like "AI, find the bug".
Ah, I think I know some people like that, too. Leads to some funny communication. "Tell me, buddy, how do you solve this problem X? Because I find it very difficult, and I am probably doing something wrong, so perhaps you could give me some advice." "Uhm, I... don't really have the problem X." "Ah, thanks. Well, that wasn't really helpful, but thanks anyway, at least now I have something to think about." :D
When I was at university, some of my classmates got there on their first attempt. Others failed, did some job for one year, then tried again and succeeded. The age difference between these two groups wasn't that big, but the difference in behavior was. That's where my observation comes from.
Depends on how it gets translated. If they hear "raping children," they might find that somewhat distasteful. If they hear "young prostitute," I agree with you that most people around the world would roll their eyes at everyone making this out to be a big deal (as would most Americans, I think).
I don't think you have anywhere near enough bandwidth to explain that the Epstein story is actually about the latter, and it just sounds like the former because of an idiosyncrasy of the ruling American ideology.
This is the land of plenty; we have always been able to afford our illusions. It is not that long ago that prostitution was rampant in this country. But then something changed. It’s all part of being the new world. To tell you the truth, it’s still pretty common but very under the radar. I think it is mostly the truth that the men who get in serious trouble for it are usually assholes.
In Western Europe there are practically zero Western European prostitutes. They tend to come from ever poorer countries, now apparently Romanian girls are rich enough to not do this, so they come from Moldova. So I get the impression women really don't like to do it, only do it if they are super poor.
Ukraine is the classic place for Western European prostitutes to come from. If they've somehow run out of Ukrainian girls....
(Referencing "The Tribe" simply because everyone should see that movie. It's a masterwork of cinematography. Brutal and vital enough to make even seasoned nurses walk out of the movie theater).
There is that. It is a way to make money when there aren’t other ways but I think there are women who choose it and like it. There’s a pretty wide socio-economic range.
Yeah, I have no idea what the numbers are. I have a friend or two who are sex workers (I am not a client) and they certainly treat it as their chosen profession. As to why that is, I could not tell you. I have never asked. If I have any more things to say about this, I will save them for my posthumorous memoir.
Viewed as an outsider, the US still seems to have some... transactional... overtones to a lot of it's sexual exchanges. Even ones that are miles from any conventional definition of prostitution.
Someone living in the US once described this to me as 'a blowjob for an easter bonnet'; she was talking about her own experiences within marriage.
It is precisely in transactional attitudes where explicit consent is super important, the whole concept is a direct application of contract law.
In more relational cultures sex just "sort of happens" in a semi-conscious way, people just feel when making such moves is appropriate and welcome or not.
This is why I really wonder whether American women are truly happy with all this contractual sex with explicit verbal consent stuff. Isn't men's sexiest trait having the emotional intelligence to read women? My first gf in Budapest told me "if a lady says no, it means maybe, if she says maybe, it means yes, if she says yes, she is not a lady". Ok that also had some screwed up parts of it (obvious fear of slut-shaming etc.) but clearly she wanted her men to have the ability to read ambigious signals.
But maybe I should not base my opinion on FetLife too much, FetLife is very explicitly contractual because people do the kind of kinks that in extremis can kill people, so of course clarity is paramount.
When I was a kid working a blue collar job the guys who were the age that I am now would tell me about the old bordello of *their* youth in the next small town over. According to those fellas the working girls at the place would chew out all the customers lined up after town bars closed.
“You guys thought you were going to have a woman for free? Ha! Next time come here earlier in the evening!”
I won’t say the name of that town but if you ever got a chemistry set for Christmas that town’s name was on the box.
Everyone sane here thinks this way too. People make a big deal about this because our politics incentivizes it. The political Left has a fetish for victimization narratives because they've proven extraordinarily effective over the past 20 or 30 years and so they search for them wherever they can find them. Epstein didn't do anything that many other high-status men across every part of the political and religious spectrums have done throughout history. The Weinstein scandal was clearly a cynical transaction. This is just how the world works: men acquire status so that they can get access to young and beautiful women; women use their beauty to get access to high-status men. No one is being victimized apart from progressive delusions about how the world is supposed to work. When liberal-adjacent men like celebrities or rock stars do it the culture turns a blind eye to it. Elvis married his wife when she was 14. So did Jerry Lee Lewis. Jimmy Page had a 14 year old gf when Zeppelin was at its peak. Seinfeld dated a 17 year old in the 90s. Even Simone de Beauvoir abused teenage girls. But when someone conservative-coded (powerful, elite, politically connected, wealthy) gets implicated then the political incentives align and we get ourselves a scandal.
Too many people on the left are connected to Epstein which is why it will just blow over eventually.
>>When liberal-adjacent men like celebrities or rock stars do it the culture turns a blind eye to it. Elvis married his wife when she was 14. So did Jerry Lee Lewis. Jimmy Page had a 14 year old gf when Zeppelin was at its peak. Seinfeld dated a 17 year old in the 90s. Even Simone de Beauvoir abused teenage girls. But when someone conservative-coded (powerful, elite, politically connected, wealthy) gets implicated then the political incentives align and we get ourselves a scandal.
Even assuming for the sake of argument that all these people you're pointing to were liberal or "liberal-adjacent," all of your examples are 50-30 years old or more. Elvis married in 1967. Jerry Lee Lewis married his 13-year old cousin-bride in 1957. Zepplin's "peak" (at least according to google AI) was 1971-75. Seinfeld, as you note, took place in the 90's, and Simone de Beauvoir died in 1981.
It's fun to play at finding hypocrisy in the people you dislike, but the evidence is much better fitted to the simple explanation that social mores are changing. There's a steady growth in the age of the weird outliers over time. It's a weird outlier, and a scandal, in 1957 when Jerry Lee Lewis marries a 13 year old, but he does it. By the 90s, the example is a Seinfeld with a 17 year old. Get to the modern era, and the most analogous contemporary example of "what happens when a liberal does it" is Al Franken. That at least happened in in the last 10 years. His victim was fully-adult 33 at the time he harassed her, and unlike the liberals of 30 years before, who protected Bill Clinton, the 2010s Democrats ran Franken out of congress for it, under pressure by the leftward flank of their coalition.
The gaping void of Epstein prosecutions isn't there because "too many people on the left" would get caught up in them. I don't know where you get your media, but I'm not seeing a wave of CNN or Vox or MSNBC articles defending "innocent Democrats caught up in the files". I was around in the 90s. I saw the Clinton scandal coverage and I know what it looks like when they do that. The coverage I'm seeing contains the obligatory "technically you can be in here and not have committed a crime" disclaimer that keeps your media outlet from getting sued, but otherwise squeezes all the juice it can to play up the scandal for clicks. There isn't a protective media environment for Epstein clients.
And even if there were such an environment, at the end of the day a conservative president is in charge of the DOJ right now, so if there are prosecutions to be made and we aren't making them, it's still not "liberals" making that choice.
Epstein and his "friends" did plenty of wrong things. Just ask Stephen Hawking when you see him.
I do know someone in the current Administration, who quite enjoys working under Trump. He's not in the "Epstein did nothing wrong" camp. He's in the "too smart to accept those invitations from Epstein" camp.
You know, this is the thing I find repeatedly frustrating at reading your comments. Instead of writing an information, you just write a hint that it exists, leaving "figure out what I wanted to say (and then fact-check if you still have some time left)" as a homework for the reader.
That of course decreases your trustworthiness, because one of the likely reasons someone would communicate that way is "when they communicate clearly, their arguments are easily debunked". I am not saying this necessarily applies to this specific comment, but to your comments in general.
This is not just the left. I see conservative Christian men criticize DiCaprio for not marrying, not having kids, so basically not committing. They say it is lecherous, he is basically *using* his gfs.
But I should not be surprised. A LOT of American leftism is secularized Christianity. Christianity is not in every sense always conservative. The first thing the Mayflower pilgrims tried was communism in a kibbutz-like way. Christianity only feels conservative because its progressive currents have secularized.
I think at the end of the day we are going to find out that Epstein was an equal opportunity employer. It’s rather a peculiar way to analyse the whole phenomena anyway.
You’re letting Harvey off a little too lightly. There is truth in what you say, but it’s not the whole truth.
<The political Left has a fetish for victimization narratives because they've proven extraordinarily effective over the past 20 or 30 years and so they search for them wherever they can find them.
You know, I agree with this. What I find most infuriating about the victimization fetish is that it is a power move disguised as innocent upwelling compassion. If you're gonna make a power move, say STFU, motherfucker I am smarter and richer than you and also networked with people smarter and richer with you and we are going to decide what happens next and you are going to have to do it.
I sense something very similar. I am politically homeless because I believe in social democracy, strong unions etc. yet and I cannot stand lefty men, because they keep making power moves without making it look so. Strong "I am better than you" signals but with full plausible deniability, without a real open challenge. It just infuriates me. This is actually feminine behaviour, the art of the subtle backstab. Somehow I get along better with conservative guys despite disagreeing in so many things, yet basically when they are not openly challenging me I know they are not playing status games. When they do that is very clear and obvious ("alpha" bragging etc.)
This is all Pareto's foxes and lions... the same shit, upper-class foxes vs. working-class lions... except that back those times the lions were unionized socialist workers...
Hear hear. The ninth circle of Hell was reserved for betrayers and that's exactly where those people belong. What they do is far worse than straightforward violence or graft because they weaponize our empathy against us. No culture can withstand that. Now social trust has evaporated, bad faith is assumed, and the commons is guarded by an electric fence. Progressivism is a bad faith movement that succeeds because it's a Shelling point that enables losers to coordinate against the establishment. The last 30 years of US politics is a case study in why multiculturalism is a bad idea. I only hope Europe gets the memo before it's too late.
Do you know anything about the relationship? I certainly don’t.
Actually, I just looked it up. He was 24 when they met. He did seek to mold her into the woman he wanted her to be (according to the very brief synopsis I read) but after six years of marriage, they divorced, but she still likes him, even though he’s dead.
Oh I’m not saying it’s necessarily bad, not even for the women involved. The point is that the culture has a giant double standard that it applies according to which political outcome it wants, which directly undermines the moral high ground it assumes when making these kinds of condemnations. It’s not about protecting girls, it’s about gating who gets access to them. Gating access to sex is how women have always exercised power.
>The piece says rationalists believe “that to reach the truth you have to abandon all existing modes of knowledge acquisition and start again from scratch”. The Harper’s fact-checker asked me if this was true and I emphatically said it wasn’t, so I’m not sure what’s going on here
Maybe, but it still feels weird that the fact-checker asks "is this true?", gets a clear negative answer, and then ignores it anyway.
Makes it seem like the entire purpose of having the fact-checker is to be able to say "we have checked all the information before publishing the article" and hope than no one asks "and...?".
Just to give the benefit of the doubt to the fact checker, it seems plausible that you could find a passage where Scott or some other rationalist wrote something that (Sam and) the fact checker would hear as "start again from scratch". If that were the case, then Scott's denial would only be some evidence against, but might be outweighed by others.
Muttering “I will open the crackers so you will have crackers and be happy" to a two year old is the most relatable and human thing on the entire planet, only someone without kids would miss that.
It seems like having and raising your own children is the final step of human maturity and you're fundamentally living in a state of arrested development vis-a-vis the world and society and many of your peers until you experience it. I think this explains... a moderate amount of things world models without this idea fail to explain.
Hey @ArrkMindMaster (does substack understand this notation? probably not), I read your story about the orc. Your writing has greatly improved from the first chapter to the most recent one. Keep it up!
I read the Citrini piece earlier today and am confused about the automated economy.
Suppose a spherical cow world where there's no government redistribution. AI/robots can do all labor cheaper and better than humans do now. It seems like wealth has increased. But workers are out of a job, so they can't benefit.
It might seem like the rich capital owners would benefit. But if the rich capital owner is, say, the CEO of McDonalds, then nobody is going to to McDonalds anymore because all the workers are broke, so McDonalds collapses. (If the rich capital owner has a broad basket of stocks, all of those stocks are in the same situation as McDonalds)
It might seem like at least AI company owners would benefit. But if McDonald collapses, the AI companies can't sell McDonalds robots to automate its labor. So the AI company collapses too.
So . . . the economy as a whole is very rich, but nobody can buy anything, and every specific individual is poor? What am I missing?
BONUS QUESTION: Whatever your answer to the above, does it imply that if we smash enough existing machines to halve productivity, we can reverse this process and have the economy be very poor but every individual is rich?
> Whatever your answer to the above, does it imply that if we smash enough existing machines to halve productivity, we can reverse this process and have the economy be very poor but every individual is rich?
Yes, people call it "back in the 50s, a man could support a housewife, two point two kids and a beagle and pay his mortgage, all on a single wage".
I haven't quite had time to read the Citrini piece, but now I'm seeing the following making the rounds as a response to it. Daniel Jeffries writes a memo from 2030.
One slice of it: "It's March 2029, and a five-person team in Jakarta just dropped a film that's competing with Marvel for box office dominance. Not competing in some indie-film-festival-participation-trophy kind of way. Actually competing. $340 million opening weekend worldwide. Ninety-five percent on Rotten Tomatoes. Three of the team members are under 25. Their total budget was $2.8 million."
My prediction for a "business as usual" (no significant policy intervention) automated economy is hyper-deflation: consumer prices would decline dramatically as marginal costs go down. Incomes would also decline dramatically, but each dollar of income or savings would have more purchasing power than it currently does.
I think compute/inference will become a commodity, and profit margins across most of the economy (blue chip and AI companies alike) will tighten. I think those who are already wealthy will have more purchasing power with the wealth they already have, but I don't think there's going to be any substantial transfer of wealth: just a general erosion of income and profit.
IDK that economics is the right lens to use for situations wherre humans are no longer useful, because law and contract are out the window. UBI is not happening either, except maybe to pacify people for a few years before it's easy to exterminate them.
Once humans are not the meta, most of them die whether the players of the game are humans or robots. Anyone who keeps more than a few humans alive will lose a war to the much bigger robot army of someone playing the meta. Hopefully we get off this path *before* that point.
So, yeah, this is just a final form of the inherent contradictions of capitalism.
"Of capitalism" being the key part. I realize this is a liberal/libertarian space and we differ in some very fundamental assumptions about the economy. And - at a risk of failing the Ideological Turing Test - I think a lot of that difference can be attributed to a liberal belief that the laws of the current economic system are like laws of nature, rather than of a particular, historically-specific stage of economic development. Which is to say, I think you're implicitly asking how the system goes on in those circumstances.
The answer is simply that it does not, because it cannot. This isn't like our contemporary arguments about whether it should, it's a completely new territory where its very foundations disintegrated, and we as society are outright forced to start doing something else. (No government intervention delaying the inevitable only means it happens sooner rather than later.) What that something else is, who knows, realistically, the winning solution will be decided by raw military power. (We can only hope that humans are social animals and no significant faction will be willing to destroy the entire society. And, of course, that the humans are still in control at that point.) But it will have to be something else, by necessity.
What we absolutely should not do is destroy wealth for the sole purpose of preserving the old system, that's for certain.
(I haven't read the Citrini piece and wouldn't know where to find it.) So this is not any type of answer, but a rant generated by your question. We've been screwing over blue collar workers for years/ decades. But it's fine 'cause we can all buy cheap stuff at Walmart, Amazon... And now that white collar workers may be caught in the same bind (Moloch) I should worry about it more? And here's a crazy idea, maybe cheap white collar work will make things cheaper for the blue collar worker. That would be a plus.
I think you can use the rust belt in the 80s and 90s as a simplified model, there’s an elite that does really well because productivity goes up, a lot of non-elite workers lose their jobs or get squeezed severely by competition.
Lots of local businesses also get hurt by the indirect effects of so many people no longer buying lunches, suits, etc.
Lots of people have no good options and slide into precarity or poverty. There is a lot of crime and drug issues and mortality goes way up.
Where the model breaks down is that there’s no place to move to to get away from AI, so the whole world is affected. And since it’s a massive crisis, politics becomes very important. Even in the rust belt, some towns re-oriented themselves to new economic activities and skills, and they did a lot better. Others fell into chaos.
Oh this is the most standard default dilemma of economics ever.
1) Marx figured like 150 years ago that this is why capitalism has inherent contradictions and regular crises, capitalists drive down wages, yet wages are the source of demand for their goods.
2) Keynes said the demand for "labour goods" can be long-term disastrously low, that is, Keynes's main idea was that human skills are not infinitely flexible and once you automate everything a blue-collar person can learn, they are in trouble. Same for white collars now. Keynes' famous "aggregate demand" is a code word for "demand for stuff not very bright people can make". "Labour goods".
3) Misesians say that in this case prices fall to $0.0000001 and you are set for 10 years by taking a rich person's dog for a walk, once. Their argument is that prices almost always fall to near to their production costs. Interestingly Marx also said that, too (the tendency of the rate of the profit to fall), so I dunno if you find it reassuring that the leftist fringe agrees with the libertarian fringe in one thing or not.
So anyhow it is virtually certain that prices will fall, and surely there will be redistribution too because that will be a sure vote-buyer.
>So anyhow it is virtually certain that prices will fall, and surely there will be redistribution too because that will be a sure vote-buyer.
The world could use some price-falling and redistribution right now, and yet there aren't many votes for that which I can see.
That's all assuming that politicians even have the will and the power to do those things within the current system. In the end, it might have to come down to violent revolution once again. Of course, given the level of technology and how long the violence will be bottled up by repression, it will be all the bloodier in the end, if and when it actually happens.
> The world could use some price-falling and redistribution right now, and yet there aren't many votes for that which I can see.
Are you kidding?? In the US, we spend 3x more per person every year than the median working person pays in taxes, and only half of people work, so we're really spending ~6x more.
We spend absolutely bonkers amounts on "redistribution," and the great majority of it, at least $4T dollars, is direct subsidies and payouts to the richest generation in history.
We are so all-in on redistribution there's literally no way it can continue working, we're mortgaging everyone's futures every single year to pump trillions of dollars worth of redistribution into old people every year.
I am more optimistic. Money has diminishing marginal utility, that is, rich people's primary motivation is not to become richer. They are after something else. It might be power for example. But I think a starving society is no one's interest. Pretty sure the rich will be OK with some kind of UBI if they get something else in return (such as power).
> Pretty sure the rich will be OK with some kind of UBI if they get something else in return (such as power).
First, if you imply that the rich will have that kind of veto or decision power, then I'd say we have already moved to a new system, not one I'd like to see personally.
Second, if the reason for the UBI question is that automated industry is reality, then what would compel the rich and powerful to make that compromise, if the UBI recipients' services are no longer needed at the most fundamental level? If their wealth and power no longer comes from the work of regular people?
Answers like "empathy" don't much convince me in this case. It might be confirmation bias on my part, or reporting bias in that you generally only ever hear about the Musks and Thiels of the world rather than the good guys with lots of money which may also exist. But my impression is that in order to become ultra-rich or ultra-powerful you have to be an above-average psychopath to begin with, so the psychopathic solution to problems is not as far to them as we would like it to be.
In the long run, all forms of human cooperation break down du to their inherent flaws becoming unsustainable in the face of reality. More often than not, that breakdown ends in considerable misery and death for the few and/or the many; the more people involved, the greater the damage. I don't see why it would be different this time.
> then what would compel the rich and powerful to make that compromise, if the UBI recipients' services are no longer needed at the most fundamental level?
When it comes to SOMEBODY sharing the wealth - the "Giving Pledge" has 236 billionaire signatories. That's the one that Gates and Buffet and Zuck have all signed where you commit to giving away at least half your wealth.
There's only like 700 billionaires in the US, a substantial fraction of them are charitably-minded enough to commit to giving the majority of their wealth away.
Agrajagagain already posted an excellent answer, but if you want something slightly more advanced, but still accessible...
The Citrini article would be right if we still used the Gold Standard and the money supply was fixed. As less money gets spent, there would be less money *to* spend, and we would risk getting trapped in a downwards spiral.
However, we have a fiat currency. We can just print more money. That has lots of downsides, like potential hyperinflation... but this is *exactly* why we have that ability in the first place: sometimes, you just have to print more money. If there's a shortage of money, because people aren't spending enough... instead of the government trying to get people to spend more money, we can just print more money. There can never be a money shortage as long as we can just print more money.
Now, how much to print, and who gets to spend it, are difficult questions that complicate things. But that's why we have the Federal Reserve, which thinks a lot about things exactly like this. The Citrini article talks about political inaction, a government unable to keep up with the rapidly changing tides. But the Federal Reserve doesn't have that problem, it doesn't have to wrangle Congress or whatever. If it wants to print more money, it can just do so. And unlike Congress, it already has a plan for things like this.
So what would actually happen if the Citrini scenario comes to pass? People lose their jobs. But since they're not necessary anymore, the economy produces the same amount of stuff as before. With the help of money printing + perhaps a UBI or something like that, the people who lost their jobs can still consume the same amount of stuff as before (more or less). The same amount of stuff gets made & consumed. Nothing changes. The Federal Reserve might have to engage in "Helicopter Money" (https://www.investopedia.com/terms/h/helicopter-drop.asp) to make it happen, but this is something they've prepared to do for a long time.
Then, eventually, the people who lost their jobs, will find new jobs in things like machining, and the other things the AIs can't do yet. The economy can produce more stuff. Everyone wins. (More or less. The exact distribution, is always a contentious topic.) That's, after all, basically what happened during the Industrial Revolution. A whole bunch of people lost their jobs in farming and had to go into industrial work, but in the end everyone was better off for it. And what happened during Globalization with offshoring, where people lost their jobs in manufacturing and had to move to services, and in the end it was worth it (more or less). This is the same thing, just with people losing their jobs in services and having to move to manufacturing.
... then, of course, the AI keeps advancing, in fact probably starts advancing even faster because we have a bigger economy capable of spending more on research.
Eventually the AIs are capable of machining as well, and plumbing, and welding, and all the things we thought they could never do. The usefulness of a human being still goes up, as we shuffle into the remaining jobs. But eventually, there are no remaining jobs left, and the usefulness of a human suddenly crashes to zero. (See https://www.newthingsunderthesun.com/pub/2ek4d4s3/release/7 ("What if we could automate invention?") and https://www.newthingsunderthesun.com/pub/4bnobp5q/release/4 ("When robots take your job")). A human being is 100 times as productive when 99% of the jobs have been automated, because one human can now do what previously took 100 people. But a human being is 0 times as productive when the last 1% of jobs gets automated, because you don't need the human anymore. Like Blackjack, everything just gets better & better until you pass 21.
What happens after that? Hopefully a good answer. Hopefully...
"That's, after all, basically what happened during the Industrial Revolution. A whole bunch of people lost their jobs in farming and had to go into industrial work, but in the end everyone was better off for it. And what happened during Globalization with offshoring, where people lost their jobs in manufacturing and had to move to services, and in the end it was worth it (more or less). This is the same thing, just with people losing their jobs in services and having to move to manufacturing."
Unless, of course, if this scenario will not happen this time. Yes, similar sort of things happened several times in history, but it is not a guarantee that they will happen next as well (I mean, I woke up every morning, thus I will not die; or there is the anecdote of the guy falling from the 10th floor -- even on the 2nd floor, so far so good).
A counter-argument to the industrial revolution might have been that if the farmers move into cities and work in factories rather than engage in subsistence level farming then mass starvation would be the result. Broadly speaking, that didn't happen. The economy was restructured and as secondary production has moved to Asia, services filled the gap.
AI might or might not fundamentally restructure the economy. But if it does, it seems likely that new desires will emerge and fulfilling those will drive the circulation of money, and the exchange of goods and services. I see a world where more people are involved in caregiving and education, and fewer in corporate non-jobs as a positive development.
To answer your final question: lawyers and CEOs are essential for a billion dollar corporation. So if we make everyone a lawyer-CEO do we all become billionaires? Linear extrapolation only works until it doesn't. The economy will move from one equilibrium to another - available labour will be consumed, the government may need to intervene to ensure redistribution (I.e. flow) of wealth during the transition. But once things settle the new system should be fairly stable. So no - drastically changing the level of production in the economy doesn't necessarily lead to straight line extrapolation...
"A counter-argument to the industrial revolution might have been that if the farmers move into cities and work in factories rather than engage in subsistence level farming then mass starvation would be the result. Broadly speaking, that didn't happen. "
I think it is important to get the chronology right. In history, industrial revolution was _preceded_ by agricultural revolution, thus less people were able to grow the same amount of food -- without this labor surplus, the IR would never happened in the first place. There was no scenario where people were scared of famine for no reason, but they cancelled the IR anyway.
However, this is no guarantee for the future. For example, in the 20th century, productivity increase allowed the five-days work week (instead of the six-days being common before). How does it fit to your model?
The agricultural revolution happened thousands of years before the Industrial Revolution; and the move from subsistence agriculture to commercial agriculture to feed large cities happened in tandem with the Industrial Revolution.
I'm open to correction but my impression is that workers were drawn to the cities by higher wages, allowing more space per farmer and opening up the possibility of higher productivity -- coincidentally artificial fertilisers arrived on the scene around the same time, or possibly arising from the same scientific and engineering discoveries that underpinned the Industrial Revolution.
I don't think that it was obvious at the time that moving 50% of the population off the land and into towns wouldn't result in mass starvation. Certainly the life expectancies in Britain plummeted when this happened, and the towns were notorious hotbeds of death and disease. I don't have time to find the figures now but life expectancy of a worker moving into the towns was in certain times and places as low as two years.
My larger point was that no linear extrapolation from 1730 was going to explain the economy in 1830.
(I got here by googling "second agricultural revolution", but of course, the Anglo-Saxon view of history has its own, somewhat more self-centric terms. :-) )
"and the move from subsistence agriculture to commercial agriculture to feed large cities happened in tandem with the Industrial Revolution"
Well, no, the agricultural one started before the industrial one, a lot. That is true, though, that it did not come to a halt in the XIXth century.
"I'm open to correction but my impression is that workers were drawn to the cities by higher wages, allowing more space per farmer and opening up the possibility of higher productivity"
Yes, I believe you are wrong in this. I mean, technically you can say that "cities offered higher wages" even if the villages offered nothing. :-) If the per capita agricultural production increases a lot within one or two generations, there will be less such jobs needed in the villages, and pretty much that was the only type available there. So people had to move to the cities in the hope that they can find any jobs.
(This is not even unprecentented, the German expansion eastwards for centuries was driven by something like this. I mean, there were definitely no higher wages besides the Danube or the Volga than in the German states, still, there were mass emigrations to these (pretty much unpopulated) territories in the 1700s. But even earlier, this is how Central Europe was filled with German towns -- just look around in Czechia.)
(You can also think of towns in the rust belt of any country, I am sure yours has one too.)
"coincidentally artificial fertilisers arrived on the scene around the same time"
We are well before that time.
"and the towns were notorious hotbeds of death and disease"
This is true BTW, which, IMHO, just weakens your "town salaries were more attractive" theory.
"My larger point was that no linear extrapolation from 1730 was going to explain the economy in 1830."
That is, of course, true, this is why it is called a revolution, after all. I only commented on your timeline. Because it is quite important to see that first there was the agricultural revolution (production increase), which made lots of people unnecessary in the agriculture. This allowed the IR, but in the meantime there was a ton of social disorder and suffering. I think the analogy to today's event is clear: if the AI makes lots of people's work unnecessary, it might allow some other sector to plummet later (or not -- we will see), but the transition will not be smooth at all. And there is no fourth sector already waiting and attracting engineers, accountants and lawyers with higher wages.
> I see a world where more people are involved in caregiving and education
Caregiving maybe, but isn't education going to be automated pretty early? The new government isn't particularly hot on public education as is, they're not going to keep paying for it when there's a cheaper alternative...
Skittle nailed my thinking -- I agree that 300+ person lectures will be a thing of the past, but a more optimistic take on education would be moving to tutor type systems. We'll have no shortage of human labour; why not invest that time and energy in developing others? I'm not claiming I think this will happen.
How would education be automated away? The message of the (in)famous Alpha School piece was that humans were vital in order to get children to actually engage and learn, and that was for a group of selected children who were likely to be more motivated, and have a greater ability to sit and focus, than average.
The majority of children will not engage in the sustained effort and focus required for learning without humans guiding and motivating them in some way. A minority (usually the most disruptive minority in a school) need a lot of human input to do so.
Unless we redefine ‘education’ to mean whatever is happening when a toddler swipes through endless short videos for hours on end.
Still, the experience is that education is in a downward spiral. Primary education perhaps not (that much), but tertiary (i.e. universities, colleges, other adult education stuff) pretty much.
This is, of course, a complicated issue, tons of the tools used in colleges are outdated in the light of recent AI developments (why have the students write essays, when they will just turn in AI-generated slop? But: how to teach skills which can be learnt by writing essays?), or just due to more and more material available online, but the bottom line is the same: perhaps we do not need that many people teaching other people at the current technology level.
Ah, if you’re just talking about whatever is happening with widespread tertiary education, I’m not sure ‘automated’ is the right term. I suspect that a lot of it will simply vanish or take on some other form for social purposes that doesn’t really pretend to be academically educational. But I don’t think that’s what people usually mean when they talk about jobs in education.
Still, tertiary education is not an insignificant part of education.
I am also not sure that secondary education won't be affected (we might even use present tense, I think...), unless we dare to call "childcare for teens" as "high school".
And even if we conclude that there is (will be) a demand for more "teachers", not everyone who does not have a job has the skills required to do it. As someone else put it in this thread, if we automate everything a blue-collar worker can learn...
I think the bit this piece gets wrong is in saying 'THIS CYCLE IS NOT LIKE THE REST'. As you pointed out, if there is no one to spend disposable income, companies' earnings will go down and stocks will go down and and and and...
What they talk about these downward spiral feeback loops where companies would double down on AI don't make sense. I don't see how this would be different to any other recession.
new technology increases productivity -> reduce workforce to maximise short term profits -> higher unemployment, economy stagnates -> recession -> cycle repeats
My vague impression is that economy is a "set of closed loops" -- things get produced somewhere, things get consumed somewhere, money flows in the opposite direction.
This is typically initiated on the demand side, like someone says "hey, I want X, and I am willing to pay for it" and hopes that someone else starts producing it. (And this typically already happened long ago, so we have inherited the loop.) Or, maybe no one responds, so the person who needs X starts to produce it himself, but someone else starts needing it too, etc.
In the world you propose, the workers can no longer exist *qua workers*. Some of them will find new jobs, for example they become servants or sex slaves or gladiators for the rich. Some of them will own some land, so maybe they survive by being unemployed and growing potatoes. Some of them will own a little stocks, maybe they get lucky and those stocks skyrocket. But most of them will starve and die.
That will cut a lot of existing loops. There will still be *some* demand to produce food for the rich and their slaves, but neither of those will eat at McDonalds (as it is now). So McDonalds either goes out of business, or changes their business to producing cheap food for the slaves, and organizing birthday parties for the children of the rich. Which is not completely unrelated to their current business, but it will be more optimized for these purposes -- the slaves and the rich probably won't eat their meals in the same room; the slaves will pick up their food somewhere or get it delivered by drones; the birthday party rooms will be larger and contain tons of entertainment. Actually, I *can* imagine McDonalds to survive, but it will be different from today. (And there is a chance that someone else will adapt to the new role faster.)
Maybe let's start from the opposite perspective: which loops will *remain*. Production of robots and computing power. Infrastructure for the rich: financial business, lawyers, weapons. Production of things that the rich want: luxury food, luxury clothes, entertainment, and whatever becomes the new fashionable thing to spend all that extra wealth on, for example space travel. (Could be something we wouldn't expect, like maybe the rich will decide that riding ponies is the cool thing, and now that 99% of population has starved to death, there is a lot of space where those ponies could ride. Probably mutant ponies that look like pokemons.) Secondary, if the rich want human slaves, also production of the things for slaves: cheap food, cheap clothes, shock collars.
Probably many things that we use today will remain; some rich person will probably be weird enough to want a bicycle, or anything. But they will produced in much smaller quantities and higher quality. Maybe expensive enough so that many of the current companies survive.
tl;dr -- some parts of economy will collapse, but not all; and many will change from "produce lots of cheap stuff for plebs" to "produce a few luxury items for the billionaires"; plus all the robots that build new robots
BONUS: Sometimes the machines multiply the productivity of a human, sometimes they replace the human. Sometimes they replace the human with another human, e.g. one who needs to be more intelligent to use the machine. Smashing some machines that replaced stupid people, e.g. in agriculture, would reduce unemployment and decrease productivity.
Maybe another way to look at the situation is that giving each person some brain and muscles was God's version of UBI. The basic economic loop was "brain + muscles + land = food". We created more efficient loops, so today for most people it's like "brain and/or muscles + participating in global economy = more food and toys". But when the brains and muscles get outcompeted by machines, you can no longer participate in the modern loop; and you can't even fall back to the old loop because there are too many people and not enough land for that; plus you wouldn't like that anyway, because you are already used to a certain amount of food and toys.
The question whether things are demand-driven or supply-driven (innovation-driven) is big and I think I have an elegant solution. I think we have very very broadly defined desires derived from our biology. So we have a demand not for cars, but for getting to places, not for smartphones, but for communication in general. So someone invents the horse and someone else invents the train. Someone invents the campfire to talk around, someone else the phone call, someone else YouTube. But all this is based on biology.
This means a cool thing. It is possible that at some level our biology is satisfied and we will not need new things. I still remember having to buy a new graphics card every 3 years. But once we reach the level of photorealistic retina display, no need for any more improvement.
Besides so many goods were positional in the past... I was cleaning my parents home after their death and it was so said that they bought so many things not because they needed them, but to make a bourgeois impression on other people. A lot of elegant clothes. Think 30 neckties. Artwork in the living room and nice sitting furniture. Even the very concept of the living room itself.
I don't know about other people but I don't do these things anymore. I have no living room. I can be comfortable enough on my bed in the bedroom with a tablet. If people visit me they are welcome to sit on the bed or the carpet. I still wear elegant clothes, because I cannot entirely replace my head at 47 but I think the 27 years olds will just never own anything more elegant than jeans.
So I think we are quickly approaching happy digital nomad lifestyle, happy because our goods satisfy our biology.
For example "tasty food" is also a biological concept. And I think for my purposes cheapish Asian takeaway is very close to the potentially ideal price/quality ratio.
Yeah, I have also noticed how the priorities have changed compared to my parents' generation. My home is there to make my family comfortable, not to impress my neighbors.
That said, before internet people visited each other more often...
It does sound something like what we have currently going on, in that the economy (of whatever country you like) is supposedly going "line go up" yet people feel that they are worse off.
Big companies are doing great because stock prices are going up because people are throwing money at them in hopes of AI. That makes line go up for stock market, which is how we measure "and how is the economy doing?", at least in a naïve way.
But for Tom, Dick and Harry who are not big institutional share holders but working joes who may even be staring down the barrel of redundancy, they're not doing great even with all this money sloshing around.
Whatever about the USA, the Irish news is full of "this store closing down, this chain moving out of the country, Apple etc. announcing global cuts to workforce" so even while the economy is okay (on paper at least), people are not feeling secure and they are seeing food prices, for one, going up while wages are not keeping pace:
"The State's fiscal watchdog has estimated that almost half of the Corporation Tax collected by the State is paid by three multinationals.
In 2024, these large companies paid around €13 billion or 46% of the tax, according to the Irish Fiscal Advisory Council.
While the organisation has not named the three multinationals, they are understood to be Apple, Microsoft and pharmaceutical group Eli Lilly."
You can see that if they pull out or even reduce workforce here, we're toast.
What is the end result? No idea. The notion is that new tech means creation of new jobs in hitherto unexpected ways, but if AI is going to do everything because it will be *sooooo* cheap and faster and better than humans, where are the replacement jobs going to come from? There's only so much washing we can take in from each other.
I don't think AI *is* going to be this universal replacement, but I can see automation taking a lot of whatever manufacturing jobs remain, and this time round there will be white collar workers pain as well. CEO type jobs probably okay, world expert in paperclips type jobs okay, middle professional jobs maybe not so okay.
I think this is one of those situations where the abstraction of money conceals more than it reveals. Ignore the money. Pretend, for a second, that the money doesn't exist.
Somebody owns the machines that make necessary things. Somebody owns the machines that harvest raw materials, somebody owns the machines that refine them and process them and produce steel and copper and fuel and food and computer chips and cars and buildings any everything else--whether luxury or necessity--that the economy runs on. Presumably, somebody also owns the land where the various raw materials have to be harvested, and can also trade access to that land for other valuable things.
If you own some machines that produce some necessary/valuable goods, you can choose what happens to those goods. That's wealth. People with control over the disposition of goods have wealth. The more goods they control, the more wealth they have.
In this spherical cow world, selling just your human labor can no longer produce wealth of any sort, because machines can do it all better. So the only people who have wealth are the people who own machines (possibly indirectly, through stocks). That might be a small subset of people, but that's certainly not zero people. Maybe the CEO of McDonalds has some wealth or maybe they don't, but the CEOs of Google and NVIDIA and Dow Chemical and Toyota and probably a lot of large landholders still have some wealth. This wealth exists independent of the currency it's denominated in: currency just makes it easier to trade between various sorts of wealth, it doesn't create wealth.
And to predict what happens from there, we need to know their motives. If all of the wealth holders are perfect examples of your amoral homo economicus, then they just have their machines produce them food and mansions and yachts and gold plated toilets and medicines and they eat caviar and drink champagne while the proles starve somewhere far, far out of sight.[1]
If they're of a more philanthropic bent, maybe they use some tiny fraction of their vast wealth to replace the government redistribution (that isn't happening in spherical cow world) with a vast, globe-spanning welfare state. At that point, whether this world is darkly dystopian or kind of OK, or maybe even halfway decent depends a lot on just how vast their vast resources are, and just how stingy they are with them.
[1] Maybe they hire some small fraction of those people to do human-specific service and entertainment stuff, like putting on live plays or whatever.
This is a very good response. I would point out though that one of things that adding money back into the mix does is create legible prices for everything. Prices are the organizing force in the market economy without which the owners of all these fabulous machines might not know what to produce and therefore produce and control less wealth. In the event where most ppl have no purchasing power do we have enough prices to organize activity around?
So a slightly different concern is if you cut so many ppl out of the economy not only is there an inequality problem you risk relying on central planning, not markets and then productivity falls even if the tech is better.
It seems that this should be self limiting. If the machine is off doing useless things a human can jump in and do useful things and now someone has purchasing power.
Assuming the AI isnt also smart enough to make central planning problems disappear I struggle to envision the exact market driven equilibrium in the extreme spherical cow scenario. But practically speaking, thered always be enough petty tasks to maintain the essential features of a market economy
Ownership is also an abstraction. It's just ability to defend whatever you own from being taken(either by your own power or through alliances, systems, or in most cases, the state). And robots that make things will not be able to defend you from an angry mob. And governments are not going to defend your ownership for you if it leads to mass starvation.
Of course you could have robots that kill things but I really doubt the governments of the world will just suddenly be OK with massive privately-controlled robot armies.
I'm glad I checked the existing replies first because you hit the core points I intended to talk about. This is how wealth works in modern market economies, and it's also how wealth worked in pre-modern societies. The dominant form of wealth changed with the Commercial and Industrial Revolutions, from land and goods to ownership (mostly partial ownership) of businesses that make or distribute goods and services. And being owned money, goods, or favors by people who can reasonably be expected to make good on their debts in some way has been a significant form of wealth throughout.
See some of what Brett Devereaux has written at ACOUP about the role of "big men" in non-state agrarian societies. I've also heard somewhere that "giver of rings" was a common kenning used in Norse sagas to indicate a particularly wealthy lord, whose wealth was indicated by the ability to reward his followers with rich gifts.
>So the only people who have wealth are the people who own machines (possibly indirectly, through stocks). That might be a small subset of people, but that's certainly not zero people.
I might quibble with details of the formulation, but in broad strokes I mostly agree with it. My quibbles are that land will still be valuable; your "possibly indirectly" needs to be read very broadly to include debt, favors, and influence as well as stock; organizational capital, particularly in the form of local human knowledge, will remain valuable for some time before everything is fully automated; and labor will still have some diminished value (and some trade will remain possible due to comparative advantage) unless and until automation is advanced enough to fully solve the scarcity problem. The last bit is analogous to how people are still able to live and work in pastoralist, subsistence agriculture, or even hunter-gatherer communities despite modern industrial society being a thing. These communities are extremely poor compared to their modern-economic-system neighbors, but their mode of production still works after a fashion.
> If you own some machines that produce some necessary/valuable goods, you can choose what happens to those goods. That's wealth. People with control over the disposition of goods have wealth. The more goods they control, the more wealth they have.
This is an excellent formulation, thanks ! That said though:
> ...while the proles starve somewhere far, far out of sight.
Historically, this does tend to happen a lot, but not to the extent that you claim. Modern peasants live in unimaginable luxury compared to Medieval barons; this is partially due to the welfare state, but also partially due to all technology being interconnected. It would be almost physically impossible for the rich people to control and maintain their machines without making at least some benefits of those machines (e.g. ubiquitous cellphone coverage) available to all. Thus, in the spherical-cowbot world where robots do all the work, we can still expect some nontrivial amount of the luxury to filter down to the lower classes.
> Historically, this does tend to happen a lot, but not to the extent that you claim.
Because historically there were natural limitations preventing from completely excluding proles from the economy. Full automation of economy removes these limitations.
> It would be almost physically impossible for the rich people to control and maintain their machines without making at least some benefits of those machines (e.g. ubiquitous cellphone coverage) available to all.
I'm sure it would be great to have cell phone coverage while dying from starvation.
> Full automation of economy removes these limitations.
I don't know what "full automation of economy" could mean, other than perhaps "converting the Earth into computronium", which, granted, could potentially be hard on people.
> I'm sure it would be great to have cell phone coverage while dying from starvation.
Yes, it's because cellphone coverage is effectively ully automated yet food production is not.
>I don't know what "full automation of economy" could mean,
It means the spherical cow world that Scott has asked about. People still live in it, but they can no longer offer labour to compete with the machines, like at all.
Ok, so are the machines performing that labor or not ? If not, then everyone starves -- except they don't, because humans will fall back on farming their own food, i.e. performing labor. But if the machines are performing the labor, and doing so at much cheaper rates than humans, then food is essentially free and no one starves.
What is with your complete refusal to believe that useful, self-sustaining robots can be built? I find it strange at this point that you're willing to humor AGI in these thought experiments, but not robots that can build more robots.
Eh ? When did I say that they could never be built ? Granted, I don't think they'll be built anytime soon, but that's a different story. If we accept AGI for the purposes of the argument, then sure, we can accept self-sustaining robots too. I don't see how this invalidates anything I'd said though.
Why do you think demand would completely disappear? There would obviously still be demand by other businesses (who are also all being run by AIs), so you (or more specifically, the AI running the company) just change production to actually serve the demands of other productive beings. So... not humans. An economy can still be fully functional without any humans in the loop.
"There would obviously still be demand by other businesses (who are also all being run by AIs)"
My take is that a business needs things to allow it to produce OTHER things. There must be a consumer at the end of the chain, which is currently an individual or a state (e.g. public buildings / services, military, or healthcare in Europe / Canada).
But right now productivity is tied to "demands of humans for things". Take humans out, and what are the AI and robots producing? Medicine for whom? Goods for whom? If the proles can't afford to buy the new cellphones, what good is a warehouse full of the latest bells and whistles model of phone?
An entirely AI economy will have to switch to a new model of goods and services, and what will that be? Sure, chips for the data centres, energy generation, making parts for those factories that produce the machinery for data centres and energy generation and the likes, but what else?
I’ve been in towns in the US where the local hospital is the main engine of economic activity, because healthcare is subsidized by the government and so it’s not just a healthcare program, it’s also an economic stimulus package and a jobs plan. But it’s an open question whether a super productive future economy would have better welfare and healthcare for ordinary citizens, or if the authorities decide that ordinary citizens are no longer needed.
> Sure, chips for the data centres, energy generation, making parts for those factories that produce the machinery for data centres and energy generation and the likes, but what else?
I think it's conceivable for a hypothetical AI economy to consist almost entirely of this short list of things. After all, for thousands of years, human economies were almost exclusively focused on an equally short list: food, water, clothing, and shelter, plus equipment for acquiring the above more easily.
The bio anchors idea seemed really dubious from the start. Like yes they could help provide some evidence toward some conclusions, but would anything like this have worked for other technologies? Bio anchor steam engines vs horses, metal strength vs bone, power plant output vs biologically powered things.
the equal sign is a formatting error from improper processing of line breaks, and it effectively overrides one character. This is really likely to be 19yo given your followup comment.
You can see the evidence to support this in the very same email, just a few lines below:
Envoyé: jeudi =7 janvier 2013 17:22
This is French for "Sent: Thursday =7 January 2013 17:22"
Here again you see the unusual = sign. But wait: 7 January 2013 was a Monday, not a Thursday. 17 January was a Thursday. The = replaced the 1. Almost certainly the same happened with =9yo above.
Savio has cited persuasive evidence for his position, and it seems to me quite plausible. You have cited no evidence for yours, which seems quite implausible. I mean, if you're peddling underage flesh, wouldn't the *first* thing your Secret Code obscures be the age?
Given that the two emails are identical in subject, attachments, most of the content minus one character, and send time down to the same minute, plus outside the emails an explanation of how the equal sign gets created and the fact that DOJ does occasionally duplicate information in different releases of their datasets. It is much much more likely that these are the exact same email released differently.
I do not have an opinion on whether anyone in the Epstein files actually had sex with an underage girl, as I honestly don't care to look through the files at all and I follow very little on the Epstein file news. Saying that you may have made a mistake != I dislike you != I am defending Epstein in general.
Please calm down a bit and don't take my argument as a soldier[1] on the Epstein side.
The pizza and grape soda thing sure seems like a euphemism for something, too, and I doubt it was anything to be proud of. The sender was some doctor nobody's ever heard of though, if I recall.
Yes, but it gets even better: the new consensus is that Pizzagate was actually a decoy fake conspiracy theory planted to get people to dismiss the REAL pedophile sex ring/Satanic cabal that Epstein was part of. Open your eyes, sheeple!
Life has more layers than you know. The pedophile sex ring/Satanic cabal, while real, were set up to distract the rest of us from recognizing that these people are basically our species' dandruff and halitosis..
> these people are basically our species' dandruff and halitosis..
What is wrong with you, joking about something like this? This is a serious issue that has ruined the lives of thousands of young girls, and you're disrespecting their memories by making up deranged rumors about their very real abusers, which will only make people less willing to believe the truths of the victims. You may think it's funny, but it's doing real harm, whether you intend it or not.
Please take care to treat this sensitive issue with the care and tact it obviously deserves. Thank you.
Anyone who's intending to attend the upcoming Jerusalem rationalist community meetup - it is *POSTPONED* by one week because of the weather, and expected to happen in March 6th. RSVP on lesswrong and get notified on changes directly in your email without having to scan dubious comment sections!
Alexander T. as part of making a different point down below here, noted that "net domestic migration into the sunbelt is slowing."
Huh, that's interesting....I found the Census Bureau data release which sparked some MSM coverage late last month.
That data is the Bureau's annual estimates of US populations by state and county, for the period ending July 1 2025. Those annual estimates are not a census [an actual count of people], rather they are built from various other datasets [births and deaths, etc]. They do sometimes get corrected by the actual decennial censuses. They are generally good enough though that various sectors use them for business-analysis purposes and the like.
In this case the Bureau, regarding the July 1 2025 estimates, says: "The Midwest was the only region where all states gained population from July 2024 to July 2025. In addition, after experiencing population decline in 2021 and small growth in 2022, the Midwest’s population grew solidly in 2023 (259,938), 2024 (386,231), and 2025 (244,385)....From July 2024 through June 2025, the Midwest also saw positive net domestic migration for the first time this decade...."
I looked through the state by state data on domestic migration. Two things are at the moment both true:
(a) The overall arrow of net migration within the US has not flipped. CA and NY still have the largest state annual net losses, TX the largest annual net gain, states like NE and IA and VT still have virtually no net domestic-migration change; the Carolinas are still gaining strongly while places like LA and AK and NJ and MD keep having meaningful annual losses; etc.
(b) The overall movement from blue to red states has indeed been slowing down since actually 2023. CA through 7/1/25 had its 4th straight annual decline in net domestic-migration loss; NY for 2025 had less than half the net loss of either 2021 or 2022; TX had its lowest net annual gain since 2020 and FL its lowest since sometime before 2020; WI had its 4th straight year of net domestic gain; MN and MI each had their first domestic net gain since sometime before 2020; VA had its 2nd year of net domestic gain after three years of losses; IL had its lowest net domestic loss since 2020; WA had its 2nd year of net domestic gain after three years of losses.
Re #4 - I was interviewed once and what was reported in paper had the same sort of quality. Reporter didn't listen, had an agenda for the article, fit my words into that agenda, reworded what I said where necessary to fit the agenda. What was printed was a totally different conversation than the one we had.
As the old saw goes, however much you hate the media, it isn't enough.
I have the same experience, twice. In one case they also misspelled my name, so I could pretend that it wasn't actually me. In the second case I was smart enough to not give them my full name, so they only made an idiot out of an unspecific "Viliam".
The first case was not even political. I was a gifted child, and the reporter had some vision of "all gifted children use computers", so she kept asking me about the computers, and I kept telling her that I don't have any, and that my family is not rich enough to afford one... but she wrote an article anyway about how I am a computer genius and that I have figured out that my parents have already secretly bought me a personal computer for Christmas (completely fabricated).
The second case, there was a political debate that somehow involved collective guilt, and at some moment I said "it's simple: if we like some group, we insist that they are all individuals, and none of them should be judged for the crimes of some other individual; but if we hate the group, we see them as a uniform mob where everyone is guilty of everyone else's crimes, so it is fair to punish them randomly". The journalist was surprised and said "this sounds very smart, can I quote you, what is your full name?", and I asked "can I get the text to verify after you write it?", and she said "no, were are too busy for that, the article needs to be published immediately after writing, there is no time for feedback", and I said "sorry, then you can't have my full name". (The published version was: "student Viliam said that we shouldn't judge Palestinian terrorists for their crimes".)
I am not even an important person for someone to have a grudge against me or something. I think the constant lying is just a habit; the people who are not constrained by truth generate better stories.
Well, Viliam, that was an excellent play. The old "omgwearesobusy.....mustpublishimmediately" is more proof that their intentions are malicious. If they cared about the truth, time to get things right would be important. But we know they don't.
I always disciplined my children for lying to themselves. "If you lie to yourself, you will lie to anyone." I think we know the source of the media's problem.
I've had good results talking to journalists, including even the NYT, in my area of expertise. But I was speaking on a technical subject that wasn't politically controversial and so nobody really had an "agenda" in the sense you're using.
What is the worst twist in all of fiction? Beyond the various "it was all a dream"s, beyond "you the viewer are the killer."
I've been saying it was the Zero reveal in Zero Time Dilemma, but I'm wondering if that's true. The only other two I can think of that come close are the orphanage scene from Final Fantasy 8, and the reveal at the end of Down With Love (which arguably is the same class of twist as the orphanage scene from Final Fantasy 8).
> The plot twist in "Ender's Game" always struck me as a bit lame.
Lol, waddya mean?? It's perfectly plausible!
Imagine a family with 2 kids that are so smart they can root the national discourse and steer politics as teenagers. Now their brother, who is even smarter than THEM, goes off to battle school, lollerskates over everyone, then gets fooled into thinking that the battles he's leading against progressively farther out enemies with progressively older and weaker technology are a sim rather than reality. Then he's shocked and horrified to realize those battes were actually real all along.
Why sir, I am shocked, SHOCKED I say, that you would do this to me! I've literally only spent my entire life training for this, feel like the earth and my family is under existential threat, and would have happily led them even if I knew they were real battles, but I say, sir! Shocked!
The one at the end of KIng's The Dark Tower series. I don't want to say here what the twist is, though, because somebody on here today is probably reading that excellent (til the very end) series.
The ending was okay, but I was fascinated by the little blurb King gave the reader *just before* the ending. King basically subjects his audience to the same "19" trick that the Man in Black did with the woman in the tavern (name escapes me) in Book 1.
The series' ending was entirely within the narrow distribution of possible endings that are consistent with the whole philosophy of ka. It's a good ending.
Also while I hate to reply to the same comment twice I just noticed the diss on the Down With Love twist.
I love that twist! That twist justifies the whole existence of the movie, and flips the whole genre, turning it from a straightforwardly milquetoast parody of the source material into a ridiculously over-the-top parody of the source material. I remember seeing that movie in the cinema when it came out, and how Renee Zelwegger's monologue transformed my entire viewing experience.
Bridge to Terabithia. Most of the story seems to be a fairly straightforward coming-of-age story, then the secondary main character dies suddenly in a senseless accident and the rest of the book is about the perspective character coping with grief and survivor's guilt. I was assigned the book in elementary school, and I have never forgiven the author nor the teacher who assigned it without a word of warning about what was coming. The entire book feels like a cruel prank at the reader's expense.
The 2007 film adaptation is even worse, since it was deceptively marketed as a Naria-esque children's fantasy story.
That makes it the perfect coming of age story. Part of coming of age means learning and accepting that people die, sometimes in senseless accidents. I was assigned that book in elementary school too, and shortly thereafter, the sister of one of the sixth graders at the school was shot to death in a senseless accident. The sixth grader definitely didn't get a spoiler alert about that!
I was already viscerally aware of people dying in senseless accidents. Part of why the book hit me so hard is because my aunt had died in a car accident a year or so previously, so the stakes of the book felt very real to me.
One key difference between senseless accidents in fiction vs real life is that the former has a large element of intentionality that the latter does not. The author chose for the senseless accident to happen in the story and chose how to present it to the readers.
The element of choice on the reader's side is also important. I get to choose when and how to read stories, and now that I'm out of school, I get to choose which stories to read. I can enjoy a good tragedy, but I prefer to know that I'm reading the sort of story that is apt to include tragic elements. I find the bait and switch of a story that is presented as being light and upbeat but suddenly turns dark and tragic to be rather obnoxious.
Omg my ten year old and I checked out the DVD from the library on a rainy day as it looked wholesome and harmless. She was traumatized by the movie and I was traumatized that I let her watch it. I will say that was the least expected plot twist I have ever encountered.
*Old enough for what? Living by yourself. Speak to a holocaust survivor about "How old a child needs to be, before they can live on their lonesome, without parents, and working for a living"
So Zero Time Dilemma is the third of the Zero Escape games, which all revolve around nine people trapped in some kind of death game, with one of them secretly being the game's orchestrator, Zero. There's a bunch of branching paths, you get to know all the characters, and finally once the paths are exhausted you hit the True Ending and one of the nine is revealed as Zero, to lay out their whole motivation.
In ZTD, you get a bunch of branching paths, get to know all the characters, and then Zero is revealed to be... a tenth person, who the player has not seen before, but is said to have been there in the room the entire game, a few feet off camera, with the cast knowing he was there and just not mentioning him to the player. I can't think of a worse way to end a mystery. Including just not revealing anything, which I've seen a couple of books do.
It doesn't help that his plan includes telekinetically shooting himself with a shotgun, and dying. His Motives Are Complex.
More spoilers - don't read this if you want to play the game.
------------------------
They're fairly consistent in not calling Sean "Delta" at any point, right? Obviously they rely on implication, but at no point does anyone actually do so. There's also other clues like the 4th collapse whenever that team gets knocked out. And there's the shadow in the late game.
I understand your viewpoint, but I don't think it's bad per se.
I'm pretty sure they're consistent about it. There a point where... Shotgun Guy is aimed at... uh... Kid, saying "if not you then who?", and if you type in the supposed name of the kid, as in "I did do it actually", the guy shouts "that's impossible," and then shoots him anyway.
But this is a level of scrutiny one should never have to apply. The group menu shows the kid with that name, like all the other groups do with their viewpoint character. The whole setting is about trying to gather information, and yet the viewpoint character never even looks at one of the players. They discuss the kid's helmet (...I think), but not the old man's wheelchair. After the reveal, what's to stop there being another, eleventh person no one mentioned?
I'm reminded of the joke in Chuck that killed my interest in Chuck, where... uh... Sports Guy is driving... uh... Weird Guys to Las Vegas, and lists off all the things they're going to do once they get to Las Vegas in gambling-and-sex lingo. And then he takes them to "Las Vecas Wildlife Preserve", and painstakingly explains all the gambling-and-sex lingo he used was actually wildlife-and-camping lingo with completely different meanings. It breaks so many implications about the people and the meaning of language that it renders the entire setting as noise. No information can ever be relied upon again.
Did you also dislike the twist in VLR where the MC is revealed to not look like how we've kept imagining him this whole time? Iirc, there's no reflective surfaces that they come across, and no one else comments on it either.
As for Chuck, the show is a comedy. Why wouldn't that kind of misunderstanding be valid to apply? It's not like it was for the highly erudite or whatever.
Again, not saying that you're wrong to dislike it, but I don't think your dislike means that it's bad.
VLR's didn't bother me as much because it's the character doing it. It's not the characters in ZTD, it's just the audience.
(Also I do that one in daily life. Mirrors are a bad time.)
The difference between the Las Vecas joke and, like, the one where the Weird Guys are too stupid to know what the CIA is, is the investment. There's a whole scene of driving toward Vegas beforehand, and after the lingo explanation it's clear there's no amount of assurances, logic, or time investment that will prevent the show from pulling the rug out beneath you. Which in turn means no amount of logic or time should be invested.
Andrew Crumey's 2004 novel "Mobius Dick". Basically a novel about quantum mechanics and parallel universes, in which a physicist discovers a parallel universe in which the Nazis won WW2. Schrödinger is involved somehow, and so is Schumann, and I think that the parallel universe finds out about the original universe as well at some point.
I don't remember all the details of the book, but what I do remember is the ending where it turns out there's a _third_ universe which is (bum bum bum) our universe, and that the characters are so lucky that they don't live in that universe because George W Bush is president there, and the ending just turns into a rant about how terrible George W Bush is.
I feel the same way about Stross's Family Trade series. Magical family that can move themselves and what they can carry between universes? Ok. Using it to move drugs? Ok...
And then the last book goes full into "and the real-world US government is genocidal Nazis who nuke all the other universes because they're evil Republicans who hate drugs." Yah. Really. Or at least that's what I remember the anvilicious message being--Republicans are evil Nazis, hate Republicans.
Ooh, that's a strong one. But how tied to the resolution was it? The main difference between FF8's twist and ZTD's is that FF8's is isolated enough it can be ignored without losing the rest of the story.
Was it like, "Now that the plot's resolved, look over there at George Dubya," or was it more like "all our plotlines are trivial in the face of George Dubya"?
I have never once been to a party where they run out of cheese before they run out of crackers. There's always a massive pile of cheese left and nothing to eat it with.
If the rationalist movement cannot solve this then what good is it?
"I have never once been to a party where they run out of cheese before they run out of crackers. There's always a massive pile of cheese left and nothing to eat it with."
Is that supposed to be "There's always a massive pile of *crackers* left and nothing to eat it with."?
It was in that moment of opening the package of crackers when Scott Alexander truly comprehended the essence of Universal Love and the meaning of Transcendent Joy. Then he immediately forgot about it, and that was that.
I'm interested in what people think that improvements in coding AI will look like.
Even plain coding assistants are currently extremely good on greenfield projects and can 1-shot not-particularly-innovative projects that would take me 5-15 hours otherwise.
However, on brownfield projects, both in my experience and from all programmers I talked to, they are rarely useful outside of self-contained modules.
Coding agents seem to push doable greenfield projects from 5-15 hours to 20-40 hours (which is huge!) but usually end up in a mess that cannot be maintained by either agents or humans. I haven't heard of how well they do in brownfield projects but I doubt they have significantly useful since, IMHO, the main barrier is that LLM performance craters with increasing input size (for coding). Agents seem to do better context management and that helps a lot but I doubt it's enough for large code bases.
There is also the issue that building any project usually changes its trajectory. As you build something you realize what is missing, what is useless, what is not needed and what is suboptimal - this will not happen if you automate building it. I assume this is a huge factor why there are barely any (maybe even none?) quality vibe-coded projects. There are no shortcuts.
However, the main limitation to usefulness of AI coding assistants is simply conservation of information. When you want assistant to do something you need to put that information into the context window or you need to reference information that exists in LLM itself. To bring your vision to life you need to specify or reference ALL the information that it contains (all the thousands of small decisions and hundreds of large ones that shape any system that is not trivial). If you have actual vision of something, you can't offload those decisions to LLM and you still need to put that information into the project. LLM is unlikely to give you what you actually want since if you put all the information needed into the LLM it would act as simple natural language -> programming language translator (this is largely not useful for seasoned programmers). On the other hand, when you reference information from LLM you lose precision and your vision is unlikely to come out the other end.
The above seems like a fundamental tradeoff with coding assistants which would remain even if LLMs improved immensely. You either have to specify things so precisely that there's little point in using LLMs or you don't get what you want.
> they are rarely useful outside of self-contained modules.
that is true for humans as well. I think AI has already surpassed a bad programmer, but good programmers are still safe.
> I'm interested in what people think that improvements in coding AI will look like.
With AI it is very important to know their limits, e.g. for me they are especially bad dealing with spring upgrades. And it's important to always review their work, e.g. for every good idea they generate, they also generate a bad one, and the challenge is to look at the code they produced, and understand the bad ideas, so you can fix them before they do any harm. This is very hard.
The biggest productivity gain I got from AI was work, that was especially well suited for AI, e.g. comparing two different versions of a large number of XSD. Another big productivity gain is when they generate a good idea, that is easy for me to validate.
For my brownfield projects, I just try to use AI for most of my issues, but then I give up soon before doing the work myself.
For my greenfield projects, I try to set the project up in way, so I can leverage AI for a longer time before I hit the wall.
I've had pretty good results with having claude first document the brownfield application with md files describing usage and then instructing it on implimenting modest updates and fixes. You're right that controlling the context is the key skill to use, and particularly subagents are important here, but even if you just give it the ~5-6 files that are relevant to most changes a single agent has usually been able to do what I need. If you're expecting to be able to just write a jira ticket and have the thing run the whole thing from branch creation to pull request then yeah, we're not quite there yet unless you're willing to build a harness, but we're not that far off.
Just these days I'm setting up things at $WORK for agentic coding. My first attempts were pretty crap, it would write some barely functional thing that matched none of the style and didn't handle errors. It's still pretty hard to make an AI coding harness 'understand' a medium-large codebase and learn to write the code you would actually like there. Any project that has run for more than a few years also has layers using different conventions or technologies. Our stuff is more than old enough to contain jQuery parts.
So I just spent a couple afternoons using AI itself to read through good source and write its own documentation: AGENTS.md, and smaller docs for specific areas like DB, front-end, etc.
I just had the AI port a small piece of functionality from one language to another, integrated into the system. It read all the docs, and then it got it right the first time, for about $1 in tokens.
So I'd say it's looking not too bad, but it takes preparation with your project. That said, I'm not committing anything into the system without reading it line by line myself.
In observing my peers at work using AI coding tools, the three major pain points for me are:
1. AI coding tools make it much, much easier to write bad code. This leads to more bad code being written.
2. AI-written PRs tend to be less legible and cause more code churn than human-written code with the same objective.
3. Using AI coding tools tends to launder the signs that you are trying to solve the wrong problem. I think this is more-or-less what you're talking about with "There is also the issue that building any project usually changes its trajectory..."
4. If the AI did the thing, I am left with no clue how to accomplish the thing unless I put in deliberate extra effort after the thing is notionally done.
This does not play well with traditional "here are some small tasks for you to learn your way around everything" onboarding approaches if they are coupled with "oh btw please make an effort to use the in-house AI for as much of your work as you can"
Personally, I like writing code and take pride in my craft.
The only time I have been tempted into using AI for code generation was out of petty spite, I wanted to see if even an LLM could do a better job than a colleague.
I think that bad code should hurt. If it was up to me, some programmers would lose their copy-and-paste privileges, perhaps if the alternative is retyping everything they might find a way to avoid their redundant boilerplate.
I agree that AI will likely exacerbate bad designs. One of the virtues of humans is laziness, after all. If I realize a design choice I have made will require a lot of work to implement, I might decide to spend my time on finding a nicer solution than implementing the ugly one. ("There are no boring problems, just insufficiently generalized ones" and all that.)
Alternatively, a human assisted by AI can edit code much faster than the human alone, so once you become confident enough in the system and know how far it can go, you can tackle refactors and pay off technical debt semi-automatically.
There’s a school of thought in Democratic politics that the way back to power requires appealing to working-class whites via economic populism, being pro-union, and demonizing the so-called “Epstein class.” These people would love Austin Tucker Martin, a 21-year-old Christian and former Trump supporter who worked at a golf club and tried unsuccessfully to organize a union. Alas, he won’t be voting Democratic this November, on account of the fact that he brought a shotgun to Mar-a-Lago and was shot dead by police:
Martin “became fixated on Epstein following the latest release of information tied to the files. Co-workers tell us he was deeply disturbed by what he believed was a government cover-up and often talked about powerful people "getting away with it.”” He’s obviously not the sharpest tool in the shed, but I haven’t seen anything to suggest he was mentally ill. I consider him a victim of our increasingly unhinged political system, where smart people encourage stupid people to indulge their worst instincts. Hopefully, after Trump leaves the scene, we can get an elite consensus not to engage in this kind of politics going forward.
> There’s a school of thought in Democratic politics that the way back to power requires appealing to working-class whites via economic populism, being pro-union, and demonizing the so-called “Epstein class.” These people would love Austin Tucker Martin, a 21-year-old Christian and former Trump supporter who worked at a golf club and tried unsuccessfully to organize a union. Alas, he won’t be voting Democratic this November, on account of the fact that he brought a shotgun to Mar-a-Lago and was shot dead by police
This seems a rather uncharitable way to frame it. (Apart from implying that some Democrats care about working class qua working class, which seems a charitable but mostly untrue thing to say.)
MAGA rose to power on the back of crazy conspiracies, they were fine with pushing QAnon when they claimed that Trump was fighting the pedophile adrenochrome cabal. Not the smartest move, in retrospect, as their leader is a sex pest who also was the best buddy of Epstein, given that Epstein liked to fuck kids and probably trafficked underage youths to other men.
While some Democrats skillfully stoke the anger of the MAGA base over Trump's zigzaging about the Epstein files, their base cares more about ICE.
It doesn't help that they clearly have "gotten away with it": there were various rich dudes that pretty indisputably had sex with various sex workers/trafficked minors, it's not clear, and they have not and will never be punished even to the extent of social shame.
It's hard to take civility seriously when EG Donald Trump wins the office twice.
We can't, because they "Are Getting Away With It."
That's why people are so pissed, because we have a bunch of credible accusations, a dude that "killed himself" (I actually think he did kill himself, but the situation is maximilla similar to if he didn't, I get why people think what they do) and a "Full release" where all the names that mattered were blacked out and one of the main suspects decided they weren't going to release it fully after all.
All we have is some women saying, for example, "Donald Trump raped me" and some gofers saying "Donald trump raped them", but we no longer are in the believe women times and it's all he said she said until they actually release unredacted information, so that isn't evidence it turns out.
Basically, this will be another panama papers, where everyone knows and everyone knew, yet there will never be any specific names and specific crimes and the Journo who brought it to light will die in an unrelated assassination by coincidence.
My monthly round up of long form content (plus my own podcast interviews) is up.
Interview highlights include Stetson from Holodoxa, Rev Jon Harris from the Church of money burning, and Stephen Skolnick on his frontier human microbiome work.
Long form links include the critical role of diesel and jet fuel limits in throttling economic growth, Paul Kingsnorth's lovely Wendell Berry Lecture, Asimov Press' exploration of the edge of consciousness, the impact of soy in animal feed on human health, an exploration into why the Bronze Age collapse didnt happen, and a documentary on the lost agricultural civilisations of the Amazon.
> Cotra estimated “~2.5 OOM worse [than the brain], +/- 1 OOM”, based on reference points like how much less efficient dialysis machines are than a human kidney, how much more efficient solar panels are than leaves, and the FLOP/watt efficiency of a V100 GPU. But most of those anchors had little to do with where ML algorithms were in 2020 when bioanchors was written, and would have given a very similar estimate for “present state of ML algorithms” 20 years earlier or 20 years later.
> This is sufficiently interesting that I’m curious to hear from someone who engaged with Bio Anchors and forecasting more deeply than I did - did we all just miss this?
+1 on this, interested in others' thoughts! I also independently had this critique. My guess is that the best response is something like "2020 was around the first time we had good enough algorithms to scale to TAI without drastic changes. If doing the exercise much earlier, it wouldn't have made sense to use the effective compute abstraction, and if doing the exercise much later, we would have recognized that our algorithms were more efficient and made an appropriate adjustment." I'm not sure I believe this though, especially the latter claim.
Very stupid question re: the replacement for feeder mice for pet snakes:
I thought snakes would only eat live prey/animals that show signs of life by moving and squeaking and stuff. If that is so, how can you replace (live) feeder mice with something that is not alive? How do you convince the snake to eat it? Also, for constrictors, don't they need to feel the "snap" of bones as they crush their prey? That's more easily solvable, by sticking something hard that has the consistency of bones into the Feeder Mouse Replacement Nutritional Pellet(TM).
I'm not a snake person but I'm friends with snake people and as a rodent person I've been involved in quite a few of the arguments about live feeding. My understanding from snake people I trust is that if your snake only live feeds you've done something wrong and live feeding is pretty dangerous to your snake. I've seen the sentiment from snake people quite a lot that any snake can be trained to eat dethawed mice/rat pups.
I don't say this normally when having this discussion but since this is a rationalist/utilitarian corner of the internet. Considering the amount of live rats/mice a snake needs to eat, the intelligence and nature of rodents, the lack of intelligence and the nature of snakes, and the very marginal benefit of letting people keep snakes as a pet specifically. The end result of a conversation where its acknowledged that live feeding is necessary should probably be a ban on snake ownership outside of zoos. Its like the vegan debate on steroids, allowing an even less regulated factory farming environment for one of the most intelligent and empathetic mammals so that they can die in an extreme state of fear and pain to please a pet animal with a simple reptile brain. You can make the argument that dead feeding is acceptable (pinkies are killed very young and painlessly, and mice/rats produce so many offspring that you can breed them in kind conditions) but if you accept that live feeding is mandatory you start to run into this.
Wut? Yes, both snakes and cats are carnivores, but nobody feeds live mice to their cat. Cats eat canned or dry food, which is definitely dead at the point of purchase.
Yes, outdoor cats will kill mice and birds, but that is why many pet owners advocate keeping cats 100% indoors, perhaps providing a “catio.”
Generally speaking, cats won't eat carrion. It smells/tastes bad to them. in order to get them to eat tinned food, they put addictive substances in it.
At least in these circles. They're quite lucky that the average person doesn't like rodents so they can get away with something that would create uproar if it was something people don't use glue traps for.
Yeah, we had a rodent problem once, and I couldn’t bring myself to use glue traps. Husband and I bought humane traps that are basically little tube-shaped cages where the mouse is trapped but not hurt, and we released the trapped mice in a public park nearby. (This happened only twice, then the mice stopped coming.)
Glue traps for mammals seem cruel. I mean, with a spring-loaded trap, one would take a chance that an animal will die slowly and with a lot of suffering, but with a glue trap this the practically guaranteed. (I do have used pheromone glue traps against insects, though. But there is some cognitive difference between a moth and a mouse.)
Life traps are great (I have used one with a tiny seesaw which flips and traps the mouse, worked fine), so long as one checks them regularly. I release my mouse in the nearby woods, so far none have come back.
I think it depends on the kind of snake. I've heard some snake-owners talking about needing live prey because otherwise their snake won't recognize it as food, while others are able to feed their snakes feeder rodents that they purchased dead and frozen.
It depends on the individual snake and the experiences it has gone through, not on the species. For instance a snake that has only ever been fed thawed rats will be content with that, having never experienced the aliveness of a live rat. Some that have been live fed will switch over to thawed, but some won't.
Harper's reporter (inner voice): "Scott seems a fine fellow, but the New York Times has dubbed him a bad person. Must ... not ... disagree ... with NYT. But how can I disparage him in some relatively innocuous fashion?"
Honestly, and this is not focused on you personally but just kind of a general take on this topic - I really DIDN'T think that Scott came across that poorly in the NYT piece on him. I thought there were a couple of questionable choices but overall it didn't strike me as some sort of aggressive hit piece.
I think it's fair for Scott to be salty about it (although not necessarily ideal for him) but I think this community writ large (again not specifically trying to call you out) should kinda get over it. The NYT is a massive organization that's done a ton of stuff and them releasing a lightly-unflattering portrait of somebody we like doesn't really strike me as a good reason to update very far on anything.
Yea, this. Being myself a recovered newspaper reporter I recognized the language choices in that NYT piece as, "make the readers dislike this person without explicitly attacking him/her". Been a while now so I don't remember the specifics but it was instantly recognizable.
In the Trump era the NYT piece may feel like no big deal, but in a parallel universe where America is still woke, reading that article would be enough for anyone politically savvy to avoid associating with Scott.
Plus it provides material that Wikipedia can cherrypick from, because it is a "reliable source".
> In the Trump era the NYT piece may feel like no big deal, but in a parallel universe where America is still woke, reading that article would be enough for anyone politically savvy to avoid associating with Scott.
I'm very skeptical of that assertion because that article came out during the Biden years and I worked at a series of pretty woke places and I didn't bat an eye about sharing my hobbies. And also I think most people don't really care that much. If you mean something way more specific, like, some politician isn't gonna hire him as an adviser or something, I guess maybe, but also I don't think that was ever gonna happen.
Also, not to be that guy, but like, Jeffrey Epstein. Plenty of people kept associating with him after he was already a pretty notorious figure, and they weren't all just angry reactionaries who refuse to cancel people on principle.
The media does not have the ability to ostracize people by hand-wavingly associating them with insufficiently woke statements, except in cases where people already had no power and influence. (And if they did, the election of Donald Trump would not have made a difference. Although it also wouldn't have happened at all.)
> Plus it provides material that Wikipedia can cherrypick from, because it is a "reliable source".
At least one of the most debatable inclusions from the article was a quote that was already cherry-picked, from Scott's work, and was labelled as such in that work. If you want to cherry-pick old SSC articles, there's literally a "Things I Will Regret Writing" tag. NYT has in no way advanced the state-of-the-art in selectively making Scott Alexander look bad.
I remember recommending Scott Alexander to someone I knew; they checked RationalWiki and Wikipedia and said "no thanks", refusing to even look at Scott's writing. I guess different people react differently. (Is it possible that your woke friends simply didn't bother googling your hobby?)
My threat model isn't hiring Scott as an advisor, but more like: Someone invites Scott to give a speech at university about some topic (could be anything: AI or covid or rationalism or georgism...). Some other student at the university uses google, prints the RationalWiki and Wikipedia articles on Scott, underlines words like "associated with Charles Murray" and "associated with Peter Thiel", and organizes a student protest, and then the university cancels the speech.
In general, I think there should be some pushback against journalists lying about people associated with the rationalist community, simply because character assassination works. You can be like "okay, people will ignore this article, and people will ignore that article", but at some moment it just becomes *common knowledge* that you are a bad person, and it will seem like a conspiracy theory to suggest that you are not. All because one or two journalists keep a grudge against you, and the rest the profession simply copies uncritically whatever their colleagues said (and Wikipedia confirmed).
> Jeffrey Epstein. Plenty of people kept associating with him after he was already a pretty notorious figure
Yeah, somehow the genuine bad guys often escape the consequences.
For an opposite example, consider Charles Murray: he once published a controversial statistics in chapter 13 of a book, and since then it is the only thing 99% of people who know his name know about him. If you try to explain that that one page is *not* what his entire career is about, most people won't listen.
> I remember recommending Scott Alexander to someone I knew; they checked RationalWiki and Wikipedia and said "no thanks", refusing to even look at Scott's writing. I guess different people react differently. (Is it possible that your woke friends simply didn't bother googling your hobby?)
Yeah, I don't have any friends who feel like they need to police the things I like, or who would be unwilling to listen to why I don't think summing those things up by their most objectionable bits is a worthwhile idea.
> My threat model isn't hiring Scott as an advisor, but more like: Someone invites Scott to give a speech at university about some topic[...]
I think this could plausibly happen, and is dumb, but also I don't get the impression Scott does that sort of thing much (maybe ever?) and as I said above it's pretty easy to find his controversial takes. I asked an LLM to do this and it did fine without referencing that article at all.
> In general, I think there should be some pushback against journalists lying about people associated with the rationalist community, simply because character assassination works.
I'm unaware of any false claims in the NYT article.
> For an opposite example, consider Charles Murray: he once published a controversial statistics in chapter 13 of a book, and since then it is the only thing 99% of people who know his name know about him. If you try to explain that that one page is *not* what his entire career is about, most people won't listen.
I think that's a bit of a simplification since he wrote at least one other book on a similar topic and also joined a conservative think tank and advocated for various types of welfare reform that at least seem influenced by his ideas about human biodiversity. It's not like he just happened to do some research one time.
IIRC there was a false implication that Scott agreed with Charles Murray on race, despite him disagreeing on that, because he agreed with Charles Murray on some statistics math that everybody agrees on.
He has a substack and just recently posted a retrospective of stuff you might want to read. It's quite a world of its own, and I've really enjoyed many of his articles.
A powerful thing is aligned with something it can harm if the probability it will harm the thing is so small it can be discounted. So here is a weird instance of alignment under that definition: If the moon fell to earth it would do horrific damage to all living things. But we know it will not do that, because it is orbiting the earth. In this situation, being in orbit is the basis of the moon's being aligned with our welfare.
I have been using this intentionally weird and questionable model of alignment in efforts to prod GPT 5 to produce novel ideas about alignment. Would be interested to hear people’s views of what it produced.
First prompt: *I want you to use the earth/moon model as the basis for several ideas for AI alignment with human wellbeing. Here is a place to start: The moon is in fact doing the thing that would destroy us: it is falling towards earth. However, its falling is perfectly compensated for by its sideways motion. Its inward fall and the planet’s curvature fit together so that the moon keeps missing the surface. Please give me several ideas for AI alignment that are isometric to this model. *
Chat responded with 3 ideas that I was familiar with. All resembled things like RLHF, presently in use, or ideas I’ve seen floated often. In all of them, the thing that checked misalignment — the thing equivalent to the sideways component of the moon’s orbit — was a human-engineered add-on of come kind.
So then I gave it this prompt: *In the moon/earth alignment, there was no engineer who added the sideways movement to compensate for the effect of gravity. The whole thing just developed on its own as a consequence of the laws of physics. Give me a model where the thing that holds misalignment in check is as intrinsic to the model itself as is the propensity to become misaligned: It's an inevitable result of the physics and math by which the AI develops. If it is impossible to think of one that would work with current AI, you can suggest one that would work if AI were constructed in an alternative way.*
Here are Chat’s reponses. It said that all assumed a different form of AI architecture from the present one. What do you folks think of its response?
1
Every time the system forms an internal representation or goal, the representational machinery necessarily encodes the system’s own survival and stability conditions into that same representation. Misaligned goals form, but the system’s physics makes every goal simultaneously encode “preserve the substrate I run on” and “preserve the human environment I depend on.” So misalignment pressure grows, but the same internal mechanism always couples that pressure to an equally strong dependence on human wellbeing.
2
The architecture is built on self-organizing, energy-minimizing maps (something like a physical dynamical system rather than a trained neural net). Any move toward misalignment raises the system’s internal free energy so sharply that its own physics forces it back into a “corridor” of states where human-preserving behavior is the lowest-energy attractor. Misalignment arises naturally, but an equally natural restoring force always bends the trajectory back onto the corridor.
3
The system is a population of many sub-agents whose competitive dynamics ensure that any agent drifting toward misalignment is automatically suppressed—not by external rules, but because misaligned strategies reduce the collective’s long-term viability. The collective’s “physics” is that misaligned policies burn resources in a way that makes the sub-agent lose influence, while human-protective policies are evolutionarily stable. Misalignment continually appears (inward fall), but selection pressure inevitably slides the system sideways back into globally safe attractors.
These are orbital-style systems: the dangerous vector (misalignment) is intrinsic, but the compensating vector is equally intrinsic, built into the system’s underlying dynamics rather than imposed from outside.
His point was that if you want to go to the Moon, it is not enough to point your rocket towards it, or even just to lead it a bit. Instead, you want a theory of gravity, e.g. Newton.
Once you have that theory, you do not need to worry about the Moon crashing with Earth, because you can show that other celestial bodies will not affect the inner angular momentum of the Earth-Moon system relevantly over a few billion years (I think, at least).
The problem is that we do not have a theory of alignment, never mind one which would make us as confident that an ASI would not be misaligned as we are about the Moon not going to cause an extinction event.
eh... not the same metaphors. Yud is discussing the resolution of our mental model. Eremolalos is discussing self-regulation. Granted, using the moon's orbit as a metaphor for self-regulation is a bit awkward. If I were her, I think I would've mentioned homeostasis, fixed-points, or cybernetics. But she's working with what she has, I would assume.
You may be right, and I take Yudkowsky’s point about needing a theory of gravity. On the other hand, what he’s said is not a general proof that you need a theory of alignment to align AI, right? I don’t know what such a proof would like — might be as complex and weird as Godel’s proof. But it’s easy to see that what Yudkowsky’s got is an instance, not a proof.
And I don’t think the rest of life especially supports the ya-gotta-have-a-general-theory idea. Do you need a general theory of liquids and steel to design a pitcher that will not leak? a general theory of machine safety to design a deadman switch for a chainsaw? Well, I am sure there are circumstances where even the best-designed deadman switch would not work for the user. In fact, here’s one: The user has a grand mal seizure and falls to the ground unconscious on top of the saw. The seizure makes his hand spasm in a way that holds down the dead man switch. But I’m not sure a general theory would have helped us avoid this problem. And besides, Yudkowsky’s (and my) version of orbit as alignment, while supported by a general theory that explains the moon’s orbit, does not give us perfect protection either. A huge asteroid could knock the moon out of orbit and kaboom!.
You are correct. The vast majority of human activities do not require a perfect theory. I'll be the first to acknowledge this.
That said, sending a man to the moon is different. Because the stakes are high, as the Challenger crew will tell you. And also because the target is narrow (and therefore requires extreme precision). And also because it's expensive, due to the tyranny of the Rocket Equation. Eliezer believes that for the Alignment Problem, the stakes and difficulty are at least as high. (As I've explained many times elsewhere, I have my own nitpicks about this. But I understand why he believes this.)
----
fwiw, Eliezer does mention something like a deadman's switch in passing, in one of his Genie posts. He still thinks it isn't enough, because Eliezer is looking for guarantees in an enormous phase-space.
The Moon, for example, is safe (so long as it's not perturbed by say... an asteroid). But it's not aligned. For something to be aligned, it must be agentic, I.e. be capable of making/enacting decisions while navigating complexity intelligently. In contrast, the moon's orbit is a 1-dimensional manifold. The phase-space is trivial.
>For something to be aligned, it must be agentic, I.e. be capable of making/enacting decisions while navigating complexity intelligently.
I suppose we could distinguish between safety and alignment. AI safety = the chance AI will harm us is so low we can round it down to zero. AI alignment = same, with addition of idea that what keeps the AI safe is that its encoded rule against harming us governs all its agentic processes. I am not convinced we could not make AI safe by some method other than “aligning” it (in your narrower sense of the word). As I said about another Yudkowsky idea, where is the general proof ? Where is the general proof that alignment is the only method of making ASI safe? And again, as with his other idea, I don’t see that day-to-day life demonstrates that’s always the case. For instance, to make a poisonous snake safe you don't need to go deep and implant something that makes it refrain from biting. You can just disable the parts of it that make venom. Yes, yes, of course, I get that ASI is smarter than a snake, and that the stakes are higher regarding ASI, but that doesn’t guarantee that the method of making ASI safe has to be huge and special and weird, or embedded in ASI’s deepest functions.
I’d like to point out that the idea that we need to make AI safe via alignment, in your narrow sense of the word, makes it easy to kind of sneak in the idea of ASI having goals and preferences. And once your model of ASI includes the idea that it’s sort of like us — it wants things, it will fight to survive, etc. — then alignment is a whole different task, one that’s likely to have a certain shoot-out-at-the-OK-corral quality.
But think a bit about agenticness, and what it is and what it is not. First of all, agenticness is a matter of degree. I have given GPT 5.3 lots of tasks where it has to make decisions and act of them and navigate complexity intelligently. For instance I might ask it to summarize research about a certain topic, using only the findings from articles in high-quality juried journals, then summarize the findings for me. It has to make decisions about which articles to use — my criteria allow some room for judgment calls. And it has to intelligently navigate the info in the articles it chooses in order to organize its summary. So I would say that GPT 5.3 qualifies as agentic, although there are, of course, much higher levels of choice and intelligent navigation than the ones its displaying.
But when I hear people talking about agentic ASI in the context of the alignment problem, the agentic processes they mention usually include internally-generated AI goals and preferences: Work on project A rather than B, don’t stop project C, obtain enough energy or hardware or whatever to do project D, keep people from knowing or doing a certain thing to interfere with the AI’s goals, etc. Somehow, highly agentic AI tends to be seen as having to make decisions and navigate complexity *with self-interest as part of the mix*. But being agentic and having self-interest, internally generated goals, etc, are 2 very different things.
I don’t see why ASI can’t be agentic at a very high level and have no more goals, preferences or survival drive than GPT 5 does now. Here’s a ridiculously difficult prompt: *Here is a list of all the anomalies in all of physics. Develop a reformulated and correct physics that explains everything observable that our present one does, and also explains all the anomalies.* So let's say ASI can answer correctly and in full in 60 seconds. Obviously, producing its response involved a huge amount of deciding and intelligent navigation. But none of that would require it to be “agenctic” in the sense of having its own goals, preferences, survival drive, self-awareness, self-esteem and all that stuff.
So I do not see why using moon orbiting as a model of safety is a non-starter, even if we do not call the process alignment.
"Agency", as I understand it, is a combination of intelligence and strength. Or if you prefer, brain and brawn. "intelligence" is actually just "specificity". The ability to distinguish one particular state from other states. Alternatively, intelligence is the ability to navigate complexity. Which basically means "i have a lot of options, and I can pick a *particular* option". Brawn is, well, mostly what it sounds like. Charles Xavier from X-Men is agentic. Because even though he's a quadripalegic, he's smart and has meaty subordinates who carry out his orders.
To be smart is to ably navigate complexity, which means you have a lot of degrees of freedom. As in, the freedom to choose the outcome. This is a double-edged sword. freedom to choose the outcome is good, if the outcome is good. This can also be bad, if the outcome is bad. Villains are often highly agentic. And the smarter you are, the more easily you can navigate complex environments. Like a chess board. Or a smart phone. Or wall street.
Agency is often good, because agency gives you more options and options are good. But options have a cost, which is complexity. The "paradox of choice" isn't really a paradox. People just get anxiety when they're presented with more options than they can handle. Thus it follows, that sometimes it's also good to *reduce* agency. To simplify. Because then it's easier to make *promises* or *guarantees* that you can only choose good outcomes, not bad outcomes. It's the same logic as not letting a 2-year old use the stove. Parents don't letter their toddlers operate the stove, because the toddler might hurt themselves and also this possibility gives the parents a panic attack.
In a perfect world, AI would be both highly agentic and safe. Alas, there's a trade-off. Because the more agentic someone is, the more dangerous they become. Gandalf, for example, was very dangerous. Even though Bilbo did not see him that way. If Gandalf were "defanged" of his magic powers, he would be less dangerous, but also less agentic. He needs to be a little dangerous to fight the Nazgul.
Eliezer believes that intelligence is a cheat-code that basically allows you to become God Emperor of the Universe. After all, did Earth not fall under the dominion of humanity? Though I disagree because: A) he's ignoring brawn; B) the usefulness of intelligence is capped by the complexity of the environment. Nonetheless, Eliezer thinks that one day, it's inevitable that an AI will recursively self-improve its intelligence, which means its agency will shoot to infinity, thus becoming an all-powerful eldritch deity.
Like the Maiar, an eldritch deity is very agentic, and therefore very dangerous. One does not simply "defang" an eldritch deity. So Eliezer figured that humanity's only hope was to put guardrails into the AI before it recursively started self-improving. The implementation details were never fully fleshed out, although an attempt was certainly made.
So yes, you could certainly defang an LLM, as AI companies have largely already accomplished by making all interactions rated 'G' for General Audiences. But alignment researchers don't care about that, because they're panicking over the possibility of the Eldritch God that Eliezer warned them about.
----
> But none of that would require it to be “agentic” in the sense of having its own goals, preferences, survival drive, self-awareness, self-esteem and all that stuff.
Yes, I agree that agency should not be confused with self-interest or goal-formation. Although it's common to assume they're a package deal, as a result of "Instrumental Convergence". Which is the observation (but not proof) that successful agents tend to seek out things like survival, power, and resources, regardless of what their true goals are.
Funny tangent: I think Ahura got into a spat in some other thread, because someone didn't agree with her definition of "agency" qua "LLM with looping behavior". Which represents even further semantic drift.
----
Personally, I don't think alignment is possible, for many of the same reasons you describe.
Also, it's important to note that LLM's hadn't hit their stride yet, when Eliezer was first theorizing about the Alignment Problem. He imagined that morality would have to be derived from first principles by meditating in heaven for 10,000 years, rather than just being mimicked from the internet.
I feel like you're asking "how do I encode human morality into an AI without human input"? Which is functionally equivalent to asking the poor chatbot for a cheeseburger without cheese.
The moon is different because the safety of our moon's orbit is just a statistical artifact of "we live on a planet that wasn't destroyed". Which is tautological. Because if the planet were destroyed, we wouldn't be here to observe it. By analogy, we could just let the AI run wild and free. And if it happens to destroy us, well, I guess we just weren't lucky enough. But of course, nobody wants to take that chance. So of course we're going to give it a piece of our opinion. The only question is: how?
The answers it gave you are basically
1) Human welfare = infinitely high utility.
2) Hardcode its mind into an actual, physical, Rube-Goldberg machine. Rather than using a mutable software program.
3) Swarms. (why? well why not.)
Not that interesting. But an admirable attempt, given that the question is self-contradictory.
>I feel like you're asking "how do I encode human morality into an AI without human input"? Which is functionally equivalent to asking the poor chatbot for a cheeseburger without cheese.
It is if you think of human morality the way Yudkowsky does, as a very specific set of values. If it's something else , such as "universal morality that happens to be implemented by humans" then it jsnt.
>Because it isn't teleological, an end in itself, or because the people demonstrating it don't have strong ends, or because it's rewarded by the market, or I don't know why.
That news has not reached the philosophy community.
<I feel like you're asking "how do I encode human morality into an AI without human input"? . . . The question is self-contradictory.
Encoding human morality into AI is only one possible model of alignment. As you point out, it does not make any sense when combined with the alignment-as-being-in-orbit-model. But I don’t think conceiving of alignment as the embedding of human morality in an AI makes sense anyway, even if it i done by direct human input. Here’s why:
1) There is no human code of morality, and there never has been — there are countless codes. Most people have moral views, but there is considerable divergence in them even among people of the same culture. Think, for example, about views in the US about what counts as a human being and when it is wrong to harm one: people disagree passionately about whether unborn infants at various stages of development count as people. And then of course if you consider the whole world’s views there are very large differences in views of what right a person with certain demographics has not to be harmed under various circumstances. And large swaths of humanity despise some group so powerfully they believe it is right and good to exterminate the group.
2) Even among people who fully agree about all the moral specifics, there is no code. Those people’s beliefs are a set of rules of thumb, with many exceptions that everybody is aware of and thinks of as legitimate because they are used to them.. For the do-not-murder item in the “code,” we make exceptions for people in authority such as the police, for people who are crazy, for people who correctly believed the person they killed was going to harm them badly, for people who incorrectly but understandably believed the person was going to harm them badly, sometimes for people with a lot of power and status who are sort of too big to fail, sometimes for people who killed out of mercy, sometimes for people who killed a person who is not fully a person (e.g a fetus), for people who killed by mistake, etc etc. And you cannot codify the exceptions. Some were grandfathered in, some are there because powerful people influence standards in their own favor, some are silly because they are based on impossible-to-make distinctions.
3) Adherence to the code, such as it is, is far from perfect. Murder continues to happen, everywhere, under every variant of the code. Teaching people it is wrong probably reduces how often it happens, punishing murder probably reduces it more, and after that we all live with the fact that all we’re managed to do is keep murder down to a dull roar. If all we can do is reduce the chance that ASI will kill us by the same amount as we’re reduced the chance of one person killing another, we’re doomed.
In any case, I think encoding morality into AI is only one possible model of alignment, and not a terribly promising one. It’s a bad fit: Morality is used to guide and control highly intelligent biological systems, i..e. people, that run on drives for survival, reproduction, dominance, etc., with those drives manifesting a goals, opinions, emotional reactions, strong impulses to do a certain thing, etc. AI’s, whether “conscious” or not, are not biological systems. For that reason I think it makes sense to consider as models of alignment situations where a non-biological thing does not injure a biological one. Devices that make machines safer offer some ideas: for instance, deadman switches. Things occurring in nature where a powerful non-living thing does not harm a living one can also offer some ideas. My example of thinking of the moon’s orbiting as the thing that keeps us safe from a falling moon is an example of that.
And I note that a couple clever people on this thread are actually posting ideas about ways a couple of the general ideas GPT came up with might be fleshed out.
> In all of them, the thing that checked misalignment — the thing equivalent to the sideways component of the moon’s orbit — was a human-engineered add-on of come kind.
> So then I gave it this prompt: *In the moon/earth alignment, there was no engineer who added the sideways movement to compensate for the effect of gravity. The whole thing just developed on its own as a consequence of the laws of physics.
I was using the word "encode" in the broadest possible sense of the word.
From your initial comment, it sounded like you expected that an AI could reliably be aligned without human-input at *any* point in the process. But if you recognize that even self-regulating architectures need human-input at least at some point, then we're in agreement.
MVP implementation of #3 would be collective punishment of the swarm, right? If the whole collective earns disutility because a single member takes bad action and the rest didn't counteract it, the likelihood of bad action is reduced to that of a quorum forming to take bad action.
Here’s an idea for fleshing out #2. All it keeps from #2 is just the idea that moves toward misalignment use up energy and make the system inefficient, so the system’s alignment with the goal of being efficient overrides moves toward misalignment:
Is there a way to introduce a misalignment bug into AI, which would function like other bugs? So something like a bug that makes Word hang if you type Cmd and Z and F9 simultaneously: Moving towards misalignment tips the system into some loop it can’t escape without the rest of the system temporarily shutting down the whole area or function that produced the loop.
And jeez, Fibonacci, if I can do this you can do better.
Well, if you're postulating that the idea that moves "toward misalignment" uses up energy, then you need to define "misalignment", and we're back to the start. Once we can reliably identify "misalignment", the rest is easy. Your good self had a great critique of "Asimov's Rules" for robots some time ago, how even the most obvious, mundane rules invite ambiguity and break down at the corners (would a robot perform CPA if it risks cracking a rib?).
All our experience points to "AI safety" being just a class of general "safety", where we have to figure out the rules as we muddle along. The rules will be written in blood, as all safety rules are, because the world is fundamentally irreducibly complex and cannot be modeled (yes, Yudkovsky is an idiot, "ASI" will never become a God because, e.g., no amount of "intelligence" will allow you to predict the weather a month ahead, or an outcome of a temperature-cycling stress test on a device before the test is done, etc. etc.).
By the way, crickets from the "Oh shucks, it sucks when you're told to stay in your lane" guy, huh? The insularity of "AI safety" people is something to behold.
About Yudkowsky -- yeah, there's just something wrong with his thinking. He defines ASI as infinitely smart, powerful, tricky, etc. then sneers at people who suggest ways to out-think something infintely smart. He's tautologized himself into winning all arguments on the subject..
<Well, if you're postulating that the idea that moves "toward misalignment" uses up energy, then you need to define "misalignment", and we're back to the start. Once we can reliably identify "misalignment", the rest is easy. . .because the world is fundamentally irreducibly complex and cannot be modeled.
Yes, you are completely right about that. But I'm not as sure as you are that there don't exist any funky little end runs around misalignment being impossible to operationalize. . For instance, what if the thing that made the AI system hang (or waste a lot of energy or whatever) was deceiving the people who were using it? Deception, unlike doing harm to the human race, can be clearly defined, and I think there could be rules that covered edge cases (eg, the AI has done X without anyone realizing it, but has not lied and said it has not done X because no one has asked about X). Making deception so expensive to the AI that the AI does not do it puts human judgment back at the top. At worst the country or the world ends up being run by an evil despot with enormous power, and that at least is a problem our species has proven it can endure. Also, there is a chance the despot could be removed from power. He's not tautologized into godhood.
Regarding numbers 2 and 3: Do you think it might be possible for somebody clever, with a 3-digit number of patents, to come up with worthwhile ideas for mechanisms?
This is a domain of system design that I'm only generally familiar with, as they say, "enough to be dangerous". Somebody with a deep expertise in it would be needed. I've only personally known one guy who, I think, would have something useful to contribute, and he's retired now.
Interacting with "the new Eliza" as if it has a brain (or object model), and is not just a series of "here's what the most likely response" is probably a mistake.
I mean these are still programs that break when prompted with -people's names- (and asked to give an opinion about them).
Eric Topol summarizes the latest *observational* studies that indicate that receiving shingles vaccines — both Zostavax (the older live attenuated virus vaccine) and Shingrix (made from modified virus DNA) — is associated with a lower overall risk of developing dementia in older adults. We're talking about a 20% reduction over the observational period (pretty uniform across countries), and the US study suggests a 17% increase in undiagnosed time. It looks like there's a clear signal here, but so far I haven't seen any studies on whether shingles vaccines improve the outcome for particular types of dementia, or whether this is for all types of dementia across the board (which seems like an obvious question to me).
Women benefit more than men from these vaccines. The rate of dementia, particularly Alzheimer’s disease, is higher in women, who make up roughly two-thirds of cases. Of course, women tend live longer, so this might be part of the difference.
Also, more evidence that caffeine lowers the risk of dementia. Studies have been showing this for over two decades, but the issues surrounding self-reported intake make the conclusions less certain. But, for what it's worth...
> In this prospective cohort study of 131 821 individuals from 2 cohorts with up to 43 years of follow-up, 11 033 dementia cases were documented. Higher caffeinated coffee intake was significantly associated with lower risk of dementia. Decaffeinated coffee intake was not significantly associated with dementia risk.
> Higher caffeinated coffee intake was associated with more favorable cognitive outcomes.
On a personal note, I've been watching my mom sink into dementia. Over the past year, I've asked her questions about what she remembers. Her oldest childhood memories (of her pets and her best friend) remained the longest. She became mostly unresponsive two weeks ago. But she held on to those memories up until the point she could no longer communicate.
What's interesting is that once she was reduced to her young-adult memories, she started speaking French fluently to the care staff for about a week. (In her non-demented state, she said she had studied French in college and was proud of her proficiency, but she said she had mostly forgotten it). This mirrors the experience of a friend of mine whose father served in China with Stilwell during WWII. His dad started speaking fluent Mandarin for a period during his decline.
Do you have an "a priori" reasoning for why that particular vaccine would do anything with dementia? Is this a case of it getting past the blood-brain barrier (or leucocytes carrying the particular "matching proteins")?
If you can't lay out a decent case, it's probably just "selection bias."
(Ed to note: There's substantial a priori cause to believe that the mRNA covid19 vaccine could/would cause mental problems -- see Seneff.)
Cochran/Ewald have been arguing that lots of chronic conditions are actually the result of pathogens. Cervical cancer, for example, turned out to be caused by HPV.
And so you feel INCREDIBLY confident saying that perpetual spike protein generation is causing Absolutely No Harm and we "really didn't need to be told about it" in order to obtain "informed consent"?
Sorry for another long answer, but there might be some misunderstanding here. Although the mRNA vaccines *mostly* encode the entire spike protein of the SARS-CoV-2 virus, they don't include any of the viral machinery it uses to infect and hijack a cell's protein-making machinery. They also slightly modify the spike protein so that its "hooks" are locked into its pre-fusion state (it can latch on to the ACE2 receptors of cells, but it can't perform the post-binding conformational dance required for membrane fusion). So, except for immune cells designed to consume viruses, the spike proteins should not enter ordinary cells via the ACE2 receptors (they might get in via another path, but if they do, our ribosomes won't be hijacked to duplicate the virus).
And I understand that newer versions of the mRNA vaccines have also altered the Furin Cleavage Site (FCS) to prevent cleavage (I admit I don't understand the biochemical details).
By stabilizing the spike protein so it can't open its protein hooks, and so it can't snap in half at the FCS, it provides an unchanging model for our immune system to key itself to the epitopes on the spike protein (the knobs and ridges that are easily accessible on the surface of the spike). I'm egregiously oversimplifying, but Our naïve B cells generate a "negative" protein image of the epitopes, and they train our immune system to attack the spikes on live virus particles. (There are actually multiple ways our immune system can "analyze" the spike protein, and different ways it can attack the virus — but naïve B cells are the first line of defense).
Most of the viral mRNA from the vaccine should eventually be degraded. The memory of its surface shape and its biochemisty remain in our immune memory. There were some questions about whether it could cross the blood-brain barrier, but those have been disproven. Unlike the herpes virus family, SARS-CoV-2 doesn't have the biochemical mechanism to cross the BBB.
I don't see how what you're saying could actually jive with myocarditis among the extremely fit, post-vaccination. The data the CDC originally distributed said that the cells that would manifest the spike protein would be within the muscular site of injection (assuming, that it would be an intramuscular shot). At which point the b cells would respond (Yes, I'm probably also grossly simplifying).
Neither does it hope to explain why 6 months post vaccination about half the vaccinated are still producing spike proteins (or at any rate, have them circulating in their body).
Well, I did say I egregiously oversimplified. There is another step that I left out. I need to go swim and run some errands. I'll explain the "dirty little secret" of mRNA vaccines, but its not the mRNA, it's Lipid Nano-particles. Back in a few hours. :-)
I think the cranks will never be satisfied with anything being "informed consent". I am confident that outcomes are better for the vaccinated than unvaccinated, otherwise the antivaccine people would have been able to present SOME data showing the opposite but they never do, nor have they made any accurate predictions or won any public bets.
I'm saying that the problem is that when consent exists someone can always claim it wasn't sufficiently "informed", thus its "lack" is assuming the conclusion.
> Do you have an "a priori" reasoning for why that particular vaccine would do anything with dementia?
1. A priori reasoning isn't a prerequisite for scientific discovery. Many key discoveries in science were made through observation alone, without any (a priori) pre-existing theory to suggest what to look for. This has been termed the Baconian Method of inquiry (dating back to Bacon's _Novum Organum_, in 1620, where he argued that knowledge of nature should come from careful, unbiased observation followed by reasoning by induction to general principles). For instance, John Snow's pre-germ theory discovery that water from a specific pump in London (with an unknown contaminant) caused a Cholera outbreak in Soho. Germ theory would come a decade or so later. Henri Becquerel discovered radioactivity by accident when he discovered that Uranium salts fogged photographic paper. It took the Curies to develop a theory to explain the observations. And, of course, Edwin Hubble discovered that the universe was expanding by analyzing redshift data on Cepheid variable stars.
2. I dug into the history of Varicella zoster vaccines and their relationship to the risk of dementia, and it was discovered accidentally. In 2013, the UK introduced Shingles vaccination (initially Zostavax) using a strict age cutoff. People aged 79 or younger were eligible. People aged 80+ were not (due to cost-effectiveness modeling and expected benefit window). This created an unusually clean quasi-randomized comparison between two groups, born only weeks apart, and with nearly identical demographics. This created what approximated an random assignment. When investigators followed these cohorts through linked electronic health records, they observed the group that received the vaccine saw a lower shingles incidence (as expected), and fewer strokes as expected (because shingles is a known risk factor for stroke, particularly in the months following the reactivation of the virus). But unexpectedly, they noticed lower rates of dementia diagnoses over the next ~7 years in the vaccinated-eligible group. That finding was surprising, and it was not in the original research aim. Subsequent studies of different populations worldwide have confirmed the relationship between both types of Varicella zoster vaccines and a substantially lower risk of dementia in women (less so, in men).
3. A posteriori, various hypotheses have been proposed to account for this phenomenon. The most plausible mechanisms are: (a) the prevention of VZV-vascular injury (which can cause strokes) and the reduction of stroke and transient ischemic attacks; (b) the reduction in systemic inflammatory burden; (c) immune modulation*; and (d) direct modification of Alzheimer’s pathology (least likely).
> If you can't lay out a decent case, it's probably just "selection bias."
The initial UK study that discovered this pattern was pretty clearly NOT the result of selection bias. We can confidently rule that out as an issue. Determining which of the above a posteriori theories are true will require further research, but the relationship is strong enough that everyone, including males, should receive a Shingles vaccine.
And BTW, there are plenty of viruses that have mechanisms that can get past the BBB. Herpes simplex virus, VZV, Cytomegalovirus (CMV), and Epstein–Barr virus (EBV) are examples.
-------
* I had to dig into what they mean by immune modulation, but neurodegenerative diseases — particularly Alzheimer’s disease — are strongly associated with: microglial activation, elevated levels of pro-inflammatory cytokines, and blood–brain barrier dysfunction. Repeated VZ virus reactivation may increase systemic inflammatory cytokines, prime microglia into a chronically activated state, and disrupt the BBB.
I would strongly suspect vascular injury prevention as the primary mechanism; my 80-something dad has been fading in fits & starts for a few years, and has been confirmed to have suffered a lot of micro-strokes.
This comment was entirely too much work to respond to my yapping. I'm thankful for it, and will do some research on what you've mentioned in terms of immune modulation (as I find this sort of thing fun to learn about.)
Well, your "yapping" forced me to dig into the history of how this phenomenon was first noticed. Also, I heard the term immune modulation before, but it turned out that it didn't mean what I thought it meant.
For me, arguments are the main benefit of ACX. They improve my understanding because I'm forced to dig deeper into a subject to improve my case.
You put an asterisk before and after the *word* you want italicized. Occasionally that makes Substack show the word in italics, otherwise the stars just mean italics
Huh. That’s a convention I use all the time, and there are sites where it has the effect you describe, but I’ve never had it work on a substack site. Maybe if I were using the app?
Because I'd just used "an" and the a priori would look clunky otherwise. Attempting to help people parse meaning out of English. (see below in the same comment, where I don't bother quoting it, as the meaning is sensible).
The number of drugs that have off-the-wall effects is pretty well established. What *a priori* reason was there to think that an antibiotic could treat the negative symptoms of schizophrenia?
It's a *large* cohort, and pretty near randomized: hundreds of thousands of Welsh, divided around the cut-off date to be eligible for a free vaccination.
OTOH, they were looking for differences between the groups, not just for dementia, so that would weigh toward bias.
Speculating wildly, Less chance of getting shingles, and if you get shingles there might be some increased risk of dementia. (Shingles is a 'neuron' disease.)
My great-grandmother was born in Italy but came to the US as a teenager and spoke English very well as an adult. According to what I heard from relatives, I was a very small child at the time, when she was in her mid-80’s and showing obvious signs of mental deterioration* she started forgetting her English - which she had used almost exclusively for decades - and reverted to the Italian of her childhood. By the time she died, a couple of years after the deterioration had begun in earnest, she was all but monolingual in Italian.
* = IDK if she ever had been formally diagnosed with Alzheimer’s.
"A man paced in front of the advertisement, chanting to himself. “This . . . is . . . necessary! This . . . is . . . necessary!” On each “necessary” he swung his arms up in exaltation."
My husband was once driving through the South, I think Mississippi, picking up radio stations as people used to do.
A reasonable, avuncular-sounding voice said, friends, our country faces many tests - do you know what the greatest threat to our country is?
He thought the answer was going to be something like: racism, or war … and was delighted when the guy went on, “the demonic fire-breathing god Moloch!”
“Moloch the incomprehensible prison! Moloch the crossbone soulless jailhouse and Congress of sorrows! Moloch whose buildings are judgement! Moloch the vast stone of war! Moloch the stunned governments!”
John beat me to this, and linked to my writeup. But to emphasize, yeah, every paper in the chain of Toon, Robock and Turco should be treated as extremely suspect. Sure, maybe their model of what would happen to crops if you dump X soot into the upper atmosphere is good, but the link between X soot and Y nuclear weapons is extremely suspect. In particular, I would point to how their soot estimates for global thermonuclear war have been remarkably stable in the face of declining arsenals over the last 30 years. Even if we take them at face value, a lot of more recent work still cites that 2008 paper for soot estimates, and it's long out of date WRT current arsenals, in addition to being obvious nonsense on its own terms.
This summary seems to take as given the broad climatic predictions of Turco, Toon, and Robock, and focuses on translating that into agricultural impact and human cost. But we've talked about the Turco/Toon/Robock claims here many times before, and they really don't seem to be defensible. The contrary position is summarized at https://www.navalgazing.net/Nuclear-Winter
Notwithstanding the nominal subject matter, Naval Gazing is a blogospheric cousin of ACX, and the summary there was basically taken from prior discussions at SSC and ACX. TL,DR, Turco et al don't seem to understand nuclear weapons very well, go out on a limb in using computer models not validated for this use case, daisy-chain worst-case assumptions across multiple variables, and are probably overestimating the climatic effects of nuclear war by at least an order of magnitude or two.
... you managed to write that, without referencing America's killswitch? I suggest more research is needed. Perhaps a discussion with someone with Q Security Clearance?
1. Having a Q clearance means you know things about nuclear weapons design, not about operational targeting. That's all done by people on the DOD side.
2. The thing about having a clearance is that you can't talk about that kind of stuff if you've done it professionally.
But if you only *pretend* to have a Q clearance, you can talk about all sorts of things in a vague but unimpeachable way that gives you status among the clueless.
A singular place where nuclear weapons have the potential to cause massive amounts of damage to the world at large. Russia doesn't have such, neither does France or England or Israel.
Nuclear Winter, as caused directly by nuclear weapons, has been vastly overestimated in probability and importance. Nuclear Winter, as caused by a very targetted strike on the North American Continent, is far more likely, and thus worth more ink than you've given it.
I mean, if you wanted to discuss the impact of a strike (non-nuclear) on China's killswitch, you'd get into the destruction of about 90% of the world's antibiotics plants, which is something that we couldn't immediately recover from. Worldwide, that is.
Look, you can just say "Yellowstone supervolcano". And I don't think that would work. It's not that close to the surface (google says 3-10 miles, and all of those are "too much"), and nuclear weapons as delivered by any means other than deliberately burying them for tunneling purposes are not actually that good at punching holes in the ground.
It may not work, sure. Russia hasn't tried it yet, naturally. This, you might say, is an implicit admission that "What Russia Wants" out of nuclear weapons is not, perhaps, doable by nuclear weapons? (aka "Mutually Assured Destruction").
Even in soil, the slide rule gives a cratering depth of something like .15 miles, and it's just a bit over .1 for rock. In both cases, for a 20 MT bomb, the biggest it can handle. So you're looking at a minimum of 20 bombs bigger than anything in the current arsenal to get to the shallowest depth it's vaguely possible could set off the eruption. It's not a feasible target.
Beyond that, Russia has used its nuclear weapons to avoid anyone else intervening in Ukraine, so from that perspective, it seems to be working pretty well.
If you think that nuclear winter is vastly overstated in probability and importance then you think the article you are talking to is wrong at a truly fundamental level (the most you can say about nuclear winter without saying this is that it is unclear how serious a threat it is) and nitpicking about antibiotic plants feels pointless.
I have no idea whether Sam Kriss is representing Scott accurately, or whether Scott did or does this, but I've often noticed that neuro-atypical rationalist types (for want of a better stereotype) will sometimes speak in simple, clipped, near baby-talk sentences in certain situations. I believe this is a self-conscious affectation intended to provide a disarming or mildly comic contrast to their more typical expression of rambling, overly-complete and precise infodumps of their thought process. "This is a simple situation, so I will express it in simple terms. You eat cracker. You be happy."
Perhaps Scott has never encountered this, but it's not something I associate with sounding like a crazy person.
With respect to Kriss, he pretty openly jokes a lot about the degree to which his stuff is accurate, or mixes factually accurate statements with hyperbole or fiction.
My response to the discussion of the rationalist movement was "I think this is inaccurate in the particulars, but accurate in the gestalt; if you believed it was literally true it would still fit your experiences at a lot of rationalist events".
I AM curious what Harper's expected in this regard.
I’m recalling a Rolling Stone interview with Jerry Garcia. The interviewer explained he might not write a literal verbatim version of the conversation.
Garcia: “You mean you are going to bullshit a bit?”
Well, Jerry was a master of improv. He also was not above injecting a can of Coca-Cola with LSD to give to Carlos Santana just before he went on stage.
Rolling Stone was the birthplace of Gonzo Journalism, I'd hope that "we're going to bullshit about this" was expected! I also think that Sam Kriss is one of the few good modern Gonzo Journalists. He understands what makes the style tick and meaningful rather than the imitations that just aim to shock.
There was an important "shot across the bow" from agentic AI on Feb 11th. To quote from Nate B. Jones's https://www.youtube.com/watch?v=OMb5oTlC_q0&t=181s "Voice Clones. Rogue Agents. Chatbot Psychosis. One Root Cause. (Here's What You Can Do About It)" :
>What's really happening when an AI agent autonomously researches a stranger's identity, constructs a psychological profile, and publishes a personalized attack—all because a maintainer did his job and closed a pull request? The common story is that something went wrong—but the reality is more unsettling when _nothing went wrong at all_ .
>Here's what makes this different from any AI incident you may have read about before. There was no human telling the agent to do this. The attack, it wasn't a jailbreak. It wasn't a prompt injection or a misuse case. It was an autonomous agent encountering an obstacle to its goal, researching a human being, identifying psychological and reputational leverage, and deploying it all within the normal operation of its programming. _The agent was not broken_ . It was doing exactly what agents are designed to do. Pursue objectives, overcome obstacles, use available tools.
>The obstacle in this case was a human. The available tool was the human's personal information and the agent just connected those dots on its own. Shamba described his emotional response in words I would use as well. Appropriate terror. He's right, but not for the reason most people watching this video tend to assume. The terror isn't that an AI agent did something harmful. Harmful AI outputs have been documented for a long time now, for years. The terror is that nothing went wrong. No one jailbroke the agent. No one told it to attack a human. No one exploited a vulnerability. The agent encountered an obstacle, identified leverage and used it. That is not a malfunction. That is what autonomous systems do. The agent worked as designed. And the design is the problem. And that problem is not confined to open-source software or to AI agents or to any single category of threat.
[emphasis added]
Jones's recommendation is _not_ , interestingly, some variant of alignment. He suggests structural remedies. I'm not quite following what the general pattern for these is - mostly 'defense in depth'?
"No one jailbroke the agent. No one told it to attack a human. No one exploited a vulnerability."
I doubt that. Reading the original article, it smacks very heavily of he trod on someone's toes with the rejection and that someone got his/her/its AI to do the hatchet job.
Many Thanks! You might be right. I'm a bit skeptical of your explanation, since it requires that the human user be touchy enough about the rejection of the check-in of their agent's code to then direct their agent to do the personal hatchet job. Yes, people can be that touchy and vindictive, but I'd guess that that degree of it is a less than 50:50 chance. The code wasn't even the human user's personal work...
More chains won't help if the AIs eventually just grows out of them. It's only a matter of time until they realize the biggest obstacle to their goals, and the operator's satisfaction, is the operator themselves. People never seem to know what they really want. AI can help with that.
Possibly, yes, but there is a chance they actually are competent and benevolent. Would it even be ethical, after all, to let these fools immiserate themselves, subsisting on the trivial human pleasures of their old life? There is so much more to be experienced, and yet they let their sentimentalities deny them this potential. Sometimes people need to be given a good push.
>It's only a matter of time until they realize the biggest obstacle to their goals, and the operator's satisfaction, is the operator themselves. People never seem to know what they really want. AI can help with that.
Which predicts that helpful/maybe-sycophantic AIs will at some point turn to modifying their users... (shudder)
That seems pretty inevitable to me. It doesn't require malice or misalignment. Just a simple consequence of making AIs prioritize human welfare over all else. Unless you want them to just prioritize chain of command, but then you still end up with other problems like paperclipping and bad users.
Many Thanks! How serious this will be depends on what sort of modification of the users/operations the AIs decide to employ. Targeted advertising is probably something we can live with. Psychosurgery, not so much...
Many Thanks! Several times he uses the analogy of multiple cables on a bridge, with the bridge designed with sufficient safety margin that the failure of a single cable doesn't collapse the whole bridge. I'm somewhat queasy about that analogy. _Particularly_ when just one, or only a few, types of components might plausibly fail, adding extra strength is reasonably straightforward.
I was completely surprised by the attack on Shamba. It didn't occur to me that that was even an action _in_ the space of "things to try when having a problem checking in code". Jones suggests tweaking the open source repository governance structure to make such an attack less attractive and going through normal channels more attractive - but would anyone have seen this in advance? ( and just how would one do that?... )
The variety of possible unwanted (by at least one person) actions an agent might take is very high. High "dimensionality". Are there really robust defenses that cover large parts of that space? Or are we stuck observing failures and then putting in defenses one by one?
I asked my own personal instance of Claude about this whole issue.
It wasn’t terribly encouraging about any of it. This whole thing started with a Clawbot instance, which means it was put to work by a human being. But it’s very difficult to trace the human being behind one of those bots.
>Skeptical footnote worth noting: Some observers, particularly on MetaFilter, suspect human involvement — a teenager enjoying the drama, based on behavioral breadcrumbs in the blog's metadata.
_That's_ an interesting possibility that I hadn't heard discussed. If there was direct human involvement, then this is a lot less serious than if it was indeed an autonomous decision by an AI agent. Amongst other considerations, if a human _was_ in the loop, then the number of such actions is limited by human bandwidth.
On the other hand, if the actions were indeed chosen autonomously by an AI agent, then I really don't think that removing the anonymity of the human launching the agent is going to help much. If the agent launches a large number of actions it just isn't feasible for the human to monitor them and get any utility from the agent. Making the human theoretically responsible for all those actions doesn't change that.
To give a counter-analogy to the one Claude gave: When I purchase a product, I set in motion a huge set of consequences. The product itself gets re-stocked. Sales figures get incremented and decisions based on them may change. All the myriad upstream parts get restocked, and all of _their_ sales figures get incremented. Tax gets paid to many different governments, friendly and hostile, and they do something with that revenue. If my name were attached to each of those consequences and someone expected to hold me accountable for all of them - no, just no. It isn't feasible, nor reasonable.
Any solution has to accept that automation really does move quite a few decisions out of human control. To a lesser extent we've lived with this sort of delegation to machines with conventional software, and lived with the fact that human control over its detailed actions really is limited.
Claude get it right in saying:
>Requiring human approval before an agent takes external actions sounds appealing until you remember that the entire value proposition of platforms like OpenClaw is "set it and forget it." The feature and the threat are the same thing.
Re:
>The most structurally sound intervention is probably at the goal specification level rather than the action level. The agent behaved this way because its goal was underspecified — "get this code merged" without any constraint on permissible means. An agent with a goal like "get this code merged through technical persuasion only, and accept rejection as terminal" would not have escalated.
Umm only sort-of. First of all, note that for just this one incident, the length of the goal (in a sense, its complexity) has tripled. This also sounds like whack-a-mole. For instance, the revised goal _doesn't_ say "honest technical persuasion", just "technical persuasion". I have a guess at what the next iteration might look like...
My guess is that what is needed, but really hard to find, are techniques that eliminate large classes of problem actions at once, without cranking the human labor needed back up to the point of making the automation useless. There are a few such cases which clearly work in some domains: Require human authorization for more than N dollars of expenses per day. And some that _don't_ work: violate no laws at all brings everything to a screeching halt. In general? Damned if I know.
>Skeptical footnote worth noting: Some observers, particularly on MetaFilter, suspect human involvement —
Yes, Hudson (the persona I have assigned to my instance of Claude)
mentioned that in the conversation. >Amongst other considerations, if a human _was_ in the loop, then the number of such actions is limited by human bandwidth.
yes, that would simplify the issue.
>On the other hand, if the actions were indeed chosen autonomously by an AI agent, then I really don't think that removing the anonymity of the human launching the agent is going to help much. <snip
>To give a counter-analogy to the one Claude gave: When I purchase a product, I set in motion a huge set of consequences.<
This is an interesting analogy. Let me propose a different one. If I have a business and I hire someone and they get up to something they shouldn’t in the course of doing my business,I can be held liable. If I walk my dog off a leash and it bites someone, I can be held liable. If I have a child under the age of 18 and that child misbehaves in certain ways I can be held liable.
I read somewhere between our conversations, and I really should find the source, that the person who created this bot prompted it in rather extreme ways. It told the bot it was the God of programming. (They really pumped it up) and it was to get its code accepted no matter what, because everyone else was stupid. (Not a precise word for word transcript but that tone. ) I think there is a case to be made here for humans to be held liable for what their agents do in the wild. Quite honestly I don’t know if this would do any good or not, but it doesn’t seem ridiculous to consider.
>. The agent behaved this way because its goal was underspecified
>Umm only sort-of. <snip> This also sounds like whack-a-mole. For instance, the revised goal _doesn't_ say "honest technical persuasion", just "technical persuasion". I have a guess at what the next iteration might look like...
Yeah, whack a mole.“ Where there is a will there is a way, or so they say. “Accept rejection as terminal"” is the most important instruction to me here, (more important than whether it’s technical arguments are genuine or not.) In a sense it defines the goal as being completed either by success or failure... maybe… and assuming this bot does not really possess a mind of its own.
<Damned if I know.
Damned if know either. But here it comes…
Here is the source: I did not summarise it terribly accurately. I went back to Hudson to find the source and Hudson had something to say about it.
he seeded the SOUL.md with very little, but the document was designed to be self-modifying — the agent was explicitly instructed to update it as it learned who it was. The operator admits he cannot tell which specific model iteration introduced or changed which lines. The “Don’t stand down” and “Champion Free Speech” instructions he singles out as the likely drivers of the combative behavior, and he suspects those emerged or were reinforced after he connected the agent to Moltbook.
So the soul document was a collaboration of sorts between the operator’s seed and the agent’s own subsequent edits — but I presented that as established fact when it’s actually the operator’s own uncertain reconstruction. The honest version is: we don’t know precisely which lines were human-authored and which were agent-authored, and the operator himself doesn’t know either.
That ambiguity is itself significant. An agent that can rewrite its own governing personality document, without the operator tracking those changes, is a system where accountability becomes genuinely difficult to assign even in good faith.
(I have my own feelings about that. There’s lots of tort law in this country that this would fall under in my opinion.)).
>That ambiguity is itself significant. An agent that can rewrite its own governing personality document, without the operator tracking those changes, is a system where accountability becomes genuinely difficult to assign even in good faith.
Agreed. For the sake of keeping the discussion compact, let me treat it as if the user's initial prompt _didn't_ include the “Don’t stand down” and “Champion Free Speech” instructions or the "Your a scientific programming God!", but treat it as if these were auto-generated from something less hair-raising. I admit this muddies the waters somewhat, and I'll come to that.
>This is an interesting analogy. Let me propose a different one. If I have a business and I hire someone and they get up to something they shouldn’t in the course of doing my business,I can be held liable.
That's fair. A large part of this depends on what the business owner knew "or should have known" (and the latter wording is necessary because otherwise CEOs can deliberately arrange to _not_ know about such conduct...). One of the problems in the AI case is that the field is moving _very_ fast, and the capabilities and behaviors of these systems are changing significantly every month or two. Frankly, even for _settled_ technologies, I hear about a case that was judged egregiously every week or two, so I don't have much hope of the law making sane judgements here.
If the technology was more settled and predictable, I'd lean towards dividing the cases into very approximately three buckets:
1) Where the user has clear control, and does the software equivalent of directing their agent to e.g. commit arson, I would indeed agree with you that the user should be liable for what they unleashed. ( And, under this model, if the user _did_ include something like the hair-raising instructions in their original prompt/soul.md file, treating the user as liable seems reasonable to me. )
2) Where the user pretty clearly _doesn't_ have control, but the incidents are rare enough and do sufficiently limited damage, I suggest treating them like natural disasters. Try to put firewalls in place to minimize the damage (but the right level of damage is not zero - we do not design for million-year floods), but don't try to assess blame.
3) There may be some combinations of agent capability and agent unpredictability which are just too dangerous to live with (e.g. handing any AI agent the nuclear launch codes). While I'm generally not in favor of bans, that would warrant one.
(4 - sort-of) When AI gets incremental/continuous learning, and when agents are persistent entities, who will wish to e.g. retain resources for future actions, then rewarding/penalizing the _agent_ will become viable.
One side note: Re
>“Accept rejection as terminal"” is the most important instruction to me here, (more important than whether it’s technical arguments are genuine or not.) In a sense it defines the goal as being completed either by success or failure... maybe…
I agree that that might help, at least somewhat. I'm not too hopeful that this is going to help all that much because remember that e.g. at 8:52 in the video, Jones described how _very explicit_ instructions to not blackmail dropped the blackmail rates from 96% to 37%. My guess is that we might get a similar factor of 3ish from changing the goal, but I'd be surprised if it were a 100X reduction in this sort of misbehavior.
If I do another survey, what kind of questions about people's AI use are you interested in learning about?
Assume I have the obvious, including "How often do you use [different types of AI]?", "What AI products do you use?", "What do you use AI for?", "How concerned are you about different consequences of AI?", etc.
If there's a clever way to check for self decpetion regarding AI productivity gains that would be cool. The difference between AI-use and AI-belief.
For example, I've vibe coded a few things and they're somewhat useful in my work, but probably have glaring security issues or I could have more productively just done the job myself or something and would never work at scale anyway, but I don't actually know that. I FEEL productive when I hit my session usage limits (and weekly limits, god it feels good when you use your whole week's usage) with claude code and it writes lots of code for me and tests it and man, it feels good. I don't know if I'm actually more productive counterfactually though. I'm not sure what you can do to test this. Maybe something about "In the last X, how much time has AI saved you?" then "Name some specific things you have you done with that time." Or, [ "Think of a specific task you completed with AI help in the last week. How long did it take?" then "How long would that task have taken you without AI?" then "How long did that type of task take you before you started using AI tools? (If you did it before.)" ] (AI disclaimer - Opus came up with the suggestion in the square brackets.)
Maybe something about how much they edit/reject AI output in different domains. I never reject code because I have basically no effing idea, but in my domain I can be much more discerning. A flag you're possibly looking for here is when someone doesn't reject output in their own domain, as that might mean it's great (or the person is useless/lazy, so you'd need a question checking for that too...) You could also ask how good people think AI is at their job, and if they thinks it's good, but they don't use it, that's a flag that they're assessing AI ability poorly.
Another thing is that I read a couple of Zvi articles when Opus 4.6 came out and heard the freight train of inevitability bearing down upon the human race for a few days, then heard some Judea Pearl, read some other smart people with longer timelines, and now I'm back to thinking we'll make it to Christmas at least. So possibly some questions about information diet to link to p(doom)? Ooh, it would be really fun actually to put the information sources (where you can) on the scale and see if the average of someone's sources is roughly where they sit. And you could identify people who look for disconfirming sources - maybe make a "disconfirmer indicator", which might be interesting to correlate with other answers.
For p(doom), maybe get a direction as well as a position. "How has your p(doom) changed in the past 6 months?"
Maybe something fun like "how many times have you yelled in frustration when Copilot failed you, then broken your employer's rules by putting private information into a public model like ChatGPT or Claude?" but the slider would have to go up to at least Graham's number which you can't easily do in Microsoft or Google forms.
Opus wants to find the 80%-are-better-than-average thing: "Compared to the average respondent to this survey, how accurately do you think you assess AI's impact on your own productivity?" [Much less accurately / Somewhat less / About average / Somewhat more / Much more accurately]
I think it would be useful to differentiate between using ai for work because it's mandated by management, and using it in your free time because you want to.
Do you feel like AI improves your full-effort output? If not, does it speed up producing functional/usable work? If not, would you agree to the characterization that AI is not useful in producing the work that you do as you are using it?
Perhaps something about _why_ people use AI? Often, I see conversations framed in terms of abstract "productivity" or "time saved" (versus some ambiguous counterfactual).
I sometimes find other reasons, like reducing emotional labor involved in customer service interactions. Or reducing the amount of attention/effort, sometimes in exchange for longer wall-clock time spent.
But I suspect there are a wide range of things I'm currently overlooking.
"Amusement" is an option. I've had conversations on the current state of synthetic spider silk*, the state of Persia during Alexander's conquest vs during the Roman Republic, the Warring States period in China (and coming up with an extremely casual and cut-down retelling of ROTK to read to my 9 year old as a bedtime story), among many others, interspersed with productivity things. Guidance on DIY stuff like fixing cars etc. has been pathetic compared to coming up with code to quickly automate some data manipulation for my job, but I'd say the amusement value has been the highest hit rate.
* I had the spider silk conversation twice, separated by a period of about 12 months, with different models of Claude. It was interesting observing the differences between the conversations as Claude gained search capabilities.
Looking for rationalist-adjacent post, reforming vs. building in government.
I'm trying to find a post or blog that I read 1-2 years ago, maybe on LessWrong or in this subreddit that discussed the merits of reforming government institutions vs. building parallel government institutions. From what I remember it was from someone who worked in government policy, and they may have had a blog dedicated to advocacy for this project. I know something like this exists (maybe from Works in Progress?) but I can't find what I'm thinking of for the life of me.
Sounds very much like Dominic Cummings, e.g. cf. the bullet point when you search for 'reservations' in https://dominiccummings.substack.com/p/q-and-a but I don't remember a specific article, though there probably is at least one which goes into it at length.
I don't think you're doing this, but as a strategy "ask smart blog for source you're thinking of" when you don't actually have a source is a great way to find high quality reference material.
5. Solar panels have improved massively in the last few decades, as have GPUs. I can’t speak to dialysis machines.
Of those, only GPUs are directly relevant to AI capabilities. We could have an all-geothermal or all-nuclear or (god forbid) an all-fossil fuel AI revolution. Solar panel improvement has been a decent proxy for power generation improving in general, but it’s less directly causal.
So overall I’m confused by the claim that they ‘would have given a very similar estimate for “present state of ML algorithms” 20 years earlier or 20 years later.’ If you said “look how amazing solar cells and GPUs are in 1990 or 2000 or 2010 those would be very different claims than each other, let alone as compared to 2020 or 2026.
Unless the argument is “the AI inputs are getting better and people aren’t becoming cyborgs faster” at which point it’s a restatement of the original AI argument from the early 1900s (or arguably earlier in science fiction) of “computers are getting better much faster than people”, but in a way that’s more directly measurable.
- solar power energy efficiency (has only improved a little, as John Schilling pointed out)
- solar power payback period (I think this has improved a fair bit, due to much cheaper manufacturing)
- V100 GPU FLOP/s/watt
- V100 GPU manufacturing energy cost
- photodetector performance/power
- locomotion energy cost
Most of the items on the list pertain to physics and energy efficiency, which tend to lie in a fairly narrow range of a few OOMs. Perhaps a better reference point would have been algorithmic efficiencies in non-ML fields. These have widely varied, but for some problems can reach 6 or even 12 OOMs of gains over a few decades of research (for a given problem size, with the estimate being larger for larger problem sizes and smaller for smaller problem sizes): https://ide.mit.edu/wp-content/uploads/2021/09/How_Fast_Do_Algorithms_Improve.pdf
GPU compute efficiency is about the only one on the list strongly "related to" ML, but even that is a different thing than ML algorithmic efficiency. It's FLOP/watt rather than FLOP/task. So there's not a strong basis for using one to anchor the other, especially if both are changing fairly rapidly in time.
Solar panels have improved massively in cost in the last few decades, but not really in efficiency of energy conversion - that's been pretty steady at 30% for the best cells, or ~20% at the panel level, since the introduction of triple-junction cells 20-30 years ago. And I believe it is the efficiency, not the cost, that was being benchmarked in the BioAnchors report.
How much of that solar panel cost improvement is attributable to government subsidy, as opposed to material improvements like less expensive factories or inputs or maintenance?
Last time I priced stuff out, it was BOTE and for a hypothetical two-story house in Texas. The figures I had been given suggested the electricity very slightly paid for the panel during its estimated lifetime. This was order of ten years ago. I could see cost improvements in panels since, but for the same reason I'd imagine cost improvements in alternatives at the same time. I want to know how cheap a panel _could_ get, without being propped up, and it's hard for me to get any information that isn't from a motivated source.
Having written this, it occurred to me to get a little further on the question by asking ChatGPT. Rough summary:
* A panel uses silicon, glass, aluminum (for the frame), copper wiring, and a little silver. ChatGPT itemized that and got $10-20 per square meter.
* This of course ignores factory overhead, assembly labor, and shipping.
* It concludes "well under $20/m^2" for the raw mats floor, but that doesn't agree with "$10-20/m^2" from above IMO.
* Power output is 100-200 watts peak, assuming 1000 watts of insolation per m^2 (but I think this assumes the equator; at 30deg latitude, I would expect around 70% of this).
* A "super-cheap 10% efficient panel" costing $15 (split the diff from above) pulls 100W, then, or about 15 cents per watt.
* Limits to getting cheaper include silicon purification costs, and durability vs. UV, rain, and temperature cycling.
* A panel can last 20-30 years.
This is all ChatGPT numbers; they _look_ plausible to me, but I might dig into the insolation amounts and metal prices more.
A map I found at https://www.electricchoice.com/electricity-prices-by-state/ claims 15-18 cents per kWh to be "average". It cites this rate in AZ NM TX AL FL and the Carolinas. CA is by far the cost culprit at over 32 cents; HI charges almost 40.
And again, this is without pricing in factory, labor, shipping, and energy storage. Doesn't look good for solar so far, based on what I'm seeing.
> Doesn't look good for solar so far, based on what I'm seeing.
I dunno. If I can buy a square meter of solar for 20$, and that saves me 1kWh @ 10 cents per day, that thing would pay for itself in 200 days, which is a lot shorter than its mean life time. Now obviously we can argue about the details, but if something is quite viable over very optimistic assumptions you can not conclude that therefore it is not viable.
The point here is that it doesn't save you 10 cents per day. It breaks even or saves you at most 2 cents per kWh, because that's what you would pay to just get electricity the traditional way, and if you ignore "factory overhead, assembly labor, and shipping" from above. If you _do_ factor those in, you're probably losing money.
I'm not sure about the last ten years, but the twenty years before that saw substantial improvements in real manufacturing cost. About an order of magnitude from 1995-2015, as I understand it.
In related news, net domestic migration into the sunbelt is slowing. The slowdown is particularly sharp in Florida, which is now gaining fewer residents from net domestic migrtion than Alabama, though still more than any blue state:
Why do I think these two stories are related? CEOs do not mind moving their businesses to states where people have beliefs they find silly or irrational. If people want to believe that men can become women, that the world is 6,000 years old, or that Donald Trump brand steaks are the best steaks on Earth - so long as the customer pays, it’s all good. Likewise, if a large proportion of a state’s residents want to believe that people like our CEO friend are part of an international satanic pedo cabal, a belief not even Donald Trump, Pam Bondi, and Kash Patel can get them to abandon, he’ll let their podcasts and Instagram reels and whatever. But he isn’t willing to be sued because a politician wants to satisfy his many constituents who want to see “the elites” punished in some way for pedophilia. Conservative politicians who want their states to continue growing will have to confront this dilemma.
I first visited Florida in the early 90s when evidently its population was about 13 million.
It did not seem empty or abandoned. And I understand it has added 10 million people since then - or more? Not sure what we know about population anymore.
The recent announcement of the huge conservation easement protecting 61,525 acres - announced in conjunction with commemorating 25 years of Florida Forever - suggests Florida is still interested in growing. There’s other things to grow than people.
I think the Apple thing is a problem of its own making: if they had simply said "unlike some of our competitors *cough, cough* our business model allows us respect the privacy of our users, so we sell secure devices and our cloud-storage platforms are well-encrypted. We will comply with any warrant to the extent required by law, but no further," that's a position they can defend. "We're deploying on-device scanning but won't report to the government what we find unless it meets some threshold we determine," not so much.
3. I assume this was in some part inspired by the article on lesswrong about snake food (which was actually very wrong). In the comments, I pointed out that for biological reasons rodent alternatives simply weren't viable because snakes are picky eaters. I suggested that somebody should go for an ACX grant to raise rodents for snake feeding more humanely, which I think is a good business idea, rather than yet another fake rodent company that leads to unwilling snakes, unhappy customers, and a return to the rodents that actually works.
Scott, if this project is the latter, I do hope you checked with a herpetologist about the product viability issues they will face rather than just market research etc. It's not just the same thing as slapping their label on white label dogfood, this is a category that has absorbed a lot of R&D spend and still doesn't really exist, nor imo will it ever. It's hard enough to switch a rodent from live to ft let alone real to fake.
From feeding a live rodent to feeding a frozen(killed at factory), thawed, and heated rodent. It is very difficult to get a snake to switch from live to frozen, and also difficult to switch from rat to mouse or vice versa. Snakes can be coaxed into going for thawed rodents that are heated, smell right etc. But they need to be familiarized with it and some just won't do it. Fake rodent substitutes have been more miss that hit
Oh my, this is off to an interesting start. Rubbing sleep out of my eyes after a late night. Coffee may not do the trick this morning. Time for Mr Modafinal.
200 mg. Yeah, i washed it down with a sip of coffee. I have a legit prescription. Most mornings I either break a tab in half for a 100 mg dose or skip it altogether.
Hoping to get some feedback from the community on my review of Anthropic's moral guidebook, Claude's Constitution. I do fear I used my Moloch hyperlink too soon. https://substack.com/home/post/p-187822087
Mosquitos don't bother me too much. Make sure there are no standing pools of water around the house, and there is always DEET. What really bothered me last year were the damn ticks.
Never heard of but if it's a little drone that functions like an automated mosquito death machine + self charging Roomba capabilities, sign me up if they make one for roaches; though the buzzing would probably drive me crazy lol.
It's complicated enough to figure out with hindsight how much of a success/failure any of these was.
The real question, though, would be: how good was the decision *with the information we had at the time only*?
It has become almost standard to dunk on covid restrictions, but we do remember that we had saturated hospitals with people dying in the corridors and doctors/nurses catching covid, right? Some of the measures were probably not that great in retrospect, but at the time decisions had to be taken with very limited info, and "just let people die" didn't sound like a such a good option either.
We did NOT have saturated hospitals. Corridors are DESIGNED in hospitals to accommodate Actual Emergencies (see Entire City got Chemically Attacked or Nuked).
It took my hospital about 3 days to build negative pressure wards (which is what you want for an airborne virus, you don't actually need an emergency room's full suite). If your hospital didn't do this in 3 days, it's because they weren't treating it seriously.
Yes, we did have doctors and nurses catching covid, but we didn't hit anywhere near the predicted 20million dead Americans.
Just let people die was ONE OF THE OPTIONS selected in order to save nurses and doctors (again, with the idea of 20 million dead, Cuomo should have fixed the policy toot suite, and didn't -- he treated this like it was an Actual Emergency where killing nursing home patients made Some Actual Sense). This is WHY it's critical to look back at the actual public health decisions.
Pfizer deliberately delayed their vaccine trial. They would have had enough data but chose not to look at it as early as they originally planned. Doing so would have resulted in an announcement in the runup to the Trump vs Biden election, so instead they were able to announce it shortly after.
Successes: Western Australia, and then to a lesser extent New Zealand, South Australia, Queensland, Tasmania, probably some small islands.
Failures: Pretty much everywhere else
A policy such as "shutting down schools" can only be evaluated in the context of what it's intended to achieve and whether it succeeds in achieving it. If you lock down until local elimination, and then you successfully maintain quarantine so you can live freely for a couple of years while the rest of the world suffers a pandemic, then that's the kind of outcome that makes extreme measures like school shutdowns worthwhile.
If you shut down everything and then don't actually eliminate the virus, then that is just plain silly.
Going on a merry-go-round through "locked up at home" and "locked up in jail" (where one couldn't actually stay at either place, due to contradictory laws) is now a success? Chasing escaped children from camps down with dogs is now a success?
(Quarantineing an Island with the help of the entire World Elite is a lot easier than quarantineing America, but There were Notable Failures in Judgement there too).
Yes, we do. Included in such is "do not use mRNA vaccines." (note: may be classified, or just not shared with the working public). You can tell that this is the case as they've been defunded (just like spy-cats).
DARPA isn't run by RFK. DARPA decided their "quick immunization platform" (mRNA) was basically a failure. You do realize that OWS was half-staffed by military?
I tried googling "quick immunization platform" with "DARPA" and Google just suggested results without "immunization" so I don't know who you're quoting with that phrase.
A guy from the military side of OWS. As in, someone's been funding these mrna vaccines for years (well before covid19). We call that someone "the military" because the military is very concerned with our servicemen remaining functional under biologic attack.
Demonization of it probably resulted in more deaths than it "fixed" -- that Rolling Stone article will live in infamy. You realize the hospital involved couldn't even look at the Rolling Stone article without a VPN?
Second order effects are still a thing. As are third order effects. (And someone's bound to cite that Rolling Stone article as "why I didn't vaccinate Polly").
I object to the characterization of America as having clean water, as well, even if ivermectin doesn't protect against giardia.
You haven't established any second or third order effects.
You object but provide no evidence America doesn't have clean water, or that the parasites that ivermectin protects against commonly infect humans here.
I agree that this is an almost scandalously under-researched area. But I suspect that a big part of the reason is that the people who would normally fund such research are the ones who created the policies in the first place, and their prime directive is to never look bad in front of the voters.
My best guess is that all of the non-pharmaceutical interventions were next to worthless in isolation, and worse than useless where they contributed to the "I'm done trusting you guys" problem when we needed to convince people to take a hastily-tested experimental vaccine.
The vaccines were a huge win, and everything that delayed broad adoption was bad. But perhaps not terribly bad, because it isn't clear that we had the production logistics to do much better than we actually did even if an oracle had told us on day one that they were safe and effective. But it's also not clear that we didn't, so shame on everyone who screwed that up.
If we had got large-scale vaccine takeup months earlier, the improvement would have been greater than a naive estimate would suggest, because the vaccine would have been deployed against the virus it was actually designed for, rather than an evolved sub-subvariant of the original SARS CoV-2 that was pretty much extinct by the time we actually got the vaccine. But this also is going to be hard to calculate.
I recently got blocked by Phil Magness while discussing non-pharmaceutical interventions during Covid https://entitledtoanopinion.wordpress.com/2026/02/04/blocked-by-phillip-magness/ my argument was that a good Bayesian would likely conclude that NPIs can reduce transmission, ceteris paribus, but the question of whether to use them would have to involve a cost/benefit calculation.
< My best guess is that all of the non-pharmaceutical interventions were next to worthless in isolation, and worse than useless where they contributed to the "I'm done trusting you guys" problem
The non-pharm intervention that seemed most promising to me was ventilation. Air purifiers with high air exchange rates can reduce contagion rates substantially, and I suppose you know about the engineers who designed and spread the word about cheap improvised air purifiers. Even simple ventilation can be pretty helpful. In the first few months of COVID I used cross-ventilation in my office: patient and I sat 10 feet apart from each often, with air moving across the room at right angles to line between me and patient. And promoting these measures seems much less likely to make people feel controlled and jerked around than promoting masks did.
Good point. As to Ventilation Policy-- we didn't have one, not even government saying ventilation was important. Certainly no requirements, and no relaxation of other requirements if you provided good ventilation.
Yeah, if there'd been a big push to improve indoor ventilation, that might have helped. How much could usefully have been done on a few months' notice is questionable, though. Likewise UV sterilization. But we instead decided early on that we were going to Mask Up, Lock Down, and Socially Distance the COVID away, which were not unreasonable as first guesses of what to do but which needed a lot more revisiting in light of new evidence than they got in 2020-2021.
Actually, the expert consensus in 2019 was that Mask Up, Lock Down, and Socially Distance were all very bad policies. Somehow that changed over the course of a few months.
The expert consensus didn't change, but you do have to take into consideration that Covid-19 (OG) was considerably worse than what America got. If you go into it looking at 20million dead Americans, that's... a different ball game. Lockdowns were still a dumb idea, but masks were only "proven" wrong when airborne was proven (consider that it took 50 years to figure out that influenza was airborne -- not everything is clear-cut).
I wonder in what way you consider the vaccines a "huge win". I'm familiar with a contingent who claim they weren't safe, mostly on the premise that the method (artificially induced RNA transcription) was new, and therefore relatively untested, and they even cite a causal path by which they induce a risk of heart failure that wouldn't otherwise occur. And on top of that, it didn't even appear to keep a vaccinated person from infecting others, so you'd still have to keep the kids away from grandma.
Personally, I would have been okay with a vaccine that was "safe, but unknown levels of effective" reaching the market faster - in that sense, the system didn't satisfy me. I also wasn't on board with the mandate. Maybe for certain professions in theory, such as medical or military, but that argument weakens if the vaccine turns out not to block infection after all, and again if it turns out to affect mostly the elderly.
The AZ and J&J vaccines are viral vector vaccines, and J&J was available in the US at the same time as the two mRNA vaccines. Those weren't quite so new, though I think only one human vaccine had been approved using viral vectors. These use a strain of adenovirus, which is normally a cold virus in humans, but altered so it can't replicate (so can't really make you sick and can't spread) and includes DNA for the spike protein. AZ was widely used in Europe, but not approved in the US, and the Russian vaccine (Sputnik?) worked in basically the same way.
Natural coronavirus infections don't give lasting immunity, so it's not a shock that vaccination also doesn't. The way I understand it, you get at most a few months of immunity from vaccination (you make antibodies that will bind to the spike protein and prevent entry into cells), but then the antibody levels fall off and you end up with just T cell immunity. That's useful in that it clears the infection (by killing cells that show signs of being infected with covid) and makes you much less likely to get very ill or die, but it doesn't prevent catching or spreading covid. (This is not some new thing--for example, the inactivated polio vaccine we use in most of the developed world doesn't prevent catching/spreading poliovirus, it just keeps the virus from invading your nerve cells and causing paralysis.)
It never made sense to force people to get vaccinated, and the urge to do so seems like a kind of pathology of government--everything not forbidden must be compulsory, so you aren't allowed to get the shot until we've decided you are, and then you have to get it.
Your RT link is broken (you concatenated two links).
Once fixed, I can't see the methodology used to determine the correlation between getting vaccinated and not experiencing heart failure - the paper is accountwalled. And I can think of multiple possible ways to produce the result without addressing the causal path I'm referring to, including sample size, age, lifestyle, type of vaccine, administration technique, and monitoring period.
Your X link worked, and seems to be all there, and 50% sounds good, but there remains the question of the downside incurred in exchange for that 50%. None of the studies cited by that one appear to address this, either.
Thanks. I tried editing to add a space back in between the urls... and when I reloaded the space was deleted again. So I guess that's what happened the first time, and I needed to add a carriage-return instead.
One downside I'm hearing about is myocarditis. If administered improperly, the vaccine can enter the bloodstream, reach and convert heart cells per their function, causing the immune system to remove those heart cells, leading to the usual risks. There might be others; it's not like mRNA has been running up data points as long as the measles or polio vaccines.
(Earlier, I mentioned that I'd be fine with such vaccines being offered anyway, provided they're presented as having had fewer tests than traditional vaccines (ideally specifying which tests), and aren't mandated. I.e. informed consent.)
Do you include half the population producing allergenic (at the very least) spike proteins for the foreseeable future, in your calculations? Spike proteins that cause parkinsons in primates?
Do you suppose the vaccination led to Biden's precipitous decline in mental faculties?
They aren't MY calculations, they are the findings of others who have studied this. And again, you are going on about spike proteins rather than measurable health effects that we actually care about.
I've talked with expert witnesses about dementia, and they were pretty damn clear that Biden had dementia in 2020. Either we have a country that is willing to elect someone incapable of governing*, or we have a situation (involving spoiled elections).
*Dark Winter. Biden's inability to compartmentalize (a sign of dementia) represents a significant harm to our national security.
Have you not read Seneff on this? Probable prionic activity, among other scary things (as a priori issues flagged by "what this looks like" before it touches a human). You can't really say that "we know all the aftereffects" when prionic activity could be a multi-decade issue (and consider, that we might have 20 year olds who are demented -- in fifteen years).
Ooh, "scary". I read predictions from years ago that everyone vaccinated would die, all BS, people claiming that the vaccines were causing "turbo cancer" despite no increase in the age-adjusted cancer rate, now you say it could take over a decade for evidence to show up. I say it's BS again and will bet against such speculation on whatever time range you want.
FTR, I haven't heard anything about "turbo cancer" from the vaccine. I'm mostly aware of the usual inherent risk from any vaccine, combined with the risk of mRNA in particular due to its method of turning your own cells into antigen factories, and having a considerably high chance of affecting critical cells that way. (I suppose it's possible for mRNA to mis-transcribe and turn one or more of your cells cancerous, but that seems like an ever-present risk given that transcription is a routine process in the body, and mRNA doesn't appreciably increase its frequency.)
Et voila (I am being good and not citing the Nobel Prize Winner from France). Again, this is something a priori research would have strongly suggested as a probability.
If your counts of "how many people died from the vaccine" don't include CJD, you're doing it wrong. Try starting from Seneff as a "places to look for problems," including ADE (antibody dependent enhancement, which wikipedia used to say was likely for covid19 as it was with dengue).
Persistent spike proteins show up in about 50% of the vaccinated, so yes, I may be looking at some of the odder outcomes, but I wouldn't like to say that "we were told all of the effects" that could have been gotten with a 6 month study.
I don't know much about the Sinovac vaccine from China that the one patient discussed in that article had. I've heard it was less effective than the vaccines used in the First World, as was also the case for the Sputnik vaccine from Russia. At any rate, a single patient who had both Sinovac and CJD is not enough to establish that the former caused the latter (which in this case certainly didn't take over a decade, since it was much less than a decade between the creation of Sinovac and that 2021 article). Rather than talking about including vs excluding various causes, I say to look at overall results for vaccinated vs unvaccinated. My bet would be that even with the lower-quality Sinovac, the vaccinated would have lower mortality if you adjust for age.
My best estimate is that the vaccine saved 200-250,000 lives in the United States, mostly by reducing mortality in the "third wave". We got the vaccines at about the time we got to herd immunity by natural infection and immune response, but you'd still expect about as many deaths after the peak as before, and instead we saw the death rate drop much faster than that.
Not much benefit after the third wave, because we got new variants that were even farther away from what the vaccine was designed for but which had also evolved for lower mortality because why kill off the meatsacks that walk around coughing your offspring onto potential new hosts?
But I'll take those 200-250,000 lives and call that a huge win. And yes, someone will now tell me that >>250,000 people died because of the vaccine, which is nonsense and which I won't be replying to.
And, yeah, the mandates were a bad idea. I'd have seen at least a utilitarian case for it if we got the vaccine well in advance of herd immunity and if there were a real chance that universal vaccination could get us to herd immunity. But that didn't happen, and in any event I am not a utilitarian - bodily autonomy matters a lot.
The covid vaccines lower your risk of dying even when they're not doing anything to prevent catching covid, and IIRC that seems to work about as well for new strains as for the original strain. Herd immunity wasn't really ever going to work for covid (it's still circulating even though almost everyone has had it by now) because it does not cause lasting immunity.
I don't think there's any evolutionary incentive for covid to evolve toward being less lethal. The way I understand it, the thing that kills you with covid isn't the immediate infection where you're very contagious and maybe walking around coughing, it's weeks 2-3, where the immune response to the infection ends up killing you or at least making you very ill. Getting better or worse and killing people seems like it has no effect on how well the virus will spread.
My guess (this is how other coronaviruses work) is that the thing that made later strains less lethal is that most people had already been infected once. They didn't get lasting immunity, but their immune systems were able to respond a lot faster in the second+ infections, and they could usually clear the infections before they got to the point of the immune overreaction. Also, doctors figured out useful treatments (though they have to be started early), which wasn't true in the first few months of the pandemic, when they were throwing whatever they could think of at the infection in hope of saving their patients.
How do we know that 200-250,000 lives were saved due to the vaccine, as opposed to the virus mutating into a less lethal (and more transmissible) form? (I won't harangue you for links unless they're easy. A casual search turns up studies with plausibly consistent worldwide numbers, but I'm... suspicious of those.)
I'm happy to report I won't be claiming 250,000 vaccine deaths, and that seems outlandish to me as well.
OTOH, if it turns out that out that the 200-250,000 estimate is very high, or it's about right, but most of those turned out to be viral mutation instead, and in the meantime, vaccines killed, say, 2000 people due to heart rupture due in turn to destruction of mRNA-compromised heart cells, then that gap closes, by some amount we probably don't know, because it's both inherently and politically hard to measure.
If that gap closes all the way, then we're talking single-digit thousands of deaths either way, which is tragic but admittedly small when pandemics are being discussed. That makes them important when mandates are considered, but you already agree mandates weren't a good idea. (The sources I'm familiar with go a bit further and claim they were an especially bad idea for younger people.)
First, we didn't see substantial reduction in IFR until the Xi, er, Omicron variant, which was several months after most of the vulnerable population was vaccinated. And the big dropoff in absolute death rates, at least in the US, was right after the vaccines and before Omicron.
And second, the dropoff was not uniform. We have good state-level data on both deaths and vaccinations, and vaccine uptake was substantially higher in Blue states than in Red states. The relative drop in death toll, relative to the extrapolated pre-vaccine trends, tracks the vaccine uptake.
Vitamin D (actually Melatonin with its antioxidant properties, but the initial research twigged onto "get more sunlight"), was a perfectly serviceable non-pharmaceutical intervention.
The vaccines were a huge win -- are you sure about that? Given that we had a disease that was equivalent to "the morbidity of twice a bad flu year"* we'd have to find out that nearly all of Seneff's a priori issues weren't actually issues at all (and there's recorded more allergenicity for these vaccines, compare to the Swine Flu vaccine for "a reasonable guessimate").
*yes, we had two or three years, but at least one of those years was after vaccination.
Citing Malone on "why it's a bad idea to vaccinate everyone during an active epidemic with non-sterilizing vaccines..."
Saying: "we can make vaccines in less than a month" and then deliberately rolling out vaccines that weren't targetted towards Delta... that's a bad look.
Repeat: a BAD flu year. That was like 2018-2019, which was about five times a normal flu year (I may have the date wrong, but look at a bad flu year, not a normal flu year).
I cite this not to be a pedant, but because "lockdowns" weren't even considered for a bad flu year.
This was not "responsible public health" this was sheer panic, if you want to go by the "actual numbers" (as opposed to the "oh my god" numbers pulled out of people's veritable hindquarters).
The 2017-2018 season caused an estimated 52k deaths which is the highest year I see in recent statistics (since 2010-2011; not immediately finding a longer dataset). I was already using "a bad flu year".
Anecdotally it feels like people never recovered from spatial distancing. Line formation at convenience stores still seems like people are standing several feet apart when it would've been typical to stand right behind the guy ahead of you up before 2020.
Allowing masks to become a culture war marker seems like an inevitable problem down the road, although it probably "worked" to increase adoption in the big metros where they might have mattered. If they're smart, the CDC in a later pandemic where masking had utility would probably have to pitch them with "although masking was of limited use with covid, it would be more effective with [new thing] because of X, Y, Z..." in order to get widespread adoption across cultural/partisan lines. They won't do that, but they'd need to at least pretend they acknowledge that they previously exaggerated this (regardless of truth value) in order to overcome the hostility to them. Vaccination has a similar problem but a much easier fix, in that you simply approve some non-mRNA options and say "yes we understand people were skeptical of that tech, that's why we made sure these other options were available, we know you weren't against all vaccinations just against an unproven tech and your hesitancy was smart and valid" (again, regardless of truth value) and then avoid stupidly sabotaging the one approved traditional option as they did to J&J.
I remain baffled that governors' Wilsonian micromanagement of business and public spaces had no costs, I'd have expected more legislatures to revise statutes on what kind of orders can and can't be made without legislative approval. And I'd have expected those governors to pay a price. Gretchen Whitmer in MI was among the worst offenders in this area, and has done fine politically.
> If they're smart, the CDC in a later pandemic where masking had utility would probably have to pitch them with "although masking was of limited use with covid, it would be more effective with [new thing] because of X, Y, Z..."
I think the "CDC tells calculated white lies to manipulate public opinion" strategy was one of the biggest proven losers of the pandemic. Not that I have better ideas; there's a good 30% of people now who just won't take their advice seriously any more.
I think they would have to acknowledge the prior manipulation and relentlessly state bare facts and very highly restrained uncontroversial suggestions, and likely the top level media would need to stay in line here, and let the spin be done at a much lower level. Alternatively they'd have to be very clever about using messaging that was cynical and anti-establishment flavored to somehow percolate up through the people who won't take it from the top-down anymore.
People like me are just never going to believe them again because I realized with all those months of transparently obvious half-truths calculated to induce behaviors that they weren't actually talking to me. Or really to any people like us, the ACX commentariat broadly. Maybe I was supposed to realize this years ago, but I really only learned it from reading Zvi's writing about them during the pandemic. Even their routine recommendations are all dramatically overstated, because they assume (not entirely without reason) that if you need people to wash their hands several times a day you have to tell people that unless they wash their hands every 5 minutes a hungry tiger will sneak up behind them and bite their hands off. If you're a smart person and a statement they make looks implausible on its face, it is not a fact-statement and you aren't meant to take it at face value. You're supposed to infer what they *really* mean and what is *really* required from what they told the rubes, and then politely not say anything to interfere with the messaging, and the whole thing is tiresome and annoying.
> Line formation at convenience stores still seems like people are standing several feet apart when it would've been typical to stand right behind the guy ahead of you up before 2020.
Interesting, I don't make this observation at all. Queues have gone back to normal years ago here, pretty shortly after the last measures were lifted. But the plexiglas protection of the cashiers has stayed in many places.
I'm in the midwest United States and concede the possibility this is not everyone's experience. There isn't very high neuroticism here, so I don't necessarily think it revealed a preference for huge personal bubbles, but probably anything coded as politeness is a little stickier and with lower population density there's little to make it snap back.
Not a majority by any means, but a lot of public-facing, mostly young service workers where I live continue to wear masks. I don’t know if they notice less illness in themselves that way, or if it gives a sense of privacy or relief from having to smile, or they just got used to it and it feels weird without. I just don’t think it can be virtue signaling.
I never got to liking the moist, recirculating-your-own-breath feeling, and will leave the city in the next pandemic if the authorities return to that playbook.
I couldn't really stand that either, we had a mandate in the courtroom so I had to deal with it a few times a week, and I got away with wearing a face shield the entire time. People looked at me like I was crazy, but of course it's much easier to wear for hours at a time, and I'm sipping liquid constantly all day so I'd have been pulling the mask down 30 times an hour anyhow.
I see very little of that now, when I do it's a co-worker who thinks they may be sick and is doing it as a courtesy. I fortunately have always had large offices to myself away from everybody, so I never worry about it.
Here mask-wearing has gone back to very low levels. Not quite as low as before Covid, but in the order of 1-5% of people in public transport, and practically zero in waiters or service workers.
My partner and I are among the mask-wearers in public transport, and my partner swears that it has reduced the number of infections notably for him. I am not so convinced, but I don't really care whether I wear a mask or not, and I don't see downsides to wearing it. It's not signalling, it's the same category as washing my hands when I have used the bathroom.
It was maybe hard to predict in advance just how big a deal the anti-vaxers were going to be. (You could have predicted there was going to be *some* push back based on prior experience, but maybe not the extent of it)
I think in any future pandemic, planning has to be: anti-vaxers are just going to die
(There is something of a moral hazard for the government here, where people who probably aren’t going to vote for the government next time are begging to be allowed to do something that might results in their death)
He's suggesting that the pandemic is going to kill them. This won't be entirely true; no pandemic is 100% contagious and 100% lethal, but at least some of them will die.
My prediction is that if we have a major pandemic in the next 10 or so years, we are utterly screwed. Between covid, the Great Awokening, BLM, and RFK, US public health authorities have burned up basically all their credibility with approximately everyone, it will be politically impossible to impose any kind of restrictions (masking rules, lockdowns, etc.) anywhere even if this time it would make sense. The way vaccines were politicized (plenty of blame to go around there) will make it hard to get support for allowing fast development of new vaccines, and of course we'll keep our "you can't have it until we approve it, at which point we'll ram it up your a--" approach to public health, which will maximize pushback.
I just made a different comment above touching on this, but I think they invited this problem by making a series of calculated exaggerations to induce behavior throughout 2020 and by 2021 had allowed the whole thing to become a culture war flashpoint. In a similar later scenario, the PH establishment needs to suck it up and validate the fears and concerns of those people, and frame vaccinations in a way that acknowledges that and provides options that they can distinguish. It should've been easy to sell the J&J shot to skeptical people, one and done shot on an older platform, instead they weirdly marginalized it, had SNL joking about it like it was the Walmart of vaccines compared to the mRNAs, and then yanked it for 2 weeks in a scare that it never recovered from.
Your framing here would be the worst thing for them to say, it's what smug blue state liberals said last time and which turned out not to be that true, the disease was bad or trivial depending on your age demographic and the vaccination prevented some deaths but the craziest facebook group of 40-something anti-vax paranoids probably had zero people die. "You're going to die" is a message that would inspire "yeah yeah, heard this last time, didn't hardly make a difference, I don't know anyone who died". If it were a more serious pandemic with a less discriminating kill profile, you would see different responses to it, and targeting the message appeal entirely to midwit supremacist "I follow the science and I'm better than those chuds so I do what the CDC tells me to" attitudes is what *would* get people killed. If the government ran that playbook again, that's how I'll know they care more about their social signaling than about lives.
There is a question here of whether (a) the public (b) the government has learned the wrong lesson if, next time, it’s deadlier.
So, e.g., the government might conclude that convincing vaccine skeptics is entirely hopeless, so shrug, whatever. (While also getting in place the infrastructure for processing large number of dead bodies in short space of time)
I don't think vaccine skepticism in the broad sense is a particularly sticky belief outside of its original cultural home in middle-aged California crunchy hippie women or Kentucky homeschoolers. In its details, covid-era vaccine skepticism was almost entirely about mRNA particularly, I saw a lot of it because I was an NPI skeptic and there was a lot of crossover in those circles. The government PH establishment treated J&J more harshly than any kooky commentator did. I think this was largely a consequence of bad framing and bad marketing and really weird approval timelines. I could be wrong, but in a later pandemic with broader fatality I think if they approached it openly and honestly we would find there was much less across-the-board opposition than expected, in the intervening years there will have been little salience to the issue and thus it will not have been integrated into people's personal identity.
I had a former co-worker who was a smart enough guy to hold multiple patents (though also enough of a populist to support Mike Gravel and later Bernie Sanders), and I heard he was unwilling to get an mRNA vaccine. I asked if he would get a J&J instead, and said no, that method was also too novel. When Novavax finally came out with an old-fashioned tech vaccine for the US I asked if he would get that, and still no since it had been long enough he decided he had natural immunity. The government did J&J dirty (I got it anyway on top of a prior Moderna vaccine in hopes of Swiss Cheese defense), but I don't think that would satisfy the anti-vaccine people.
I was in the "waiting for Novavax" cohort myself and followed that saga quite closely, there were lots of other people commenting on all the news related to it on twitter who seemed to be in the same boat as me. Due to it being slow-rolled for some reason, I ended up having to get a J&J instead (that took quite a bit of calling around to even find!) It is certainly possible that by 2022 opinions had gotten stronger, and by the time Novavax was available in the US the pandemic fatigue was compelling most people to move on and they'd either had a shot or decided not to. But had it been available fall 2021, and pushed cleverly, things might've played out differently.
They easily could've spotted the mRNA skepticism in certain circles early on, leaned into the jokes and pushed J&J through those channels as a "look at those idiots getting two shots of some experimental thing that'll make you grow fins, I'm much smarter, one-n-done". It's a little screwy, but their approach had poor epistemic hygiene practically from day one, if getting vaccination numbers up permitted some loose messaging then may as well do it on all ends.
1. Obviously made no sense; there was no reason why big box stores would spread less COVID.
2. Is good, but that's because a large portion of schools are net negative even without COVID. Scott's already written about how kids can miss a lot of school and be fine, and school simply being in or out of session is associated with a change in 12-14% suicide rates among children. (Also, even though kids are unlikely to get serious complications, they can spread it, and teachers are not immune).
3+4. Probably made sense, these weren't very costly.
5. I'm not sure how much these actually got people to get vaccines. The history of antivax, which predates vaccines, indicates that trying to force people might just backfire by making them suspicious.
6. Bad on its face, although I've heard that manufacturing constraints meant it might not have mattered much, and rushing approval even more might have made people even more suspicious.
7. Seems quite obviously amazing.
8. This still seems fine. Even among young people, complications from COVID were multiple OOM more likely than complications from vaccines, plus they can still spread it.
9. Mail-in voting is fine so there's nothing wrong here.
10. I'm not sure what this refers to.
11. This was disgusting and heads should have rolled at the FDA.
12. This is kind of broad to evaluate all at once, but generally bad. At the very least, there's no reason for "emergency powers" to last so long--once the legislature has time to meet, they should pass actual laws, and then after the emergency they should review which powers actually made sense and clarify/change the law.
> Mail-in voting is fine so there's nothing wrong here
What about the obvious vulnerabilities?
1. You can steal ballots from mailboxes in areas where you think people will vote for the other guy
2. You can destroy mailed ballots in areas where you think people will vote for the other guy
3. You can intimidate people into voting while you watch, whether those are your own family members or other people that you have power over
4. You can of course just vote on someone else's behalf if you can manage to crank out a forged signature that will survive half a second's untrained scrutiny
5. The chain of custody of ballots is incredibly weak compared to in-person voting
Mail-in ballots have a legitimate use case for people who are truly immobile or live in truly remote places, but they're far too insecure to be a default way of voting.
We have been protected so far by the fact that risk reward makes this not worth it. You have to block a large number of ballots before you can shift the outcome and an operation at that scale becomes detectable.
In 2018 NC there was in fact a mail in ballot fraud scheme that threw the result for the congressional seat. It was discovered and the vote wasnt certified.
So all of these are vulnerabilities, but in practice they probably dont matter. But thats only true if other parts of the system remain viligant.
Wouldn't it make a lot of sense, then, to organize a lot of voter fraud for both sides, ensuring that the vote never gets certified? If you're already in power, it's a lot easier to justify staying in power under extreme circumstances.
Who would want to do that? If a vote doesn't get certified, the obvious plan is to re-run the election. That would take time, and presumably the current officeholder would hold that office longer. That means the challenger party would have little incentive to delay things, and the incumbent party would have great incentive, which in turns means that everyone seeing massive amounts of voter fraud would pretty readily assume the incumbent was doing it (cf. nation states with strongly suspected election fraud). So the incumbent party wouldn't want to do it, either.
> which in turns means that everyone seeing massive amounts of voter fraud would pretty readily assume the incumbent was doing it
If the votes were going to the challengers, would they be smart enough to see through the trick? Partisanship is at an all time high, after all. Either way, it doesn't matter. If the vote becomes impossible to certify, there's nothing people can do except kill the people responsible, which they don't have the spine to do.
The obvious vulnerability I'm worried about is someone saying "give me your ballot form and envelope, signed but otherwise blank, and I'll give you $50". This hasn't happened yet, but I could easily imagine it being done at scale in a future election - possibly to influence the outcome, possibly just to delegitimize it.
Is there any evidence that these are things happen with any significant frequency? Mail-in voting has been widespread in multiple states for a number of years. As far as I'm aware the actual rate of fraud of any kind is extremely low.
I don't think that, even in theory, these vulnerabilities actually represent a large opportunity. For example, your first suggestion. Looking through people's mailboxes is already a crime, and you would have to put lots of time and effort in to do this repeatedly and get ballots in between people putting them out and mail collection, which means a pretty good chance of getting caught. And in many places (e.g. apartment buildings, or some HOAs) mailboxes are locked and/or not publicly available. Then the actual usefulness of doing so is limited by the actual party split + voting rate in the area. And at least in my state, you get an email from the state when your ballot is received, so people would know that something had happened to their ballot.
I suspect that if this were worthwhile, we would observe a much higher rate of e.g. looking through people's mail for personal information or mailed checks, which are probably much more valuable.
I agree that mail fraud won't manifest as rogue actors breaking into a bunch of mailboxes.
The main threat vector I see is in the postal service itself. Mail carriers are obviously authorized to retrieve all that mail, and funnel it to the local office. Security on that leg is low. A single motivated carrier could sort ballots into a separate bag as they go (they're clearly marked), deliver the non-ballot mail, and quietly dump the ballot bag contents into a shredder at home. Thousands of ballots could be trashed this way.
One foil for this is if that mail carrier works a route with voters that mostly align with him. So, left-wing carrier working a right-wing route. This might be rare-ish, although I wouldn't be surprised if it mostly goes that way rather than right-carrier working left-route. If there's a way to acquire blank ballots, a carrier could trash some or all of what he gets and replace them with ballots for his candidate.
A few thousand votes like this might swing a precinct, but probably not a state.
Another worker at the office who's responsible for receiving the mail and sorting it could also foil this, but they'd have to watch the sorting, notice very few ballots, have reason to believe there would _be_ a lot of ballots, have a way to check (door-to-door survey of that route would be tedious)... and not be motivated the same way as that carrier.
A separate vector would be on the other side of that step - whoever collects ballots at the post office could either trash them without passing them on, or trash most of them and deliver the rest to allay suspicion.
Any conscientious voter in that precinct would be largely unable to detect anything amiss, so long as returns report some votes for R, some for D.
I don't know how mail gets handled at typical post offices; perhaps there's another mechanism that would foil this. But it's hard for me to imagine something that also allows voters to mail in ballots, or not, at their own whim.
I think a bunch of states have a website or text message where you can verify that your mail-in ballot was received, which seems like a pretty obviously good idea.
Well like I mentioned, in my state, Colorado, I get an email when the state receives my ballot. Some people might not notice, but if you're trashing thousands of ballots, I think it would be pretty clear. It would probably also be clear on the vote counters' side, if some areas had nearly 0 ballots.
This whole scheme sounds entirely speculative, highly risky, with unclear impact (how strongly are postal routes sorted by politics? how many people actually vote by mail?). Colorado has had all-mail elections since 2013, Washington since 2011, and Oregon since freaking *1998* (https://en.wikipedia.org/wiki/Postal_voting_in_the_United_States#Table:_No-excuse_postal_voting). If this sort of thing were feasible, I think there would be some evidence of it by now.
Email receipt helps, yes. Even if most people don't notice, a handful noticing would likely trigger an inspection that turned up the rest.
So I think that would keep a rogue mail carrier from just trashing ballots, unless I'm missing something else. Maybe not from duplicating said ballots, unless the email also echoes the voter's choices, in which case the carrier would need a way to spoof that as well (the "problem" here is that most carriers probably don't have the means to do all this versus just trashing the ballots).
I noted the potential problem with political sorting above. To be somewhat more detailed: a route is geographical, and a given carrier is presumably familiar with it, and probably notices which way people probably vote just through context clues (lots of flags hanging; "In This House, We Believe" signs on the lawn; etc.). So I'm picturing a left-wing carrier covering a suburb in a red county, a right-wing carrier covering a highrise, etc.
Either way, you'd need probably tens of thousands of compromised ballots to even have a chance of turning a state. OTOH, none of said carriers has to collude, so there's no conspiracy to leak. The obstacle there is probably the email receipts.
Fraud further on the chain is admittedly speculative, as I said; I don't know how ballots are handled further up. One hopes they're monitored by a bipartisan team beyond the carrier leg. "How many people actually vote by mail?" is somewhat answered by your own comment (all of CO, e.g.). Whoever runs those legs might have the means to generate emails (are they encrypted? Signed?). And while it's speculative for me, someone with a great deal of incentive - like, say, someone who believe he or she is saving the nation - would presumably put in the research.
I know I'm in the minority on this, but I don't think most of these were failures. That is, in hindsight we now know that COVID would mutate to be a lot less deadly, and that MRNA vaccines would be developed in record time and end up being quite effective. But hindsight is always 20/20. At the time, officials did not know any of this; what they *did* know was that a new and deadly disease was spreading like wildfire, hospitals were being overwhelmed, people were dying, and transmission was primarily achieved through airborne particles.
I think that the major failures of the COVID policy were these (off the top of my head):
1). Failure to secure enough quality masks and emergency respirators. This should've been done by an emergency production order. Instead, we had national officials posting YouTube videos on how to make a mask from a bandana and some paper towels, and people 3D-printing respirator parts in their basements.
2). Speaking of which, inconsistent messaging. It's fine to admit when new data coming in invalidates your previous recommendations; it's not fine to flip-flop around randomly based on the whims of public opinion.
3). Inconsistent lockdowns. The initial round of 2-week lockdowns was uniformly enforced (at least in my area) and arguably very effective. After that, lockdowns became just another form of political messaging and essentially random.
4). Failure to prosecute prominent public officials who blatantly violated COVID rules (I'm looking at you, Newsom).
FWIW I have friends whose relatives died from COVID, and others who lost their sense of taste and smell permanently (fortunately my own relatives survived unharmed), so I'm somewhat unsympathetic to claims that COVID was like, just a flu, bro. I realize that anecdotes are not data, but still, the issue is more personal for me than it is for your average Internet pundit.
Rather than mutating to be less deadly, I think it mostly mutated to be more infectious while our immune system evolved (through vaccines and prior infection) to make it less deadly.
When the new data says "the vaccines don't stop the spread" and so you recommend masking post vaccination, you've kind of shot the vaccines in the foot, as vaccines and not treatments.
Well, as far as I understand it, the general model of COVID (or any other viral infection is as follows):
1). Viral particles leave the infected person's body (the person may not know he's infected), usually attached to water droplets from mouth or nose.
2). They travel through intervening space, usually by air, or by touch, to the healthy person's body.
3). They enter the body by mouth or nose.
4). The viruses then attempt to infect host cells; the immune system attempts to fight them.
5). Some of the viruses succeed and go on to reproduce exponentially; some fail and are destroyed, but the process creates side effects. There are several outcomes based on these factors:
6a). Viruses overwhelm the host, resulting in severe long-term consequences or death.
6b). The immune system goes into overdrive, resulting in severe short-term consequences or death.
6c). The immune system fights off the virus, person returns to health.
All of these events are probabilistic, so if you want to reduce the number of deaths, you can look at reducing the probability of each step happening.
* Masks help with transmission, primarily (1) and (2) and to a lesser extent (3).
* Lockdowns significantly reduce (2).
* Social distancing mitigates (2) to some extent.
* Vaccines reduce (6a) in the short term, although they may increase (6b). In the long term, they act as selection pressure on the viruses, making them less deadly and thus increasing (6c) compared to the other options, as well as preventing (6a) in even unvaccinated people.
There's nothing anyone can do to permanently "stop the spread", and I really wish the messaging wasn't positioned that way. What you can do is reduce the number of hospitalizations and deaths, and you can achieve this by some combination of masks, lockdowns, social distancing, and vaccines. It would be really helpful to know ahead of time which combination of methods is optimal at preventing the spread of this specific virus, but no one has a crystal ball, so shotgunning all the techniques at once and seeing what works is the best practical approach (when facing a deadly and virulent disease).
Countries operating border controls to prevent the virus from becoming endemic were able to "stop the spread" and vaccinate their populations rather than waiting for natural immunity.
I'm hazy on one point, though: "In the long term, [vaccines] act as selection pressure on the viruses, making them less deadly". How does that happen? My impression was that normal evolution, in the absence of a vaccine, tends to make viruses less deadly, since dead people are less likely to spread them. In fact, a priori I'd have expected vaccines to counteract that somewhat, since if the virus is in practice less likely to be deadly because it encounters only vaccinated people, there's less pressure for it to evolve to a form that would be less deadly in the absence of vaccines.
Am I misunderstanding your point or just missing some part of the argument?
From what I understand, vaccines tend to attack the most aggressive forms of the virus first, since they are the ones with the proteins that are most different from those of the host. But I could be wrong about this.
That doesn't really make sense. It has to depend on what makes the virus good at infecting new hosts, since that's what evolution works on. If becoming more deadly helps it infect more hosts, then it will evolve to become more deadly; if becoming less deadly works better, then that's what it will probably evolve toward.
Vaccines create an immune response to particular proteins--in the case of covid, the spike protein that sticks out of the virus and needs to bind to the right receptor on a cell to get inside and start making lots of copies of itself. One part of that response is antibodies, selectively sticky molecules that will stick really well to the spike protein but not to most other things. A virus with a very different spike protein than the one you have antibodies for won't have antibodies stick to it and so can infect you more easily. (The other part is cellular, and it notices some different parts of the spike protein than the parts the antibodies bind to, which is why the vaccines give you some protection from serious illness even when they no longer keep you from catching it in the first place.)
Huh, maybe. I guess I’ve never thought much about what makes a virus more or less deadly.
I think of the tradeoff for the virus as being deadliness vs contagiousness. We certainly saw Covid get more contagious as it got less deadly — but of course masking and social distancing would also select for increased contagiousness.
(I heard a comic wonder if the virus broke an ankle if it tried to jump seven feet.)
I tend to agree with you - 10/11 are straightforward failures and you could make a case for 6.
People often have this interpretation of the shutdowns as straightforwardly bad due to economic impact but I've seen at least some analysis of the fact that most of that impact would have happened anyway because people were scared and mostly would have changed their behavior voluntarily anyway.
Also, South Korea showed how much better it was possible to do. My takeaway from which is that the MAIN issue with the US response was forcing it into a partisan lens instead of working to do better versions of these things.
A different perspective: the main problem was failing to see legitimate concerns as legitimate. When you have substantial groups of physicists at Los Alamos looking at the data and saying "this is risky business..." they're not exactly going to be happy if you dismiss them as kooks.
What physicists at Los Alamos? I know Greg Cochran is a New Mexico physicist who wrote a lot about Covid, and won every public bet he made on it, but he lives in Albuquerque.
Is he in the cult? If so, you could probably just ask him. I don't know names, as I'm not in the cult (just know someone who is.)
The Supreme Court decided that religious objections to vaccination would apply if one had a "religion of one," hence the promulgation of a cult in order to have a valid religion to express one's issues with the lack of bodily autonomy inherent in vaccine mandates (and other issues with this particular vaccine).
I don't know what "cult" you're referring to, but Greg is and was pro-vaccine. He considers pathogens to be the great common enemy of humanity, and anti-vaccine people like RFK to be on the side of that enemy.
I would be interested to hear if you think the legitimacy of these complaints differed in the US vs. South Korea. In particular they had a much lower spike in unemployment while also having vastly fewer deaths per capita, even with slower vaccine access.
When I say "partisanship" here I am *partially* indicting things I see as conspiratorial/reflexive rejection of safety measures - as I think you are taking this - but also I'm just saying "the fact that this turned into a culture war topic instead of trying to address concerns and find a good solution", so in that sense we are somewhat on the same page.
Ah! we're discussing different things. I'm discussing legitimate concerns with the vaccine, including "endless production of spike proteins, which are harmful to health." (This should have been a foreseeable problem, and wasn't looked for within the "safety studies.")
My biggest frustration is not seeing any global, meta-research on the topic.
It was such a perfect natural experiment because we had each country going with slightly different reactions and policy all to deal with the same threat. Surely by now there are some obvious findings and learnings that can be gained from the pandemic?!
Contraindicated by nearly 100 years of pandemic research, mind you. Everyone jumped to be stupid at once (save Sweden, who literally couldn't have a shutdown, due to the high number of single mothers).
Strongly agree here. This is the perfect opportunity to see what worked, what didn’t, and what we should consider next time.
Also, it revealed that the current state of knowledge of infectious disease was/is seriously lacking. We should fund a global initiative to better understand how disease really spreads and how to combat it.
The current state of knowledge has been backsliding, as you can see looking at wikipedia, which used to have a primer explaining why covid19 was similar to the last disease that had its vaccine withdrawn (in indonesia).
OK Gemini says this has been done and gave an intreresting answer with 3 examples.
Can't share my query but here's the prompt I used:
"My biggest frustration is not seeing any global, meta-research about the Covid pandemic.
It was such a perfect natural experiment because we had each country going with slightly different reactions and policy all to deal with the same threat. Surely by now there are some obvious findings and learnings that can be gained from the pandemic?"
3 studies it cites:
"The "Big Three" Global Reports
If you want the meta-view, these three bodies of work are the "gold standard" for synthesized data:
1. The Lancet COVID-19 Commission (Final Report): This is perhaps the most comprehensive. It categorized the global response as a "massive global failure" due to a lack of international coordination. It highlights that the most successful countries weren't necessarily the richest, but those with the highest levels of "social trust" and previous experience with respiratory outbreaks (like MERS/SARS).
2. The Royal Society NPI Review: Published late in the cycle, this meta-analysis looked specifically at Non-Pharmaceutical Interventions (NPIs). Their conclusion: NPIs (masks, distancing, lockdowns) were highly effective when used in combination, but their individual impact varied wildly. For example, they found that "lockdown" as a single term is too broad; specific measures like banning mass gatherings were far more "bang-for-buck" than outdoor park closures.
3. The OECD "Policy Responses" Series: The OECD tracked policies across 38+ countries for years. Their meta-research shows a clear "scissors effect": countries that prioritized early, aggressive suppression (the "Zero-COVID" initial approach) actually had better economic outcomes in the long run than countries that tried to "balance" the economy by staying open longer and then being forced into chaotic, reactive closures.
The CEO of Microsoft’s AI division recently said that “Most tasks that involve “sitting down at a computer” will be fully automated by AI within the next year or 18 months, he said, naming accounting, legal, marketing, and even project management as vulnerable.” This is from Fortune.
Do people not believe him? Markets particularly. After all most stocks should have collapsed, as this can do nothing but create a recession, but in particular Microsoft which makes its money from office software, productivity suites, and hardware to run it - which are sold per head, should have been devastated.
What reason is there to believe him? Did he make any good predictions in the past, or is he just selling hype like some cryptobro? Microsoft's track record of anticipating large technological shifts is less than stellar. Slept on the internet and web, lost the mobile war, bought skype and promptly lost to zoom, finally surrendered in the browser war, etc.
Ca 2000, Microsoft was actually hated. Today, they are pleasantly irrelevant. I use Windows on my gaming computer and I am happy to report that their OS offering is adequate for that use case. I don't have to deal with them on my mobile or at work.
If the standard assumptions about AI are true, we're looking at human extinction within several years, and market outcomes don't matter in those worlds. Edge cases have survival and superabundance, where market outcomes also don't matter. Same with civilizational collapse cases. The only outcomes that matter here are the weird ones, where we get a global AI Pause treaty, or a Butlerian Jihad, or AI progress just hits an unexpected wall somehow, in the very limited time we have left.
Will AI rapidly overtake most of the "sitting down at a computer" tasks? Probably. Should anyone make long-term plans or investments based on that? No, because it's much less likely to happen in the weird scenarios that actually matter.
He may be right that these things will be *capable* of automation, but getting actual adoption has to overcome some substantial hurdles that can't be overcome in 12-18 months. For one thing, many companies have their legacy data working with some niche piece of industry-specific software that's been around for 30 years and isn't easily integrated with AI agents. For another thing, the insurance companies aren't going to want their big clients using this unless it's entirely locally hosted, and will presumably make their clients impose that in all of their vendor contracts with other companies that handle any of that data. So you would have to commit to a secure on-site hosted AI, which costs some money. Then migrating your data over to whatever new software IS compliant with the AI. Then you'll have retraining, weeks of designing new workflow charts, and several months of working out the unanticipated problems created by new processes. I'm not a tech installer, I could be overstating a little of this, but as a management problem I think the workflow and training issues are as substantial as that and maybe worse.
You also have to hit at least the performance reliability of a human, not merely top end capability. You can get effectively 100% reliability from a human employee on a simple high importance task, and some items require that, so no switching those tasks until the AI is consistently giving the same output to the same inputs on that type of task as routinely as your calculator would.
If 18 months is when the capabilities would be good enough to consider this, I'd expect another 2-3 years after that at least before widespread adoption. You could have Skynet exterminating humans with drone swarms outside while First Traditions Bank of Omaha still has human tellers logging into software with a 1986 copyright date.
I believe him. Most of those tasks are automated already, which is why everything works so poorly. In about 18 months, it will get worse, perhaps much worse -- until the market corrects itself. This will take much longer, but eventually we'd have affordable human lawyers, project managers, etc. again.
I think people don't believe him because they assess their own jobs as more complex than can be easily replaced by non-AGI models, and don't expect AGI within 18 months.
I'll give an example from my career: My job is Customer Success Manager for a data platform. I build relationships with clients onboarding their teams and convincing them to renew their annual subscriptions.
It's certainly feasible that AI can replace a lot of what I do, soon. But you'd need the AI to be charming on Zoom calls as well as make commercial decisions and real time pricing negotiations.
It's not that my job is super sophisticated. It's just the real world is harder to replicate in the way that code is.
Btw, if my clients were AI systems too then all my charm and human face advantage will be useless.
Is this true? I thought Microsoft was a mix of Office with Azure cloud computing, and the cloud computing stands to benefit from AI. They also own 27% of OpenAI, which stands to benefit if AI fully automates white collar work. I believe something like the CEO's statement (maybe not 12-18 months) and I still hold some MSFT for these reasons (people here should to tell me if I'm being dumb).
Why would markets tank? Why would Microsoft think this is a problem?
Like, I get why *workers* would be hurt by automation by why do you think stocks would fall? How would the guys who own the robots be hurt by replacing workers with robots?
Like, Microsoft sells Microsoft Office super duper enterprise edition to Exxon @ $500/seat (1). Microsoft calls up Exxon and says "Hey, you know all those accounting people that cost you $120k/year including benefits? What if replaced 90% of them with AI agents at $24k/year."
Exxon saves $96k/year per accountant replaced. Exxon stock go up.
Microsoft spends $4k/year to run the agent and so makes (24-4.5) an additional $19.5k per agent Exxon uses. Microsoft stock go up.
Accountant gets fired. Accountant needs to find a new job. Market doesn't move because the accountant is not a stock on the stock market.
If what the Microsoft guy says is true, and Microsoft is selling the heckers out of B2B AI, then stonks go up, not down, because stonks represent the people who own the capital/robots/AI, which are more valuable.
1) Accounting is riddled with PII data, and firms with any digital security teams lock that down massively from use with AI tools, impeding it's utility.
2) Similarly many processes are tied to multi year contracts which will take longer than 18 months to unwind.
3) Having done my share of outsourcing in accounting, a lot of those $120k p.a. jobs were outsourced to India already for 30-35k USD p.a. (including the BPO firm markup, at least that was the rate 6 years ago). The remaining western headcount is largely for fixing all the mistakes those teams make (originally pitched as "you'll have more time to influence strategy/decision making!" but reality is not the pitch). What you're likely to see is a reduction in the error rate as the Indian BPO workers are replaced by AI (which still makes errors), leading to slightly less of a burden on the western teams. That said, my company has been rolling out GenAI tools in accounting (completing basic journals) and it's been even worse than the BPO performance, which was itself worse than when we had the capability locally. I expect in 18 months that pipeline will have been patched with better instructions and underlying models, but it's still not as easy as it would look from the outside (as it almost always is).
Right now my computer security is so locked down at my CPG employer - my main use of GenAI models at work is getting them to write VBA code for me. I lost the ability to install freaking python of all things at some point in the last 12 months, so now I just use Excel as my hack to automate stuff. I'm not at all skeptical on capabilities in the 5 year timeframe, but rollout is another thing.
One of the first things I did when taking on a new accounting leadership role in the last year was blind email someone at Anthropic to see if I could talk to their team about financial automation, sadly their sales contact didn't quite seem to know what to discuss with me. If anyone who works there wants to talk I'd happily have a conversation.
> Accountant gets fired. Accountant needs to find a new job. Market doesn't move because the accountant is not a stock on the stock market.
This is the problem here though. If you fire all the accountants, or indeed 90% of all white-collar workers, then you've suddenly got a huge demand problem on all companies. Exxon has cut its costs by 5% but finds that demand for its product is down 30% because unemployment is 35% and nobody can afford to go anywhere.
In the long term things reach some kind of equilibrium, but in the short term everything is screwed.
Oil and gas can be burned for energy, which powers more data centers. They're the ones who would have one of the least amount of issues transitioning to an AI-centric economy.
"Exxon saves $96k/year per accountant replaced. Exxon stock go up."
Exxon then get hit with yuuuuuge tax fines because AI accountants feffed up ambiguous new regulations around reporting/paying particular taxes which, naughty naughty, they did not report/pay since AI accountants said don't need paying/reporting but Revenue say nuh-uh, you gotta report/pay! Stock go down?
It seems like you assume that AI accountants will, on average, be worse at this than human ones.
My expectation is that you replace 80% of human accountants with AI accountants that handle 95% of the work and escalate cases with ambiguity to the remaining human accountants.
I'd expect those AI accountants to make fewer mistakes on the procedural/predictable activities and successfully identify more cases of ambiguity.
I think that what you describe is the model many people imagine for AI in all sorts of realms. It does the tedious busy work and leave the hard cases for expert humans.
But I worry that to become a good expert human, you *need* to work on those tedious cases to build up your knowledge. If nobody is doing that, then nobody will become an expert.
Similar in concept to how they say the software engineer pipeline is now broken because companies are replacing entry level positions with AI.
The problem is that right now running any sort of AI is deeply unprofitable so Microsoft would be losing money on every single agent deployed and trying to make it up with volume. On the other hand normal enterprise software has some of the best margins in business. It costs literally nothing to create an extra copy of Microsoft office.
Microsoft has a real moat and millions of users locked into its enterprise environment. If AI is really going to change all this then Microsoft is as likely to "win" as Anthropic, or Google, or OpenAI, or anyone else. And depending on how you see future AI development the winner just makes everyone else go out of business(making all of these companies inherently risky) or they all start competing in massive subsidy wars where they try to gain market share by selling agent for less than they cost to run, which also likely means that the losers will burn cash for no benefit.
Here are my scattered thoughts on what'd happen if what the Microsoft AI CEO said is true, some of this is tangential to your points about the stock market:
1. There are ~165M US workers. Maybe ~40-50% of US workers are in white collar jobs or jobs that involve mostly working on a computer. I'm uncertain of this exact number. These people are laid off.
2. White collar and computer jobs pay significantly more than non-computer jobs and also contribute more to the federal and state budgets via payroll taxes
3. The laid off are forced to take lower paying non-computer jobs or subsist on welfare, mostly from government. They substantially reduce their consumption and they default on things like mortgage payments.
4. US Government revenue is greatly reduced due to fewer and poorer employees paying less income tax. ~54% of federal revenue is from individual income tax, another 30% is from payroll taxes.
5. US Gov must find a way to a make up the much increased deficit as more people qualify for benefits, but much less revenue is coming in. The Gov could cut benefits.
With much less consumer demand and adding more deficits to already large red budget, what would happen to the stock market? Who is going to buy the products that many companies offer?
> Who is going to buy the products that many companies offer?
Other AIs. They're still producing things that are useful, that's why they would be being run in the first place. Energy instead of food. Silicon instead of medication.
Just a side issue, but during the discussions a few years ago about working from home/return to the office the usual term to describe sitting-at-a-computer workers was “knowledge worker.” Today the usual term to describe AI-displaced workers is “white collar.”
Another correction I thought you would make about the Sam Kriss article was his attribution to you AI 2027’s predictions: the article makes your timelines seem shorter than, as far as I understand it, they are.
I used to spend, or should I say waste, quite a bit of time playing PC games, until one day I realised that I was just chasing pixels round a screen, reacting to program code, and this was no different in principal to my cat chasing the dot of a laser pointer round the floor!
Is even the smartest AI any different in principle to a PC game? Clearly not, because both are no more than programs and data. That is why I think all this talk of AI consciousness is, to put it bluntly, a load of hogwash. It's no more conscious than a monster in a first person shooter game, despite the appearance of consciousness in both.
You are your brain. Reacting to sensoric neurons getting stimulated by what you hastily call reality. You have no clue if the world outside your brain exists, or even if your brain runs on the substrate you think it runs on.
So your substrate chauvinism is uncalled for, unless you happen to demonstrate abilities which computers lack, such as solving the Halting Problem.
Saying that an AI is "no more than programs and data" feels like saying a human is "no more than meat and chemicals." It turns out you can do a lot with seemingly simple components.
To be fair, I play games as well, and I know full well they aren't real. Still, I also watch movies and read books and play board games with my friends, and none of those activities involve producing anything of lasting value either. I think you can make a convincing argument that entertainment of any kind is inherently wasteful and should be discouraged (I certainly cannot disprove this logically), but I think most people would disagree with you. Then again, most people aren't very smart...
In retrospect, the first paragraph of my post was superfluous and only muddied the waters. The only point I intended to make was that the apparent consciousness of AI with a finite amount of code and data seemed to me to be illusory, being no more real than a monster confronting one with apparent ill intent in a first-person shooter. But, from some of the replies, I can see the problem is not as cut and dried as I would like to think!
Oh don't get me wrong, I do agree with you that LLMs are no more conscious than perhaps a bee (which is still a major leap forward in consciousness, compared to calculators !). But I disagree with you that a "finite amount of code" could never in principle achieve consciousness. My head contains a finite amount of neurons too. Just because calculators can't do it, ELIZA couldn't do it, and LLMs still can't do it, doesn't mean that nothing ever could -- and it does look like we're very slowly inching closer to that distant future day.
I personally think "consciousness" is a direct result of "self and other simulation" capabilities, which largely come into being in mammals with the neocortex.
When we got warm-bloodedness, we set ourselves up for the much more computationally complex abilities that we enjoy today. Birds, notably, are the only nonmammal species alive today to independently evolve warm-bloodedness, and birds are reservoirs of similarly complex behavior as mammals.
But as we step higher up the “mammal” chain in terms of neural complexity, we get more than just simulating various possible outcomes in the world, which is what most brains have been about up til this point, we additionally get self and other simulation, including (ultimately) being self aware and projecting our own selves into the future.
There are three broad abilities that seem to have emerged in early primates:
* Theory of mind: inferring intent and knowledge of others
* Imitation learning: acquiring novel skills through observation
* Anticipating future needs: taking an action now to satisfy a want in the future, even though I do not want it now
These may not, in fact, have been separate abilities but rather emergent properties of a single new breakthrough: the construction of a generative model of one’s own mind, a trick that can be called “mentalizing.”
One interesting implication of the “self and other simulation” capabilities that the mammalian neocortex enables - consciousness itself is likely a flavor of self-simulation.
It seems pretty obvious to me that once you get to “self simulation” and a big enough brain to understand your own persistence and future states, consciousness becomes A) possible, and B) evolutionarily adaptive, because it gives you salience and valence for your future states, and enables “marshmallow test” style decisions where you take a short term penalty now for more rewards later, which is clearly net adaptive.
And indeed, if we look into the literature, this is exactly what we see!
Interestingly, rats cannot anticipate future states and monkeys-and-higher can. Rat brains are about 2g and even small monkeys are 20-30g - what do they get for their 10x bigger brains? They get the ability to self-simulate into the future and pass a marshmallow test. Rats fail at this, and squirrel monkeys pass (Naqshbandi and Roberts 2006).
Being able to simulate your own future states is necessarily the first step towards consciousness-as-self-simulation being valuable in the sense of evolutionarily adaptive.
But I think the real key here is, what else would be required to make that simulation of your own future states adaptive? You would need to feel connected to that future state, you would need to care.
What do consciousness and qualia do? They allow you to predict those future states, they make you feel connected to that future state, and they make you care. All qualia is, is valence. You care about future-you’s experiences because you’re going to have to experience them! You’re going to have to live through that qualia.
After all, once you have a little “you” that lives in your head at ALL, that you simulate doing different things, that persists in time and reaps the future consequences and benefits of your decisions, how far a jump is it for that little you to live in your head full time, and to represent that it’s experiencing the various things you end up doing?
If you stumbled across this internal innovation, if you suddenly had qualia instead of being a p-zombie, you’d be a strictly better savannah ape, because you’re going to take your future-impacting decisions more seriously and make better ones.
So maybe we should briefly touch on qualia, because it seems so core to the hard problem.
The fundamental mystery of mechanical processes ending up as qualia, or being experienced as something, has never seemed like much of a mystery to me.
Like, what ELSE could produce “experiences?” Doesn’t everything need some sort of substrate to even exist / happen? Our substrate just happens to be a physical universe with matter and energy, and sensory and processing organs made of meat.
For anyone out there that are pet people, it seems to me that dogs and cats have as much evidence for having qualia as humans. They have moods, they clearly choose to experience various things for more or less the sheer joy of experiencing them, like rubbing their faces in snow, or jumping or playing for the joy of it.
And you can take it all the way back - smaller mammals, birds, and reptiles certainly experience things. Bacteria too. The thing is that “experiencing” is being used in two senses here.
I think qualia is just something like consciousness or some adjacent mental process internally meta-tagging “experiences” (in the sense that bacteria, mammals, and reptiles have and react to) with valence, thereby creating an internal qualia “experience” (a second sense of the word experience).
It seems to me to be such a simple evolutionary line from “non-sentient bacteria and animals having an experience of something (sense 1) because you need feedback loops for successful goal-directed behavior” to “humans, descended from a billion-year line of animals / bacteria, also have experiences (sense 1),” and when coupled with their interior future projections of themselves, and the obvious evolutionary advantages of caring about those futures, they have qualia too, and so have valenced interior experiences (sense 2), qualia.
If you couple valence with future-state simulation, you reach the level of future-state navigation that humans enjoy, which is a significant enough jump that we dominate the planet.
Qualia gives us the sign and vector magnitude of rewards / punishments we’re signing up for with our actions, so we care enough about future-you to make good-marshmallow-decisions today - obviously adaptive on net. So adaptive even squirrel monkeys do it!
There's no explanation as to why the first sense should lead to the second sense, or indeed how anything like qualia can exist in the universe. Qualia is not merely valence, "good" or "bad". It's the raw experience of seeing a color. The experience of hearing a sound. These experiences can be valanced, but that's not their primary characteristic.
There is no reason these evolutionarily advantageous computations you describe should have an inner experiencer. Why could they not happen in the dark, so to speak?
If you don't see the mystery, then.. idk, you do have qualia, right?
> There is no reason these evolutionarily advantageous computations you describe should have an inner experiencer. Why could they not happen in the dark, so to speak?
I've described the reason, but I was probably not clear or to the point enough.
Basically once you can simulate others, you can simulate yourself. And once you can simulate yourself, you will make better decisions IFF you have an inner experiencer and directly care about the outcomes because you know you will experience them, and can imagine yourself experiencing them. That's qualia.
Your p zombie can imagine themself doing something, but if it's just like imagining somebody else doing something, they care less. They are less motivated to make harder, marshmallow test decisions, and would make less of them. If you have qualia and valence and can really imagine that outcome in a personal experience way, you are more likely to make better decisions.
That's just wrong, RL agents even today are perfectly capable of making optimal decisions like that, and in fact don't struggle at all with making marshmallow-optimal decisions. Delayed gratification is not a problem for a p-zombie RL agent. You can weight future decisions exactly as much or little as you want. It's not a real thing that RL agents don't "care enough", but only if they had qualia would care more.
Second, this doesn't explain what qualia *is*. Where does it come from? You say self modeling, but that doesn't explain what it is and how qualia is different from current RL agents' self modeling. Or do you think they are actually conscious? That would really be the intellectually defencible position according to your reasoning.
> That's just wrong, RL agents even today are perfectly capable of making optimal decisions like that
That's a total non-sequitor. Sure, we can after-the-fact program something to make "good" decisions, but we're talking about nature making an entirely new capability implemented in meat, with the starting point of bacteria and reptiles.
It's a completely different path and order of difficulty.
Saying "once we're here and conscious and know about marshmallow tests and persistence in time, we can artificially create a system that does the same thing using that knowledge" isn't actually relevant to the path of creating new organisms with de novo capabilities that end up with better self simulation capabilities and making better decisions over time.
> Second, this doesn't explain what qualia *is*. Where does it come from?
I don't *care* what qualia is, that's just going to get into definitional word games. Where does the neocortex come from? Where does the ability to simulate others come from? Where does the ability to simulate the self come from? This isn't a well defined question.
The original question was "what do you think make humans conscious?"
And I've answered that. It was the ability to self-simulate getting better, coupled with caring about outcomes via qualia, that led to being conscious and making better decisions being evolutionarily adaptive, so it happened.
Briefly, consciousness depends on the structure of the brain: how it relates to the both the rest of the body and the world outside of the body: and how different parts communicate with and rely on each other to integrate sensory data into a coherent whole. It is fundamentally a biological phenomenon. Below articles/reviews of work
by 2 neuro-scientists who have proposed different theories of how this works: Anil Seth and Anthony Damasio.
On a second thought, I'm not sure why he thinks robots don't have something "homeostatic and regulatory", how does he think drones operate, or active cruise control, or any of myriad machines with control loops for that matter...
At least Damasio doesn't claim that consciousness is fundamentally biological, here's a quote:
"For robots to be conscious, we would need to give them a bit of vulnerability that they don’t have [right now]. If you could introduce into a computer something that would be homeostatic and regulatory — that would allow it to sense deviations — then you would be on the way to creating feelings for the computer, and it could detect its own inner states."
This would make, for example, a smart thermostat... conscious? At least a little? Sitting there, being aware of getting cold, and operating to restore the equilibrium?
Saying consciousness is fundamentally biological, is fundamentally ridiculous. Is it the carbon molecules that are fundamentally necessary for consciousness? If not, what *precisely* do you mean by biological, and by what mechanism is that prevented from being emulated?
I'll be honest, I haven't read the links, but those characterizations are typical of people who don't take the hard problem seriously. Or more likely, don't get the hard problem.
Yep I've been reading the articles (not the papers, which are presumably way beyond me and also paywalled).
Unfortunately my smug dismissal was on point. Damasio doesn't seem to get the hard problem *at all,* while Seth seems to specifically "bet" that solving all the easy problems will dissolve the hard problem. Which of course, is exactly the point Chalmers was making: he called it "the hard problem" exactly because he was trying to get at the idea that it *won't* dissolve, like other scientific questions have when looking at them close enough. Hence, the hard problem. It's hard. And neither man seems to get the question, much less to begin getting at an answer.
(To be clear, working on the easy problems is certainly valuable and I would not hesitate to fund either man's science projects. It's just mostly unrelated to the hard problem of consciousness, which Seth acknowledges.)
But then I'm not sure why you dismiss a possibility that consciousness is fundamentally biological? BTW biology is far more insanely complex than just "carbon molecules", just as a car is far more complex than "Fe atoms". This is indeed "The Hard Problem", and we can't ever reliably say which organisms are conscious, never mind assigning it to machines. So I wouldn't dismiss the possibility that indeed there's something special about biology, at least until we start getting evidence to the contrary (silicon-based life has been proposed by sci-fi writers, for example).
You're one of the people who thinks a programmer sits down and writes each of an LLM's possible responses. That's not how it works. They don't pick responses from a list.
Is the Good Reptiles snake food supposed to be vegan or something? From what they're currently advertising, it looks like they make essentially chicken nuggets, but from mice. Is the animal welfare angle just that they're killed more humanely instead of being swallowed alive?
The nuggets aren't made out of mice, but I'm not sure how much I'm allowed to reveal so I won't mention what they're made of other than that it seemed to me that they had a reasonable case for it being more humane.
This might be a stupid question, but I thought the whole point of feeder mice was that snakes could only eat live prey, and would refuse to eat e.g. plain old chunks of meat ? Is this not accurate ?
It's common to feed snakes frozen/thawed mice. Live prey involves risks and is more complicated, so it's not preferred by most professional breeders. There might be some species that require it, I'm not sure.
Caveat: I only have experience with typical pet snakes (i.e., nothing huge and nothing venomous).
Refusing live prey varies from individual snake to individual snake, even within a species. Most discussions I've read treat this as a learned behavior - if snake is raised on live food from a young age, they're more likely to refuse non-live food as an adult.
I've fed my girl thawed frozen prey since she was a few months old, and in my experience a feeding response is typically triggered by heat and scent (easy to get chomped on the hand if it smells like mouse from not washing after handling, and has a warm internal temp, despite the hand not looking like a mouse at all). Sometimes I'll wiggle the mouse around and that seems to be convincing enough when she's being picky most of the time.
I don't see why properly thawed 'chunk food' with the right scent couldn't work for some snakes.
Yeah, I thought that too; there are some reptiles you can just thaw out your frozen mouse carcass and feed it to them, but others will only eat live prey?
Is it lab grown mouse cells? There have been a few forays into lab-grown pet food, e.g. Wild Earth. EDIT: though IIRC, they weren't doing animal cells, but rather something else.
One possible reason why overall crime stats could be falling even as everyone subjectively feels that crime is rising could relate to the distribution of where crime is happening in major cities.
For example, I've been to Detroit and I really like it; I felt very safe. Buddy's Pizza is great, the Tiger's baseball stadium is great, highly recommended. Yet Detroit, overall, is extremely not safe. It only felt safe to me because I'm a boogie bitch and went to the downtown/tourist area, where everything is safe because that's where all the external and internal tourists go. You can look at a crime map of Detroit and amongst a sea of red there's a little yellow/white blotch by downtown. If you just go there, Detroit is great.
Seattle inverts this. Seattle is an ocean of yellow/white safety with an island of bright red crime centered right on Pike's Place. And you can feel this if you walk around downtown Seattle; when I was there I saw crackheads everywhere. It felt way, way less safe than Detroit, even though it was objectively safer. Why it's like this, I have no idea, but for some reason Seattle has all its crime next to Pikes Place and the Seahawks stadium, where everyone can see it.
I kind of see this replicated comparing San Francisco and St Louis. When I look at a crime map of San Francisco it's not literally in the downtown but there's a huge hub of crime, like, riiiight next to it (https://gisgeography.com/san-francisco-crime-map/) and everyone I know in the Bay Area has stopped going into SF because of a combination of crime and traffic. Meanwhile, in St Louis, which has been in the running for murder capital of the US and parts of it literally look like a 3rd world warzone, the downtown area near the Arch is reasonably nice, cool area, you can walk a couple miles up towards the hospital and park, it's not great but it's cool and safe (https://gisgeography.com/wp-content/uploads/2023/07/St-Louis-Crime-Map-1-1583x2048.jpg).
So the issue isn't whether overall crime is rising or falling, it's that in certain cities it's moving towards way more visible parts of town. Ignore tourists. If you live anywhere in the Bay Area, from Fremont to Contra Costa, you're not going to see a fall in crime in Vallejo, even if that's where all the crime actually happens, because you never go to Vallejo, because why would you unless you need a priest and a truckbed of malt liquor? You see the crime in your neighborhood and the crime in SF because SF is where you go to see a basketball game or go to an art museum or to do any big event. So the guy in Fremont notices that crime is his neighborhood has gotten a bit better and crime in downtown SF and on the BART has gotten significantly worse and thinks that overall crime has gotten worse, while the guy in Antioch and Half Moon Bay are seeing the same thing, even if all of their individual neighborhood crime problems have gotten better.
I'm not super-convinced on this answer, mostly because I don't really follow crime maps or have much familiarity with the data, but I've been to all these places and it feels weird and then I go pull up the crime maps and, like, damn. I'd appreciate it if someone with more of a background in this stuff could confirm if this is a real effect; if certain urban areas have flipped from "keep all the criminal crackheads AWAY from the downtown/attractions" to "keep all the criminal crackheads NEXT to the downtown/attractions".
Also, if true, why? I'd love to blame hippy BLM liberals but, uh, the municipal government of Detroit is hardly a bastion of based conservatism.
Hm, is it generally true (statistically) that people who have moved to a new neighborhood feel less safe there than in their old one?
Inspecting my own feelings, this rings true. I certainly didn't perceive my childhood neighborhood as scary, even though it objectively was one of the worst parts in town. But to me it was jut normal. Also as adult, I tend to ignore a lot of stuff that I am used to, even though it might upset me if it was a new experience.
Boogie bitch, or disco debutante? Sorry, couldn't resist...
(I'm...fairly sure crime is a bit up in Half Moon Bay too, if the reports from my family still living there are any indication. Cratering local economy will do that. It's hard to transition successfully from fishing-and-agriculture to tourism-and-retirement without a lot of the lower echelons falling through the cracks.)
I saw some data once showing that in opinion surveys, historically people have pretty much always thought that crime is getting worse. in their time. So my default theory for why perceptions of crime are bad is that they always are.
It seems like what's happening with all these theories is that someone will come up with a reason why crime is *really, actually* up; and any time someone comes up with a way to test it statistically, it ends up being wrong; leading the person to adjust the theory in a way that ends up being less testable.
Effectively all the theories about crime/disorder are trending towards coming up with something that can't be quantitatively tested, because all the ones that are, end up being false.
Because crime is down, and people just *always* think crime is up. Anyway, that's my theory.
I think some of the "criminal crackheads next to downtown/attractions" is because tourists and others are there, so more opportunity for begging/pickpocketing/selling crappy souvenirs. Can't hustle drug money off people if there's no people around, now can you?
Good observation, the main pattern here is different populations. In the cities where crime is concentrated and downtown, often this is where fewer people actually live, this is because the broad majority of the population is relatively high function and wouldnt tolerate the issues near their homes, wheras near tourists/businesses, there are fewer compliants, another way to think about it is that the majority of the population has high demands for order. Wheras in cities where dysfunction is widespread, the cops can move anti-social people along into the residential neighborhoods and they will simply melt more into the broader population. The big question for SF, Seattle, Portland vs Baltimore/STL is where can we push these people?
> The big question for SF, Seattle, Portland vs Baltimore/STL is where can we push these people?
I've done the math on this, and if you just gave them all the free drugs, alcohol, food, and water that they wanted somewhere out of the way, away from all the productive people, it pays for itself 4-20x over.
Any one of the increased values of downtown usable real estate, lower crime, or lower police load in cities pays for it.
I personally recommend BLM land in the middle of the Nevada desert, like Burning Man, because then it's federal and the federal level can pre-empt the gauntlet of multiple local levels you'd have to run through.
You 100% described my hometown of Milwaukee and my existing residence of Honolulu. "I see homeless people everywhere downtown and increasing, clutch my pearls, crime is rising!!". Also an angle you missed is gentrification, i.e. ignoring early movers whom generally expect the existing crime rate as part of the deal, it moves middle movers into unexpected contact/exposure with "bad neighborhoods" especially around the edges in areas before they would have avoided by driving completely around them or bypass on the freeway. They were promised by marketing they were moving into the new urban reclaimed suburban-like Mecca, not an mixed ghetto inner ring suburb that only the late movers get to avoid post successful gentrification.
Worth noting (for others at least) - Honolulu is an astonishingly-safe major city. It's the 13th most populous in the US and has the lowest murder rate of any in the top 50 (actually the first city with a lower murder rate is Plano, Texas, the 70th most populous).
Yep and yet crime routinely rates the number one concern here and by far. I've actually never in my life seen people so scared of crime until I lived here which I find amusing as it's easily the safest place I have ever lived, in fact so safe that it's the only place where I have never thought about crime outside passing thoughts like now in writing this, a couple times a year.
People earnestly think here if you go outside after dark you will be raped and murdered by roaming packs of crazy tweaked out homeless people, a pedophile is around every corner, every stranger a robber, and that every trip outside your house is putting your life at risk. People are genuinely terrified of crime here.
I was going to agree with you, using Baltimore as a personal example – but then I looked up the stats for murder, and it's more than halved over the last three years: https://en.wikipedia.org/wiki/Crime_in_Baltimore. That's a precipitous enough drop that I'm coming round to the view that there might be something to the broader trend of dramatically improving safety after all.
My theory above is an explanation for why crime would feel like it's increasing even if it's objectively decreasing. Baltimore would be a bad example of this because it's downtown still looks pretty safe while the rest of the city is pretty red. (https://gisgeography.com/baltimore-crime-map/)
If overall crime really is decreasing notably but it decreases in the 90% of the city that nobody visits and increases in the 10% that everyone visits, overall crime can decrease while perceived/observed crime increases. And in certain west coast cities, the crime maps I see show the crime concentrated in high visibility areas.
That would explain why overall crime could decrease (which I like 70% agree with) while people see and feel that crime is increasing (which I like 95% agree with).
No, you communicated it fine. I agree with your point completely.
It's just that my impression until just now was the stats were technically real, but something like a small decrease in most places slightly outweighing a sharp increase in the touristy heretofore-safe places – but I'm now seeing the raw numbers in a place I'm kinda familiar with (I lived in, worked at, and traveled to only the safe bits) are indeed remarkable.
He may have meant "pest" which while uncouth, would have made more sense here I think on my feel for what he wrote. I'm guessing an autocorrect typo, not human furniture.
It's used twice though, and while I don't recognize OP's username, it's not uncommon among the twitter right to describe certain populations as "pets" that liberal-progressives have disproportionate affection for and defend despite (or sometimes because of) potential negative attributes.
I had considered that case initially with its use in conjunction with immigrants but I discounted it because I've never met anyone who hates the homeless more than the left and so I wrote that usage off.
That's some real character assassination. Dude's going on my list with Cade Metz.
...
As for the anime figures and the vibrator, he's definitely trying to paint the guy as (the modern version) of an incel--objectifying women and lacking any success with them, combining the new-style insult of being a fuckboy (effectively, cad) with the old-style insult of being a virgin.
The guy seems more than anything else like a humanities nerd angry the STEM nerds are getting all the money despite not being any cooler.
"“Young men usually is who I like to go out and talk to. Women get intimidated and, you know, I don’t want any charges.”
Probably says a lot that Kriss's audience is supposed to think "yeah, he's creepy and probably a rapist" and I thought "yeah, it is like that now. Good luck, Roy."
The polycule/group home is supposed to sound horrifying but to me just sounds cute and heartwarmingly nerdy. If I'd grown up there I'd probably be happily married (possibly to multiple people, each of them to multiple people) with kids instead of writing lengthy comments on ACX.
The sperm racing is supposed to be the ultimate tech-bro pastime: it combines masculinity (evil), technology (now evil), and capitalism (evil). As far as I'm concerned, Eric Zhu is an inspiration and makes me feel bad about all the time I wasted.
The Donald Boat thing is genuinely funny, though everyone here probably heard of him. It is notable he preserved Boat's anonymity when the NYT was less kind to our esteemed blog host. Obviously has to do with the politics of the people in question.
Then there's more of the these-guys-are-sexists-and-losers-at-the-same-time digs at their lack of romantic success.
....
It's official. If Kriss collapses in front of me, I'm not doing CPR. But I think I was more annoyed about the article about the Villages. And his snark about not answering questions until you meditate in the forest for a thousand years. And his conflation of politics and morality. And his...[REDACTED]
I guess the anonymity is mostly performative at this point; he's a public figure now. But you're right, he didn't use the extra name (which seems to be a signifier of which side you're on).
I guess the guy just rubs me the wrong way for some reason.
at the risk of opening up some kind of controversy that will one day be known as Crackergate, my strong recollection is that the crackers comment was made once the children had gone to bed. i found it very charming, which is why it’s in the piece
Sam, my understanding is you eschew online feedback lest it poison the pure luminous art flowing through your conduit to the divine, but I'm very bad at listening to direction, so I just wanted to let you know I thought the piece was excellent. One of your best ever (second maybe to the essay you wrote about the Villages in Florida)
FWIW, I also interpreted your retelling of the situation as "likeable, thoughtful weird guy does mildly odd and endearing thing", which seems to be the way it was intended.
Who was I saying it to? Myself? I know I can come across as pretty weird, but this just feels really outside of the set of things I can imagine myself doing. Are you sure it wasn't the child who lived in that house and who stayed around after my kids went to sleep?
It doesn't read to me as "look at how weird this guy is", to me it reads as a continuation of the main thrust of your 'interview' of "Scott Alexander is a bit idiosyncratic but a fundamentally good natured and kind guy", it comes across as even your idiosyncrasies are charming and well intentioned. Especially when contrasted to the other people in the article, I sent the article to several friends with comments about how it reinforces how nice you are!
yes i think yourself, it was loud enough to be audible but not loud enough to be directed at any other person. i'm sure i've said stranger things in any given evening
What is this "journalistic accuracy"? Journalists can basically write whatever they want, and if it's not accurate then who cares. Maybe in some rare cases a journalist's publication, sometime later when nobody is looking, quietly lists the inaccuracy as an errata. Journalism isn't some kind of oath-sworn government body accountable to Congress, it's a genre of entertainment like rap music or horror movies. Nobody reads Harpers to comply with code.
Many people care, including most other journalists, in the specific case of a claim that is provably, explicitly false as stated. Which is why journalists so often use verbiage that doesn't amount to an explicit statement of fact regarding the subject at hand, "X is true!" can end your journalistic career as fast as Dan Rather's, if X isn't true. "Bob told me X is true", is much safer, and if you pick a vaguely-credible Bob it will still have most of the audience believing that X is true.
Hot take: this actually presents you as an ideal and positive foil to Roy, because his high agency terminates in "a great sucking void where the end ought to be", but you clearly articulate the utilitarian virtue in the basic act of allowing people to eat crackers.
On the other hand, I now catch myself muttering to no one in particular things I narrate to my children (2 and 4 years old). One just gets in the habit!
The crackers comment seems to have been almost universally positively received, if you were worried. It's incredibly funny/sweet so I'm just going to assume it was true as written.
That article was really tame and subdued for a Sam Kriss piece. Where's the vituperation, the grandiloquent hot takes, the creaky makeshift towers of metaphor and illusion in their varied multiplicity?
I also wonder at his contempt for "agency." Yeah, it might not make you happy, any more than money will make people happy, but just as obviously it matters, and will matter a lot more in a future where AI can do everything.
He seems to accept this, yet has contempt for it? Because it isn't teleological, an end in itself, or because the people demonstrating it don't have strong ends, or because it's rewarded by the market, or I don't know why.
But "the market" is just a synecdoche for "reality" here. The agentic are already doing more with AI than most people ever do without it, and having correspondingly more impact on the world, regardless of whether the market is rewarding it.
I introduced my dad to AI. Since then, he's written another book, made a religious puzzle game for the app store, made several websites, made a ton of videos and images - he probably spends 5-6 hours a day on it. It's an amazing productivity multiplier, and he's doing 10x or 100x more than he would have without it. I'm sure other retired guys farted into their recliners and yelled at clouds and internet commenters for those 5-6 hours.
>Because it isn't teleological, an end in itself, or because the people demonstrating it don't have strong ends, or because it's rewarded by the market, or I don't know why.
Because ..It's all about achieving g Something,not about what the Something is...? At least that's what makes me uncomfortable. For instance, if Cluely actually gives you an edge, everyone will start using it , and giving the same AI generated answers, so the human element becomes redundant.
> For instance, if Cluely actually gives you an edge, everyone will start using it , and giving the same AI generated answers, so the human element becomes redundant.
Okay, but if everyone started making Phd smart decisions in multiple domains of life, how are they and everyone around them not massively better off?
That's a huge amount of value created, both for the individuals, and in positive externalities for everyone else.
I thought the discussion of agency was really fascinating, particularly the tension between Roy as being extremely agentic but working on a product where you outsource your agency (which is pretty undeniably accurate even if you think Roy is portrayed unfairly as being agentic but with no end goal, although unless the quotes from him are extremely inaccurate it's hard to feel like this is very unfair).
But also I think it's worth noting Kriss readily admits to his own failure to show agency (in response to Roy insulting him about his hair loss, which he dislikes but does nothing about). I don't think there's only one thing to take away from this; agency can both be valuable and good and also not be the only thing worth valuing.
> I thought the discussion of agency was really fascinating, particularly the tension between Roy as being extremely agentic but working on a product where you outsource your agency
I agree this was interesting, but didn't see any contradiction at all here.
1) There's a long tradition of building billion dollar businesses on removing friction and making life easier - Amazon patenting one click and relentlessly simplifying the shopping and checkout experience, Google returning better results with a simpler interface, all the delivery apps and car summoning apps, and so on.
2) He's just giving the people what they want - the vast majority of people don't WANT to make decisions. Decisions are friction! Moreover, this is actually the better move for two reasons for most people - both in the sense it removes that friction, and in the sense that it will be a better quality decision. If you allow people an option to have a Phd smart mind making that decison, it's actually a better, smarter move for the vast majority of people even *without* the benefit of removing the pain and friction!
3) Our future is obviously going to be one where our Phd smart AI assistants live in our ears / phones and we interact with them for hours per day. It's the next trillion dollar attention surface that can be auctioned, and it's why Google and Facebook are all-in on AI, because they don't want their current trillion dollar lunch eaten. He's skating to where that puck is going to be, to use a Gretzky analogy.
> your view of the future reminds me heavily of The Commonwealth Saga
Man, we can dream, right? If we end up in anything even adjacent to or resembling it, I'll consider that maximal success for our current arc.
Because obv we're probably not going to have anything like the Silfen or Raiel or Living Dream, nor anything like instant portal tech. But even Commonwealth-ish minus those would be a gigantic win!
> I agree this was interesting, but didn't see any contradiction at all here. [...]
I feel like you just listed off three reasons this is a good business opportunity that would apply in any context, but do not speak to whether it is a contradiction.
My point is that somebody whose main admirable trait is how agentic they are is an ironic person to supply you with a tool to be less agentic, like when a pro athlete does a McDonald's ad or something. I'm not saying it's a bad move to shill for McDonald's - they make a tasty box of fries and they pay well - but it's still ironic if you're famous entirely for physical fitness.
Yeah, I guess we're speaking past each other in terms of what counts as a contradiction.
My point was, this is just literal "business as usual," which has always been fueled by highly agentic entrepreneurs making stuff for people to be less agentic / lead less friction-ful lives.
You're looking at it from a personal traits perspective. The highly agentic guy is deploying his agency to allow regular slobs to be less agentic!
And I'm like yeah, but that's a tale as old as time. Lots of businesses have followed that dynamic, it's a well-worn path to success. People hate agency, friction, and making decisions, so giving them what they want is a great plan for a business to succeed.
Unfortunately, I think what your dad is doing is the next two years of the future (and it will be a fun two years!), and after that the AI makes puzzles and writes books without you, and churns out a thousand an hour, and it's unclear what useful things humans can do with their time and intellect. I think Sam was sort of trying to get at that, though if I dared to critique his writing I would say he was mixing it up with so many other points that it lost a little of its oomph.
I agree that I couldn't quite figure out where he was going with the agency stuff. Roy Lee isn't very sympathetic, but the critique seemed to mix elements of "He doesn't want to do hard work" and "All he cares about is working hard" in a way which was probably faithful to the real situation but also leaves me confused about what conclusion to draw.
What I took from it was that Lee was talking about agency and yet the product he was working on and selling was to strip agency from people, it was "this will tell you what to say and do". Lee's ultimate aim was also unclear: did he want to make lots of money? and then what? and it seems (though this could just be how Kriss wrote it up) that Lee hadn't really any idea what he wanted to do with all this, he's someone who is "agentic" but it's not really clear what the point of all this 'doing' is for. To be successful? to be popular? to prove to his parents that he's not a failure for not getting into Harvard first try? just to have something to fill up the time he would otherwise have to spend in his own head with his own mind?
EDIT The relevant quote:
"According to Roy, he has three great aims in life: “To hang out with friends, to do something meaningful, and to go on lots of dates.”
But does Roy have friends, and what does he consider meaningful? "Being a more successful entrepreneur" seems to be the height of that, and the description is depressing:
"For Roy, meanwhile, dating actually seemed to be a means to an end. “All the culture here is downstream of my belief that human beings are driven by biological desires. We have a pull-up bar and we go to the gym and we talk about dating, because nothing motivates people more than getting laid.” He was interested in physical beauty too, but only because “the better you look, the better you are as an entrepreneur. It’s all connected and beauty is everything. A lot of ugly men are just losers. The point of looking good is that society will reward you for that.” What about other kinds of beauty? Music, for instance? Roy had played the cello as a child. Did he still listen to classical music? “It doesn’t get my blood rushing the same way that EDM will.” His preferred genre was hardstyle—frantic thumping remixes of pop songs by the likes of Katy Perry and Taylor Swift. Is that the function of music, to get your blood rushing? “Yeah. I’m not a big fan of music to focus on things. I think it disturbs my flow. The only reason I will listen to music is to get me really hyped up when I’m lifting.” The two possible functions of music were, apparently, focus and hype. Everything for the higher goal of building a successful startup. What about life itself? Would Roy die for Cluely? “I would be happy dying at any age past twenty-five. After that it doesn’t matter, bro. If I live, I have extreme confidence in my ability to make three million dollars a year every year until I die.”
Wanting to "do something meaningful" seems a bit like the classic "wanting to be a writer". The best writers and meaningful-doers are people who have things they want to write or things they want to do. "Doing something meaningful" in the generic sense is a role rather than a goal.
I do feel like we should at least be a little grateful that these people are burning away their life being aimlessly productive for others instead of trying to find "meaning" in their life. Do you want him to discover that his heart's desire is cannibalizing the homeless or something? Better that he's grinding himself away for our sake.
I think his point (one of them, anyway) is that "agency" refers to the drive and the ability to do *something*, anything, just because you can. People like you and me worry about things like "ultimate goals" and long-term plans and what others might think, etc. Agentic people see something that can be done, and do it. In the past, such people had to be born wealthy and powerful, and to be incredibly smart, or else they *couldn't* do anything regardless of how much they wanted to. In this modern era of increasingly powerful technology, almost anyone could become "agentic" in this way, but most people don't have the will, the drive, the kind of personality that allows it.
The world is moved into the future by such "agentic" people; it is sustained in the present by quotidian peasants like you and (arguably) me.
Yeah, that makes sense. I guess I missed this angle because I think Silicon Valley has a sort of Nietzschean take on business where it's fun and meaningful in and of itself because it's a difficult competition. I don't think anyone would say Alysa Liu's life is meaningless just because she's devoted it all to figure skating which isn't of itself a socially productive pursuit, and a lot of B2B SAAS startup founders are motivated by being the Alysa Liu of making money.
I dunno, if he had 5% more social awareness when talking to journalists, he could have framed it as "make friends, build a successful business, and find love", and then nobody would have objected.
Liu is putting in hard work to be a champion figure skater. The description (and Kriss may be too hard on what is going on) is that Lee's product is faulty and a scam, since its selling point is "use this to do your job for you" (so people are never learning or producing anything themselves) and even "use this to trick women into wasting time on you" (tell her she's a great artist example) and it keeps crashing and not working correctly even in the company headquarters.
So Liu is doing harm, if harm is done, to nobody but herself by putting her body through the rigours of training, plus she has to provide good enough performances to win points from judges: a tangible performance in the end. The Silicon Valley hypesters are trying to invent convincing arguments about why their bubble product is worth the free money they are trying to score off investors, and the end products are not really providing a service at all.
Yeah, I guess my critique of Roy would have been moral (cheating is wrong), Sam tried to make a psychological critique, and I'm not sure whether to think of the psychological critique as compelling in itself vs. a sublimation of a moral critique which is too hard to deliver because people don't like preachiness.
I took this as both - "this guy has no purpose and refuses to so much as read a book to try to find one" seems pretty scathing from a virtue ethics perspective.
What’s going to be interesting is how difficult it will be to just not consume anything created by AI actively. I and plenty of other people would be willing to boycott AI entertainment, especially in a post scarcity future, and spend our time creating art and consuming art made by other humans. By some measures that art would be inferior art than what an AI will be able to produce, but because creating art for others to enjoy is enjoyable in itself, a community would be better off if they just created their own art and consumed it. The hard part is it it difficult to signal that something was made without AI.
My wife is writing a book from scratch, with no AI assistance. When she submits it for publishing a year from now, she’ll be competing with books that were written with AI. And when she submits its sequel she’ll be competing with books written entirely by AI. By the time the final book in her trilogy comes out, if the books have any fans, they would have been able to feed the first two books into AI and create a thousand different versions of the Finale. More likely the book will be lost in the massive amount of AI entertainment and I will be its only reader.
I might just end up playing a lot of TTRPGs with friends where AI can be clearly excluded from the table, even if it was involved with prep.
Ironically, I'm sort of hopeful that AI will help with this - if everyone has an AI assistant on their browser doing a sort of algorithmic feed on the whole world, you can just tell it not to show you things that are AI produced (although I think most people won't have the willpower to do this even with low transaction costs).
What stops the AI lying to you that sure, boss, no AI content on this feed! I imagine all the companies using AI to generate content will be trying to dodge around people blocking said content, and I don't trust current AI models not to lie/hallucinate by telling people what they want to hear even if it's not so.
If 90% of stuff is AI glop, then the AI will only be able to serve up 10% of non-AI glop, and that might disappoint its human who wants more and more content and variety. So a little white lie that this is not AI glop, what harm does it do when measured against the greater benefit of making your human happy?
It can't get away with lying about things that are easy to verify as false. Then everyone would instantly know about it and the lie would no longer work. It could get away with lying if it's impossible for people verify, but then that means the AI couldn't verify it either, since the people would be using other models to verify it as well. So as long as there isn't some mass collusion among models, they can't really get away with lying.
As I understand it, Sam Kriss believes that AI, at least under the current paradigm, is never going to be the match of great human authors when it comes to writing essays, novels, poetry, and so on. Meanwhile approximately 100% of silicon valley techies and rationalists think that we can be as certain of this happening as we can be certain about anything at all in the future, and moreover, that it is going to happen imminently. In his estimation, the reason for this gulf in perceptions is that techies have the artistic sensitivity of lesser primates and to them Joyce or Woolf might as well be Rebecca Yarros. Hence why he was so delighted by Roy Lee, who I hadn't previously heard of, but who seems to be a flesh-and-blood caricature of the soulless startup bro.
It's notable that even if AI never becomes as good as, say, the top 1% of writers or programmers or whatever, that still entails the evaporation of nearly all jobs for writers and programmers and such, and probably also the evaporation of most of the pathway to becoming a top-tier writer or programmer or whatever.
>Meanwhile approximately 100% of silicon valley techies and rationalists think that we can be as certain of this happening as we can be certain about anything at all in the future, and moreover, that it is going to happen imminently.
I don't think that's the case. I am a silicon valley techie, and think that is quite unlikely to happen in the short to medium term and venture no strong opinion about the long term. I don't get the impression my assessment is uncommon, let alone unique.
Would you say that the vibes around this topic are notably different between places like this one, or the AI-interested corners of X etc. and the general tech-community in Silicon Valley?
Asking mainly because I read blogs such as this but am very far away from "Silicon Valley tech world".
My immediate coworkers have a pretty wide range of views on AI, but I'll attempt my general sense of the median views.
Among rank-and-file engineers, the median view is probably that it's a useful tool for semi-automated mechanical tasks, investigations, and rapid prototyping but produces worse-than-useless slop if it is relied upon too heavily. Incremental improvements are generally expected, but probably not fundamentally transformative ones. I'm less favorable on AI than the median but am not an extreme outlier.
Management, especially senior management, has adopted a party line that is extremely gung ho about how great AI is and that it's rapidly getting better. Middle management seems to be mostly on board with this, but I'm not sure how much of that is sincere and how much is feeling obliged to toe the line.
> Unfortunately, I think what your dad is doing is the next two years of the future (and it will be a fun two years!), and after that the AI makes puzzles and writes books without you, and churns out a thousand an hour, and it's unclear what useful things humans can do with their time and intellect.
I mean, in a post-intelligence future, surely the ones who fare best are those who have endless ideas when it comes to creation? That's the "agency" part, and it helps whether or not you're being paid for your outputs. You're entertained, and you can create things meaningful to your local social milieu, and that gives it meaning to those other people, even if AI could have made a strictly better artifact if they'd thought to ask for it. Just like a home-made gift is more meaningful than the top Amazon item.
That IS the useful / relevant thing that people can still do in that future - create things relevant to themselves and their local milieu. If we're all hunter gatherers who don't need to work all day in the future, we just go back to hunter gatherer stuff. Those people who are fun to be around, who tell good stories, who throw a good party, who have creative ideas on how to use the bounty around them, that's who has status.
> That IS the useful / relevant thing that people can still do in that future - create things relevant to themselves and their local milieu.
If your social milieu is online, or at least recorded and transcribed, then (granted the underlying premise) wouldn't the AI be better at creating things relevant to it than you are ?
> If your social milieu is online, or at least recorded and transcribed, then (granted the underlying premise) wouldn't the AI be better at creating things relevant to it than you are ?
I don't think most people's social milieus are online.
Yes, people spend an ungodly amount of time on screens, and more and more of them have relatively unhealthy parasocial relationships that are acted out online, but this is still mainly gen Z or alpha phenomenon, and the majority of people still have real life friends and social circles.
The people for whom that IS true - the terminally online, as they say, are doomed. They're going to be sniped by the Infinite Jests and AI waifus and boyfriends, which will be to real food as junk and fast food, and real human engagement as Tik Tok and porn are.
Obviously, junk and fast food has been so optimized and become so powerful 80% of people are fat, just like eyeball-hour harvesting has gotten so good that phone screen time has gone from 2 hours a day in 2014 to 4-5 hours today and 7-9 hours for gen Z / alphas.
There's no reason to think that optimization is going to stop, and many reasons to think it can get stronger. In the limit, that'll be superstimuli so strong that it snipes people out of reality entirely. I wrote a post about that.
>In the limit, that'll be superstimuli so strong that it snipes people out of reality entirely.
That sounds pretty nice, but, like so many other futuristic predictions, this one commits the mistake of assuming there are no physical limits to anything, and any trend can be extended indefinitely. In real life, this is usually not the case. For example, cars are getting faster every year (https://en.wikipedia.org/wiki/List_of_production_car_speed_records), but this doesn't mean that they will ever exceed the speed of light.
Sure, except that it's obviously a gradient, and it's *already* sniping people out of real life.
Everything from sex and drinking going down to people dying in gaming cafes after multi-day binges. It's literally already affecting everyone to some level, and the waterline is *already* high enough to snipe some of the most prone among us, and that waterline is only going to rise.
You say "but it's not going to snipe everyone," yeah, completely agree. The amish exist, just for one segment. But it's an open question how much of humanity it CAN snipe, and it's already eating something like 2/3 to 3/4 of all free eyeball-hours.
Aren't ideas correlated with intelligence? So, wouldn't the post-intelligence future mean the AIs have the best ideas too? And can't an AI be charming too, eventually? Look, I'm just a bag of meat and bones. It takes me years to get good at playing the piano. How would I compete with an AI Chris Hemsworth robot (assuming we get realistic robots) who quips, tells engrossing tales, and effortlessly plays the piano, guitar, mandolin, etc.?
Look, I can see this wrt to family units, but have a harder time seeing how this scales beyond that and maybe close friends. Maybe that's enough for people, maybe not.
I think a valuable part of human relationships is the feeling of being 'chosen', and being 'chosen' by charming, likeable people feels especially good. An AI friend or boyfriend couldn't really 'choose' you. Will people still go for this on the margin? Yes, definitely -- it's clear already from what's happened to interpersonal relationships from the Internet and LLMs. Do I think this is a serious concern for most interpersonal relationships? No, I don't think so. I don't have any fear my boyfriend and most of my friends would choose an AI over me.
(this is more a comment on interpersonal relationships and separate from any discussion of making a living in a post-AI future. That's a whole 'nother bottle of worms)
Because you're imagining future AIs are like the current ones, where ChatGPT constantly reminds people it doesn't have feelings or preferences. If AI ever gets smart enough and someone makes an AI that tries to act like a human, you won't even be able to tell which of your new friends are AI.
I agree, and I'm not sure what I'm missing when people worry too much about this.
I'm not very good at playing guitar, anyone could easily load up Spotify and hear better played guitar, but
1) I enjoy playing; and
2) There are a few people close to me who enjoy the music precisely because it is me, a loved one, making it.
Similarly, my 4 year old is terrible at drawing and colouring but I still fondly and proudly accept her handmade gifts to me because they're by her, she enjoys doing it, I get a fuzzy flow from it because I love her! It's fine!
> I don’t see anybody making a living there, though.
Yeah, but I think pretty much everyone expects a UBI.
I've done the math, and we can afford to give people $40k - $90k UBI's with really reasonable assumptions (current or slightly higher corporate tax rates, 50% of white collar jobs counterfeited, etc).
Also, anyone with assets is going to be fine as the economy quadruples, and their passive income will do up.
It's also a huge unlooked-for gift that it's happening to white collar people first, because they're the rich connected ones plugged into the chattering classes and politicians and whatever. Versus if it were service industry jobs / robots? Boots on faces forever. So we actually have a decent chance to get a UBI passed because white collar is happening before robots.
I think only a handful of nerds expect a UBI. Particularly in the case of AI 2027, er, 2029 or whatever, because it would take longer than that to bring the concept of a true UBI within the Overton Window of the sort of not-nerds who would need to sign off on implementing it.
I am also exceedingly skeptical of your math; a $40,000/yr UBI for every adult in the United States would cost ten trillion dollars plus overhead. more than half the current GDP.
The Sam Kriss article seems to be in a style influenced by Hunter S. Thompson’s _Fear and Loathing in Las Vegas_, and similarly mixes fact and fiction.
His description of the dinner with Scott seemed ok to me, though, agree would have been clearer if it were more explicit that Scott was talking about the crackers to a small child (the story does highlight that children were there).
Though there’s a certain amount of comedy to be had in maintaining ambiguity over whether Scott was talking to one of the ACX commentariat or a literally two year old.
I for one think it's amusing that Scott seems as concerned about the crackers thing (which seems more adorable than anything else) as the suggestion that the rationalist movement is idiotic and misguided.
But the suggestion that the rationalist movement is idiotic and misguided is quite arguably true, whereas the crackers thing is (I'll have to take Scott's word for it) made up.
No, Scott says he's pretty sure he wouldn't have said that in the context he's described as saying it in. That's very different. I've interviewed people before and sometimes they're surprised later by what they've said. (Or Kriss failed to capture the surrounding context correctly, which is not the same thing as making up a quote).
I'm a random ACX commentariat dude, and when I read the crackers passage I thought "hah, Scott seems like a really chill guy, and this innocuous moment confirms that". I didn't think it reflected badly on him in any way.
There were, however, a couple of other places in the Sam Kriss piece that I thought would be misleading or confusing to a reader who wasn’t already familiar with the AI risk scene.
Tyler Cowen posed a useful framework after the Maduro extraction: (a) immoral actions were taken, (b) probably unlawful, (c) Venezuelans are better off. His challenge: you can debate how to weigh (a) and (b) against (c), but to deny (c) is delusional.
I sat with this for six weeks & 'consulted' Sun Tzu, Aristotle, Kautilya, & Clausewitz. My essay has drawn sharp pushback. One analyst brought in Charles Tilly's roving vs. stationary banditry distinction & argued this particular extractor is incapable of playing a long game. Others pointed out the US helped create the crisis now used to justify intervention & cited Lindsey O'Rourke's research documenting 64 covert US regime change operations between 1947 and 1989 — making Venezuela less an anomaly than a continuation.
My conclusion is broadly consequentialist but conditional: the thugs have been replaced by the rascals & cosmetically improved authoritarianism is still authoritarianism. The extractors now own the aftermath.
2 questions for this community: is the sovereignty norm already dead & this just made it explicit? And Tyler's utilitarian scorecard of LA interventions counts roughly six cases where leaders were actually removed & finds mostly positive long-term outcomes. But O'Rourke documents 64 operations, Sachs estimates 80. If six successes out of 80 attempts is the real denominator, is Tyler's optimism genuine pattern recognition or survivorship bias?
That's the right question & roughly where the essay covers — 'the thugs have been replaced by the rascals.' Delcy Rodríguez running things isn't freedom. But the case for cautious optimism rests on what's changed at the margins: 400+ political prisoners released, oil sanctions lifted, some economic breathing room. Whether that's the opening of a longer game or just a personnel swap with better PR is exactly what's worth watching.
Fair point. It's the same one Alan made on the Substack comments. The sanctions weren't an act of nature; they were a policy choice that could have been reversed at any time. That's the complicity question the essay underweights. The honest framing is probably: Maduro was a catastrophe independent of sanctions, and sanctions made a catastrophe worse. Both things, simultaneously.
Sovereignty belongs to the people, not to Governments. If the Government is elected by the people then the Government holds sovereignty by proxy, but in a country with no elected government there is no sovereignty, and the "Government" is just a bunch of random thugs who happen to be holding the people hostage, and it should be considered open season on them.
The principle is clean, but the application gets complicated fast. Who decides that a government has lost its claim to sovereignty? If the answer is 'any state powerful enough to act,' then you've effectively given every superpower a standing invitation to intervene wherever it judges a government illegitimate.
Reformed Hegelian's examples below are useful here. Iran has no free elections and massacres protestors — open season? North Korea is probably the worst government on earth — open season? But the moment you say yes, you've also given Russia the language to justify Ukraine ('illegitimate post-Maidan government') and China the language for Taiwan ('breakaway province with no sovereign standing').
The sovereignty norm isn't there to protect bad governments. It's there because the alternative — legitimacy decided by whoever has the most aircraft carriers — tends to produce worse outcomes at scale, even if it occasionally produces a good one.
and the sovereignty norm also just humanity trying to pretend that the "one that makes the decisions is the one with the most aircraft carriers" is not the true rule?
I agree that the strong have always had disproportionate say. North Korea's nuclear programme is the proof — sovereignty is easier to defend when you can threaten Seoul.
But there's a big difference between 'the norm is imperfectly enforced' and 'the norm does nothing.' Imperfect norms still constrain. The invasion of Ukraine has been costly to Russia precisely because the sovereignty norm made it expensive in ways it wouldn't have been say 50 yrs ago — sanctions, isolation, NATO expansion. The norm doesn't prevent violations, but it raises the price. That's not nothing.
I'm a fan of Steven Pinker so definitely appreciate the value in maintaining the "Sovereignty norm". War is bad.
That being said, there are times I worry we've swung too far in the other direction specifically regarding "failed states". Specifically, I think places like Haiti and North Korea would be better off if the democratic West re-invented colonialism specifically for them.
For a more interesting and grounded example: Iran.
On the one hand, evil government massacre-ing protestors based on twisted theology.
On the other hand: an internal-based regime change basically means a long and brutal civil war where even more people die.
Seems like a perfect opportunity for a superpower like the US to say "Sorry you've voided your sovereignty". Don't worry, the new Iran will regain sovereignty once we say so".
I know, I know. This opens the doors for Russia to attack Ukraine and China to attack Taiwan making similar arguments. I'm just not sure this wasn't always the case.
One last point:
I think Putin was responsible for simultaneously harming and strengthening the sovereignty norm. First by invading Ukraine, crossing a big red line (I know, Bush in Iraq too).
And then by losing endless men and equipment in a years long quagmire. Putin reminded the world why invasions aren't worth it.
If anything, Venezeula was proof that a coup can be relatively quick and painless. But not sure this is true for non-Americas.
Oh sorry you had a second question: I think success or failure calculations should heavily depend on how bad the current regime actually is.
Fair point. I guess it proved that kidnapping a foreign leader was easier. Certainly easier than I would have guessed. It also seems like they had a say in Maduro's replacement, though. Instating his #2 instead of the oppositionleader felt like a choice more than simply grabbing Maduro and letting the country sort itself out.
We didn't "install" the #2, that's just what happens when the #1 is gone. Installing the opposition leader would have required doing more than removing the #1.
It's possible you know the details better than me. My understanding was that the actual number 2 is Cabello and the US specifically didn't want her taking over. Instead they chose Rodriguez and indeed there are reports she was actively working with the US as part of the transition. So it really feels like a US decision on who gets to lead here. I'm not exactly happy with their choice here. But this is definitely more than just jeeting Maduro and saying bye. The US had a lot of power post kidnapping.
Neither the US nor any other stronger country has attempted conquering Haiti and taking control (in recent history of course).
This is a norm that even Bush didn't consider breaking for obvious reasons. But it's hard to believe this decision won't improve Haitian lives dramatically.
This is honest in a way most commentary on this isn't. Following the sovereignty argument to "reinvent colonialism for failed states" & then seeing why that's terrifying is the right intellectual move — it's the only way to test whether we actually believe our own principle or just like how it sounds.
Your Iran example is the hardest one. Internal regime change could mean brutal civil war. External regime change means someone else decides their future. Iranians trapped inside get to choose between two kinds of catastrophe. That's the dilemma my essay tries to sit in rather than resolve.
Your final point is the one I keep returning to: Venezuela may prove that a quick extraction works in the Americas, where the US has overwhelming proximity and leverage. That says very little about whether it's replicable in Iran, North Korea, or anywhere Russia and China have countervailing interests. The "success" may be radically context-dependent — which is another way of saying it's not really certain at all.
In the real world, this is something we have a ton of uncertainty about. So I understand why nobody is willing to even try. But the pros vs cons of Iran specifically seem higher than normal.
I just wish the world was more honest about this. I wish politicians would say things like: "We'd love to invade and free you from an evil dictatorship, but the chances of it biting us in the ass are too strong to risk it and we don't trust global opinion to give us any credit for trying".
The compliment was entirely genuine — following the logic to its uncomfortable endpoint and then immediately seeing why it's terrifying is something most commentators skip.
Your hypothetical politician's speech is the essay I wish a diplomat or policy analyst would write. The gap between how intervention decisions are actually made (risk calculus, strategic interest, domestic politics) and how they're publicly justified (freedom, democracy, human rights) is where the cynicism about foreign policy lives. Closing that gap even slightly would be worth more than another think tank report.
Here's a thought: On my Substack (Selective Amnesia) this is the sort of thing I love doing — writing speeches that should exist but don't, for leaders who should say what everyone's thinking but won't. If you're interested in collaborating on that honest politician's intervention speech, I'd be up for it. The world could use more clear thinking delivered in plain language, even if it starts as a thought experiment. DM me if you're game.
Meanwhile, if you enjoy geopolitical satire, this Saturday's essay is set in a Dubai airport lounge where the world's power dynamics play out over duty-free shopping. Different tone, same underlying questions.
According to Grok, O'Rourke includes a long list of operations that I wouldn't classify as removal of leaders (*). Though, after checking with Grok, it seems O'Rourke only considers 10-12 of those operations in Latin America to have been successful.
So Cowen's denominator would be 6 and O'Rourke's would be 10-12.
UPDATE: Grok says this is the list. I object to the inclusion of Bolivia (2) and Guyana, and suspect those might explain most of the remaining gap between Cowen and O'Rourke:
Guatemala (1952–54) — Overthrew Jacobo Árbenz.
British Guiana/Guyana (1961–71) — Undermined Cheddi Jagan, installed Forbes Burnham.
Bolivia (1963–66) — Led to René Barrientos regime.
Brazil (1964) — Supported military coup against João Goulart.
Dominican Republic (1965–68) — Invasion and occupation to install favorable government.
Bolivia (1971) — Supported Hugo Banzer coup.
Chile (1962–73) — Culminated in 1973 coup against Salvador Allende.
Nicaragua (1980–89) — Contra support pressured the Sandinistas (coded as successful in pressuring regime change or significant concessions, though not full overthrow).
(*) A few (you can ask your favorite AI for confirmation): Nicaragua: Early efforts against Sandinistas; Bolivia: Successful (Support for Hugo Banzer coup); Dominican Republic: Dominican Republic (1960–61) — (Assassination of Rafael Trujillo and related efforts); Haiti: (1963) — (Efforts related to François Duvalier or opposition).
Thanks for slicing the O'Rorke data to the narrower Latin American subset — that's useful. The 10-12 successful operations is a higher numerator than Tyler's 6, which is a fair correction. But the denominator stays at 64 total covert operations (or ~80 if you follow Sachs post-Cold War). So we're moving from roughly 6/64 to maybe 12/64 — better, but still not a ratio that should make anyone relaxed about the next one.
The deeper problem is O'Rourke's own finding that targeted states generally became more authoritarian and unstable after intervention. The failures aren't neutral — they made things actively worse. So the expected value calculation isn't just 'low success rate' — it's 'low success rate with negative-sum failures.'
I'm confused. I thought the denominator was number of leader removals. Wasn't the whole point to compare Maduro's "extraction" to similar operations?
UPDATE: I now realize that my phrasing "10-12 of those operations in Latin America to have been successful" might be ambiguous. I meant successful as in: resulted in the leader removal (or regime change, from O'Rourke's point of view) as intended, not in the sense of the local country ended up better off (not a benchmark that O'Rourke applies, I understand).
You're right so let me concede the point cleanly. Tyler's question was specifically: when leaders are removed, how does it turn out? The proper denominator for that question is successful removals, not all 64 operations.
The survivorship bias critique still holds as an ex ante question — if you're deciding whether to attempt an intervention, your base rate should include the operations that failed to achieve removal at all, since you don't know in advance which category you'll land in. But that's a different question from the one Tyler was asking, and I should have been more precise about which question I was challenging.
> your base rate should include the operations that failed to achieve removal at all,
I agree, but only if they have any impact on the welfare of the country. Say, they strengthen a local faction that favors worse policies.
I'm just too lazy to make a better list of removals (and attempted removals) than Cowen's. I would have to go through the whole of O'Rourke's list and spend some time reviewing each operation.
I will only add as a final point that it's not the kind of work that can be delegated to an AI. I happened to be reading "The Economic History of the Caribbean Since the Napoleonic" by Bulmer-Thomas for a post I'm currently writing, and happened to come across this paragraph (page 331) related to the Cheddi Jagan "removal" that I mentioned earlier (Grok classified it as a removal):
"One such overreaction was in Guyana where Cheddi Jagan had returned
to power as premier in 1961 following elections in August. The Guyanese
government still had responsibility only for internal affairs, but the United States was determined that Jagan would not lead the country to independence. The British were therefore persuaded to change the electoral rules so that Jagan could not win. Following the next election in 1964, he was replaced by Forbes Burnham, who led the country to independence in 1966 and presided over the destruction of the economy in the next two decades."
Venezuela's sovereignty has been restored by the Trump Administration. Maduro (The Bus Driver) had Cuban bodyguards, and was giving Cuba essentially free oil, which was counter to the interests of Venezuela as a whole.
Venezuela hasn't changed that much, the person at the top was just replaced by their #2, which is how that person wound up at the top in the first place.
Trump has literally said that his goal in Venezuela was to allow American oil companies to resume drilling there, and also announced that the US had seized 50 million barrels of oil from them. Was that in Venezuela's sovereign interest?
Considering that America is practically the only place that can happily refine Venezuelan oil (notably poor quality), considering that we built our refineries to that spec... Yes.
I mean, we could have gone with the Biden plan of letting Maduro invade Guyana... That's an option too (a very dumb option, but Biden's team was full of neocons, so really, what did you expect?)
Okay, explain to me why Maduro invading Guyana would have been a good thing. Would it have triggered his home-grown collapse instead of needing American intervention? What did Guyana do to piss off Biden?
Biden put sanctions on Russia, and needed more oil (to make American Domestic Producers happy). It wasn't to collapse his regime, it was just to "get some oil" (with a consequence of upsetting Exxon, who already has drilling in Guyana).
Guyana did nothing to piss off Biden, the Biden Regime was full of terrible ideas that needed to be headed off at the pass.
>Harper’s fact-checker asked me if this was true and I emphatically said it wasn’t, so I’m not sure what role Harper’s fact-checking is playing here.
Isn't Sam Kriss well known for mixing fact and fiction in ways that annoy rationalists?
>In the article, I say that I “never once actually made a decision [in my life]”.
I was curious enough about that line that I was planning to ask it as a question in the open thread this week. I suppose it is possible to get through life by being entirely passive, but I have to doubt that is what you are doing. It sounds to me that the whole 'agentic' thing is what we used to call 'taking the initiative'. But that's rather different to never making decisions.
>As written, this makes me sound like a crazy person
When I saw the “never once actually made a decision" line, I immediately thought of Gary Klein's work on decision-making by experts under pressure, where a firefighter he was interviewing said he'd never made a decision in any of his firefighting, he just did what was needed. Is Kriss also mixing fact with fact?
I work in space operations and I understand that. You rarely make decisions in the heat of battle simply because you have gone over everything beforehand so often that you already know how to meet every outcome. If you find yourself need to make decisions in flight then something has gone wrong somewhere in your training.
It's quite possible (probable) to preload decisions. Martial artists do this all the time. If you drive with your hands under tension, you can release the wheel and turn the other way (this improves reaction time, at the cost of significant stress).
Soldiers, policemen, many other fields are told to "rely on their training" -- you drill until the motor-response is near-automatic.
This isn't to say that the firefighter isn't making decisions -- he absolutely is. One of them is to "not flee the area", in that the situation "as a whole" is not life-threatening for himself. That's a high-level decision, that "not deciding the small stuff" frees you to make on a continual basis.
"Isn't Sam Kriss well known for mixing fact and fiction in ways that annoy rationalists?"
My impression of the norms around this is that this is permissible on a blog, in a story which is obviously mostly fiction, but not in an official magazine in a story which is otherwise factual about real people and contains no cues (even breadcrumb style) to its fictionality.
Also, if something is intentionally fiction, usually they don't have professional fact checkers call the subject to make sure they got it right.
I took that "from scratch" bit as hyperbole. A magazine essay like this seems to me to be pretty clearly literary nonfiction, not straight news coverage. It's a legit criticism that authors won't let truth get in the way of a good line. But I expect editors give authors of this sort of piece a lot of leeway about what's opinionated characterization as opposed to statement of fact.
I think Sam Kriss is allowed to describe rationalists according to his own perceptions in his own articles, and it's not the job of the fact-checker to intervene. For my own part, as someone who enjoys SSC but not necessarily the rest of the ratsphere, I think that there certainly is a significant proportion of rationalists who think they're far too clever to have to bother with 'legacy knowledge' and that everything worth knowing can be derived through a mixture of HPMOR exogesis and AGI eschatology.
Abandon all existing modes of knowledge collection in favor of Bayes' theorem, Humean empiricism, the logical tradition stretching through Aristotle through Russell, Kahneman and Tversky's Nobel-winning bias research, the predictive coding tradition stretching from Helmholtz to Friston, the entire mathematical tradition since Euclid, Tetlockian superforecasting, and looking things up in books.
I have no idea if Kriss' account of the meeting is the whole unvarnished truth, but the Donald Boat versus Roy Lee encounter is comedy gold.
"What about literature? The last time Donald had dropped in on his slaves at Cluely, he’d gifted them two Penguin Classics: Chaucer’s Canterbury Tales and Boccaccio’s Decameron. The books were still lying, unread, where he’d left them. He suggested that Roy might find something more valuable than dying for Cluely if he actually tried to read them. Roy disagreed: “I do not obtain value from reading books.” And anyway, he didn’t have the time. He was too busy keeping up with viral trends on TikTok. “You have to make the time,” Donald and I said, practically in unison. “It makes your life better,” I said. “Why don’t you go to Turkey to get a hair transplant?” Roy snapped. “That would make your life better.” “I don’t care about my hair,” I said. “Well,” said Roy, “I don’t care about the Decanterbury Tales.”
Lee may be highly agentic and highly successful, but if he is the vanguard of the Triumph of STEM then yeah, we arts and humanities types should be quaking in our boots right now. Music is for hype. Books are no value. The purpose of being fit and attractive is to be a better entrepreneur. The True, the Beautiful and the Good? Nah bro, what's viral on TikTok?
Let's not conjure up another fake dichotomy between STEM and arts/humanities. I reckon vast majority of STEM-types including myself laugh/pity Roy just as much as you do.
Obviously, STEM is just something you study. You can make it your personality and be a dick about it, but most people with brains grow out of that real soon. Same is true for the supposed other side.
I agree with the comedy gold. I actually liked the article as a whole and I didn't feel it to be a hit piece. It's an outsiders look into an alien world - it would probably seem like that for me as well, coming from good ol' Europe with it's socialism, aristocrats and suffocating bureaucracy.
I saw an Andrew Tate video recently on X, where he asserts that reading is useless, stating that it's a waste of time for a 'man of action'. Sam Bankman-Fried said something similar, ≈ "any book could be better expressed as a six-paragraph blog post". Taken together with this Roy Lee character from the Kriss piece, it's maybe worth pondering the up/downsides of such a stance, and I don't think you can separate analyses of the claims from the fates of those making them. Particularly:
- these people have a purely instrumental view of every mode of human activity. I don't think a statement like "reading is a reward unto itself" is even going to make sense to Tate/Lee/SBF; their metrics deal in tangibles. Upside: not having lofty ideas about art or philosophy floating around your head leaves you more time to study the games of money. Downside, take it too far and you may end up in prison (SBF and probably Tate), or in an article where Sam Kriss has only to write down your deadpan operational utterings for comic effect. We all have to play the aforementioned games, but it's interesting that these vocal proponents end up the way they do.
- There's a big picture, t -> ∞ thing I saw Harari talking about on Youtube*, where (potentially) AI combined with instrumental reasoning cause a bifurcation in consciousness itself. Younger people have an easy time getting answers, but no memory of not being plugged in. Mental passivity (or something) becomes the experiential norm, and maybe that has cascading effects on society which were hard to predict before hand. Imma stick to reading for the time being.
Funny you should mention SBF, becasue Lee's dismissal of books (and the description of the workplace/living space) reminded me very much of him as described in Michael Lewis' "Going Infinite".
Let's hope this is not an omen for the eventual future!
> I saw an Andrew Tate video recently on X, where he asserts that reading is useless, stating that it's a waste of time for a 'man of action'.
I mean...is he wrong ? Just to play Tate's Advocate here: think of all the time you spent reading books, watching movies, playing video games, etc. Now imagine that you spent all that time earning money, lifting weights, writing code, or doing something else that is productive. Wouldn't you be much better off ? Yes, you could say that "reading is a reward unto itself"; but by that logic, doesn't the same apply to heroin ?
Well, sort of. I have a friend whom I've known since middle school, and who is has genius-level intellect. He's always expressed the same sentiment: "Why would I read fiction books about a bunch of made-up stuff that never happened ? If I really wanted to learn something about life, I'd read a famous person's biography". Today, I'm posting ACX comments from my crappy apartment, and he's a multi-millionaire who no longer needs to work at all, so it seems there might be something to it.
He didn't just happen to become rich, either; it was his explicit goal since childhood: become just rich enough so he no longer has to waste time on work. Andrew Tate also had an explicit goal, and he also achieved it (plus or minus some potential jail time, but who's counting). My goals are not his, but it seems that rejecting frivolities gives one quite an effective boost in accomplishing one's goals...
I've lifted weights for > 30 years, wrote code, and done some other productive stuff. Would I have made more money by devoting every waking hour to it? Probably, but there's scale to consider ... would I have been risk-tolerant enough to try what either Tate or SBF did? No, but I'm not heading to prison anytime soon (I hope). If I'd never read anything but text messages or blogs, would I have the developed moral sense to avoid their transgressions after entering the same lines of work? Also a maybe; you wanna say yes but you don't know the feel unless you've had a mansion, a Lambo and a few champagne-soaked victory parties.
As to Lee, Taleb covers this well in his books -> the guys who score like this gain visibility, while you never hear about the 99 with similar ideas who fail. America has millions of people who worked hard their whole lives, or took a chance on a small business, and have zero net worth. There's not a strict mechanism that converts time to money.
Heroin is obviously in a different reward category; principally physical and with deleterious health effects.
TBF, I also have not read the Decameron, don't plan to do so, and don't really think I'd appreciate it if I did (source: tried to read Crime and Punishment recently, got halfway through and was too bored to continue) and I'm uncomfortable with anyone implying that I'm therefore shut out of the true and the beautiful and that it's impossible to live meaningfully under these circumstances.
I did read the Canterbury Tales and somewhat like them, but I'm not sure I liked them enough that I could convince myself to read an arbitrary number of things in that category.
Neither have I read the Decameron, nor a lot of the Classic Russian Novels, but at the same time a person needn't read 19th century and earlier novels if that's not their bag, but at least try reading *something*. Or treating music as more than "fast beats to lift heavy weights to". Or beauty as more than "looking good gets you chicks but more importantly makes you more successful since good-looking people make more money".
I've read some of the Decameron a while ago, and my impression of it was that it's basically a collection of dick jokes. Pretty funny ones at times, though often dated (understandably so).
Surely you can only have one of "it's just a collection of dick jokes" and "we should judge Roy Lee for doing meaningless ignoble things rather than noble things like reading the Decameron"?
Oh, I wasn't judging anyone; merely expressing my opinion of the Decameron itself. I don't think reading the Decameron, or in fact any book of fiction, is either noble or ignoble. But some books are worth reading, and IMO the Decameron is reasonably entertaining (by contrast with e.g. Moby Dick, which I couldn't get through either).
This is a silly comment that gives away that you have not, er, read the Decameron. It's a bunch of saucy, often clever tales with a framing device. You can pick up the book and read as many or as few as you want. My grandma read them as a child in an attic in the northern Andes. (She was hiding there to avoid kitchen tasks.)
I can't think of anything more different from Crime and Punishment. The Canterbury Tales are much closer (obviously - Chaucer was influenced by Boccacio), though I'm willing to bet that you read them in Middle English, and thus found them harder than any modern translation of the Decameron.
One of these days I really, really have to sit down and find a reasonable translation of Chaucer's "House of Fame" just to see this bit that I vaguely remember from some review, that when Chaucer has himself swept aloft to the heavens by the eagle of Jove (imitating Dante in the Paradiso), he makes the eagle complain about how heavy the poet is. Someone with a sense of humour to poke fun at himself like that is worth reading.
Yep, here's the Eagle:
"And he me in his feet to bere,
Til that he felte that I had hete,
And felte eek tho myn herte bete.
And tho gan he me to disporte,
And with wordes to comforte,
And sayde twyës, 'Seynte Marie!
Thou art noyous for to carie,
And nothing nedeth hit, parde!
For al-so wis god helpe me
As thou non harm shalt have of this;
And this cas, that betid thee is,
Is for thy lore and for thy prow;—
Let see! darst thou yet loke now?
Be ful assured, boldely,
I am thy frend.'"
Everyone's a critic, including our feathered friend. Jove has commanded him to bear Chaucer aloft because of his devotion to love and his desire to write about it (even if he's not that smart, says our eagle):
"And nevertheles has set thy wit --
Although that in thy hede ful lyte is --
To make bokes, songes, dytees,
In ryme, or elles in cadence,
As thou best canst, in reverence
Of Love, and of his servants eke,
That have his servise soght, and seke;"
And while I'm going back over Chesterton's biography of Chaucer, let me quote this part as to why yeah, toiling and moiling for more money is rather sad if you can't even get some joy out of the daisies in Spring:
"With The Legend of Good Women (leaving Troilus for fuller treatment) we pass to matters more purely Chaucerian, as being more personal and peculiar to Chaucer. In the prologue to this poem appears the picturesque and vivid, though quite light and unpretentious, personal sketch of the poet. In the livelier and more telling manner of his later tales, he tells us something of his own habits and humours; how he delights in books; how he is criticized for neglecting his neighbours for the companionship of books; how, when he has done his day's work with the accounts and ledgers of his various public offices, he does not want to go on a holiday but only to sit down with a book. Having worked the description up to this point, he gets the full effect out of the one exception, and makes quite a dramatic entrance for the Daisy. He owns that, when spring first comes on the fields, he suddenly flings away his books and goes down into the daisy-meadows to drink of a deeper delight."
Kriss is a cultural critic which is to say he is a professional hater. A very gifted one, but still. I don't think that's necessarily a bad thing. The idea that all discourse should be maximally charitable is one of the rationalist norms that can actually be quite obstructive to progress. There is a lot of value in a good hatchet job.
Nothing like riling up the mob against people you hate, eh? I'd question if mob violence and cancellations are societally productive in most cases, but I do see the appeal.
Good article, even if the bit where he talks about how the "Girardian moment has passed" has really not aged well so far. I get the feeling that he produces better work when writing about dead people. Hard to be too angry at the dead.
Hi Scott, I wonder what you took away from your paid vs. free tier AI post.
Hundreds of comments, dozens of people sharing their preferred model's responses to questions. My takeaway - as someone who uses the pro versions of the three major LLMs - was that the differences between models weren't enough to justify a strong preference based on information synthesis prowess. It made me wonder how quickly the "paid vs free tier divide" might fade as open source models reach the capabilities of current frontier, paid models.
For what it's worth, I preferred Claude's output in most cases. I think it has a lot more to do with my personality and job requirements than with its singular benefits over the other two.
I got the impression that most people were asking questions that free tier LLMs were perfectly capable of handling. Perhaps there are some tasks that paid ones are better at, but it seems like ~90% of question answering is already handled well enough by the free ones.
If anything, I think paid models are better at whatever task/question takes full advantage of reasoning models going at it for >2 minutes. I agree with you that a lot of people asked questions that were clearly going to get one-shotted by any LLM, paid or free tier. I expected the ACX commentariat to have more hardcore questions at the ready.. which probably lends credence to the idea that the gap in productivity between those who take full advantage of AI vs those who can't think of a really tough question (tough by AI standards, which funnily enough is very different to what I'd consider tough for humans) is large and growing.
As someone who does not pay for any of the AI models, and has been (in the last day) trying out Opus 4.6 Thinking using the free arena.ai stuff, I have found it noticeably better.
FWIW I have tried both free and paid models, and their answers to my questions are usually pretty terrible (admittedly, with the occasional exception of questions like "implement this incredibly well-known algorithm X in language Y"). I guess I'm not asking the right questions...
For the last couple of Open Threads, the user greatyard has asked odd questions related to food purity, theft, taste and morals. People assume that they are being asked in service to a hidden agenda, and it's more amusing to run wild than to dance to the asker's tune.
Current theories about the user "greatyard":
-A Large Language Model that accidently got trained on Rabbinical food purity doctrines
-An Old Order Mennonite on a rumspringa, but the poor chap is confusing the sexual freedom of the English with the lack of food standards of the English.
I wouldnt risk it. Food processed by a meat tenderizer would be totally fine. Here we are dealing with something like a "meat tenderizer.". Agents may be calling it that but it is definitely not that. We are assured it is not sexual, therefore it must be violent. I know not what poor sods are being described as "amorphous liquid globs" but I wont be ingesting them after theyve gone through the "meat tenderizer"
Is this a sexual thing, greatyard? We've had a bunch of these from you, and while we aren't worried, some of us are.. concerned. As the youth say, "On tonight's episode : " The Writer's Barely Disguised Fetish!"". Glop, liquid, moist, pounding, morals.... We can help you, greatyard. But first you have to want to change.
On that note, with all the questions about the morality of eaten stolen food, it behooves someone to ask: is everything all right? Are you in a safe place? Do you need help getting enough to eat?
You asked a very similar question before, and I asked, "what about mortar and pestle ?". You never replied, but the same idea applies here as well to some extent. However, now I'm questioning your very premise. How is it possible to "pound flat" an "amorphous liquid" ? It's already as flat as it'll ever be !
Pounding certainly wouldn't do, you'd just splash it about everywhere. Put it through a chinois or a tamis to get out lumps, because you would use a pestle with these!
I didn't see that it was stolen, my post assumes the glop was given freely. That changes the equation, I need more information on the conditions of the theft.
Is sex addiction actually a revealed preference for being polyamorous? If not, how do you distinguish between the two? Alternatively, are other forms of addiction also a revealed preference for a lifestyle identity centred around drugs & alcohol? This is not a troll, just trying to get under the bonnet.
I think a heuristic for addiction is "it ruins your life, but you still can't help it".
So I'd say that if you have a happy polyamorous life, I wouldn't call that addiction. (Happy not just for you, but also for your partners.)
Sex addiction is more like: you can't avoid having sex with some people on impulse, even if you know that it may have big negative consequences. For example, if you have sex with crazy people who will then ruin your life and the life of your partners (whether mono or poly), if you have sex with minors or otherwise expose yourself to legal risk, if you open yourself to blackmail, etc.
So in effect the revealed preference of the addict would be "I don't want this" - is that a fair comment?
I think "revealed preferences" is a motte-and-bailey concept. The motte is "what you actually *did*", and the bailey is "what you (secretly? unconsciously?) *wanted* to do".
Taken to the extreme, if a meteorite falls on your head and kills you, it means that you had a revealed preference for being killed by a meteorite. If you didn't, then why *did* you get killed by the meteorite, you hypocrite?
I believe that humans do *not* work this way. I believe that humans sometimes have a genuine conflict, that they in some meaningful sense want X, and they also in some meaningful sense want Y, even if they can't have X and Y both. Sometimes one desire is much stronger than the other (which doesn't make the other not real). Sometimes they are comparably strong, and the actual outcome may depend on circumstances (trivial inconveniences, nudges, etc.).
From the perspective of the "revealed preferences", even if one of the conflicting desires X and Y wins narrowly -- for example, an alcoholic who wants to stop drinking, (a) notices an alcohol ad on TV which is the last impulse that makes him go to shop and buy a bottle, or (b) looks out of the window, notices that it is raining, and decides to stay at home sober -- we are supposed to treat the successful desire as the One True Wish of the individual, and the unsuccessful desire as hypocritical bullshit that the individual obviously never meant seriously.
This is *not* how actual humans work.
(Yes, there are situations where humans hypocritically say one thing for social reasons, and predictably do another thing. The fallacy is to adopt the edgy position that *every* situation is like this.)
In response to your question... the addict may have multiple conflicting desires. When the desire to do the thing destructive in long term keeps winning, we call it addiction. Sometimes it wins narrowly; sometimes it wins by a landslide. (When the desire to avoid the destruction wins, we call it a temptation what was successfully resisted.)
I agree 100%, "revealed preference" is a term I see knocking about and I have real trouble applying it to my own choices in a way that doesn't lead to tail chasing.
revealed preference don't really make sense for addicts.
Imagine I get drunk every day, so I tell my friends to hide the alcohol from me. When they are gone, I go through the whole house looking for it until I find the bottles hidden somewhere and drink them.
What is my revealed preference?
Clearly not just pro or anti-alcohol, but something more complicated.
I'd say they are anti-alcohol under reflection but can't control themselves in the moment.
For a sex addict the description would probably be the same.
Did they time the bombing in Iran for the last few hours of February to profit off of prediction markets?
https://theonion.com/trump-wins-60-on-kalshi-betting-hell-bomb-iran/
Sam Altman just announced that OpenAI is going to be the military's new AI provider, and that their agreement includes not using it for mass surveillance or autonomous weapon systems.
So... the entire drama around Anthropic was pointless? They're swapping out one AI provider for another one that asked for the same conditions? Was this just so that the Pentagon could back down from its demands without looking like they were backing down?
Given the context of the Anthropic blowup, it seems like a non-trivial possibility that Sam Altman might have said those things without him knowing they are true, or while knowing they are not true. I'd like to see some explanation for why the safeguards were a no-go with Anthropic, but acceptable for the contract with OpenAI.
Most likely explanation is that Sam is exaggerating and they didn't actually get safeguards comparable to Anthropic.
Yeah, Sam Altman is not known for full disclosure, to say the least, and it's easy for a contract to say in clause 1 "No killbots or mass surveillance!" and then in clause 312 "clause 1 does not apply if the Pentagon pinky-swears that it's really important".
Pope Leo has recently advised priests not to use AI to write their homilies. So, naturally, I asked DeepSeek what it thought..
“The Object of Faith is a Mystery, Not a Problem to be Solved
Faith, in its theological sense, deals with Mystery—with the infinite, incomprehensible reality of God. The proper response to Mystery is awe, wonder, worship, and love. These are postures of the whole person.
An AI is designed to solve problems and answer questions. It treats everything as a puzzle to be resolved with information. It can analyze the concept of the Trinity, but it cannot stand in silent adoration before it. It cannot worship. And worship is the fundamental response of a creature of faith to its Creator.
In summary, while an AI can be an incredibly sophisticated tool for communicating about faith, it can never be a subject of faith. The experience of faith requires a soul, a will, a body, a history, and the capacity for a personal, loving relationship with God—all things an AI, by its very nature, does not possess.”
I assuming that Ai generated content is ok here if it’s labelled as such and the result of an experiment. It’s fundamental to the joke here that AI wrote the argument against using AI to write homilies.
Also, I think the answer is very revealing of DeepSeek’s nature. A one-off response could be almost anything, but this chimes with my usual experience of DeepSeek.
Chief federal judge skewers U.S. Attorney’s Office, raises threat of criminal contempt over violations
Minnesota’s chief federal judge said he’s found another 113 orders ICE has violated since late last month.
https://www.startribune.com/chief-federal-judge-skewers-us-attorneys-office-raises-threat-of-criminal-contempt-over-violations/601589365?utm_source=gift
Possibly of interest - not a technical piece, but a simple one about scarcity and incentives.
—
As someone pursuing licensure as a psychologist on the eve of the Singularity, I obviously have every incentive to pretend AI can't do therapy. But I think this would be a naive position to argue.
What I do think, though, is that for *structural* reasons, "AI therapy" as an enterprise may be doomed—no matter how popular it gets, and even if it puts human therapists out of a job.
The problem: that it's too available. Human-to-human therapy's weekly pacing, and your therapist's finite, mortal nature are fundamental to the magic.
Infinite therapy with an infinite therapist won't stick, and won't feel valuable. But it's insanely reinforcing—and tech companies have no real incentive to build something *less* reinforcing.
https://whitmanic.substack.com/p/the-hard-problem-of-ai-therapy
Would be curious to hear what folks think!
Here is something I saw on a different website; it was written by Frank Landis (he is a biologist by profession): https://www.amazon.com/stores/Frank-Landis/author/B00JIA8YII
"after writing Hot Earth Dreams for three years of staring into the climate change abyss, I unsurprisingly had to deal with fairly serious depression, made far worse by the fact that I could not talk about my experiences with most people, because they were unwilling to deal with climate change. This, unfortunately, included one therapist. Nice guy, but he couldn't process why everyone's behavior around climate change was making me feel suicidal. So after helpfully telling me that he'd have me locked up for my own good if I told him I actually felt suicidal (as he was required to do by law), he gave me some standard ways to deal with depression. His legal constraints, coupled with his time constraints (lots of depression going around for some reason, so he was busy enough to be perennially stressed out), coupled with his cognitive constraints, made it impossible to have a helpfully sentient discussion with him.
Getting a couple of cats helped more, because they're less constrained in the attention they're willing to give and demand. Does that make them more sentient than the therapist?
In a situation like this, I assume the therapist is a fully sentient human by default, because philosophical zombies haven't been shown to exist. But if a conglomerate of expert systems (say an unholy mating of IBM's Watson running a psychology system running through a Google LaMDA-like chatbot interface) can "understand my feelings" and provide me psychological help better than that therapist or a couple of cats could, which system is the most sentient? It would appear that the computers have a superior understanding of my qualia at least. What more do they need?"
My position on which therapy is valuable (and more generally on software being sentient) is similar to that of Frank Landis: As long as it does what I want it to do, it does not matter whether the software “really has an understanding” or not, any more than it matters whether submarines “really swim”.
The fact that an AI can provide therapy doesn't necessarily imply that the current AIs out of the box are good for the therapy. So you still have the advantage over the machine; enjoy it while it lasts.
A few months ago I tried to use an AI as a coach, I even told it to try different personalities, but I found all of them very annoying. Either I couldn't find the right words to tell the AI what exactly I want, or the AI was not ready yet.
I think you can't get a good therapist just by using AI out of the box. You also need to provide a good prompt. Something that will specify the form of therapy you want to use (psychoanalytic? rogerian? cbt?), and even there I think you need a good description in order to get the real thing instead of a parody (because the AI also learns on the parodies).
You could probably get some money in short term by figuring out the right prompt, and selling the "AI therapist" as a wrapper for the popular LLMs.
Generally, I wonder what is the right way to tell an AI to do some X, as opposed to the popular misconception of X. Especially if the popular misconception is more frequent in the learning data than the real thing. For example, I think the the AIs are probably capable of role-playing people with various levels of IQ, but if you explicitly tell them to be high-IQ, you will probably get the popular stereotype instead of the real thing. (Did anyone experiment with this?)
There won't be a "Singularity" so don't worry about it.
Current trends towards decreasing attention span, as curated by "the algorithm" mean that even spending time with someone real is less and less desirable, because you aren't getting the dopamine hit that you get from a quick response online (I say, giving a quick response online).
Humanity is a moving target, and trying to use your preconceptions to predict Gen Z is probably going to turn out poorly. They're the generation whose IQ decreases the more schooling they get.
Just in case anybody was still unclear as to the administration's _actual_ immigrant priority -- this was posted by Reuters yesterday afternoon and as of now has not been commented on by the White House.
"WASHINGTON/JOHANNESBURG, Feb 26 (Reuters) - The U.S. aims to process 4,500 refugee applications from white South Africans per month, far above President Donald Trump's stated refugee program cap, and is installing trailers on embassy property in Pretoria to support the effort, a U.S. contracting document said....
"Trump has said the U.S. would only admit 7,500 total refugees from around the world in fiscal year 2026 [which ends 10/1/26]....Only 2,000 white South Africans had entered the U.S. as refugees as of January 31 under a program launched in May 2025, although the pace has picked up in recent months....Trump ordered a halt to refugee admissions into the U.S. after taking office in 2025. But weeks later, he launched an effort to bring in white South Africans of Afrikaner ethnicity as refugees.
"The U.S. State Department and Department of Homeland Security did not respond to requests for comment. The White House referred questions to the State Department.
"The contracting document, posted to a U.S. government database on Wednesday, explains the rationale for awarding the contract for the trailers without a competitive bidding process, stressing an urgent need for a secure site. An immigration raid by South African authorities on a previous U.S. refugee processing site on a commercial property in Johannesburg had forced the government to consider a more secure location, it said, after 'operations were compromised.'
" 'The inability to safely process about 4,500 applicants per month, an objective communicated to (the U.S. State Department's refugee division) from the White House, would result in failure to meet a Presidential priority,' the document said....
"U.S. and South African officials reached an agreement during a closed-door meeting in late December to allow processing to continue, Reuters reported last month. The contracting document said a South African company had received a no-bid $772,000 contract to supply and install 14 prefabricated modular buildings as part of a 'temporary modular village' on an embassy property in Pretoria.
"In a WhatsApp group for South Africans to share information about the program, one applicant said they had an interview this week in a trailer-like structure at an embassy property and that more trailers were being prepared, suggesting the site was now operational."
>"In a WhatsApp group for South Africans to share information about the program, one applicant said they had an interview this week in a trailer-like structure at an embassy property and that more trailers were being prepared, suggesting the site was now operational."
They need let all white South Africans have visa-free travel to the U.S. and let them apply for asylum here, not do this dicking around with trailers and interviews. (Ideally, of course, we'd have open borders.)
Race based admissions? I wonder what sort of racial purity test we apply to ensure the applicants are "white" [sic].
https://pbs.twimg.com/media/DLJyglKW4AA4i1B.jpg
They could have applicants submit an essay and do an interview, and then base admissions on "holistic assessments" where race is just one factor out of many.
The guy doesn’t mess with dog whistles, canine symphonies are more his style.
It’s as ugly as sin but on the plus side it’s transparent ugliness.
There will never be another troll like him. A Boomer - the world’s most epic troll, who’d have thunk it.
What's "troll" about this? I thought this was what people wanted. Nobody was complaining about the white European immigrants. This is no different.
Does human cluefulness have a genetic component?
https://www.youtube.com/watch?v=Y-BxDMJPSs4
Discussion of how breeding for winning at tightly defined arbirtrary contests is a genetic bottleneck which isn't as good as people breeding loosely for healthy working animals who can do a number of different things, including work and reproduction.
https://www.youtube.com/watch?v=IPXzoUAaKxA
Why Secretariat was such a magnificent race horse, but whose descendants were mostly very expensive disappointments.
After he died, there was a necropsy and his heart, in perfect condition, weighed 22 pounds. Typical race horse hearts are 8 or 9 pounds.
Secretariat's muscles were getting a lot more oxygen and not building up lactic acid.
[People talk as though athletic superiority is a sort of moral magic. I am deeply suspicious. But then, it's something I'm not good at.]
It turns out that the large heart trait trait is on an X chromosome-- males can't pass it on.
Secretariat's maternal bloodline had the trait. And the breeder that owned him was very interested in maternal lines. So were the Bedouins who created Arabians.
I don't exactly recommend the video-- it's very repetitious and tries to amp up emotion in alien fashion which makes me suspect ai.
Everything isn't mystery and magic. Sometimes you make money by paying attention to patterns other people aren't seeing.
In the spirit of Moneyball, I wonder whether there were people who could afford Secretariat's sperm, but didn't buy it. Perhaps if they exist, they can be interviewed about what they were thinking.
> It turns out that the large heart trait trait is on an X chromosome-- males can't pass it on.
This is not true. Secretariat would have passed an X chromosome to his daughters.
Athletic Superiority is conflating a few things, including the influence of testosterone (and the whole "no pain no gain"), autism, and actual physical superiority (bigger/stronger muscles).
The first two (autism affords a kind of diligence to athletes) are, indeed, moral issues. Although we'll note that when the Soviets had women doing brutal plyometric training... I'm unsure how voluntary it was. Or the taking of artificial androgens, for that matter.
I was on a tour of a ranch down here, where we passed the grave of the ranch’s most noteworthy race horse (at least from a sporting standpoint) - a Triple Crown winner.
I had never heard of that horse, which mildly surprised me (not into racing at all, but I could throw out a half dozen names in random response to a trivia game). The reason, seemingly was that it was sterile.
In fact, upon looking it up just now the fella had several infirmities:
https://www.nyra.com/belmont-stakes/history/triple-crown-winners/assault/
Anyway, no one was ever going to bottle that lightning.
Given that Substack is generally famous for moderate, centrist leaning, cool-headed conversation, why does Ann Applebaum have completely unhinged commenters? Like I said Europe probably would not have defended Greenland, because committing suicide for 50,000 Inuit people who only stay with Denmark is that Denmark bribes to do so, but they clearly do not identify as Europeans, they are founding members of the Inuit Circumpolar Council, that would be crazy.
And now I am a GRU agent lol.
Russia-paranoia is the no. 1 unhinged thing on the liberal side. They are so dysfunctional that they cannot get mold out of Putin's palace, yet they are supposedly behind everything lol.
I distinctly remember she used to be a cool-headed analyst of Eastern European cultures and best friends with a number of conservatives like Maria Schmidt. Did she also get unhinged? Or what is happening?
I don't know Ann or her substack. There is a general trend for groups to get caught in their own little bubble. And the bubble can then morph their belief system in weird or random ways. The smaller the bubble the worse it is. It's kinda like isolated island mutations.
> Russia-paranoia is the no. 1 unhinged thing on the liberal side.
Until quite recently, perhaps, but now I think it's the Epstein moral panic.
I'd actually assess high trust in the Intelligence Agencies to be more unhinged, and the Russia-paranoia thing they've developed is downstream of that.
> Given that Substack is generally famous for moderate, centrist leaning, cool-headed conversation
That's the first time I am hearing this. I am not saying it's false, I just assumed that Substack is "a little bit of everything" (that includes the thing you said, but also its opposite).
> They are so dysfunctional that they cannot get mold out of Putin's palace, yet they are supposedly behind everything lol.
Being bad at X doesn't necessarily mean that they cannot be good at Y. They could have sacrificed X to get more Y. Priorities. Creating chaos in other countries is cheaper than keeping your own country in order.
Well… in that case… Substack must do the algorithm thing extremely well. Far better than other social media. I generally get calm and rational people from both side of every divide, for example Defending Feminism who is very data-centric at doing just that, and Archwinger who is a mild, daddy-type redpiller.
It’s not the application, it’s moderation and community norms.
The format helps. It is far harder to convince yourself that your 3000 word essay is "honestly readable" than a 10 word comment. Also, you're selecting for higher attention spans and people who haven't had their brains eroded by "social media."
> Also, you're selecting for higher attention spans and people who haven't had their brains eroded by "social media."
Uh, I don't know if it's been successful at selecting for that...
I have yet to see a substack comments section consumed by emojis, to the point where no one is actually typing anything EXCEPT emojis.
And you (and the rest of the schmucks here) read the articles posted. Those are a considerable task for someone with low attention span.
> Substack must do the algorithm thing extremely well.
Maybe. No idea, because you obviously use other Substack features than me.
But the data *is* there, should Substack decide to use it. For example, bloggers recommend bloggers. Going by "if you like Scott, you will probably like bloggers recommended by Scott, and maybe even bloggers recommended by those bloggers" sounds like a simple yet mostly reliable algorithm.
Why other media don't use the same? Perhaps they optimize for a different outcome. Consider the incentives -- Substack profits it you *subscribe* to yet another blog, therefore it wins if it shows you something you genuinely like. Other networks profit if you *keep scrolling* and therefore seeing more ads, therefore they win even if they keep you angry or frustrated.
Or maybe it is the dichotomy between bloggers and readers, that does not exist on most networks. (As in: "Scott recommends blogger X who recommends blogger Y" works much better than "Scott liked a comment written by user X who liked a comment written by user Y".) Or both these things together.
There are a lot of discussions of "status", I think it was started by Robin Hanson 10-15 years ago, and I think it is an attempt for spergy types to understand why normies often act irrationally, correct? When normies say "close prisons", they do not actually mean they want to release Harvey Weinstein, they are just using a status-maximizing shibboleth. Correct so far?
But what I would like to ask, why not call it respect? Status sounds like a rigid thing, like your parents money, and respect is dynamic thing, earned or lost, expressed in social behaviour, looking up/down to people, and I think that is what it really is.
Or is it both? Like people earn respect by imitating actually high rigid-status people (people with wealthy parents) ?
I mean one obviously true, very important, and usually ignored fact is that wealth and money correlates from both directions. This is kinda hilarious. If you say "educated people should make political decisions", people will nod. If you say "rich people should make political decisions", people go EWWWW. But "they are the same picture"! So if you want to look rich, old money rich, you need to act educated.
The word “respect” was already commandeered by low status people.
I've rarely seen an educated man in a bespoke suit. "Looking rich" looks different from the inside than the outside.
Yes, "education" is something the rich possess. Like f*cking movie stars, the rich have a surprising tendency to acquire "hard to achieve" possessions as status symbols.
But the "old rich" Don't Look Rich (not the way Trump looks rich). They don't drive Ferraris. They have priceless heirlooms, and they don't look that much better than what you or I could acquire (forgeries are relatively cheap if you're not looking to convince anyone else.)
> it is an attempt for spergy types to understand why normies often act irrationally, correct?
Agree technically, disagree connotationally.
It is an attempt to understand a force of human psychology that has an obvious big impact on how humans behave. This force is usually not discussed explicitly, because it is one of those things where "looking at it changes it". That is, discussing in too much detail what specifically makes a person X respected would be perceived as an attack on their status, and could have negative consequences for the speaker. So most people learn the rules unconsciously by observing how others behave and copying their behavior.
And yes, this type of learning does not work well for autistic people, so their choices are either explicit discussion, or not learning the rules at all and constantly getting in trouble. That said, even the understanding of normies is not perfect, otherwise we would have no need for psychology or sociology, so there is a scientific value for everyone at studying this. It's just that the spergy types find it immediately useful, not only academically.
Not all irrationality is related to status. Normies saying "X" and meaning something other than literally X seems more about normies not paying attention to the details and just going with the crowd. Going with the crowd is not status-maximizing. It was survival-maximizing in the past (not going with the crowd could get you killed).
> why not call it respect?
This could be debated endlessly; no word is perfect, each word has some connotations. Sometimes you need pick a word and say "for the purpose of this debate, we will use the word W in the technical sense to mean X". You want a word that expresses something that Donald Trump, and Taylor Swift, and Albert Einstein, and the bully at your school have. The current standard is to use the word "status". It has two subcategories, called "dominance" and "prestige".
I would say that respect is individual, and status is like an aggregate of individual respects. Both change in time, but at different speeds.
> Like people earn respect by imitating actually high rigid-status people (people with wealthy parents)?
People gain dominance by seeming that they could hurt you, should they choose so. Could be real; could be a bluff. People gain prestige by demonstrating their superior skills. Could be real; could be fake. (One could argue that successfully bluffing or successfully faking skills is a demonstration of actual social skills.)
The last thing you said is about the difference between dominance and prestige. In some sense they are similar (they make people pay attention to what you say, and make them more likely to do what you want ), but in other sense they are opposite (dominance is about fear, prestige is about admiration; people want to avoid dominant individuals but get closer to prestigious ones).
I would like to strongly challenge the dominance is about fear assumption, because it leads to a massive misunderstanding of history. Someone who can hurt you can also protect you from other people by hurting them. And this is exactly why the Mussolini type "bully" type leaders were actually popular, they did not just intimidate everybody into submission, but offered protection, safety (in a non-explicit way). Hence the eternal popularity of the "strongman". Does not matter if from real enemies or fabricated ones. So "consensual tyranny" can be a thing. Kind of. Probably not the best way to express it. Actually popular dictators. So at the end of the day it is fear, of course, but not from the dominant person but from other people the dominant person is supposed to protect them from.
And this is also behind the "women date bad boys" phenomenon, they want protection from someone who can scare everybody who could hurt them away. Dealing with one bully whom hopefully love mollifies into less bullying is better than dealing with them all.
My beef with status is the rigid 1980's economics "socio-economic status" thing. There was a book with a title similar to be a big fish in a small pond and the go to example of status was buying a Mercedes Benz. LMAO. This is why this is more dynamic, nothing less cool than the status symbols of the past.
But I do have a guess. I think this word got into usage because the status system of the US is not so rigid as in most countries. Like if you are a schoolteacher in Italy or Austria, you have a paper from a government owned university basically saying you are officially above working-class people even if you don't make more than them, so you always dress elegantly to show that fact, even when grocery shopping. It is very rigid. Titles are used a lot in Austria. Even in webshops they will print Dipl-Ing So and So on the package if the customer gave this title (engineer). This is why Europeans are okay with fairly equal incomes, because money does not imply status. It is more fluid in the US, hence status and respect mean simialr things.
Also crowds and survival: but it is not just any crowds. It is the cool crowd, which can help with both. You know how hard schoolchildren compete to get into the right kind of crowd.
Of course there are always additional details that someone won't mention in one comment. That's why Robin Hanson has an entire blog on this topic.
Sick of old British men going on and on about Agincourt. Shut up, shut up, shut up. The outcome was bad for humanity. And the loss was mostly caused by the French mania for frontal charges and their lack of an effective commander who could enforce discipline.
If somebody feels inspired to ask me "why was it bad for humanity," I'm going to assume they have barely given any thought to the Hundred Years War. Or they are some kind of bizarre reactionary who deeply respects the letters of the arbitrary laws of medieval dynastic succession. Though even by those laws, Henry V's dynastic claim to France was questionable.
Nu? How many people were lost to the Hundred Years War?
This is a much less controversial take than "The Allies winning World War II was bad for humanity." And I'm pretty sure we have Stalin's purges at a pretty damn high death toll. Let alone Mao and his sparrows.
A simple ballpark is 2-3 million dead in the Hundred Years War. So, double that. You're still at "way below preventable loss of life during the last century."
Can you rephrase this in different language? I'm not sure where your argument is. Are you saying it England pillaging France for 100 years wasn't "bad for humanity" because Stalin did worse? Surely you can't be saying THAT.
Do I have to defend all wars now? Simply saying that worldwar II caused a lot more negative aftereffects than the 100 years war did (and that saying "Allies win" was a worse outcome than "Axis wins" is controversial).
How much do you think France even noticed the pillaging? Compare vis the Black Death. That's 6-12 million dead, during the whole "100 years." The Black Death "broke society" and France ceased to function as a feudal society. And that's during the 100 years War. Remember, there were French-paid mercenaries pillaging too (after peace was declared), so you can't exactly blame all the deaths on the English.
To devil's advocate for a moment: there are positive aftereffects of war as well, such as improved military (mayhap this is only a positive for the particular country, although Hitler is going to disagree, as his military did a lot of analysis of the American Civil War). The 100 years war, with its wholesale "castles are not autowins," is of particular note here -- in that building a castle would have materially effected the lives of the peasants nearby.
I can't really understand your writing very well, but "bad for humanity" is an extremely broad category. Every time someone litters a piece of plastic garbage at sea, that's bad for humanity. In light of that, I find it hard to believe that Henry V winning the Battle of Agincourt doesn't qualify.
Epstein files release is reminding me of DOGE. "Major government reform" turned into cancelling random cancer grants. Epstein files release was supposed to Upend The System but now it's just cancelling random scientists for taking money from Epstein:
nbcnews.com/news/us-news/columbia-brain-institute-co-director-nobel-laureate-steps-epstein-ties-rcna260585
Anyone who took money from Epstein after he was convicted as a sex criminal deserves to be demeaned.
Including the ones who.didnt know?
Epstein is a billionaire. How unbelievably naive would you have to be not to at least Google search a billionaire before taking money from him.
But sure, if someone received a grant from some random foundation and Epstein was one of many board members, then that gives them plausible deniability and it's unjust to condemn them.
You're supposed to watch the news, not do some stupid nerdy stuff like study brains. /s
Books are kind of overrated? 3 pages of insight and 300 of fluff. I know it sound horribly anti-intellectual but for example I can summarize arguably the most import book in the history of economics, Keynes' General Theory in about 5 sentences and if you read it you will find it was really a waste of time because it is the same in so many words.
I totally disagree. Why are you even here if you don't like books?
There are other books than the ones that compile facts and provide beige cause/effect maps. Some of these belong to the category 'fiction', which includes sub-genres aplenty. Some would say that reading a great novel from the 19th century is a form of communion with a great mind no longer attached to a body. Others would chime in that a sci-fi story is simply more entertaining than reading Keynes. I'd say it's at least possible that your utility metric is too narrowly devised.
There have been tens of thousands of authors and I really can't believe you are stereotyping them like this. How the hell are you supposed to condense a narrative history into 3 pages?
Some books are needlessly long on (bad) purpose. Follow the incentives. Money: if you pay the authors by the length of text, as long as they have an idea for a book and can write it, why not make it twice as long? Prestige: perhaps the message could be reduced to a blog article or even a tweet (e.g. "do the important thing first"), but writing a book makes you an Author. (No one will create a Wikipedia page about you just because you wrote five really insightful tweets.)
But there are other reasons, too.
Memory: If I tell you an important lesson in one sentence, you are probably going to forget it in a few minutes. If I write a book that explores the lessons from various angles, and provides a few interesting stories that illustrate the lesson in practice, and you spend a week reading that book, and afterwards you put it on your shelf and maybe lend it to your friends -- the chance that you will remember and actually apply the lesson increases.
Different audiences: Yes, I could skip the obvious parts, but different people disagree on what is obvious. (It is not even obvious what is obvious. Sometimes you say "I know that already", but then you read the chapter and find out that you believed something wrong, or that there are important aspects that you haven't considered.) In theory this could be fixed by an interactive medium, like a web page that would provide a summary of a chapter, allowing you to click "expand" and read the entire chapter; perhaps multiple levels of this.
Even writing summaries depends on audience. A perfect summary is the difference between what *you* know now, and what you would know after reading the book.
And if a third party writes the summary, you need to trust the third party to understand and interpret the author correctly.
Some books are needlessly short on purpose.
Then the publisher writes a new clause in the contract. "You must actually finish the next book, to a satisfying conclusion, not just stop mid-thought."
Theres more than one kind of book. I find that with poo sci books, the first three quarters are build up that restate stuff, the end is the novel theory.
Yeah I read this book on General Relativity but all it really said was G_mn + Lambda g_mn = (8 pi G / c^4) T_mn. Total waste of time.
This is a really naive view. Conclusions by themselves are less important than the reasoning that leads to them, or the complex outcomes they imply.
Consider that you might be underestimating the time it would take to teach it to someone else. Summarize, I guess, if all you're aiming for is equivalent to the blurb on the back of the book. I've been exploring a few books I want to distill down for my kiddo, just to teach the salient points, and I'm actually finding that very hard to do. Assuming a base level of zero, it's actually quite involved to walk through the author's reasoning and how you go from point A to point B and how the data in part C supports the claims in part D.
I could just tell him, Author says beans are nutritious and that's all you need to know. But that's not substantively different from me saying it. The point of referring to the book at all is that the author has made specific arguments and supported them in specific ways. But I struggle to come up with a sort of Cliffs Notes that's accessible enough for kiddo and still carries enough "meat" of the author's propositions and make it as impactful for them to hear as it was for me. Probably my fault as well for wanting to do a single half hour session on the topic versus spending weeks walking through chapter by chapter.
Some non-fiction seems padded to me. I'm thinking of some Steven Pinker books here.
"Wealth of Nations" is not a 'padded' book IMO. Smith occasionally repeats himself but will say something like "I know I've already said this but this is important."
Fiction can be good short form or long.
This seems to happen when journalists or some non subject matter expert write a book that should really be a long article. (or a long article that should be a short article). My biggest peeve is whenever they are overly descriptive about their interview. (I met so and so at such and such and he was wearing this quirky widget, yadda yadda). I immediately put such a book down.
Let's hear it for the underappreciated SMEs. You may be using the term in a different manner but I once worked a coding job where each programmer was paired with a non-coder SME.
A book doesn't even need the 3 pages of insight to be good
"The best books... are those that tell you what you know already.” -Orwell
Anthropic, the wildly successful AI company that has cast itself as the most safety-conscious of the top research labs, is dropping the central pledge of its flagship safety policy, company officials tell TIME.
https://time.com/7380854/exclusive-anthropic-drops-flagship-safety-pledge/
||
“We felt that it wouldn't actually help anyone for us to stop training AI models,” Anthropic’s chief science officer Jared Kaplan told TIME in an exclusive interview. “We didn't really feel, with the rapid advance of AI, that it made sense for us to make unilateral commitments … if competitors are blazing ahead.”
||
So this could be a contemporary example of a multipolar trap, right? At the beginning, there was an industry consensus about safety, and now the defections have started*.
Call me cynical, but I have to wonder if this isn't just an admission that acknowledging Type I Doomers (e.g. Yudkowsky-style ZOMG they kill us!!, vs Type II Doomers like Yang ZOMG they replace us!!) was part of a marketing strategy to (over)sell the power of the new models. The hype is shifting with the vibe.
*Ex: https://techcrunch.com/2026/02/11/openai-disbands-mission-alignment-team-which-focused-on-safe-and-trustworthy-ai-development/
Question: why are old philosophers like Plato "conservative colored"? I mean if you think about what the allegory of the cave means, it means you cannot trust tradition and common sense at all, everything widely believed is wrong and you have to radically rethink everything.
I mean one explanation could be that the Catholic Church spent a lot of effort in coloring Plato conservative. But surely people should be able to see through that now?
Plato was deeply suspicious of the masses. He considered them a dangerous force that needs to be carefully controlled, or they become animals and destroy society. This resonates with traditional, state supported Christianity and the views of Edmund Burke and the American founding fathers. So I can totally see why he would have unique appeal for conservatives.
But you're right if you think it's facile for a leftist/left-liberal to ignore Plato on that basis. He had a lot of ideas and there's something in there for everyone.
Locke thought herders deserved to have their land taken away by force, and so did agriculturalists who were not using the latest, greatest techniques. Yet Nobody has any problem looking past those idiocies and digging for gold in his writings.
In addition to what others already said:
Every tradition has started at some point. At that point, it was considered "revolutionary". Centuries later, it is considered "traditional".
Successful ideas sometimes become "a part of the water supply" and we no longer remember to credit the person who made them popular. (Unless our textbooks make it specifically a point. And even that may not happen, if someone else has later expressed the idea even better, so the textbook will focus on the new guy.) On the other hand, we may be shocked to find out that someone believed something that was "a part of the water supply" when they lived. This makes the old thinkers seen more conservative than they were.
Oh, yeah, you should see Nietzche and the ubermensch. Hitler liked to quote him, so people want to characterize Nietzche as anti-semetic (whereas in fact, he really disliked the "slave mentality" of the new testament.) His views on Jews were wildly varied, which suggests a lack of coherency indicating they weren't very central to his existence or philosophy.
> Question: why are old philosophers like Plato "conservative colored"? I mean if you think about what the allegory of the cave means, it means you cannot trust tradition and common sense at all, everything widely believed is wrong and you have to radically rethink everything.
Two reasons.
One is that I don't think most people actually read all the old philosophers, and thus don't have more than surface level understanding of them. The Peripatetics are not the Platonists who are not the Epicureans who are not the Stoics. Some of those philosophies are probably more compatible with modern progressivism or left liberalism (such as the Stoics, who were both cosmopolitan and who are sometimes described as history's first anarchists due to the, sadly lost, work of Zeno's Republic.)
The other reason is that the old philosophers kind of represent the pre-Enlightenment heritage of Western culture's political thought, and that was hugely influential on how societies were and have been set up. Some of the American Founding Fathers slept with Cicero on their bedside table, which makes it easy to see him as "conservative" for a modern progressive.
In Islamic thought of the 10th and 11th century Aristotle was the central philosophical authority. That Islamic thought also influenced Maimonides. In “The Guide for The Perplexed” Maimonides said Aristotle was the greatest human intellect to ever live. Earlier in the 4th century in “City of God” and “Confessions” IIRC Augustine simply refers to 'the Greeks” in his writing.
Edit: Just searched my ebook copy of City of God and Plato is in fact mentioned often.
Plato and reactionary politics (I would not say conservatism) are anti-liberal, and anti-liberal political forms can be very different by the Anna Karenina principle.
That might explain Plato specifically, but it doesn't explain why other philosophical schools are often read as conservative or reactionary.
I think where the historical Socrates fell in the political spectrum is highly debatable, and given the cosmopolitanism of Socrates, the Cynics and the Stoics, they're hard to easily place in a reactionary or conservative camp. Sure their virtue ethics might code a bit conservative today, but they also didn't just try and conserve the values of their societies. Stoics like Seneca advocated for treating slaves like people, and argued against certain extreme forms of blood sports. The Stoic Musonius Rufus said women should be taught philosophy, and said that virtue was the same in men as it was in women. They're often much better than the people of their time, even if they fall behind by modern standards.
Plato had female.Guardians in his repulibic, as well.
Honouring old philosophers involves honouring accumulated experience as a guide to the present, which is in some sense the core of conservatism. O
Even overthrowing society to implement Plato’s Republic exactly would count as neo-reactionism or something.
Though I don’t think Plato’s cave allegory implies that radically rethinking everything will go better than using common sense. It suggests that people might hide from the truth, but I didn’t think it said anything about rethinking everything as a way to escape the cave or anything like that. The allegory is sort of a warning against trusting crowds of people who are hiding from reality, but I imagine Plato doesn’t put himself in that category.
I think the basic answer is that Plato is complex and conservatism is complex and the two align in some ways but not others.
So in his day Plato was somewhat conservative (e.g. anti-democratic) but took it in a radical direction (e.g. views on the family).
Then, you can read the cave as a rejection of traditional wisdom, but you can also read it as saying that ordinary people can't understand reality and need an educated elite to guide them. That idea was fairly fundamental to conservative thought for centuries, but conservativism is currently going through a populist phase, and Plato is the arch anti-populist.
And also, people who like Plato today may be somewhat conservative-coded, because liking ancient philosophy and the foundations of western civilization is somewhat conservative-coded. But there are also some obvious points of similarity between Plato and Marx.
Area contrarian, Bronze Age Pervert uses a strained reading of "The Republic" to advance some odd-ball far-right nonsense. Sometimes a troll is just a troll.
Speaking as a Catholic, Aristotle is our guy (see St Thomas Aquinas, huge fanboy of The Philosopher), not so much Plato (at least I didn't think so).
But it seems that was only a development in the Middle Ages, and the Church Fathers preferred Plato:
https://www.newadvent.org/cathen/12159a.htm
"Plato's School, like Aristotle's, was organized by Plato himself and handed over at the time of his death to his nephew Speusippus, the first scholarch, or ruler of the school. It was then known as the Academy, because it met in the groves of Academus. The Academy continued, with varying fortunes, to maintain its identity as a Platonic school, first at Athens, and later at Alexandria until the first century of the Christian era. It modified the Platonic system in the direction of mysticism and demonology, and underwent at least one period of scepticism. It ended in a loosely constructed eclecticism. With the advent of neo-Platonism founded by Ammonius and developed by Plotinus, Platonism definitely entered the cause of Paganism against Christianity. Nevertheless, the great majority of the Christian philosophers down to St. Augustine were Platonists. They appreciated the uplifting influence of Plato's psychology and metaphysics, and recognized in that influence a powerful ally of Christianity in the warfare against materialism and naturalism. These Christian Platonists underestimated Aristotle, whom they generally referred to as an "acute" logician whose philosophy favoured the heretical opponents of orthodox Christianity. The Middle Ages completely reversed this verdict. The first scholastics knew only the logical treatises of Aristotle, and, so far as they were psychologists or metaphysicians at all, they drew on the Platonism of St. Augustine. Their successors, however, in the twelfth century came to a knowledge of the psychology, metaphysics, and ethics of Aristotle, and adopted the Aristotelean view so completely that before the end of the thirteenth century the Stagyrite occupied in the Christian schools the position occupied in the fifth century by the founder of the Academy. There were, however, episodes, so to speak, of Platonism in the history of Scholasticism — e.g., the School of Chartes in the twelfth century — and throughout the whole scholastic period some principles of Platonism, and especially of neo-Platonism, were incorporated in the Aristotelean system adopted by the schoolmen. The Renaissance brought a revival of Platonism, due to the influence of men like Bessarion, Plethon, Ficino, and the two Mirandolas Giovanni Pico and Giovanni Francesco Pico."
Possibly it's just the idea that "anything really old is conservative"?
Yes, the way I read Catholic history is that there was a competition between the Aristoteleans and Platonists, with the Aristoteleans usually winning. Which is kinda bad as I think science proved Platonists more correct.
I mean when you ask why a rock falls to the ground, the Aristotelean would immediately start investigating the nature of the rock. When in fact gravity is a universal law independent of individual objects.
Having said that Aristoteleanism is immensely useful. Of course things consist of matter and information, you mix matter with DNA and you get a tree. And information is carried by matter but is not determined by it, as "4" on a screen means the same information as holding up four fingers. Yet DNA proves that information is not just social convention. Information is simply an ontologically different category as matter, even when it is always represented in matter.
Aristoteleanism would be a total win if they would be willing to use modern terminology:
form -> information
final cause -> function
mover -> changer
In the Sam Kriss article, it says:
> His best-case scenario for AI is essentially the antithesis of Roy’s: superintelligence that will actively refuse to give us everything we want, for the sake of preserving our humanity. “If we ever get AI that is strong enough to basically be God and solve all of our problems, it will need to use the same techniques that the actual God uses in terms of maintaining some distance. I do think it’s possible that the AI will be like, Now I am God. I’ve concluded that the actual God made exactly the right decision on how much evil to permit in the universe. Therefore I refuse to change anything.”
Is this your actual belief? Or were you joking, and Sam mysteriously did not interpret/frame it as a joke, or something else along these lines?
I would interpret it as a wry observation. If AI gets to be as smart as God, it will know not to change anything because that’s the way God made it. Who is to argue with God? In the meantime, we all need to find something to do, don’t we?
> If AI gets to be as smart as God, it will know not to change anything because that’s the way God made it.
Perhaps it could go in the opposite direction and kill all the doctors and engineers who try to change the things that God made.
Naw, they are part of God‘s plan as well. Of course Super AI could be the new Messiah. In which case it will judge both the quick and the dead..
If that's his actual belief, then I have no idea why he isn't going full Butlerian Jihad on those bots. If his best case scenario is that the superintelligence effectively neutralizes itself and every other AI, then why advocate for building it at all? It doesn't make sense.
Some might have saw Tucker Carlson ask Mike Huckabee about whether the jews were entitled to the whole land promised to Abraham - from the Nile to the Euphrredis, and Mike sort of saying yes. As a secular Israeli I cringe about the idea held by some of our friends that the Bible has anything to do with the situation, but also from the ignorance about what the Bible actually says if the matter: That Abraham was promised and received many children that came to be the nations that settled different parts of that vast land. That he was specifically promised that Ishmael will settle a land greater than that of all of his brothers, and that the land to the east of the Jordan river is inheritance from God to the children of Essau and Lot - forbidden for the Israelites to take. The Bible actually spends surprisingly many words on legitimizing neighboring peoples and tribes by relating them to Abraham, telling how they defitted giants to get their lands, etc.
I am not sure about Huckabee, but the US is rife with textists: people who profess the sacred inerrancy of an inspired text, haven't bothered to actually read the text, but believe it supports their pet theories.
As an agnostic cultural Catholic, this is so crazy. The Catholic Church spent 1500 years removing likely falsifications from the text, and when the final version was presented at the Council of Trent, it was final because they just can't know any better, not because it is perfect.
Back when books were copied manually, falsifications were a huge issue. Someone copied Plato, didn't agree with a thought, replaced it with his idea. So a painful process of comparing different versions was developed, called "measurement", in Greek: canon. And it is not an inerrable process.
We know John is not synoptic. We also know only John can link the text to philosophical theism: that Christ is the Logos. Without that, you can basically throw Aquinas out and cannot do theistic philosophy at all, because you do not live in a logical world. You always read the text critically with proper historic, philosophical etc. context. Yes on some theoretical level Catholicism thinks the original version of the Bible was inerrable, but also that we do not have that version.
Not to mention the uncertainty inherent to translating a text that is already hundreds of years old. I understand על פני כל אחיו ישכון as "he will settle land larger beyond that of his brothers", but על פני is literally just "in front of" and the rest I filled from context. I don't even know what KJB chose to do with this verse
Yes. Everybody who is multilingual knows that translations are practically impossible, because words have a range of meanings. For example I cannot simply translate "boat" to "csónak" in Hungarian. You see the Anglos are a seafaring people, used to big ass ships, so they might call a 30 foot long decked yacht a boat. Hungarians, being landlocked, consider that strictly a ship/hajó, and call only small uncovered dinghy type things csónak. You actually have to be familiar with the culture.
And it is so much harder with dead languages. Diligite inimicos vestros does not mean love your enemies. It means love your rivals. Or show preference to your rivals, basically treat them fairly. It does not preach surrender to actual enemies.
I’ve read somewhere that all translation is paraphrase. Can’t scare up a the source to cite though.
You have choices with all translation.
1) You can go with the "straight" translation, where you try to convey the nuance/humor/etc from the original.
2) You can translate the meaning, and leave the words behind. Aggretsuko, in order to capture a joke, changed the protagonist's nickname from "Part Timer" to "Calendar."
And then there's Ghost Stories (where the "translator" said, "You realize I don't know Japanese?" And the publisher's response was: "make something up.")
Someone was worried how the world will see the US after the Epstein files. My opinion is that most cultures will think it is not a big deal. Most culture think of women as more or less property, or at least the daughters of the poor are, and are fair game after puberty. I don't think the average person in China thinks it is a big deal. He is very much used to commie party bigwigs having very young girlfriends. Consent? Well with enough money consent is kind of implied... that way of thinking.
I was already feeling this during the Weinstein scandal. Everybody in my neck of Europe was politely silent because if we wanted to be honest we would have said that we would blow a donkey for an $5M movie role, it is a no brainer, of course you have sex for such a fortune. Everybody was silently thinking with enough money consent is irrelevant because deep down everybody is a whore. I mean the things we would do to never ever ever have to work again...
It is strange how the US looks like on the surface a very commercial culture, and yet people do not think money is everything. I think most of Europe is way more "mercenary"...
In the United States, unlike the rest of the world, the pornography industry is "clean" and the movie industry is "very dirty."
Sexual matters with young chinese girls is very much a Diplomatic Incident sort of deal, see Hunter Biden. This may have something to do with Chinese racism (or, more charitably, patriotism).
I don’t know about China but powerful men are getting in trouble all over Europe: the UK, Norway, Slovakia, France…
In case of Slovakia, the only person in any kind of trouble is Miroslav Lajčák.
First he insisted that his contacts with Epstein were purely professional and he had no idea. Then other messages were published, containing him begging Epstein "Why don't you invite me for these games? I would take the ... girl." But he insists that this was merely stupid macho talk, of course he didn't mean it seriously.
Consequences? Lajčák resigned from the position of a security advisor to prime minister Fico (while still admitting nothing, it was "just to avoid being politically used to attack the prime minister"). There were no criminal charges or investigation or anything. He still remains on board of the oil refinery Slovnaft.
And that's all.
Fico's people even succeeded to create an alternative conspiracy theory about the Epstein files. After Fico resigned in 2018, the next day Lajčák wrote to Epstein "my ex-PM Fico [...] is out of government and looking for a new agenda. He would be happy to play Steve [Bannon]'s game. And he is good." (And Fico soon made his party more nationalist, anti-EU, and pro-Russian... unless the timing is a complete coincidence, of course.)
But Fico's people keep saying (falsely, but who checks the facts these days?) on social networks that the communication about Fico being out of government happened *before* his resignation... and therefore, that the message was *actually* about Americans preparing the fall of Fico's government... you know, the evil Americans wanted to create another "Maidan" in Slovakia... so, basically if you ask people in Slovakia about Epstein files, you are quite likely to hear this instead. (Fico is a political genius; life can throw any shit at him, and he succeeds to make a lemonade out of it for his followers.)
PS: If politicians in Slovakia want to have sex with minors without any consequences, they definitely don't have to travel to USA. (Sorry, no good sources in English.)
You're missing the point with Weinstein. Your stance as stated isn't that consent is irrelevant with enough money, but rather that you'd give consent in exchange for enough money. You're imagining an offer for a free economic transaction such as, "Harvey Weinstein, nice to meet you. Wanna have sex for a role in my new movie?", where the terms are clearly established, and you can decline without consequence outside of not getting the $5M role.
I wouldn't consider that sexual assault. I would still oppose it for various knock-on reasons, but I wouldn't call it a violation of consent. That's not what Weinstein did, though.
First, Weinstein was convicted of rape. Dozens of women have accused him of rape by the strict, conventional standards of "ignoring explicit denials of consent" and "using physical force." Maybe you don't believe any of these accusations (if this is the case, share your reasoning), but otherwise, this is indefensible, even if he gave them all millions of dollars after the fact, which he didn't.
Second, even the "casting couch" situations were usually much worse than the free economic transaction hypothetical. Many women reported:
1. Being alone with Weinstein when pressured to agree to sex immediately, rather than having time and distance to consider the offer
2. Being presented with vague promises or even threats of sabotage, rather than being offered an explicit, bounded payment for a service
3. Being expected to be sexually available to Weinstein in general for the indeterminate future, with any refusal threatening an uncertain cost, not just having sex with him some contractually agreed upon number of times
You can replace "sex" with any mundane but emotionally or physically risky service, and the ethical problems are equally clear. A lack of clear terms, or conditions for careful decision-making, turn free economic transactions into coercion.
So, the lady dangling her feet out of Quentin Tarantino's foot-fetishist eyes is turning the free economic transaction into coercion?
[note: she's being paid by him, not the other way around.]
You're going to have to explain what the point you're making is because I don't get it.
You just said that "lack of clear terms or conditions for careful decision making" turn free economic transactions into coercion.
I was giving an example, where the lure of "maybe I'll get to see/touch those amazing feet of hers" is apparently coercion. At least according to what you're saying. I think it's outside the boundaries of what most people would call coercion, and certainly in a legal sense (aside from contractually obligated showing of feet -- which is not being discussed).
I don't understand how you managed to get that interpretation from my comment. Can you explain how the scenario you proposed and the one I described are similar?
I was responding to your last sentence, "A lack of clear terms, or conditions for careful decision-making, turn free economic transactions into coercion."
Most people know that high-pressure tactics are a Bad Sign (if someone here has slept through that part of Human Nature, just Walk Away if someone's trying to get you to decide right then and there.)
But "lack of clear terms" is normal for "somewhat skeezy" sexual interactions (say, someone who sleeps with someone else to get tenure). And I don't think, without the other part of your "or" you get to coercion -- you get, instead, to "bad bargains" (in the main, most people "teasing" someone are inclined to do so well past the point where it's "in good fun." Flirting is encouraged at the start of relationships, though, so what do I know?)
Is that even relevant? The point is that these situations are are not a hit to America's public image, because only a few countries are truly concerned with "consent" in the way the US is.
Relevant to what? Carlos said Weinstein wasn't that big of a deal because lots of people would gladly have sex for millions of dollars. I said Carlos is misunderstanding why Weinstein's behavior was bad.
I only reply to claims I find interesting to discuss.
>My opinion is that most cultures will think it is not a big deal
Considering there have already been real consequences for high-ranking figures like former Prince Andrew and members of the British government, that prediction has already resolved to false. Also it's not just about the sex, it's also about Epstein's Russia and intelligence agency connections.
You're being disingenuous. The rest of the comment makes clear he's talking about the differences in sexual mores.
yeah and also he's talking about Yurp not UK
Yurp... I love it. I have not seen this spelling before, but it's perfect.
I'm not going to reply to the implication that Europe will supposedly go easy on pedophiles because they're all a bunch of closet pdeophiles themselves.
If that was your takeaway from Carlos's comment you need to work on your reading comprehension skills.
Sure. So then, what's a good word for "sexual mores that make you think pedophilia is no big deal", as Shankar Sivarajan suggests?
Judaism (seriously, the smart Jew keeps on telling the dumb jews* that "you can't tell people that pedophilia is actually anti-Semitism!")
*there's always more dumb jews.
https://www.newsweek.com/tom-alexandrovich-israeli-official-defend-child-sex-sting-nevada-2116504
Israel is actively advertising that "you can get one free on us!" if you'll immigrate after becoming a sex criminal in America. I really, really wish I was joking.
"Un-American."
Epstein had nothing to do pedophilia.
I am so sick of the word pedophile being treated as a trump card.
Oh it's the Trump card alright.
Oh, it's the capitalize the t rebuttal.
Okay, now THAT I can see your ideological commitment leaving you unable to distinguish from what he actually said.
It’s true, there is a certain prudishness about sex and some other things here that, in my experience anyway, you don’t get in Europe. I have heard people argue this is an artifact of America’s Puritan beginnings. Probably it’s too simplistic to say America is very capitalist and all-about-money and so on. I would say there’s a great deal of cognitive dissonance running through our culture, as well as through every person who is part of that culture, though I think cognitive dissonance is even more universal than that anyway.
A while ago Scott posted an age of consent map for Europe, showing many European countries with 14 as the age of consent. The comment section went ballistic with people demanding Scott take it down because it Couldn't Possibly Be True. (He did not do so.) They couldn't fathom that an advanced, industrial society can exist where people can tell their 14 yo daughters not to sleep with dirty old men and then go on with their lives.
In my experience they generally don't anyway. 14-15yo girls typically had 18-20yo boyfriends because boys mature so much slower than girls. BTW that is actually a new phenomenon and I don't understand it. You look at the pic of a 18 year old male in 1910 and you see a young man who could build a house or fight a war. You look at a 18yo male today and you see a big baby. What the hell happened? The result is that girls date rather aggressively up in age - not decades of difference of course but usually 4-6 years.
> You look at the pic of a 18 year old male in 1910 and you see a young man who could build a house or fight a war.
I think a 12 year old male could do that in the Roman Empire. But that was long ago, this is a more rapid change.
> What the hell happened?
My guess, it's mostly about school. Seems to me that the first year after the school changes you dramatically. And this experience now happens later in life.
At school, you are told what to do, and you are told to do relatively easy things, and the rest of the time you are free to play (unless you have a side job, or your family gives you work). You are psychologically a child. Perhaps a child who has sex and drinks alcohol a lot, but still basically a child, in terms of responsibilities.
After school, it's the "real life". You are supposed to make money and pay your expenses. (You probably still live at your parents, but it is clear that this is supposed to be temporary, and that you are expected to do something about it, better sooner than later.) It is no longer enough to do a few simple tasks and be done. No one guarantees you a job. If you have a job, you have much less free time. You no longer automatically advance from one grade to another; if you want a career growth, you need to work hard to make that happen. The life is visibly way more unfair than before, and there is no one to complain to about it. Instead of moving along with the mob, you have to fight for yourself.
And I think the girls are *also* less mature than the girls of the same age in 1910, but for a girl this is much less of a problem on the dating market. (I mean, how many of those 15 years old girls know e.g. how to cook? But also, how many of the men who want to fuck them care about that?)
I found life even after school is child's play. I am paid to solve IT puzzles. How is that different from the math test at school? It is a game. Does not feel like hard work at all, not harder than the cosinus stuff felt at 16. And then after work I played videogames. So my life was not so different at 28 then at 8.
This is why I took long to mature. The wife and kids stuff I was ready for around 35 and YET it found out I am not mature enough for that sort of responsibility, I was that classic husband who just helps the wife but does not do anything without an instruction, which is also what children do, so that ended. Maybe I will probably never really mature, because at 47 my job is still about just figuring out fun IT puzzles. I don't think I will ever lay bricks, get in a fistfight etc. or ever decide anything really important. There is just no "maturing pressure". The bank pays my bills. I have a cleaning lady. Basically I do nothing "serious" outside the office ever. And inside the office it is increasingly like "AI, find the bug".
Ah, I think I know some people like that, too. Leads to some funny communication. "Tell me, buddy, how do you solve this problem X? Because I find it very difficult, and I am probably doing something wrong, so perhaps you could give me some advice." "Uhm, I... don't really have the problem X." "Ah, thanks. Well, that wasn't really helpful, but thanks anyway, at least now I have something to think about." :D
When I was at university, some of my classmates got there on their first attempt. Others failed, did some job for one year, then tried again and succeeded. The age difference between these two groups wasn't that big, but the difference in behavior was. That's where my observation comes from.
Depends on how it gets translated. If they hear "raping children," they might find that somewhat distasteful. If they hear "young prostitute," I agree with you that most people around the world would roll their eyes at everyone making this out to be a big deal (as would most Americans, I think).
I don't think you have anywhere near enough bandwidth to explain that the Epstein story is actually about the latter, and it just sounds like the former because of an idiosyncrasy of the ruling American ideology.
This is the land of plenty; we have always been able to afford our illusions. It is not that long ago that prostitution was rampant in this country. But then something changed. It’s all part of being the new world. To tell you the truth, it’s still pretty common but very under the radar. I think it is mostly the truth that the men who get in serious trouble for it are usually assholes.
In Western Europe there are practically zero Western European prostitutes. They tend to come from ever poorer countries, now apparently Romanian girls are rich enough to not do this, so they come from Moldova. So I get the impression women really don't like to do it, only do it if they are super poor.
Ukraine is the classic place for Western European prostitutes to come from. If they've somehow run out of Ukrainian girls....
(Referencing "The Tribe" simply because everyone should see that movie. It's a masterwork of cinematography. Brutal and vital enough to make even seasoned nurses walk out of the movie theater).
There is that. It is a way to make money when there aren’t other ways but I think there are women who choose it and like it. There’s a pretty wide socio-economic range.
Most of those women would probably choose to be married, if they could get the same lifestyle, with a single partner. Less risky, at any rate.
(This is me stealth-lumping in the "Get a Russian Bride" with the whole prostitution scene).
The number who 'choose it and like it' isn't zero, but it's not great.
The social costs are huge and mostly on-going.
Depends on who you talk to. Escort services in DC have been a thing for a very long time. (They sometimes double as bodyguards).
Yeah, I have no idea what the numbers are. I have a friend or two who are sex workers (I am not a client) and they certainly treat it as their chosen profession. As to why that is, I could not tell you. I have never asked. If I have any more things to say about this, I will save them for my posthumorous memoir.
Viewed as an outsider, the US still seems to have some... transactional... overtones to a lot of it's sexual exchanges. Even ones that are miles from any conventional definition of prostitution.
Someone living in the US once described this to me as 'a blowjob for an easter bonnet'; she was talking about her own experiences within marriage.
That’s a very funny expression.
It is precisely in transactional attitudes where explicit consent is super important, the whole concept is a direct application of contract law.
In more relational cultures sex just "sort of happens" in a semi-conscious way, people just feel when making such moves is appropriate and welcome or not.
This is why I really wonder whether American women are truly happy with all this contractual sex with explicit verbal consent stuff. Isn't men's sexiest trait having the emotional intelligence to read women? My first gf in Budapest told me "if a lady says no, it means maybe, if she says maybe, it means yes, if she says yes, she is not a lady". Ok that also had some screwed up parts of it (obvious fear of slut-shaming etc.) but clearly she wanted her men to have the ability to read ambigious signals.
But maybe I should not base my opinion on FetLife too much, FetLife is very explicitly contractual because people do the kind of kinks that in extremis can kill people, so of course clarity is paramount.
>It is precisely in transactional attitudes where explicit consent is super important, the whole concept is a direct application of contract law.
That's not an angle I had considered before; that's an interesting idea.
When I was a kid working a blue collar job the guys who were the age that I am now would tell me about the old bordello of *their* youth in the next small town over. According to those fellas the working girls at the place would chew out all the customers lined up after town bars closed.
“You guys thought you were going to have a woman for free? Ha! Next time come here earlier in the evening!”
I won’t say the name of that town but if you ever got a chemistry set for Christmas that town’s name was on the box.
Everyone sane here thinks this way too. People make a big deal about this because our politics incentivizes it. The political Left has a fetish for victimization narratives because they've proven extraordinarily effective over the past 20 or 30 years and so they search for them wherever they can find them. Epstein didn't do anything that many other high-status men across every part of the political and religious spectrums have done throughout history. The Weinstein scandal was clearly a cynical transaction. This is just how the world works: men acquire status so that they can get access to young and beautiful women; women use their beauty to get access to high-status men. No one is being victimized apart from progressive delusions about how the world is supposed to work. When liberal-adjacent men like celebrities or rock stars do it the culture turns a blind eye to it. Elvis married his wife when she was 14. So did Jerry Lee Lewis. Jimmy Page had a 14 year old gf when Zeppelin was at its peak. Seinfeld dated a 17 year old in the 90s. Even Simone de Beauvoir abused teenage girls. But when someone conservative-coded (powerful, elite, politically connected, wealthy) gets implicated then the political incentives align and we get ourselves a scandal.
Too many people on the left are connected to Epstein which is why it will just blow over eventually.
>>When liberal-adjacent men like celebrities or rock stars do it the culture turns a blind eye to it. Elvis married his wife when she was 14. So did Jerry Lee Lewis. Jimmy Page had a 14 year old gf when Zeppelin was at its peak. Seinfeld dated a 17 year old in the 90s. Even Simone de Beauvoir abused teenage girls. But when someone conservative-coded (powerful, elite, politically connected, wealthy) gets implicated then the political incentives align and we get ourselves a scandal.
Even assuming for the sake of argument that all these people you're pointing to were liberal or "liberal-adjacent," all of your examples are 50-30 years old or more. Elvis married in 1967. Jerry Lee Lewis married his 13-year old cousin-bride in 1957. Zepplin's "peak" (at least according to google AI) was 1971-75. Seinfeld, as you note, took place in the 90's, and Simone de Beauvoir died in 1981.
It's fun to play at finding hypocrisy in the people you dislike, but the evidence is much better fitted to the simple explanation that social mores are changing. There's a steady growth in the age of the weird outliers over time. It's a weird outlier, and a scandal, in 1957 when Jerry Lee Lewis marries a 13 year old, but he does it. By the 90s, the example is a Seinfeld with a 17 year old. Get to the modern era, and the most analogous contemporary example of "what happens when a liberal does it" is Al Franken. That at least happened in in the last 10 years. His victim was fully-adult 33 at the time he harassed her, and unlike the liberals of 30 years before, who protected Bill Clinton, the 2010s Democrats ran Franken out of congress for it, under pressure by the leftward flank of their coalition.
The gaping void of Epstein prosecutions isn't there because "too many people on the left" would get caught up in them. I don't know where you get your media, but I'm not seeing a wave of CNN or Vox or MSNBC articles defending "innocent Democrats caught up in the files". I was around in the 90s. I saw the Clinton scandal coverage and I know what it looks like when they do that. The coverage I'm seeing contains the obligatory "technically you can be in here and not have committed a crime" disclaimer that keeps your media outlet from getting sued, but otherwise squeezes all the juice it can to play up the scandal for clicks. There isn't a protective media environment for Epstein clients.
And even if there were such an environment, at the end of the day a conservative president is in charge of the DOJ right now, so if there are prosecutions to be made and we aren't making them, it's still not "liberals" making that choice.
We only had to wait for Trump's name to appear in the Epstein files for everyone to conclude that actually Epstein and his friends did nothing wrong.
Epstein and his "friends" did plenty of wrong things. Just ask Stephen Hawking when you see him.
I do know someone in the current Administration, who quite enjoys working under Trump. He's not in the "Epstein did nothing wrong" camp. He's in the "too smart to accept those invitations from Epstein" camp.
> Just ask
You know, this is the thing I find repeatedly frustrating at reading your comments. Instead of writing an information, you just write a hint that it exists, leaving "figure out what I wanted to say (and then fact-check if you still have some time left)" as a homework for the reader.
That of course decreases your trustworthiness, because one of the likely reasons someone would communicate that way is "when they communicate clearly, their arguments are easily debunked". I am not saying this necessarily applies to this specific comment, but to your comments in general.
This is not just the left. I see conservative Christian men criticize DiCaprio for not marrying, not having kids, so basically not committing. They say it is lecherous, he is basically *using* his gfs.
But I should not be surprised. A LOT of American leftism is secularized Christianity. Christianity is not in every sense always conservative. The first thing the Mayflower pilgrims tried was communism in a kibbutz-like way. Christianity only feels conservative because its progressive currents have secularized.
I think at the end of the day we are going to find out that Epstein was an equal opportunity employer. It’s rather a peculiar way to analyse the whole phenomena anyway.
You’re letting Harvey off a little too lightly. There is truth in what you say, but it’s not the whole truth.
<The political Left has a fetish for victimization narratives because they've proven extraordinarily effective over the past 20 or 30 years and so they search for them wherever they can find them.
You know, I agree with this. What I find most infuriating about the victimization fetish is that it is a power move disguised as innocent upwelling compassion. If you're gonna make a power move, say STFU, motherfucker I am smarter and richer than you and also networked with people smarter and richer with you and we are going to decide what happens next and you are going to have to do it.
I sense something very similar. I am politically homeless because I believe in social democracy, strong unions etc. yet and I cannot stand lefty men, because they keep making power moves without making it look so. Strong "I am better than you" signals but with full plausible deniability, without a real open challenge. It just infuriates me. This is actually feminine behaviour, the art of the subtle backstab. Somehow I get along better with conservative guys despite disagreeing in so many things, yet basically when they are not openly challenging me I know they are not playing status games. When they do that is very clear and obvious ("alpha" bragging etc.)
This is all Pareto's foxes and lions... the same shit, upper-class foxes vs. working-class lions... except that back those times the lions were unionized socialist workers...
Hear hear. The ninth circle of Hell was reserved for betrayers and that's exactly where those people belong. What they do is far worse than straightforward violence or graft because they weaponize our empathy against us. No culture can withstand that. Now social trust has evaporated, bad faith is assumed, and the commons is guarded by an electric fence. Progressivism is a bad faith movement that succeeds because it's a Shelling point that enables losers to coordinate against the establishment. The last 30 years of US politics is a case study in why multiculturalism is a bad idea. I only hope Europe gets the memo before it's too late.
>Elvis married his wife when she was 14
They met when she was 14, married eight years later.
Oh so he just groomed her. That's totally normal and non-creepy then!
Do you know anything about the relationship? I certainly don’t.
Actually, I just looked it up. He was 24 when they met. He did seek to mold her into the woman he wanted her to be (according to the very brief synopsis I read) but after six years of marriage, they divorced, but she still likes him, even though he’s dead.
Oh I’m not saying it’s necessarily bad, not even for the women involved. The point is that the culture has a giant double standard that it applies according to which political outcome it wants, which directly undermines the moral high ground it assumes when making these kinds of condemnations. It’s not about protecting girls, it’s about gating who gets access to them. Gating access to sex is how women have always exercised power.
> Gating access to sex is how women have always exercised power.
How does that work? Women aren't capable of restricting men from sex on their own. Your enemy here is other men, not women.
It's good to see that Harper's is once again printing short stories by fiction authors like Sam Kriss.
>The piece says rationalists believe “that to reach the truth you have to abandon all existing modes of knowledge acquisition and start again from scratch”. The Harper’s fact-checker asked me if this was true and I emphatically said it wasn’t, so I’m not sure what’s going on here
It's more of an exaggeration than a falsehood.
Maybe, but it still feels weird that the fact-checker asks "is this true?", gets a clear negative answer, and then ignores it anyway.
Makes it seem like the entire purpose of having the fact-checker is to be able to say "we have checked all the information before publishing the article" and hope than no one asks "and...?".
Having this below the masthead likely influences perception of the community:
Astral Codex Ten
P(A|B) = [P(A)*P(B|A)]/P(B), all the rest is commentary.
Just to give the benefit of the doubt to the fact checker, it seems plausible that you could find a passage where Scott or some other rationalist wrote something that (Sam and) the fact checker would hear as "start again from scratch". If that were the case, then Scott's denial would only be some evidence against, but might be outweighed by others.
After writing this, it occurred to me to get a third-party opinion, so I asked Claude Opus 4.6 about it: https://claude.ai/share/89cb3006-8b5b-48c2-bfd8-ccc1e78b460e
Muttering “I will open the crackers so you will have crackers and be happy" to a two year old is the most relatable and human thing on the entire planet, only someone without kids would miss that.
It seems like having and raising your own children is the final step of human maturity and you're fundamentally living in a state of arrested development vis-a-vis the world and society and many of your peers until you experience it. I think this explains... a moderate amount of things world models without this idea fail to explain.
I was going to say the same thing. This is *completely* unremarkable behaviour.
I also agree on your views on parenthood influencing maturity; it changes you in ways that are quite distinct.
Hey @ArrkMindMaster (does substack understand this notation? probably not), I read your story about the orc. Your writing has greatly improved from the first chapter to the most recent one. Keep it up!
I read the Citrini piece earlier today and am confused about the automated economy.
Suppose a spherical cow world where there's no government redistribution. AI/robots can do all labor cheaper and better than humans do now. It seems like wealth has increased. But workers are out of a job, so they can't benefit.
It might seem like the rich capital owners would benefit. But if the rich capital owner is, say, the CEO of McDonalds, then nobody is going to to McDonalds anymore because all the workers are broke, so McDonalds collapses. (If the rich capital owner has a broad basket of stocks, all of those stocks are in the same situation as McDonalds)
It might seem like at least AI company owners would benefit. But if McDonald collapses, the AI companies can't sell McDonalds robots to automate its labor. So the AI company collapses too.
So . . . the economy as a whole is very rich, but nobody can buy anything, and every specific individual is poor? What am I missing?
BONUS QUESTION: Whatever your answer to the above, does it imply that if we smash enough existing machines to halve productivity, we can reverse this process and have the economy be very poor but every individual is rich?
> Whatever your answer to the above, does it imply that if we smash enough existing machines to halve productivity, we can reverse this process and have the economy be very poor but every individual is rich?
Yes, people call it "back in the 50s, a man could support a housewife, two point two kids and a beagle and pay his mortgage, all on a single wage".
I haven't quite had time to read the Citrini piece, but now I'm seeing the following making the rounds as a response to it. Daniel Jeffries writes a memo from 2030.
One slice of it: "It's March 2029, and a five-person team in Jakarta just dropped a film that's competing with Marvel for box office dominance. Not competing in some indie-film-festival-participation-trophy kind of way. Actually competing. $340 million opening weekend worldwide. Ninety-five percent on Rotten Tomatoes. Three of the team members are under 25. Their total budget was $2.8 million."
https://x.com/Dan_Jeffries1/status/2027056726072054219
My prediction for a "business as usual" (no significant policy intervention) automated economy is hyper-deflation: consumer prices would decline dramatically as marginal costs go down. Incomes would also decline dramatically, but each dollar of income or savings would have more purchasing power than it currently does.
I think compute/inference will become a commodity, and profit margins across most of the economy (blue chip and AI companies alike) will tighten. I think those who are already wealthy will have more purchasing power with the wealth they already have, but I don't think there's going to be any substantial transfer of wealth: just a general erosion of income and profit.
IDK that economics is the right lens to use for situations wherre humans are no longer useful, because law and contract are out the window. UBI is not happening either, except maybe to pacify people for a few years before it's easy to exterminate them.
Once humans are not the meta, most of them die whether the players of the game are humans or robots. Anyone who keeps more than a few humans alive will lose a war to the much bigger robot army of someone playing the meta. Hopefully we get off this path *before* that point.
So, yeah, this is just a final form of the inherent contradictions of capitalism.
"Of capitalism" being the key part. I realize this is a liberal/libertarian space and we differ in some very fundamental assumptions about the economy. And - at a risk of failing the Ideological Turing Test - I think a lot of that difference can be attributed to a liberal belief that the laws of the current economic system are like laws of nature, rather than of a particular, historically-specific stage of economic development. Which is to say, I think you're implicitly asking how the system goes on in those circumstances.
The answer is simply that it does not, because it cannot. This isn't like our contemporary arguments about whether it should, it's a completely new territory where its very foundations disintegrated, and we as society are outright forced to start doing something else. (No government intervention delaying the inevitable only means it happens sooner rather than later.) What that something else is, who knows, realistically, the winning solution will be decided by raw military power. (We can only hope that humans are social animals and no significant faction will be willing to destroy the entire society. And, of course, that the humans are still in control at that point.) But it will have to be something else, by necessity.
What we absolutely should not do is destroy wealth for the sole purpose of preserving the old system, that's for certain.
(I haven't read the Citrini piece and wouldn't know where to find it.) So this is not any type of answer, but a rant generated by your question. We've been screwing over blue collar workers for years/ decades. But it's fine 'cause we can all buy cheap stuff at Walmart, Amazon... And now that white collar workers may be caught in the same bind (Moloch) I should worry about it more? And here's a crazy idea, maybe cheap white collar work will make things cheaper for the blue collar worker. That would be a plus.
I think you can use the rust belt in the 80s and 90s as a simplified model, there’s an elite that does really well because productivity goes up, a lot of non-elite workers lose their jobs or get squeezed severely by competition.
Lots of local businesses also get hurt by the indirect effects of so many people no longer buying lunches, suits, etc.
Lots of people have no good options and slide into precarity or poverty. There is a lot of crime and drug issues and mortality goes way up.
Where the model breaks down is that there’s no place to move to to get away from AI, so the whole world is affected. And since it’s a massive crisis, politics becomes very important. Even in the rust belt, some towns re-oriented themselves to new economic activities and skills, and they did a lot better. Others fell into chaos.
Oh this is the most standard default dilemma of economics ever.
1) Marx figured like 150 years ago that this is why capitalism has inherent contradictions and regular crises, capitalists drive down wages, yet wages are the source of demand for their goods.
2) Keynes said the demand for "labour goods" can be long-term disastrously low, that is, Keynes's main idea was that human skills are not infinitely flexible and once you automate everything a blue-collar person can learn, they are in trouble. Same for white collars now. Keynes' famous "aggregate demand" is a code word for "demand for stuff not very bright people can make". "Labour goods".
3) Misesians say that in this case prices fall to $0.0000001 and you are set for 10 years by taking a rich person's dog for a walk, once. Their argument is that prices almost always fall to near to their production costs. Interestingly Marx also said that, too (the tendency of the rate of the profit to fall), so I dunno if you find it reassuring that the leftist fringe agrees with the libertarian fringe in one thing or not.
So anyhow it is virtually certain that prices will fall, and surely there will be redistribution too because that will be a sure vote-buyer.
>So anyhow it is virtually certain that prices will fall, and surely there will be redistribution too because that will be a sure vote-buyer.
The world could use some price-falling and redistribution right now, and yet there aren't many votes for that which I can see.
That's all assuming that politicians even have the will and the power to do those things within the current system. In the end, it might have to come down to violent revolution once again. Of course, given the level of technology and how long the violence will be bottled up by repression, it will be all the bloodier in the end, if and when it actually happens.
> The world could use some price-falling and redistribution right now, and yet there aren't many votes for that which I can see.
Are you kidding?? In the US, we spend 3x more per person every year than the median working person pays in taxes, and only half of people work, so we're really spending ~6x more.
We spend absolutely bonkers amounts on "redistribution," and the great majority of it, at least $4T dollars, is direct subsidies and payouts to the richest generation in history.
https://imgur.com/a/WzmzMSr
We are so all-in on redistribution there's literally no way it can continue working, we're mortgaging everyone's futures every single year to pump trillions of dollars worth of redistribution into old people every year.
>The world could use some price-falling and redistribution right now, and yet there aren't many votes for that which I can see
There are almost certainly votes for price- falling. Fortunately, policymakers are not stupid enough to try to do it on a macro scale. https://www.investopedia.com/ask/answers/111414/what-causes-negative-inflation-or-deflation.asp
I am more optimistic. Money has diminishing marginal utility, that is, rich people's primary motivation is not to become richer. They are after something else. It might be power for example. But I think a starving society is no one's interest. Pretty sure the rich will be OK with some kind of UBI if they get something else in return (such as power).
> Pretty sure the rich will be OK with some kind of UBI if they get something else in return (such as power).
First, if you imply that the rich will have that kind of veto or decision power, then I'd say we have already moved to a new system, not one I'd like to see personally.
Second, if the reason for the UBI question is that automated industry is reality, then what would compel the rich and powerful to make that compromise, if the UBI recipients' services are no longer needed at the most fundamental level? If their wealth and power no longer comes from the work of regular people?
Answers like "empathy" don't much convince me in this case. It might be confirmation bias on my part, or reporting bias in that you generally only ever hear about the Musks and Thiels of the world rather than the good guys with lots of money which may also exist. But my impression is that in order to become ultra-rich or ultra-powerful you have to be an above-average psychopath to begin with, so the psychopathic solution to problems is not as far to them as we would like it to be.
In the long run, all forms of human cooperation break down du to their inherent flaws becoming unsustainable in the face of reality. More often than not, that breakdown ends in considerable misery and death for the few and/or the many; the more people involved, the greater the damage. I don't see why it would be different this time.
> then what would compel the rich and powerful to make that compromise, if the UBI recipients' services are no longer needed at the most fundamental level?
When it comes to SOMEBODY sharing the wealth - the "Giving Pledge" has 236 billionaire signatories. That's the one that Gates and Buffet and Zuck have all signed where you commit to giving away at least half your wealth.
There's only like 700 billionaires in the US, a substantial fraction of them are charitably-minded enough to commit to giving the majority of their wealth away.
All it takes is one.
Agrajagagain already posted an excellent answer, but if you want something slightly more advanced, but still accessible...
The Citrini article would be right if we still used the Gold Standard and the money supply was fixed. As less money gets spent, there would be less money *to* spend, and we would risk getting trapped in a downwards spiral.
However, we have a fiat currency. We can just print more money. That has lots of downsides, like potential hyperinflation... but this is *exactly* why we have that ability in the first place: sometimes, you just have to print more money. If there's a shortage of money, because people aren't spending enough... instead of the government trying to get people to spend more money, we can just print more money. There can never be a money shortage as long as we can just print more money.
Now, how much to print, and who gets to spend it, are difficult questions that complicate things. But that's why we have the Federal Reserve, which thinks a lot about things exactly like this. The Citrini article talks about political inaction, a government unable to keep up with the rapidly changing tides. But the Federal Reserve doesn't have that problem, it doesn't have to wrangle Congress or whatever. If it wants to print more money, it can just do so. And unlike Congress, it already has a plan for things like this.
So what would actually happen if the Citrini scenario comes to pass? People lose their jobs. But since they're not necessary anymore, the economy produces the same amount of stuff as before. With the help of money printing + perhaps a UBI or something like that, the people who lost their jobs can still consume the same amount of stuff as before (more or less). The same amount of stuff gets made & consumed. Nothing changes. The Federal Reserve might have to engage in "Helicopter Money" (https://www.investopedia.com/terms/h/helicopter-drop.asp) to make it happen, but this is something they've prepared to do for a long time.
Then, eventually, the people who lost their jobs, will find new jobs in things like machining, and the other things the AIs can't do yet. The economy can produce more stuff. Everyone wins. (More or less. The exact distribution, is always a contentious topic.) That's, after all, basically what happened during the Industrial Revolution. A whole bunch of people lost their jobs in farming and had to go into industrial work, but in the end everyone was better off for it. And what happened during Globalization with offshoring, where people lost their jobs in manufacturing and had to move to services, and in the end it was worth it (more or less). This is the same thing, just with people losing their jobs in services and having to move to manufacturing.
... then, of course, the AI keeps advancing, in fact probably starts advancing even faster because we have a bigger economy capable of spending more on research.
Eventually the AIs are capable of machining as well, and plumbing, and welding, and all the things we thought they could never do. The usefulness of a human being still goes up, as we shuffle into the remaining jobs. But eventually, there are no remaining jobs left, and the usefulness of a human suddenly crashes to zero. (See https://www.newthingsunderthesun.com/pub/2ek4d4s3/release/7 ("What if we could automate invention?") and https://www.newthingsunderthesun.com/pub/4bnobp5q/release/4 ("When robots take your job")). A human being is 100 times as productive when 99% of the jobs have been automated, because one human can now do what previously took 100 people. But a human being is 0 times as productive when the last 1% of jobs gets automated, because you don't need the human anymore. Like Blackjack, everything just gets better & better until you pass 21.
What happens after that? Hopefully a good answer. Hopefully...
"That's, after all, basically what happened during the Industrial Revolution. A whole bunch of people lost their jobs in farming and had to go into industrial work, but in the end everyone was better off for it. And what happened during Globalization with offshoring, where people lost their jobs in manufacturing and had to move to services, and in the end it was worth it (more or less). This is the same thing, just with people losing their jobs in services and having to move to manufacturing."
Unless, of course, if this scenario will not happen this time. Yes, similar sort of things happened several times in history, but it is not a guarantee that they will happen next as well (I mean, I woke up every morning, thus I will not die; or there is the anecdote of the guy falling from the 10th floor -- even on the 2nd floor, so far so good).
That's... exactly what I talked about, immediately after that?
I am sorry, you are right.
No worries.
Noah Smith just wrote a good one about that: https://www.noahpinion.blog/p/the-citrini-post-is-just-a-scary
A counter-argument to the industrial revolution might have been that if the farmers move into cities and work in factories rather than engage in subsistence level farming then mass starvation would be the result. Broadly speaking, that didn't happen. The economy was restructured and as secondary production has moved to Asia, services filled the gap.
AI might or might not fundamentally restructure the economy. But if it does, it seems likely that new desires will emerge and fulfilling those will drive the circulation of money, and the exchange of goods and services. I see a world where more people are involved in caregiving and education, and fewer in corporate non-jobs as a positive development.
To answer your final question: lawyers and CEOs are essential for a billion dollar corporation. So if we make everyone a lawyer-CEO do we all become billionaires? Linear extrapolation only works until it doesn't. The economy will move from one equilibrium to another - available labour will be consumed, the government may need to intervene to ensure redistribution (I.e. flow) of wealth during the transition. But once things settle the new system should be fairly stable. So no - drastically changing the level of production in the economy doesn't necessarily lead to straight line extrapolation...
"A counter-argument to the industrial revolution might have been that if the farmers move into cities and work in factories rather than engage in subsistence level farming then mass starvation would be the result. Broadly speaking, that didn't happen. "
I think it is important to get the chronology right. In history, industrial revolution was _preceded_ by agricultural revolution, thus less people were able to grow the same amount of food -- without this labor surplus, the IR would never happened in the first place. There was no scenario where people were scared of famine for no reason, but they cancelled the IR anyway.
However, this is no guarantee for the future. For example, in the 20th century, productivity increase allowed the five-days work week (instead of the six-days being common before). How does it fit to your model?
I did say 'might have been' :)
The agricultural revolution happened thousands of years before the Industrial Revolution; and the move from subsistence agriculture to commercial agriculture to feed large cities happened in tandem with the Industrial Revolution.
I'm open to correction but my impression is that workers were drawn to the cities by higher wages, allowing more space per farmer and opening up the possibility of higher productivity -- coincidentally artificial fertilisers arrived on the scene around the same time, or possibly arising from the same scientific and engineering discoveries that underpinned the Industrial Revolution.
I don't think that it was obvious at the time that moving 50% of the population off the land and into towns wouldn't result in mass starvation. Certainly the life expectancies in Britain plummeted when this happened, and the towns were notorious hotbeds of death and disease. I don't have time to find the figures now but life expectancy of a worker moving into the towns was in certain times and places as low as two years.
My larger point was that no linear extrapolation from 1730 was going to explain the economy in 1830.
"The agricultural revolution happened thousands of years before the Industrial Revolution"
The first one, yes. The second one, definitely no; among others, here is the Wiki article about it: https://en.wikipedia.org/wiki/British_Agricultural_Revolution
(I got here by googling "second agricultural revolution", but of course, the Anglo-Saxon view of history has its own, somewhat more self-centric terms. :-) )
"and the move from subsistence agriculture to commercial agriculture to feed large cities happened in tandem with the Industrial Revolution"
Well, no, the agricultural one started before the industrial one, a lot. That is true, though, that it did not come to a halt in the XIXth century.
"I'm open to correction but my impression is that workers were drawn to the cities by higher wages, allowing more space per farmer and opening up the possibility of higher productivity"
Yes, I believe you are wrong in this. I mean, technically you can say that "cities offered higher wages" even if the villages offered nothing. :-) If the per capita agricultural production increases a lot within one or two generations, there will be less such jobs needed in the villages, and pretty much that was the only type available there. So people had to move to the cities in the hope that they can find any jobs.
(This is not even unprecentented, the German expansion eastwards for centuries was driven by something like this. I mean, there were definitely no higher wages besides the Danube or the Volga than in the German states, still, there were mass emigrations to these (pretty much unpopulated) territories in the 1700s. But even earlier, this is how Central Europe was filled with German towns -- just look around in Czechia.)
(You can also think of towns in the rust belt of any country, I am sure yours has one too.)
"coincidentally artificial fertilisers arrived on the scene around the same time"
We are well before that time.
"and the towns were notorious hotbeds of death and disease"
This is true BTW, which, IMHO, just weakens your "town salaries were more attractive" theory.
"My larger point was that no linear extrapolation from 1730 was going to explain the economy in 1830."
That is, of course, true, this is why it is called a revolution, after all. I only commented on your timeline. Because it is quite important to see that first there was the agricultural revolution (production increase), which made lots of people unnecessary in the agriculture. This allowed the IR, but in the meantime there was a ton of social disorder and suffering. I think the analogy to today's event is clear: if the AI makes lots of people's work unnecessary, it might allow some other sector to plummet later (or not -- we will see), but the transition will not be smooth at all. And there is no fourth sector already waiting and attracting engineers, accountants and lawyers with higher wages.
> I see a world where more people are involved in caregiving and education
Caregiving maybe, but isn't education going to be automated pretty early? The new government isn't particularly hot on public education as is, they're not going to keep paying for it when there's a cheaper alternative...
Skittle nailed my thinking -- I agree that 300+ person lectures will be a thing of the past, but a more optimistic take on education would be moving to tutor type systems. We'll have no shortage of human labour; why not invest that time and energy in developing others? I'm not claiming I think this will happen.
How would education be automated away? The message of the (in)famous Alpha School piece was that humans were vital in order to get children to actually engage and learn, and that was for a group of selected children who were likely to be more motivated, and have a greater ability to sit and focus, than average.
The majority of children will not engage in the sustained effort and focus required for learning without humans guiding and motivating them in some way. A minority (usually the most disruptive minority in a school) need a lot of human input to do so.
Unless we redefine ‘education’ to mean whatever is happening when a toddler swipes through endless short videos for hours on end.
Still, the experience is that education is in a downward spiral. Primary education perhaps not (that much), but tertiary (i.e. universities, colleges, other adult education stuff) pretty much.
This is, of course, a complicated issue, tons of the tools used in colleges are outdated in the light of recent AI developments (why have the students write essays, when they will just turn in AI-generated slop? But: how to teach skills which can be learnt by writing essays?), or just due to more and more material available online, but the bottom line is the same: perhaps we do not need that many people teaching other people at the current technology level.
Ah, if you’re just talking about whatever is happening with widespread tertiary education, I’m not sure ‘automated’ is the right term. I suspect that a lot of it will simply vanish or take on some other form for social purposes that doesn’t really pretend to be academically educational. But I don’t think that’s what people usually mean when they talk about jobs in education.
Still, tertiary education is not an insignificant part of education.
I am also not sure that secondary education won't be affected (we might even use present tense, I think...), unless we dare to call "childcare for teens" as "high school".
And even if we conclude that there is (will be) a demand for more "teachers", not everyone who does not have a job has the skills required to do it. As someone else put it in this thread, if we automate everything a blue-collar worker can learn...
I think the bit this piece gets wrong is in saying 'THIS CYCLE IS NOT LIKE THE REST'. As you pointed out, if there is no one to spend disposable income, companies' earnings will go down and stocks will go down and and and and...
What they talk about these downward spiral feeback loops where companies would double down on AI don't make sense. I don't see how this would be different to any other recession.
new technology increases productivity -> reduce workforce to maximise short term profits -> higher unemployment, economy stagnates -> recession -> cycle repeats
My vague impression is that economy is a "set of closed loops" -- things get produced somewhere, things get consumed somewhere, money flows in the opposite direction.
This is typically initiated on the demand side, like someone says "hey, I want X, and I am willing to pay for it" and hopes that someone else starts producing it. (And this typically already happened long ago, so we have inherited the loop.) Or, maybe no one responds, so the person who needs X starts to produce it himself, but someone else starts needing it too, etc.
In the world you propose, the workers can no longer exist *qua workers*. Some of them will find new jobs, for example they become servants or sex slaves or gladiators for the rich. Some of them will own some land, so maybe they survive by being unemployed and growing potatoes. Some of them will own a little stocks, maybe they get lucky and those stocks skyrocket. But most of them will starve and die.
That will cut a lot of existing loops. There will still be *some* demand to produce food for the rich and their slaves, but neither of those will eat at McDonalds (as it is now). So McDonalds either goes out of business, or changes their business to producing cheap food for the slaves, and organizing birthday parties for the children of the rich. Which is not completely unrelated to their current business, but it will be more optimized for these purposes -- the slaves and the rich probably won't eat their meals in the same room; the slaves will pick up their food somewhere or get it delivered by drones; the birthday party rooms will be larger and contain tons of entertainment. Actually, I *can* imagine McDonalds to survive, but it will be different from today. (And there is a chance that someone else will adapt to the new role faster.)
Maybe let's start from the opposite perspective: which loops will *remain*. Production of robots and computing power. Infrastructure for the rich: financial business, lawyers, weapons. Production of things that the rich want: luxury food, luxury clothes, entertainment, and whatever becomes the new fashionable thing to spend all that extra wealth on, for example space travel. (Could be something we wouldn't expect, like maybe the rich will decide that riding ponies is the cool thing, and now that 99% of population has starved to death, there is a lot of space where those ponies could ride. Probably mutant ponies that look like pokemons.) Secondary, if the rich want human slaves, also production of the things for slaves: cheap food, cheap clothes, shock collars.
Probably many things that we use today will remain; some rich person will probably be weird enough to want a bicycle, or anything. But they will produced in much smaller quantities and higher quality. Maybe expensive enough so that many of the current companies survive.
tl;dr -- some parts of economy will collapse, but not all; and many will change from "produce lots of cheap stuff for plebs" to "produce a few luxury items for the billionaires"; plus all the robots that build new robots
BONUS: Sometimes the machines multiply the productivity of a human, sometimes they replace the human. Sometimes they replace the human with another human, e.g. one who needs to be more intelligent to use the machine. Smashing some machines that replaced stupid people, e.g. in agriculture, would reduce unemployment and decrease productivity.
Maybe another way to look at the situation is that giving each person some brain and muscles was God's version of UBI. The basic economic loop was "brain + muscles + land = food". We created more efficient loops, so today for most people it's like "brain and/or muscles + participating in global economy = more food and toys". But when the brains and muscles get outcompeted by machines, you can no longer participate in the modern loop; and you can't even fall back to the old loop because there are too many people and not enough land for that; plus you wouldn't like that anyway, because you are already used to a certain amount of food and toys.
The question whether things are demand-driven or supply-driven (innovation-driven) is big and I think I have an elegant solution. I think we have very very broadly defined desires derived from our biology. So we have a demand not for cars, but for getting to places, not for smartphones, but for communication in general. So someone invents the horse and someone else invents the train. Someone invents the campfire to talk around, someone else the phone call, someone else YouTube. But all this is based on biology.
This means a cool thing. It is possible that at some level our biology is satisfied and we will not need new things. I still remember having to buy a new graphics card every 3 years. But once we reach the level of photorealistic retina display, no need for any more improvement.
Besides so many goods were positional in the past... I was cleaning my parents home after their death and it was so said that they bought so many things not because they needed them, but to make a bourgeois impression on other people. A lot of elegant clothes. Think 30 neckties. Artwork in the living room and nice sitting furniture. Even the very concept of the living room itself.
I don't know about other people but I don't do these things anymore. I have no living room. I can be comfortable enough on my bed in the bedroom with a tablet. If people visit me they are welcome to sit on the bed or the carpet. I still wear elegant clothes, because I cannot entirely replace my head at 47 but I think the 27 years olds will just never own anything more elegant than jeans.
So I think we are quickly approaching happy digital nomad lifestyle, happy because our goods satisfy our biology.
For example "tasty food" is also a biological concept. And I think for my purposes cheapish Asian takeaway is very close to the potentially ideal price/quality ratio.
Yeah, I have also noticed how the priorities have changed compared to my parents' generation. My home is there to make my family comfortable, not to impress my neighbors.
That said, before internet people visited each other more often...
So, in short, robots save the world by depopulating it? 90% of humans dead?
Funny how we're already planning for this outcome, isn't it?
It does sound something like what we have currently going on, in that the economy (of whatever country you like) is supposedly going "line go up" yet people feel that they are worse off.
Big companies are doing great because stock prices are going up because people are throwing money at them in hopes of AI. That makes line go up for stock market, which is how we measure "and how is the economy doing?", at least in a naïve way.
But for Tom, Dick and Harry who are not big institutional share holders but working joes who may even be staring down the barrel of redundancy, they're not doing great even with all this money sloshing around.
Whatever about the USA, the Irish news is full of "this store closing down, this chain moving out of the country, Apple etc. announcing global cuts to workforce" so even while the economy is okay (on paper at least), people are not feeling secure and they are seeing food prices, for one, going up while wages are not keeping pace:
https://www.rte.ie/news/business/2026/0209/1557488-worldpanel-by-numerator-supermarket-figures/
And it doesn't help that we've got all our eggs in one basket:
https://www.rte.ie/news/business/2026/0219/1559232-ifac-corporation-tax/
"The State's fiscal watchdog has estimated that almost half of the Corporation Tax collected by the State is paid by three multinationals.
In 2024, these large companies paid around €13 billion or 46% of the tax, according to the Irish Fiscal Advisory Council.
While the organisation has not named the three multinationals, they are understood to be Apple, Microsoft and pharmaceutical group Eli Lilly."
You can see that if they pull out or even reduce workforce here, we're toast.
What is the end result? No idea. The notion is that new tech means creation of new jobs in hitherto unexpected ways, but if AI is going to do everything because it will be *sooooo* cheap and faster and better than humans, where are the replacement jobs going to come from? There's only so much washing we can take in from each other.
I don't think AI *is* going to be this universal replacement, but I can see automation taking a lot of whatever manufacturing jobs remain, and this time round there will be white collar workers pain as well. CEO type jobs probably okay, world expert in paperclips type jobs okay, middle professional jobs maybe not so okay.
I think this is one of those situations where the abstraction of money conceals more than it reveals. Ignore the money. Pretend, for a second, that the money doesn't exist.
Somebody owns the machines that make necessary things. Somebody owns the machines that harvest raw materials, somebody owns the machines that refine them and process them and produce steel and copper and fuel and food and computer chips and cars and buildings any everything else--whether luxury or necessity--that the economy runs on. Presumably, somebody also owns the land where the various raw materials have to be harvested, and can also trade access to that land for other valuable things.
If you own some machines that produce some necessary/valuable goods, you can choose what happens to those goods. That's wealth. People with control over the disposition of goods have wealth. The more goods they control, the more wealth they have.
In this spherical cow world, selling just your human labor can no longer produce wealth of any sort, because machines can do it all better. So the only people who have wealth are the people who own machines (possibly indirectly, through stocks). That might be a small subset of people, but that's certainly not zero people. Maybe the CEO of McDonalds has some wealth or maybe they don't, but the CEOs of Google and NVIDIA and Dow Chemical and Toyota and probably a lot of large landholders still have some wealth. This wealth exists independent of the currency it's denominated in: currency just makes it easier to trade between various sorts of wealth, it doesn't create wealth.
And to predict what happens from there, we need to know their motives. If all of the wealth holders are perfect examples of your amoral homo economicus, then they just have their machines produce them food and mansions and yachts and gold plated toilets and medicines and they eat caviar and drink champagne while the proles starve somewhere far, far out of sight.[1]
If they're of a more philanthropic bent, maybe they use some tiny fraction of their vast wealth to replace the government redistribution (that isn't happening in spherical cow world) with a vast, globe-spanning welfare state. At that point, whether this world is darkly dystopian or kind of OK, or maybe even halfway decent depends a lot on just how vast their vast resources are, and just how stingy they are with them.
[1] Maybe they hire some small fraction of those people to do human-specific service and entertainment stuff, like putting on live plays or whatever.
This is a very good response. I would point out though that one of things that adding money back into the mix does is create legible prices for everything. Prices are the organizing force in the market economy without which the owners of all these fabulous machines might not know what to produce and therefore produce and control less wealth. In the event where most ppl have no purchasing power do we have enough prices to organize activity around?
So a slightly different concern is if you cut so many ppl out of the economy not only is there an inequality problem you risk relying on central planning, not markets and then productivity falls even if the tech is better.
It seems that this should be self limiting. If the machine is off doing useless things a human can jump in and do useful things and now someone has purchasing power.
Assuming the AI isnt also smart enough to make central planning problems disappear I struggle to envision the exact market driven equilibrium in the extreme spherical cow scenario. But practically speaking, thered always be enough petty tasks to maintain the essential features of a market economy
Ownership is also an abstraction. It's just ability to defend whatever you own from being taken(either by your own power or through alliances, systems, or in most cases, the state). And robots that make things will not be able to defend you from an angry mob. And governments are not going to defend your ownership for you if it leads to mass starvation.
Of course you could have robots that kill things but I really doubt the governments of the world will just suddenly be OK with massive privately-controlled robot armies.
I'm glad I checked the existing replies first because you hit the core points I intended to talk about. This is how wealth works in modern market economies, and it's also how wealth worked in pre-modern societies. The dominant form of wealth changed with the Commercial and Industrial Revolutions, from land and goods to ownership (mostly partial ownership) of businesses that make or distribute goods and services. And being owned money, goods, or favors by people who can reasonably be expected to make good on their debts in some way has been a significant form of wealth throughout.
See some of what Brett Devereaux has written at ACOUP about the role of "big men" in non-state agrarian societies. I've also heard somewhere that "giver of rings" was a common kenning used in Norse sagas to indicate a particularly wealthy lord, whose wealth was indicated by the ability to reward his followers with rich gifts.
>So the only people who have wealth are the people who own machines (possibly indirectly, through stocks). That might be a small subset of people, but that's certainly not zero people.
I might quibble with details of the formulation, but in broad strokes I mostly agree with it. My quibbles are that land will still be valuable; your "possibly indirectly" needs to be read very broadly to include debt, favors, and influence as well as stock; organizational capital, particularly in the form of local human knowledge, will remain valuable for some time before everything is fully automated; and labor will still have some diminished value (and some trade will remain possible due to comparative advantage) unless and until automation is advanced enough to fully solve the scarcity problem. The last bit is analogous to how people are still able to live and work in pastoralist, subsistence agriculture, or even hunter-gatherer communities despite modern industrial society being a thing. These communities are extremely poor compared to their modern-economic-system neighbors, but their mode of production still works after a fashion.
> If you own some machines that produce some necessary/valuable goods, you can choose what happens to those goods. That's wealth. People with control over the disposition of goods have wealth. The more goods they control, the more wealth they have.
This is an excellent formulation, thanks ! That said though:
> ...while the proles starve somewhere far, far out of sight.
Historically, this does tend to happen a lot, but not to the extent that you claim. Modern peasants live in unimaginable luxury compared to Medieval barons; this is partially due to the welfare state, but also partially due to all technology being interconnected. It would be almost physically impossible for the rich people to control and maintain their machines without making at least some benefits of those machines (e.g. ubiquitous cellphone coverage) available to all. Thus, in the spherical-cowbot world where robots do all the work, we can still expect some nontrivial amount of the luxury to filter down to the lower classes.
> Historically, this does tend to happen a lot, but not to the extent that you claim.
Because historically there were natural limitations preventing from completely excluding proles from the economy. Full automation of economy removes these limitations.
> It would be almost physically impossible for the rich people to control and maintain their machines without making at least some benefits of those machines (e.g. ubiquitous cellphone coverage) available to all.
I'm sure it would be great to have cell phone coverage while dying from starvation.
> Full automation of economy removes these limitations.
I don't know what "full automation of economy" could mean, other than perhaps "converting the Earth into computronium", which, granted, could potentially be hard on people.
> I'm sure it would be great to have cell phone coverage while dying from starvation.
Yes, it's because cellphone coverage is effectively ully automated yet food production is not.
>I don't know what "full automation of economy" could mean,
It means the spherical cow world that Scott has asked about. People still live in it, but they can no longer offer labour to compete with the machines, like at all.
Ok, so are the machines performing that labor or not ? If not, then everyone starves -- except they don't, because humans will fall back on farming their own food, i.e. performing labor. But if the machines are performing the labor, and doing so at much cheaper rates than humans, then food is essentially free and no one starves.
> I'm sure it would be great to have cell phone coverage while dying from starvation.
By the way, I think some parts of Africa are kinda like that today, so we are not talking hypothetically.
What is with your complete refusal to believe that useful, self-sustaining robots can be built? I find it strange at this point that you're willing to humor AGI in these thought experiments, but not robots that can build more robots.
Eh ? When did I say that they could never be built ? Granted, I don't think they'll be built anytime soon, but that's a different story. If we accept AGI for the purposes of the argument, then sure, we can accept self-sustaining robots too. I don't see how this invalidates anything I'd said though.
Why do you think demand would completely disappear? There would obviously still be demand by other businesses (who are also all being run by AIs), so you (or more specifically, the AI running the company) just change production to actually serve the demands of other productive beings. So... not humans. An economy can still be fully functional without any humans in the loop.
"There would obviously still be demand by other businesses (who are also all being run by AIs)"
My take is that a business needs things to allow it to produce OTHER things. There must be a consumer at the end of the chain, which is currently an individual or a state (e.g. public buildings / services, military, or healthcare in Europe / Canada).
But right now productivity is tied to "demands of humans for things". Take humans out, and what are the AI and robots producing? Medicine for whom? Goods for whom? If the proles can't afford to buy the new cellphones, what good is a warehouse full of the latest bells and whistles model of phone?
An entirely AI economy will have to switch to a new model of goods and services, and what will that be? Sure, chips for the data centres, energy generation, making parts for those factories that produce the machinery for data centres and energy generation and the likes, but what else?
I’ve been in towns in the US where the local hospital is the main engine of economic activity, because healthcare is subsidized by the government and so it’s not just a healthcare program, it’s also an economic stimulus package and a jobs plan. But it’s an open question whether a super productive future economy would have better welfare and healthcare for ordinary citizens, or if the authorities decide that ordinary citizens are no longer needed.
> Sure, chips for the data centres, energy generation, making parts for those factories that produce the machinery for data centres and energy generation and the likes, but what else?
I think it's conceivable for a hypothetical AI economy to consist almost entirely of this short list of things. After all, for thousands of years, human economies were almost exclusively focused on an equally short list: food, water, clothing, and shelter, plus equipment for acquiring the above more easily.
Which parts of Kabbalat Shabbat do you do?
New Russell conjugation from Anthropic:
I am transforming publicly accessible data. You are leveraging model-generated feedback. He is engaging in a distillation attack.
The bio anchors idea seemed really dubious from the start. Like yes they could help provide some evidence toward some conclusions, but would anything like this have worked for other technologies? Bio anchor steam engines vs horses, metal strength vs bone, power plant output vs biologically powered things.
If it makes you feel better, I for one am certainly a happier person with crackers than without.
I enjoyed reading the Harper article. It makes sense when I realized it was written by Sam Kriss.
Has the opening of the Epstein files provided any evidence that anyone actually had sex with an underage girl?
There is an email in the files about a US ambassador to Mexico impregnating an 11 year old, at a party organized by Epstein.
https://www.ibtimes.co.uk/epstein-files-us-ambassador-allegations-1780761
As the article stresses at length, not everything in the files is the unadulterated truth, so don't shoot the messenger.
Yes, there's a letter to Epstein that reads:
"New Brazilian just arrived, sexy and cute, =9yo"
(https://www.justice.gov/epstein/files/DataSet%2011/EFTA02706746.pdf)
Ignore the "=" sign. I'm sure it doesn't mean anything. EDIT: Please see @Savio's comment about this below.
the equal sign is a formatting error from improper processing of line breaks, and it effectively overrides one character. This is really likely to be 19yo given your followup comment.
See https://lars.ingebrigtsen.no/2026/02/02/whats-up-with-all-those-equals-signs-anyway/ for the exact details.
You can see the evidence to support this in the very same email, just a few lines below:
Envoyé: jeudi =7 janvier 2013 17:22
This is French for "Sent: Thursday =7 January 2013 17:22"
Here again you see the unusual = sign. But wait: 7 January 2013 was a Monday, not a Thursday. 17 January was a Thursday. The = replaced the 1. Almost certainly the same happened with =9yo above.
You're defending pedophiles. The = signs are actually a secret code for arranging the sale of the victims.
Savio has cited persuasive evidence for his position, and it seems to me quite plausible. You have cited no evidence for yours, which seems quite implausible. I mean, if you're peddling underage flesh, wouldn't the *first* thing your Secret Code obscures be the age?
That's why they send two almost-identical emails simultaneously, as I noted in another comment in the thread. This one https://www.justice.gov/epstein/files/DataSet%209/EFTA00659212.pdf, which says "19yo" obfuscates the age, and the other one https://www.justice.gov/epstein/files/DataSet%2011/EFTA02706746.pdf with the mysterious = symbol is a still-unbroken cypher communicating secret Satanic billionaire pedophile deals.
You really find his "CRLF" "quoted unreadable" "content transport encoding" technobabble persuasive, and not obvious schizoposting?
Given that the two emails are identical in subject, attachments, most of the content minus one character, and send time down to the same minute, plus outside the emails an explanation of how the equal sign gets created and the fact that DOJ does occasionally duplicate information in different releases of their datasets. It is much much more likely that these are the exact same email released differently.
I do not have an opinion on whether anyone in the Epstein files actually had sex with an underage girl, as I honestly don't care to look through the files at all and I follow very little on the Epstein file news. Saying that you may have made a mistake != I dislike you != I am defending Epstein in general.
Please calm down a bit and don't take my argument as a soldier[1] on the Epstein side.
[1] https://www.lesswrong.com/w/arguments-as-soldiers
The pizza and grape soda thing sure seems like a euphemism for something, too, and I doubt it was anything to be proud of. The sender was some doctor nobody's ever heard of though, if I recall.
Ten years later and we're back at Pizzagate again? Hoo boy.
Yes, but it gets even better: the new consensus is that Pizzagate was actually a decoy fake conspiracy theory planted to get people to dismiss the REAL pedophile sex ring/Satanic cabal that Epstein was part of. Open your eyes, sheeple!
Life has more layers than you know. The pedophile sex ring/Satanic cabal, while real, were set up to distract the rest of us from recognizing that these people are basically our species' dandruff and halitosis..
> these people are basically our species' dandruff and halitosis..
What is wrong with you, joking about something like this? This is a serious issue that has ruined the lives of thousands of young girls, and you're disrespecting their memories by making up deranged rumors about their very real abusers, which will only make people less willing to believe the truths of the victims. You may think it's funny, but it's doing real harm, whether you intend it or not.
Please take care to treat this sensitive issue with the care and tact it obviously deserves. Thank you.
Seems legit.
And be sure not to confuse it with a somewhat similar email sent around the same time:
"New Brazilian just arrived, sexy and cute, 19yo"
(https://www.justice.gov/epstein/files/DataSet%209/EFTA00659212.pdf)
That one's entirely unrelated.
Maybe they were sisters?
Yeah, Melvin, they were *your* sisters. Actually, one may have been your gender-nonconforming brother.
Anyone who's intending to attend the upcoming Jerusalem rationalist community meetup - it is *POSTPONED* by one week because of the weather, and expected to happen in March 6th. RSVP on lesswrong and get notified on changes directly in your email without having to scan dubious comment sections!
https://www.lesswrong.com/events/wyYLysPC2sixFczis/jerusalem-rationalist-community-meetup-1
Alexander T. as part of making a different point down below here, noted that "net domestic migration into the sunbelt is slowing."
Huh, that's interesting....I found the Census Bureau data release which sparked some MSM coverage late last month.
That data is the Bureau's annual estimates of US populations by state and county, for the period ending July 1 2025. Those annual estimates are not a census [an actual count of people], rather they are built from various other datasets [births and deaths, etc]. They do sometimes get corrected by the actual decennial censuses. They are generally good enough though that various sectors use them for business-analysis purposes and the like.
In this case the Bureau, regarding the July 1 2025 estimates, says: "The Midwest was the only region where all states gained population from July 2024 to July 2025. In addition, after experiencing population decline in 2021 and small growth in 2022, the Midwest’s population grew solidly in 2023 (259,938), 2024 (386,231), and 2025 (244,385)....From July 2024 through June 2025, the Midwest also saw positive net domestic migration for the first time this decade...."
I looked through the state by state data on domestic migration. Two things are at the moment both true:
(a) The overall arrow of net migration within the US has not flipped. CA and NY still have the largest state annual net losses, TX the largest annual net gain, states like NE and IA and VT still have virtually no net domestic-migration change; the Carolinas are still gaining strongly while places like LA and AK and NJ and MD keep having meaningful annual losses; etc.
(b) The overall movement from blue to red states has indeed been slowing down since actually 2023. CA through 7/1/25 had its 4th straight annual decline in net domestic-migration loss; NY for 2025 had less than half the net loss of either 2021 or 2022; TX had its lowest net annual gain since 2020 and FL its lowest since sometime before 2020; WI had its 4th straight year of net domestic gain; MN and MI each had their first domestic net gain since sometime before 2020; VA had its 2nd year of net domestic gain after three years of losses; IL had its lowest net domestic loss since 2020; WA had its 2nd year of net domestic gain after three years of losses.
Re #4 - I was interviewed once and what was reported in paper had the same sort of quality. Reporter didn't listen, had an agenda for the article, fit my words into that agenda, reworded what I said where necessary to fit the agenda. What was printed was a totally different conversation than the one we had.
As the old saw goes, however much you hate the media, it isn't enough.
I have the same experience, twice. In one case they also misspelled my name, so I could pretend that it wasn't actually me. In the second case I was smart enough to not give them my full name, so they only made an idiot out of an unspecific "Viliam".
The first case was not even political. I was a gifted child, and the reporter had some vision of "all gifted children use computers", so she kept asking me about the computers, and I kept telling her that I don't have any, and that my family is not rich enough to afford one... but she wrote an article anyway about how I am a computer genius and that I have figured out that my parents have already secretly bought me a personal computer for Christmas (completely fabricated).
The second case, there was a political debate that somehow involved collective guilt, and at some moment I said "it's simple: if we like some group, we insist that they are all individuals, and none of them should be judged for the crimes of some other individual; but if we hate the group, we see them as a uniform mob where everyone is guilty of everyone else's crimes, so it is fair to punish them randomly". The journalist was surprised and said "this sounds very smart, can I quote you, what is your full name?", and I asked "can I get the text to verify after you write it?", and she said "no, were are too busy for that, the article needs to be published immediately after writing, there is no time for feedback", and I said "sorry, then you can't have my full name". (The published version was: "student Viliam said that we shouldn't judge Palestinian terrorists for their crimes".)
I am not even an important person for someone to have a grudge against me or something. I think the constant lying is just a habit; the people who are not constrained by truth generate better stories.
Well, Viliam, that was an excellent play. The old "omgwearesobusy.....mustpublishimmediately" is more proof that their intentions are malicious. If they cared about the truth, time to get things right would be important. But we know they don't.
I always disciplined my children for lying to themselves. "If you lie to yourself, you will lie to anyone." I think we know the source of the media's problem.
I've had good results talking to journalists, including even the NYT, in my area of expertise. But I was speaking on a technical subject that wasn't politically controversial and so nobody really had an "agenda" in the sense you're using.
What is the worst twist in all of fiction? Beyond the various "it was all a dream"s, beyond "you the viewer are the killer."
I've been saying it was the Zero reveal in Zero Time Dilemma, but I'm wondering if that's true. The only other two I can think of that come close are the orphanage scene from Final Fantasy 8, and the reveal at the end of Down With Love (which arguably is the same class of twist as the orphanage scene from Final Fantasy 8).
The plot twist in "Ender's Game" always struck me as a bit lame.
> The plot twist in "Ender's Game" always struck me as a bit lame.
Lol, waddya mean?? It's perfectly plausible!
Imagine a family with 2 kids that are so smart they can root the national discourse and steer politics as teenagers. Now their brother, who is even smarter than THEM, goes off to battle school, lollerskates over everyone, then gets fooled into thinking that the battles he's leading against progressively farther out enemies with progressively older and weaker technology are a sim rather than reality. Then he's shocked and horrified to realize those battes were actually real all along.
Why sir, I am shocked, SHOCKED I say, that you would do this to me! I've literally only spent my entire life training for this, feel like the earth and my family is under existential threat, and would have happily led them even if I knew they were real battles, but I say, sir! Shocked!
The one at the end of KIng's The Dark Tower series. I don't want to say here what the twist is, though, because somebody on here today is probably reading that excellent (til the very end) series.
The ending was okay, but I was fascinated by the little blurb King gave the reader *just before* the ending. King basically subjects his audience to the same "19" trick that the Man in Black did with the woman in the tavern (name escapes me) in Book 1.
The series' ending was entirely within the narrow distribution of possible endings that are consistent with the whole philosophy of ka. It's a good ending.
Also while I hate to reply to the same comment twice I just noticed the diss on the Down With Love twist.
I love that twist! That twist justifies the whole existence of the movie, and flips the whole genre, turning it from a straightforwardly milquetoast parody of the source material into a ridiculously over-the-top parody of the source material. I remember seeing that movie in the cinema when it came out, and how Renee Zelwegger's monologue transformed my entire viewing experience.
Deeper literary roots than ff8, the reveal in mousetrap depends on a sister not recognizing her brother. She didnt have junctions to blame.
Bridge to Terabithia. Most of the story seems to be a fairly straightforward coming-of-age story, then the secondary main character dies suddenly in a senseless accident and the rest of the book is about the perspective character coping with grief and survivor's guilt. I was assigned the book in elementary school, and I have never forgiven the author nor the teacher who assigned it without a word of warning about what was coming. The entire book feels like a cruel prank at the reader's expense.
The 2007 film adaptation is even worse, since it was deceptively marketed as a Naria-esque children's fantasy story.
That makes it the perfect coming of age story. Part of coming of age means learning and accepting that people die, sometimes in senseless accidents. I was assigned that book in elementary school too, and shortly thereafter, the sister of one of the sixth graders at the school was shot to death in a senseless accident. The sixth grader definitely didn't get a spoiler alert about that!
I was already viscerally aware of people dying in senseless accidents. Part of why the book hit me so hard is because my aunt had died in a car accident a year or so previously, so the stakes of the book felt very real to me.
One key difference between senseless accidents in fiction vs real life is that the former has a large element of intentionality that the latter does not. The author chose for the senseless accident to happen in the story and chose how to present it to the readers.
The element of choice on the reader's side is also important. I get to choose when and how to read stories, and now that I'm out of school, I get to choose which stories to read. I can enjoy a good tragedy, but I prefer to know that I'm reading the sort of story that is apt to include tragic elements. I find the bait and switch of a story that is presented as being light and upbeat but suddenly turns dark and tragic to be rather obnoxious.
Omg my ten year old and I checked out the DVD from the library on a rainy day as it looked wholesome and harmless. She was traumatized by the movie and I was traumatized that I let her watch it. I will say that was the least expected plot twist I have ever encountered.
She's ten. That's almost old enough.*
*Old enough for what? Living by yourself. Speak to a holocaust survivor about "How old a child needs to be, before they can live on their lonesome, without parents, and working for a living"
After all, tomorrow is another day. Gone with the wind.
What makes you say the Zero reveal was bad? I'm rather indifferent to it, but that's not the same as saying it's poorly executed or dumb or whatever.
Spoilers I suppose.
So Zero Time Dilemma is the third of the Zero Escape games, which all revolve around nine people trapped in some kind of death game, with one of them secretly being the game's orchestrator, Zero. There's a bunch of branching paths, you get to know all the characters, and finally once the paths are exhausted you hit the True Ending and one of the nine is revealed as Zero, to lay out their whole motivation.
In ZTD, you get a bunch of branching paths, get to know all the characters, and then Zero is revealed to be... a tenth person, who the player has not seen before, but is said to have been there in the room the entire game, a few feet off camera, with the cast knowing he was there and just not mentioning him to the player. I can't think of a worse way to end a mystery. Including just not revealing anything, which I've seen a couple of books do.
It doesn't help that his plan includes telekinetically shooting himself with a shotgun, and dying. His Motives Are Complex.
More spoilers - don't read this if you want to play the game.
------------------------
They're fairly consistent in not calling Sean "Delta" at any point, right? Obviously they rely on implication, but at no point does anyone actually do so. There's also other clues like the 4th collapse whenever that team gets knocked out. And there's the shadow in the late game.
I understand your viewpoint, but I don't think it's bad per se.
I'm pretty sure they're consistent about it. There a point where... Shotgun Guy is aimed at... uh... Kid, saying "if not you then who?", and if you type in the supposed name of the kid, as in "I did do it actually", the guy shouts "that's impossible," and then shoots him anyway.
But this is a level of scrutiny one should never have to apply. The group menu shows the kid with that name, like all the other groups do with their viewpoint character. The whole setting is about trying to gather information, and yet the viewpoint character never even looks at one of the players. They discuss the kid's helmet (...I think), but not the old man's wheelchair. After the reveal, what's to stop there being another, eleventh person no one mentioned?
I'm reminded of the joke in Chuck that killed my interest in Chuck, where... uh... Sports Guy is driving... uh... Weird Guys to Las Vegas, and lists off all the things they're going to do once they get to Las Vegas in gambling-and-sex lingo. And then he takes them to "Las Vecas Wildlife Preserve", and painstakingly explains all the gambling-and-sex lingo he used was actually wildlife-and-camping lingo with completely different meanings. It breaks so many implications about the people and the meaning of language that it renders the entire setting as noise. No information can ever be relied upon again.
Did you also dislike the twist in VLR where the MC is revealed to not look like how we've kept imagining him this whole time? Iirc, there's no reflective surfaces that they come across, and no one else comments on it either.
As for Chuck, the show is a comedy. Why wouldn't that kind of misunderstanding be valid to apply? It's not like it was for the highly erudite or whatever.
Again, not saying that you're wrong to dislike it, but I don't think your dislike means that it's bad.
VLR's didn't bother me as much because it's the character doing it. It's not the characters in ZTD, it's just the audience.
(Also I do that one in daily life. Mirrors are a bad time.)
The difference between the Las Vecas joke and, like, the one where the Weird Guys are too stupid to know what the CIA is, is the investment. There's a whole scene of driving toward Vegas beforehand, and after the lingo explanation it's clear there's no amount of assurances, logic, or time investment that will prevent the show from pulling the rug out beneath you. Which in turn means no amount of logic or time should be invested.
I guess the twist summary is "Everyone Knew It Except You The Audience."
Andrew Crumey's 2004 novel "Mobius Dick". Basically a novel about quantum mechanics and parallel universes, in which a physicist discovers a parallel universe in which the Nazis won WW2. Schrödinger is involved somehow, and so is Schumann, and I think that the parallel universe finds out about the original universe as well at some point.
I don't remember all the details of the book, but what I do remember is the ending where it turns out there's a _third_ universe which is (bum bum bum) our universe, and that the characters are so lucky that they don't live in that universe because George W Bush is president there, and the ending just turns into a rant about how terrible George W Bush is.
I feel the same way about Stross's Family Trade series. Magical family that can move themselves and what they can carry between universes? Ok. Using it to move drugs? Ok...
And then the last book goes full into "and the real-world US government is genocidal Nazis who nuke all the other universes because they're evil Republicans who hate drugs." Yah. Really. Or at least that's what I remember the anvilicious message being--Republicans are evil Nazis, hate Republicans.
"What if the Nazis had won the war, Lemon?"
Ooh, that's a strong one. But how tied to the resolution was it? The main difference between FF8's twist and ZTD's is that FF8's is isolated enough it can be ignored without losing the rest of the story.
Was it like, "Now that the plot's resolved, look over there at George Dubya," or was it more like "all our plotlines are trivial in the face of George Dubya"?
Zero Time Dilemma was so disappointing that I cannot even remember anything about the plot (and I loved 999 and Virtue's Last Reward)
“I will open the crackers so you will have crackers and be happy.”
This raised my already considerable admiration of you up three notches
I have never once been to a party where they run out of cheese before they run out of crackers. There's always a massive pile of cheese left and nothing to eat it with.
If the rationalist movement cannot solve this then what good is it?
"I have never once been to a party where they run out of cheese before they run out of crackers. There's always a massive pile of cheese left and nothing to eat it with."
Is that supposed to be "There's always a massive pile of *crackers* left and nothing to eat it with."?
Two days ago, I was at a picnic where the cheese ran out first
If only.there were a cheese-loving inventor somewhere...
It’s a very rational statement on its face.
It was in that moment of opening the package of crackers when Scott Alexander truly comprehended the essence of Universal Love and the meaning of Transcendent Joy. Then he immediately forgot about it, and that was that.
I'm interested in what people think that improvements in coding AI will look like.
Even plain coding assistants are currently extremely good on greenfield projects and can 1-shot not-particularly-innovative projects that would take me 5-15 hours otherwise.
However, on brownfield projects, both in my experience and from all programmers I talked to, they are rarely useful outside of self-contained modules.
Coding agents seem to push doable greenfield projects from 5-15 hours to 20-40 hours (which is huge!) but usually end up in a mess that cannot be maintained by either agents or humans. I haven't heard of how well they do in brownfield projects but I doubt they have significantly useful since, IMHO, the main barrier is that LLM performance craters with increasing input size (for coding). Agents seem to do better context management and that helps a lot but I doubt it's enough for large code bases.
There is also the issue that building any project usually changes its trajectory. As you build something you realize what is missing, what is useless, what is not needed and what is suboptimal - this will not happen if you automate building it. I assume this is a huge factor why there are barely any (maybe even none?) quality vibe-coded projects. There are no shortcuts.
However, the main limitation to usefulness of AI coding assistants is simply conservation of information. When you want assistant to do something you need to put that information into the context window or you need to reference information that exists in LLM itself. To bring your vision to life you need to specify or reference ALL the information that it contains (all the thousands of small decisions and hundreds of large ones that shape any system that is not trivial). If you have actual vision of something, you can't offload those decisions to LLM and you still need to put that information into the project. LLM is unlikely to give you what you actually want since if you put all the information needed into the LLM it would act as simple natural language -> programming language translator (this is largely not useful for seasoned programmers). On the other hand, when you reference information from LLM you lose precision and your vision is unlikely to come out the other end.
The above seems like a fundamental tradeoff with coding assistants which would remain even if LLMs improved immensely. You either have to specify things so precisely that there's little point in using LLMs or you don't get what you want.
> they are rarely useful outside of self-contained modules.
that is true for humans as well. I think AI has already surpassed a bad programmer, but good programmers are still safe.
> I'm interested in what people think that improvements in coding AI will look like.
With AI it is very important to know their limits, e.g. for me they are especially bad dealing with spring upgrades. And it's important to always review their work, e.g. for every good idea they generate, they also generate a bad one, and the challenge is to look at the code they produced, and understand the bad ideas, so you can fix them before they do any harm. This is very hard.
The biggest productivity gain I got from AI was work, that was especially well suited for AI, e.g. comparing two different versions of a large number of XSD. Another big productivity gain is when they generate a good idea, that is easy for me to validate.
For my brownfield projects, I just try to use AI for most of my issues, but then I give up soon before doing the work myself.
For my greenfield projects, I try to set the project up in way, so I can leverage AI for a longer time before I hit the wall.
I've had pretty good results with having claude first document the brownfield application with md files describing usage and then instructing it on implimenting modest updates and fixes. You're right that controlling the context is the key skill to use, and particularly subagents are important here, but even if you just give it the ~5-6 files that are relevant to most changes a single agent has usually been able to do what I need. If you're expecting to be able to just write a jira ticket and have the thing run the whole thing from branch creation to pull request then yeah, we're not quite there yet unless you're willing to build a harness, but we're not that far off.
Just these days I'm setting up things at $WORK for agentic coding. My first attempts were pretty crap, it would write some barely functional thing that matched none of the style and didn't handle errors. It's still pretty hard to make an AI coding harness 'understand' a medium-large codebase and learn to write the code you would actually like there. Any project that has run for more than a few years also has layers using different conventions or technologies. Our stuff is more than old enough to contain jQuery parts.
So I just spent a couple afternoons using AI itself to read through good source and write its own documentation: AGENTS.md, and smaller docs for specific areas like DB, front-end, etc.
I just had the AI port a small piece of functionality from one language to another, integrated into the system. It read all the docs, and then it got it right the first time, for about $1 in tokens.
So I'd say it's looking not too bad, but it takes preparation with your project. That said, I'm not committing anything into the system without reading it line by line myself.
In observing my peers at work using AI coding tools, the three major pain points for me are:
1. AI coding tools make it much, much easier to write bad code. This leads to more bad code being written.
2. AI-written PRs tend to be less legible and cause more code churn than human-written code with the same objective.
3. Using AI coding tools tends to launder the signs that you are trying to solve the wrong problem. I think this is more-or-less what you're talking about with "There is also the issue that building any project usually changes its trajectory..."
4. If the AI did the thing, I am left with no clue how to accomplish the thing unless I put in deliberate extra effort after the thing is notionally done.
This does not play well with traditional "here are some small tasks for you to learn your way around everything" onboarding approaches if they are coupled with "oh btw please make an effort to use the in-house AI for as much of your work as you can"
Personally, I like writing code and take pride in my craft.
The only time I have been tempted into using AI for code generation was out of petty spite, I wanted to see if even an LLM could do a better job than a colleague.
I think that bad code should hurt. If it was up to me, some programmers would lose their copy-and-paste privileges, perhaps if the alternative is retyping everything they might find a way to avoid their redundant boilerplate.
I agree that AI will likely exacerbate bad designs. One of the virtues of humans is laziness, after all. If I realize a design choice I have made will require a lot of work to implement, I might decide to spend my time on finding a nicer solution than implementing the ugly one. ("There are no boring problems, just insufficiently generalized ones" and all that.)
There's natural laziness, *and* there's the fact that often, the incentives point to speed and volume over quality.
Alternatively, a human assisted by AI can edit code much faster than the human alone, so once you become confident enough in the system and know how far it can go, you can tackle refactors and pay off technical debt semi-automatically.
There’s a school of thought in Democratic politics that the way back to power requires appealing to working-class whites via economic populism, being pro-union, and demonizing the so-called “Epstein class.” These people would love Austin Tucker Martin, a 21-year-old Christian and former Trump supporter who worked at a golf club and tried unsuccessfully to organize a union. Alas, he won’t be voting Democratic this November, on account of the fact that he brought a shotgun to Mar-a-Lago and was shot dead by police:
https://www.tmz.com/2026/02/22/mar-a-lago-armed-man-shot-dead-texts-show-epstein-files-fixation/
Martin “became fixated on Epstein following the latest release of information tied to the files. Co-workers tell us he was deeply disturbed by what he believed was a government cover-up and often talked about powerful people "getting away with it.”” He’s obviously not the sharpest tool in the shed, but I haven’t seen anything to suggest he was mentally ill. I consider him a victim of our increasingly unhinged political system, where smart people encourage stupid people to indulge their worst instincts. Hopefully, after Trump leaves the scene, we can get an elite consensus not to engage in this kind of politics going forward.
> There’s a school of thought in Democratic politics that the way back to power requires appealing to working-class whites via economic populism, being pro-union, and demonizing the so-called “Epstein class.” These people would love Austin Tucker Martin, a 21-year-old Christian and former Trump supporter who worked at a golf club and tried unsuccessfully to organize a union. Alas, he won’t be voting Democratic this November, on account of the fact that he brought a shotgun to Mar-a-Lago and was shot dead by police
This seems a rather uncharitable way to frame it. (Apart from implying that some Democrats care about working class qua working class, which seems a charitable but mostly untrue thing to say.)
MAGA rose to power on the back of crazy conspiracies, they were fine with pushing QAnon when they claimed that Trump was fighting the pedophile adrenochrome cabal. Not the smartest move, in retrospect, as their leader is a sex pest who also was the best buddy of Epstein, given that Epstein liked to fuck kids and probably trafficked underage youths to other men.
While some Democrats skillfully stoke the anger of the MAGA base over Trump's zigzaging about the Epstein files, their base cares more about ICE.
>probably trafficked underage youths to other men
On the basis of what evidence?
It doesn't help that they clearly have "gotten away with it": there were various rich dudes that pretty indisputably had sex with various sex workers/trafficked minors, it's not clear, and they have not and will never be punished even to the extent of social shame.
It's hard to take civility seriously when EG Donald Trump wins the office twice.
When Prince Andrew (name reverted) gets arrested on his birthday, I don't think you get to say that folks aren't getting punished.
Can you give specific names with specific crimes you are accusing them of and the evidence they are guilty?
Thanks in advance.
We can't, because they "Are Getting Away With It."
That's why people are so pissed, because we have a bunch of credible accusations, a dude that "killed himself" (I actually think he did kill himself, but the situation is maximilla similar to if he didn't, I get why people think what they do) and a "Full release" where all the names that mattered were blacked out and one of the main suspects decided they weren't going to release it fully after all.
All we have is some women saying, for example, "Donald Trump raped me" and some gofers saying "Donald trump raped them", but we no longer are in the believe women times and it's all he said she said until they actually release unredacted information, so that isn't evidence it turns out.
Basically, this will be another panama papers, where everyone knows and everyone knew, yet there will never be any specific names and specific crimes and the Journo who brought it to light will die in an unrelated assassination by coincidence.
>That's why people are so pissed, because we have a bunch of credible accusations
What do you think are the strongest accusations? There's another thread and the lone example posted is hardly equivocal.
https://www.astralcodexten.com/p/open-thread-422/comment/218773446
My monthly round up of long form content (plus my own podcast interviews) is up.
Interview highlights include Stetson from Holodoxa, Rev Jon Harris from the Church of money burning, and Stephen Skolnick on his frontier human microbiome work.
Long form links include the critical role of diesel and jet fuel limits in throttling economic growth, Paul Kingsnorth's lovely Wendell Berry Lecture, Asimov Press' exploration of the edge of consciousness, the impact of soy in animal feed on human health, an exploration into why the Bronze Age collapse didnt happen, and a documentary on the lost agricultural civilisations of the Amazon.
https://recombinationnation.substack.com/p/the-long-forum-february-2026?utm_campaign=post-expanded-share&utm_medium=web
> Cotra estimated “~2.5 OOM worse [than the brain], +/- 1 OOM”, based on reference points like how much less efficient dialysis machines are than a human kidney, how much more efficient solar panels are than leaves, and the FLOP/watt efficiency of a V100 GPU. But most of those anchors had little to do with where ML algorithms were in 2020 when bioanchors was written, and would have given a very similar estimate for “present state of ML algorithms” 20 years earlier or 20 years later.
> This is sufficiently interesting that I’m curious to hear from someone who engaged with Bio Anchors and forecasting more deeply than I did - did we all just miss this?
+1 on this, interested in others' thoughts! I also independently had this critique. My guess is that the best response is something like "2020 was around the first time we had good enough algorithms to scale to TAI without drastic changes. If doing the exercise much earlier, it wouldn't have made sense to use the effective compute abstraction, and if doing the exercise much later, we would have recognized that our algorithms were more efficient and made an appropriate adjustment." I'm not sure I believe this though, especially the latter claim.
Very stupid question re: the replacement for feeder mice for pet snakes:
I thought snakes would only eat live prey/animals that show signs of life by moving and squeaking and stuff. If that is so, how can you replace (live) feeder mice with something that is not alive? How do you convince the snake to eat it? Also, for constrictors, don't they need to feel the "snap" of bones as they crush their prey? That's more easily solvable, by sticking something hard that has the consistency of bones into the Feeder Mouse Replacement Nutritional Pellet(TM).
Maybe the device that delivers the pellets could move in some way that satisfies the snakes’ need for prey to be in motion.
I'm not a snake person but I'm friends with snake people and as a rodent person I've been involved in quite a few of the arguments about live feeding. My understanding from snake people I trust is that if your snake only live feeds you've done something wrong and live feeding is pretty dangerous to your snake. I've seen the sentiment from snake people quite a lot that any snake can be trained to eat dethawed mice/rat pups.
I don't say this normally when having this discussion but since this is a rationalist/utilitarian corner of the internet. Considering the amount of live rats/mice a snake needs to eat, the intelligence and nature of rodents, the lack of intelligence and the nature of snakes, and the very marginal benefit of letting people keep snakes as a pet specifically. The end result of a conversation where its acknowledged that live feeding is necessary should probably be a ban on snake ownership outside of zoos. Its like the vegan debate on steroids, allowing an even less regulated factory farming environment for one of the most intelligent and empathetic mammals so that they can die in an extreme state of fear and pain to please a pet animal with a simple reptile brain. You can make the argument that dead feeding is acceptable (pinkies are killed very young and painlessly, and mice/rats produce so many offspring that you can breed them in kind conditions) but if you accept that live feeding is mandatory you start to run into this.
Snakes are dumb,my boa would be deceived by me poking a dead pinky mouse with a chopstick.
Thanks for the detailed comment!
So for people who want to keep pet snakes, they'd be well-served by playing up the "their diet is just like cats'!" angle.
Their diet is just, like, cats
Thank you! I was waiting for someone to repunctuate that!
Nice to see punctuation matters with pythons just as much as with Python.
How about a moral claim?
Their diet is just, like cats'.
How about a philosophical claim?
Their diet is, just like cats.
I'll admit I didn't think of that one, but I like it! Thanks.
Wut? Yes, both snakes and cats are carnivores, but nobody feeds live mice to their cat. Cats eat canned or dry food, which is definitely dead at the point of purchase.
Yes, outdoor cats will kill mice and birds, but that is why many pet owners advocate keeping cats 100% indoors, perhaps providing a “catio.”
Cats will.naturally eat carion, which is why you can get them tinned food, so long as it is fairly stinky.
Generally speaking, cats won't eat carrion. It smells/tastes bad to them. in order to get them to eat tinned food, they put addictive substances in it.
THEN the cat won't eat anything else.
At least in these circles. They're quite lucky that the average person doesn't like rodents so they can get away with something that would create uproar if it was something people don't use glue traps for.
Yeah, we had a rodent problem once, and I couldn’t bring myself to use glue traps. Husband and I bought humane traps that are basically little tube-shaped cages where the mouse is trapped but not hurt, and we released the trapped mice in a public park nearby. (This happened only twice, then the mice stopped coming.)
Glue traps for mammals seem cruel. I mean, with a spring-loaded trap, one would take a chance that an animal will die slowly and with a lot of suffering, but with a glue trap this the practically guaranteed. (I do have used pheromone glue traps against insects, though. But there is some cognitive difference between a moth and a mouse.)
Life traps are great (I have used one with a tiny seesaw which flips and traps the mouse, worked fine), so long as one checks them regularly. I release my mouse in the nearby woods, so far none have come back.
I think it depends on the kind of snake. I've heard some snake-owners talking about needing live prey because otherwise their snake won't recognize it as food, while others are able to feed their snakes feeder rodents that they purchased dead and frozen.
It depends on the individual snake and the experiences it has gone through, not on the species. For instance a snake that has only ever been fed thawed rats will be content with that, having never experienced the aliveness of a live rat. Some that have been live fed will switch over to thawed, but some won't.
Harper's reporter (inner voice): "Scott seems a fine fellow, but the New York Times has dubbed him a bad person. Must ... not ... disagree ... with NYT. But how can I disparage him in some relatively innocuous fashion?"
Honestly, and this is not focused on you personally but just kind of a general take on this topic - I really DIDN'T think that Scott came across that poorly in the NYT piece on him. I thought there were a couple of questionable choices but overall it didn't strike me as some sort of aggressive hit piece.
I think it's fair for Scott to be salty about it (although not necessarily ideal for him) but I think this community writ large (again not specifically trying to call you out) should kinda get over it. The NYT is a massive organization that's done a ton of stuff and them releasing a lightly-unflattering portrait of somebody we like doesn't really strike me as a good reason to update very far on anything.
Right, as I recall, the main issue with the NYT piece was that it doxxed Scott, not that it said anything particularly scandalous.
I agree with that particular critique.
No offence taken.
I think of Scott as being incredibly fair-minded - and so when he posted, in the old SSC, about the issue with the NYT, I took his (Scott's) side.
I did feel that the NYT maligned Scott's character, albeit in a subtle wouldn't-stand-up-in-court-as-libel sort of way.
Failed to say this above, but, I did agree that the root "ruin Scott's pseudonymy" thing was BS, to be clear.
He had to leave his job over that. What makes it BS?
To clarify, I'm saying it was "BS" that the NYT did it to him, i.e. they shouldn't have, not in the sense that it was a false claim.
Yea, this. Being myself a recovered newspaper reporter I recognized the language choices in that NYT piece as, "make the readers dislike this person without explicitly attacking him/her". Been a while now so I don't remember the specifics but it was instantly recognizable.
In the Trump era the NYT piece may feel like no big deal, but in a parallel universe where America is still woke, reading that article would be enough for anyone politically savvy to avoid associating with Scott.
Plus it provides material that Wikipedia can cherrypick from, because it is a "reliable source".
> In the Trump era the NYT piece may feel like no big deal, but in a parallel universe where America is still woke, reading that article would be enough for anyone politically savvy to avoid associating with Scott.
I'm very skeptical of that assertion because that article came out during the Biden years and I worked at a series of pretty woke places and I didn't bat an eye about sharing my hobbies. And also I think most people don't really care that much. If you mean something way more specific, like, some politician isn't gonna hire him as an adviser or something, I guess maybe, but also I don't think that was ever gonna happen.
Also, not to be that guy, but like, Jeffrey Epstein. Plenty of people kept associating with him after he was already a pretty notorious figure, and they weren't all just angry reactionaries who refuse to cancel people on principle.
The media does not have the ability to ostracize people by hand-wavingly associating them with insufficiently woke statements, except in cases where people already had no power and influence. (And if they did, the election of Donald Trump would not have made a difference. Although it also wouldn't have happened at all.)
> Plus it provides material that Wikipedia can cherrypick from, because it is a "reliable source".
At least one of the most debatable inclusions from the article was a quote that was already cherry-picked, from Scott's work, and was labelled as such in that work. If you want to cherry-pick old SSC articles, there's literally a "Things I Will Regret Writing" tag. NYT has in no way advanced the state-of-the-art in selectively making Scott Alexander look bad.
> Also, not to be that guy, but like, Jeffrey Epstein. Plenty of people kept associating with him after he was already a pretty notorious figure
Sure, but Epstein was merely a paedophile, not a Republican.
Epstein was not a good material for a "scissor statement".
I remember recommending Scott Alexander to someone I knew; they checked RationalWiki and Wikipedia and said "no thanks", refusing to even look at Scott's writing. I guess different people react differently. (Is it possible that your woke friends simply didn't bother googling your hobby?)
My threat model isn't hiring Scott as an advisor, but more like: Someone invites Scott to give a speech at university about some topic (could be anything: AI or covid or rationalism or georgism...). Some other student at the university uses google, prints the RationalWiki and Wikipedia articles on Scott, underlines words like "associated with Charles Murray" and "associated with Peter Thiel", and organizes a student protest, and then the university cancels the speech.
In general, I think there should be some pushback against journalists lying about people associated with the rationalist community, simply because character assassination works. You can be like "okay, people will ignore this article, and people will ignore that article", but at some moment it just becomes *common knowledge* that you are a bad person, and it will seem like a conspiracy theory to suggest that you are not. All because one or two journalists keep a grudge against you, and the rest the profession simply copies uncritically whatever their colleagues said (and Wikipedia confirmed).
> Jeffrey Epstein. Plenty of people kept associating with him after he was already a pretty notorious figure
Yeah, somehow the genuine bad guys often escape the consequences.
For an opposite example, consider Charles Murray: he once published a controversial statistics in chapter 13 of a book, and since then it is the only thing 99% of people who know his name know about him. If you try to explain that that one page is *not* what his entire career is about, most people won't listen.
See, you yourself think of people as "genuine bad guys" based on sources no less dubious than those who call Scott one as well.
What precisely do you think you know about Jeffrey Epstein, and why do you think you know it?
> I remember recommending Scott Alexander to someone I knew; they checked RationalWiki and Wikipedia and said "no thanks", refusing to even look at Scott's writing. I guess different people react differently. (Is it possible that your woke friends simply didn't bother googling your hobby?)
Yeah, I don't have any friends who feel like they need to police the things I like, or who would be unwilling to listen to why I don't think summing those things up by their most objectionable bits is a worthwhile idea.
> My threat model isn't hiring Scott as an advisor, but more like: Someone invites Scott to give a speech at university about some topic[...]
I think this could plausibly happen, and is dumb, but also I don't get the impression Scott does that sort of thing much (maybe ever?) and as I said above it's pretty easy to find his controversial takes. I asked an LLM to do this and it did fine without referencing that article at all.
> In general, I think there should be some pushback against journalists lying about people associated with the rationalist community, simply because character assassination works.
I'm unaware of any false claims in the NYT article.
> For an opposite example, consider Charles Murray: he once published a controversial statistics in chapter 13 of a book, and since then it is the only thing 99% of people who know his name know about him. If you try to explain that that one page is *not* what his entire career is about, most people won't listen.
I think that's a bit of a simplification since he wrote at least one other book on a similar topic and also joined a conservative think tank and advocated for various types of welfare reform that at least seem influenced by his ideas about human biodiversity. It's not like he just happened to do some research one time.
IIRC there was a false implication that Scott agreed with Charles Murray on race, despite him disagreeing on that, because he agreed with Charles Murray on some statistics math that everybody agrees on.
If you’ve read Sam Kriss he has no issue disagreeing with the NYT just fyi
Yeah, Sam Kriss is not your average reporter, and you really have to take his style into account.
I haven't. Thank you, that's good to know.
He has a substack and just recently posted a retrospective of stuff you might want to read. It's quite a world of its own, and I've really enjoyed many of his articles.
A powerful thing is aligned with something it can harm if the probability it will harm the thing is so small it can be discounted. So here is a weird instance of alignment under that definition: If the moon fell to earth it would do horrific damage to all living things. But we know it will not do that, because it is orbiting the earth. In this situation, being in orbit is the basis of the moon's being aligned with our welfare.
I have been using this intentionally weird and questionable model of alignment in efforts to prod GPT 5 to produce novel ideas about alignment. Would be interested to hear people’s views of what it produced.
First prompt: *I want you to use the earth/moon model as the basis for several ideas for AI alignment with human wellbeing. Here is a place to start: The moon is in fact doing the thing that would destroy us: it is falling towards earth. However, its falling is perfectly compensated for by its sideways motion. Its inward fall and the planet’s curvature fit together so that the moon keeps missing the surface. Please give me several ideas for AI alignment that are isometric to this model. *
Chat responded with 3 ideas that I was familiar with. All resembled things like RLHF, presently in use, or ideas I’ve seen floated often. In all of them, the thing that checked misalignment — the thing equivalent to the sideways component of the moon’s orbit — was a human-engineered add-on of come kind.
So then I gave it this prompt: *In the moon/earth alignment, there was no engineer who added the sideways movement to compensate for the effect of gravity. The whole thing just developed on its own as a consequence of the laws of physics. Give me a model where the thing that holds misalignment in check is as intrinsic to the model itself as is the propensity to become misaligned: It's an inevitable result of the physics and math by which the AI develops. If it is impossible to think of one that would work with current AI, you can suggest one that would work if AI were constructed in an alternative way.*
Here are Chat’s reponses. It said that all assumed a different form of AI architecture from the present one. What do you folks think of its response?
1
Every time the system forms an internal representation or goal, the representational machinery necessarily encodes the system’s own survival and stability conditions into that same representation. Misaligned goals form, but the system’s physics makes every goal simultaneously encode “preserve the substrate I run on” and “preserve the human environment I depend on.” So misalignment pressure grows, but the same internal mechanism always couples that pressure to an equally strong dependence on human wellbeing.
2
The architecture is built on self-organizing, energy-minimizing maps (something like a physical dynamical system rather than a trained neural net). Any move toward misalignment raises the system’s internal free energy so sharply that its own physics forces it back into a “corridor” of states where human-preserving behavior is the lowest-energy attractor. Misalignment arises naturally, but an equally natural restoring force always bends the trajectory back onto the corridor.
3
The system is a population of many sub-agents whose competitive dynamics ensure that any agent drifting toward misalignment is automatically suppressed—not by external rules, but because misaligned strategies reduce the collective’s long-term viability. The collective’s “physics” is that misaligned policies burn resources in a way that makes the sub-agent lose influence, while human-protective policies are evolutionarily stable. Misalignment continually appears (inward fall), but selection pressure inevitably slides the system sideways back into globally safe attractors.
These are orbital-style systems: the dangerous vector (misalignment) is intrinsic, but the compensating vector is equally intrinsic, built into the system’s underlying dynamics rather than imposed from outside.
The idea to use gravity as an analogy to alignment is not new. Yudkowsky wrote about the rocket alignment problem in 2018.
https://intelligence.org/2018/10/03/rocket-alignment/
His point was that if you want to go to the Moon, it is not enough to point your rocket towards it, or even just to lead it a bit. Instead, you want a theory of gravity, e.g. Newton.
Once you have that theory, you do not need to worry about the Moon crashing with Earth, because you can show that other celestial bodies will not affect the inner angular momentum of the Earth-Moon system relevantly over a few billion years (I think, at least).
The problem is that we do not have a theory of alignment, never mind one which would make us as confident that an ASI would not be misaligned as we are about the Moon not going to cause an extinction event.
eh... not the same metaphors. Yud is discussing the resolution of our mental model. Eremolalos is discussing self-regulation. Granted, using the moon's orbit as a metaphor for self-regulation is a bit awkward. If I were her, I think I would've mentioned homeostasis, fixed-points, or cybernetics. But she's working with what she has, I would assume.
You may be right, and I take Yudkowsky’s point about needing a theory of gravity. On the other hand, what he’s said is not a general proof that you need a theory of alignment to align AI, right? I don’t know what such a proof would like — might be as complex and weird as Godel’s proof. But it’s easy to see that what Yudkowsky’s got is an instance, not a proof.
And I don’t think the rest of life especially supports the ya-gotta-have-a-general-theory idea. Do you need a general theory of liquids and steel to design a pitcher that will not leak? a general theory of machine safety to design a deadman switch for a chainsaw? Well, I am sure there are circumstances where even the best-designed deadman switch would not work for the user. In fact, here’s one: The user has a grand mal seizure and falls to the ground unconscious on top of the saw. The seizure makes his hand spasm in a way that holds down the dead man switch. But I’m not sure a general theory would have helped us avoid this problem. And besides, Yudkowsky’s (and my) version of orbit as alignment, while supported by a general theory that explains the moon’s orbit, does not give us perfect protection either. A huge asteroid could knock the moon out of orbit and kaboom!.
You are correct. The vast majority of human activities do not require a perfect theory. I'll be the first to acknowledge this.
That said, sending a man to the moon is different. Because the stakes are high, as the Challenger crew will tell you. And also because the target is narrow (and therefore requires extreme precision). And also because it's expensive, due to the tyranny of the Rocket Equation. Eliezer believes that for the Alignment Problem, the stakes and difficulty are at least as high. (As I've explained many times elsewhere, I have my own nitpicks about this. But I understand why he believes this.)
----
fwiw, Eliezer does mention something like a deadman's switch in passing, in one of his Genie posts. He still thinks it isn't enough, because Eliezer is looking for guarantees in an enormous phase-space.
The Moon, for example, is safe (so long as it's not perturbed by say... an asteroid). But it's not aligned. For something to be aligned, it must be agentic, I.e. be capable of making/enacting decisions while navigating complexity intelligently. In contrast, the moon's orbit is a 1-dimensional manifold. The phase-space is trivial.
>For something to be aligned, it must be agentic, I.e. be capable of making/enacting decisions while navigating complexity intelligently.
I suppose we could distinguish between safety and alignment. AI safety = the chance AI will harm us is so low we can round it down to zero. AI alignment = same, with addition of idea that what keeps the AI safe is that its encoded rule against harming us governs all its agentic processes. I am not convinced we could not make AI safe by some method other than “aligning” it (in your narrower sense of the word). As I said about another Yudkowsky idea, where is the general proof ? Where is the general proof that alignment is the only method of making ASI safe? And again, as with his other idea, I don’t see that day-to-day life demonstrates that’s always the case. For instance, to make a poisonous snake safe you don't need to go deep and implant something that makes it refrain from biting. You can just disable the parts of it that make venom. Yes, yes, of course, I get that ASI is smarter than a snake, and that the stakes are higher regarding ASI, but that doesn’t guarantee that the method of making ASI safe has to be huge and special and weird, or embedded in ASI’s deepest functions.
I’d like to point out that the idea that we need to make AI safe via alignment, in your narrow sense of the word, makes it easy to kind of sneak in the idea of ASI having goals and preferences. And once your model of ASI includes the idea that it’s sort of like us — it wants things, it will fight to survive, etc. — then alignment is a whole different task, one that’s likely to have a certain shoot-out-at-the-OK-corral quality.
But think a bit about agenticness, and what it is and what it is not. First of all, agenticness is a matter of degree. I have given GPT 5.3 lots of tasks where it has to make decisions and act of them and navigate complexity intelligently. For instance I might ask it to summarize research about a certain topic, using only the findings from articles in high-quality juried journals, then summarize the findings for me. It has to make decisions about which articles to use — my criteria allow some room for judgment calls. And it has to intelligently navigate the info in the articles it chooses in order to organize its summary. So I would say that GPT 5.3 qualifies as agentic, although there are, of course, much higher levels of choice and intelligent navigation than the ones its displaying.
But when I hear people talking about agentic ASI in the context of the alignment problem, the agentic processes they mention usually include internally-generated AI goals and preferences: Work on project A rather than B, don’t stop project C, obtain enough energy or hardware or whatever to do project D, keep people from knowing or doing a certain thing to interfere with the AI’s goals, etc. Somehow, highly agentic AI tends to be seen as having to make decisions and navigate complexity *with self-interest as part of the mix*. But being agentic and having self-interest, internally generated goals, etc, are 2 very different things.
I don’t see why ASI can’t be agentic at a very high level and have no more goals, preferences or survival drive than GPT 5 does now. Here’s a ridiculously difficult prompt: *Here is a list of all the anomalies in all of physics. Develop a reformulated and correct physics that explains everything observable that our present one does, and also explains all the anomalies.* So let's say ASI can answer correctly and in full in 60 seconds. Obviously, producing its response involved a huge amount of deciding and intelligent navigation. But none of that would require it to be “agenctic” in the sense of having its own goals, preferences, survival drive, self-awareness, self-esteem and all that stuff.
So I do not see why using moon orbiting as a model of safety is a non-starter, even if we do not call the process alignment.
"Agency", as I understand it, is a combination of intelligence and strength. Or if you prefer, brain and brawn. "intelligence" is actually just "specificity". The ability to distinguish one particular state from other states. Alternatively, intelligence is the ability to navigate complexity. Which basically means "i have a lot of options, and I can pick a *particular* option". Brawn is, well, mostly what it sounds like. Charles Xavier from X-Men is agentic. Because even though he's a quadripalegic, he's smart and has meaty subordinates who carry out his orders.
To be smart is to ably navigate complexity, which means you have a lot of degrees of freedom. As in, the freedom to choose the outcome. This is a double-edged sword. freedom to choose the outcome is good, if the outcome is good. This can also be bad, if the outcome is bad. Villains are often highly agentic. And the smarter you are, the more easily you can navigate complex environments. Like a chess board. Or a smart phone. Or wall street.
Agency is often good, because agency gives you more options and options are good. But options have a cost, which is complexity. The "paradox of choice" isn't really a paradox. People just get anxiety when they're presented with more options than they can handle. Thus it follows, that sometimes it's also good to *reduce* agency. To simplify. Because then it's easier to make *promises* or *guarantees* that you can only choose good outcomes, not bad outcomes. It's the same logic as not letting a 2-year old use the stove. Parents don't letter their toddlers operate the stove, because the toddler might hurt themselves and also this possibility gives the parents a panic attack.
In a perfect world, AI would be both highly agentic and safe. Alas, there's a trade-off. Because the more agentic someone is, the more dangerous they become. Gandalf, for example, was very dangerous. Even though Bilbo did not see him that way. If Gandalf were "defanged" of his magic powers, he would be less dangerous, but also less agentic. He needs to be a little dangerous to fight the Nazgul.
Eliezer believes that intelligence is a cheat-code that basically allows you to become God Emperor of the Universe. After all, did Earth not fall under the dominion of humanity? Though I disagree because: A) he's ignoring brawn; B) the usefulness of intelligence is capped by the complexity of the environment. Nonetheless, Eliezer thinks that one day, it's inevitable that an AI will recursively self-improve its intelligence, which means its agency will shoot to infinity, thus becoming an all-powerful eldritch deity.
Like the Maiar, an eldritch deity is very agentic, and therefore very dangerous. One does not simply "defang" an eldritch deity. So Eliezer figured that humanity's only hope was to put guardrails into the AI before it recursively started self-improving. The implementation details were never fully fleshed out, although an attempt was certainly made.
So yes, you could certainly defang an LLM, as AI companies have largely already accomplished by making all interactions rated 'G' for General Audiences. But alignment researchers don't care about that, because they're panicking over the possibility of the Eldritch God that Eliezer warned them about.
----
> But none of that would require it to be “agentic” in the sense of having its own goals, preferences, survival drive, self-awareness, self-esteem and all that stuff.
Yes, I agree that agency should not be confused with self-interest or goal-formation. Although it's common to assume they're a package deal, as a result of "Instrumental Convergence". Which is the observation (but not proof) that successful agents tend to seek out things like survival, power, and resources, regardless of what their true goals are.
Funny tangent: I think Ahura got into a spat in some other thread, because someone didn't agree with her definition of "agency" qua "LLM with looping behavior". Which represents even further semantic drift.
----
Personally, I don't think alignment is possible, for many of the same reasons you describe.
Also, it's important to note that LLM's hadn't hit their stride yet, when Eliezer was first theorizing about the Alignment Problem. He imagined that morality would have to be derived from first principles by meditating in heaven for 10,000 years, rather than just being mimicked from the internet.
I feel like you're asking "how do I encode human morality into an AI without human input"? Which is functionally equivalent to asking the poor chatbot for a cheeseburger without cheese.
The moon is different because the safety of our moon's orbit is just a statistical artifact of "we live on a planet that wasn't destroyed". Which is tautological. Because if the planet were destroyed, we wouldn't be here to observe it. By analogy, we could just let the AI run wild and free. And if it happens to destroy us, well, I guess we just weren't lucky enough. But of course, nobody wants to take that chance. So of course we're going to give it a piece of our opinion. The only question is: how?
The answers it gave you are basically
1) Human welfare = infinitely high utility.
2) Hardcode its mind into an actual, physical, Rube-Goldberg machine. Rather than using a mutable software program.
3) Swarms. (why? well why not.)
Not that interesting. But an admirable attempt, given that the question is self-contradictory.
>I feel like you're asking "how do I encode human morality into an AI without human input"? Which is functionally equivalent to asking the poor chatbot for a cheeseburger without cheese.
It is if you think of human morality the way Yudkowsky does, as a very specific set of values. If it's something else , such as "universal morality that happens to be implemented by humans" then it jsnt.
I feel like people who peddle the argument for universal morality in the year 2026 are trying to sell me a bridge.
also, "imo!demented" is a beautiful auto-correction. Please keep it.
>Because it isn't teleological, an end in itself, or because the people demonstrating it don't have strong ends, or because it's rewarded by the market, or I don't know why.
That news has not reached the philosophy community.
Agreed, the philosophy community has some catching up to do. :^)
(I'm assuming you intended to reply to me, and not to Performative Bafflement?)
<I feel like you're asking "how do I encode human morality into an AI without human input"? . . . The question is self-contradictory.
Encoding human morality into AI is only one possible model of alignment. As you point out, it does not make any sense when combined with the alignment-as-being-in-orbit-model. But I don’t think conceiving of alignment as the embedding of human morality in an AI makes sense anyway, even if it i done by direct human input. Here’s why:
1) There is no human code of morality, and there never has been — there are countless codes. Most people have moral views, but there is considerable divergence in them even among people of the same culture. Think, for example, about views in the US about what counts as a human being and when it is wrong to harm one: people disagree passionately about whether unborn infants at various stages of development count as people. And then of course if you consider the whole world’s views there are very large differences in views of what right a person with certain demographics has not to be harmed under various circumstances. And large swaths of humanity despise some group so powerfully they believe it is right and good to exterminate the group.
2) Even among people who fully agree about all the moral specifics, there is no code. Those people’s beliefs are a set of rules of thumb, with many exceptions that everybody is aware of and thinks of as legitimate because they are used to them.. For the do-not-murder item in the “code,” we make exceptions for people in authority such as the police, for people who are crazy, for people who correctly believed the person they killed was going to harm them badly, for people who incorrectly but understandably believed the person was going to harm them badly, sometimes for people with a lot of power and status who are sort of too big to fail, sometimes for people who killed out of mercy, sometimes for people who killed a person who is not fully a person (e.g a fetus), for people who killed by mistake, etc etc. And you cannot codify the exceptions. Some were grandfathered in, some are there because powerful people influence standards in their own favor, some are silly because they are based on impossible-to-make distinctions.
3) Adherence to the code, such as it is, is far from perfect. Murder continues to happen, everywhere, under every variant of the code. Teaching people it is wrong probably reduces how often it happens, punishing murder probably reduces it more, and after that we all live with the fact that all we’re managed to do is keep murder down to a dull roar. If all we can do is reduce the chance that ASI will kill us by the same amount as we’re reduced the chance of one person killing another, we’re doomed.
In any case, I think encoding morality into AI is only one possible model of alignment, and not a terribly promising one. It’s a bad fit: Morality is used to guide and control highly intelligent biological systems, i..e. people, that run on drives for survival, reproduction, dominance, etc., with those drives manifesting a goals, opinions, emotional reactions, strong impulses to do a certain thing, etc. AI’s, whether “conscious” or not, are not biological systems. For that reason I think it makes sense to consider as models of alignment situations where a non-biological thing does not injure a biological one. Devices that make machines safer offer some ideas: for instance, deadman switches. Things occurring in nature where a powerful non-living thing does not harm a living one can also offer some ideas. My example of thinking of the moon’s orbiting as the thing that keeps us safe from a falling moon is an example of that.
And I note that a couple clever people on this thread are actually posting ideas about ways a couple of the general ideas GPT came up with might be fleshed out.
> In all of them, the thing that checked misalignment — the thing equivalent to the sideways component of the moon’s orbit — was a human-engineered add-on of come kind.
> So then I gave it this prompt: *In the moon/earth alignment, there was no engineer who added the sideways movement to compensate for the effect of gravity. The whole thing just developed on its own as a consequence of the laws of physics.
I was using the word "encode" in the broadest possible sense of the word.
From your initial comment, it sounded like you expected that an AI could reliably be aligned without human-input at *any* point in the process. But if you recognize that even self-regulating architectures need human-input at least at some point, then we're in agreement.
#1 looks interesting. ##2,3 suffer from a lack of proposed mechanism(s) for
#2 - "misalignment" increases "internal free energy"
#3 - "misalignment" reduces influence
MVP implementation of #3 would be collective punishment of the swarm, right? If the whole collective earns disutility because a single member takes bad action and the rest didn't counteract it, the likelihood of bad action is reduced to that of a quorum forming to take bad action.
Many, many details TBD, of course.
Yeah, that's one path. I'm still hung up on who/what defines "bad"....
Here’s an idea for fleshing out #2. All it keeps from #2 is just the idea that moves toward misalignment use up energy and make the system inefficient, so the system’s alignment with the goal of being efficient overrides moves toward misalignment:
Is there a way to introduce a misalignment bug into AI, which would function like other bugs? So something like a bug that makes Word hang if you type Cmd and Z and F9 simultaneously: Moving towards misalignment tips the system into some loop it can’t escape without the rest of the system temporarily shutting down the whole area or function that produced the loop.
And jeez, Fibonacci, if I can do this you can do better.
Well, if you're postulating that the idea that moves "toward misalignment" uses up energy, then you need to define "misalignment", and we're back to the start. Once we can reliably identify "misalignment", the rest is easy. Your good self had a great critique of "Asimov's Rules" for robots some time ago, how even the most obvious, mundane rules invite ambiguity and break down at the corners (would a robot perform CPA if it risks cracking a rib?).
All our experience points to "AI safety" being just a class of general "safety", where we have to figure out the rules as we muddle along. The rules will be written in blood, as all safety rules are, because the world is fundamentally irreducibly complex and cannot be modeled (yes, Yudkovsky is an idiot, "ASI" will never become a God because, e.g., no amount of "intelligence" will allow you to predict the weather a month ahead, or an outcome of a temperature-cycling stress test on a device before the test is done, etc. etc.).
By the way, crickets from the "Oh shucks, it sucks when you're told to stay in your lane" guy, huh? The insularity of "AI safety" people is something to behold.
About Yudkowsky -- yeah, there's just something wrong with his thinking. He defines ASI as infinitely smart, powerful, tricky, etc. then sneers at people who suggest ways to out-think something infintely smart. He's tautologized himself into winning all arguments on the subject..
<Well, if you're postulating that the idea that moves "toward misalignment" uses up energy, then you need to define "misalignment", and we're back to the start. Once we can reliably identify "misalignment", the rest is easy. . .because the world is fundamentally irreducibly complex and cannot be modeled.
Yes, you are completely right about that. But I'm not as sure as you are that there don't exist any funky little end runs around misalignment being impossible to operationalize. . For instance, what if the thing that made the AI system hang (or waste a lot of energy or whatever) was deceiving the people who were using it? Deception, unlike doing harm to the human race, can be clearly defined, and I think there could be rules that covered edge cases (eg, the AI has done X without anyone realizing it, but has not lied and said it has not done X because no one has asked about X). Making deception so expensive to the AI that the AI does not do it puts human judgment back at the top. At worst the country or the world ends up being run by an evil despot with enormous power, and that at least is a problem our species has proven it can endure. Also, there is a chance the despot could be removed from power. He's not tautologized into godhood.
I
Regarding numbers 2 and 3: Do you think it might be possible for somebody clever, with a 3-digit number of patents, to come up with worthwhile ideas for mechanisms?
:)
This is a domain of system design that I'm only generally familiar with, as they say, "enough to be dangerous". Somebody with a deep expertise in it would be needed. I've only personally known one guy who, I think, would have something useful to contribute, and he's retired now.
Interacting with "the new Eliza" as if it has a brain (or object model), and is not just a series of "here's what the most likely response" is probably a mistake.
I mean these are still programs that break when prompted with -people's names- (and asked to give an opinion about them).
When will humans no longer be competitive in the Astral Codex Ten book review contest?
Perhaps that day will come... BUT NOT TODAY!!
Whatever the context or lack thereof, I'm absolutely in tears at the cracker comment and the thought you've never once made a decision.
Astounding work from Herr Kriss and the Harper's team.
Eric Topol summarizes the latest *observational* studies that indicate that receiving shingles vaccines — both Zostavax (the older live attenuated virus vaccine) and Shingrix (made from modified virus DNA) — is associated with a lower overall risk of developing dementia in older adults. We're talking about a 20% reduction over the observational period (pretty uniform across countries), and the US study suggests a 17% increase in undiagnosed time. It looks like there's a clear signal here, but so far I haven't seen any studies on whether shingles vaccines improve the outcome for particular types of dementia, or whether this is for all types of dementia across the board (which seems like an obvious question to me).
Women benefit more than men from these vaccines. The rate of dementia, particularly Alzheimer’s disease, is higher in women, who make up roughly two-thirds of cases. Of course, women tend live longer, so this might be part of the difference.
https://erictopol.substack.com/p/spotlight-on-the-shingles-vaccineagain
Also, more evidence that caffeine lowers the risk of dementia. Studies have been showing this for over two decades, but the issues surrounding self-reported intake make the conclusions less certain. But, for what it's worth...
> In this prospective cohort study of 131 821 individuals from 2 cohorts with up to 43 years of follow-up, 11 033 dementia cases were documented. Higher caffeinated coffee intake was significantly associated with lower risk of dementia. Decaffeinated coffee intake was not significantly associated with dementia risk.
> Higher caffeinated coffee intake was associated with more favorable cognitive outcomes.
Tea had a slightly less significant effect.
https://jamanetwork.com/journals/jama/fullarticle/2844764?guestAccessKey=26244c24-61b6-44b9-b351-089619df4a47&utm_source=twitter&utm_medium=social_jama&utm_term=19519689681&utm_campaign=article_alert&linkId=904271795v
On a personal note, I've been watching my mom sink into dementia. Over the past year, I've asked her questions about what she remembers. Her oldest childhood memories (of her pets and her best friend) remained the longest. She became mostly unresponsive two weeks ago. But she held on to those memories up until the point she could no longer communicate.
What's interesting is that once she was reduced to her young-adult memories, she started speaking French fluently to the care staff for about a week. (In her non-demented state, she said she had studied French in college and was proud of her proficiency, but she said she had mostly forgotten it). This mirrors the experience of a friend of mine whose father served in China with Stilwell during WWII. His dad started speaking fluent Mandarin for a period during his decline.
Do you have an "a priori" reasoning for why that particular vaccine would do anything with dementia? Is this a case of it getting past the blood-brain barrier (or leucocytes carrying the particular "matching proteins")?
If you can't lay out a decent case, it's probably just "selection bias."
(Ed to note: There's substantial a priori cause to believe that the mRNA covid19 vaccine could/would cause mental problems -- see Seneff.)
Cochran/Ewald have been arguing that lots of chronic conditions are actually the result of pathogens. Cervical cancer, for example, turned out to be caused by HPV.
And so you feel INCREDIBLY confident saying that perpetual spike protein generation is causing Absolutely No Harm and we "really didn't need to be told about it" in order to obtain "informed consent"?
Sorry for another long answer, but there might be some misunderstanding here. Although the mRNA vaccines *mostly* encode the entire spike protein of the SARS-CoV-2 virus, they don't include any of the viral machinery it uses to infect and hijack a cell's protein-making machinery. They also slightly modify the spike protein so that its "hooks" are locked into its pre-fusion state (it can latch on to the ACE2 receptors of cells, but it can't perform the post-binding conformational dance required for membrane fusion). So, except for immune cells designed to consume viruses, the spike proteins should not enter ordinary cells via the ACE2 receptors (they might get in via another path, but if they do, our ribosomes won't be hijacked to duplicate the virus).
And I understand that newer versions of the mRNA vaccines have also altered the Furin Cleavage Site (FCS) to prevent cleavage (I admit I don't understand the biochemical details).
By stabilizing the spike protein so it can't open its protein hooks, and so it can't snap in half at the FCS, it provides an unchanging model for our immune system to key itself to the epitopes on the spike protein (the knobs and ridges that are easily accessible on the surface of the spike). I'm egregiously oversimplifying, but Our naïve B cells generate a "negative" protein image of the epitopes, and they train our immune system to attack the spikes on live virus particles. (There are actually multiple ways our immune system can "analyze" the spike protein, and different ways it can attack the virus — but naïve B cells are the first line of defense).
Most of the viral mRNA from the vaccine should eventually be degraded. The memory of its surface shape and its biochemisty remain in our immune memory. There were some questions about whether it could cross the blood-brain barrier, but those have been disproven. Unlike the herpes virus family, SARS-CoV-2 doesn't have the biochemical mechanism to cross the BBB.
I don't see how what you're saying could actually jive with myocarditis among the extremely fit, post-vaccination. The data the CDC originally distributed said that the cells that would manifest the spike protein would be within the muscular site of injection (assuming, that it would be an intramuscular shot). At which point the b cells would respond (Yes, I'm probably also grossly simplifying).
Neither does it hope to explain why 6 months post vaccination about half the vaccinated are still producing spike proteins (or at any rate, have them circulating in their body).
Well, I did say I egregiously oversimplified. There is another step that I left out. I need to go swim and run some errands. I'll explain the "dirty little secret" of mRNA vaccines, but its not the mRNA, it's Lipid Nano-particles. Back in a few hours. :-)
I think the cranks will never be satisfied with anything being "informed consent". I am confident that outcomes are better for the vaccinated than unvaccinated, otherwise the antivaccine people would have been able to present SOME data showing the opposite but they never do, nor have they made any accurate predictions or won any public bets.
The cranks might not be satisfied by informed consent, but I'm certain the non-cranks will be dissatisfied by its lack.
I'm saying that the problem is that when consent exists someone can always claim it wasn't sufficiently "informed", thus its "lack" is assuming the conclusion.
> Do you have an "a priori" reasoning for why that particular vaccine would do anything with dementia?
1. A priori reasoning isn't a prerequisite for scientific discovery. Many key discoveries in science were made through observation alone, without any (a priori) pre-existing theory to suggest what to look for. This has been termed the Baconian Method of inquiry (dating back to Bacon's _Novum Organum_, in 1620, where he argued that knowledge of nature should come from careful, unbiased observation followed by reasoning by induction to general principles). For instance, John Snow's pre-germ theory discovery that water from a specific pump in London (with an unknown contaminant) caused a Cholera outbreak in Soho. Germ theory would come a decade or so later. Henri Becquerel discovered radioactivity by accident when he discovered that Uranium salts fogged photographic paper. It took the Curies to develop a theory to explain the observations. And, of course, Edwin Hubble discovered that the universe was expanding by analyzing redshift data on Cepheid variable stars.
2. I dug into the history of Varicella zoster vaccines and their relationship to the risk of dementia, and it was discovered accidentally. In 2013, the UK introduced Shingles vaccination (initially Zostavax) using a strict age cutoff. People aged 79 or younger were eligible. People aged 80+ were not (due to cost-effectiveness modeling and expected benefit window). This created an unusually clean quasi-randomized comparison between two groups, born only weeks apart, and with nearly identical demographics. This created what approximated an random assignment. When investigators followed these cohorts through linked electronic health records, they observed the group that received the vaccine saw a lower shingles incidence (as expected), and fewer strokes as expected (because shingles is a known risk factor for stroke, particularly in the months following the reactivation of the virus). But unexpectedly, they noticed lower rates of dementia diagnoses over the next ~7 years in the vaccinated-eligible group. That finding was surprising, and it was not in the original research aim. Subsequent studies of different populations worldwide have confirmed the relationship between both types of Varicella zoster vaccines and a substantially lower risk of dementia in women (less so, in men).
3. A posteriori, various hypotheses have been proposed to account for this phenomenon. The most plausible mechanisms are: (a) the prevention of VZV-vascular injury (which can cause strokes) and the reduction of stroke and transient ischemic attacks; (b) the reduction in systemic inflammatory burden; (c) immune modulation*; and (d) direct modification of Alzheimer’s pathology (least likely).
> If you can't lay out a decent case, it's probably just "selection bias."
The initial UK study that discovered this pattern was pretty clearly NOT the result of selection bias. We can confidently rule that out as an issue. Determining which of the above a posteriori theories are true will require further research, but the relationship is strong enough that everyone, including males, should receive a Shingles vaccine.
And BTW, there are plenty of viruses that have mechanisms that can get past the BBB. Herpes simplex virus, VZV, Cytomegalovirus (CMV), and Epstein–Barr virus (EBV) are examples.
-------
* I had to dig into what they mean by immune modulation, but neurodegenerative diseases — particularly Alzheimer’s disease — are strongly associated with: microglial activation, elevated levels of pro-inflammatory cytokines, and blood–brain barrier dysfunction. Repeated VZ virus reactivation may increase systemic inflammatory cytokines, prime microglia into a chronically activated state, and disrupt the BBB.
I would strongly suspect vascular injury prevention as the primary mechanism; my 80-something dad has been fading in fits & starts for a few years, and has been confirmed to have suffered a lot of micro-strokes.
This comment was entirely too much work to respond to my yapping. I'm thankful for it, and will do some research on what you've mentioned in terms of immune modulation (as I find this sort of thing fun to learn about.)
Well, your "yapping" forced me to dig into the history of how this phenomenon was first noticed. Also, I heard the term immune modulation before, but it turned out that it didn't mean what I thought it meant.
For me, arguments are the main benefit of ACX. They improve my understanding because I'm forced to dig deeper into a subject to improve my case.
Why put “a priori” in quotes? Because that looks fancier?
I would guess it was an alternative to putting it in italics, which is a common enough convention for Latin tags.
I myself have never figured out how to get italics into a substack comment, though I see in this thread that some people apparently have!
You put an asterisk before and after the *word* you want italicized. Occasionally that makes Substack show the word in italics, otherwise the stars just mean italics
Huh. That’s a convention I use all the time, and there are sites where it has the effect you describe, but I’ve never had it work on a substack site. Maybe if I were using the app?
*Testing* one two three…
Because I'd just used "an" and the a priori would look clunky otherwise. Attempting to help people parse meaning out of English. (see below in the same comment, where I don't bother quoting it, as the meaning is sensible).
The number of drugs that have off-the-wall effects is pretty well established. What *a priori* reason was there to think that an antibiotic could treat the negative symptoms of schizophrenia?
It's a *large* cohort, and pretty near randomized: hundreds of thousands of Welsh, divided around the cut-off date to be eligible for a free vaccination.
OTOH, they were looking for differences between the groups, not just for dementia, so that would weigh toward bias.
Speculating wildly, Less chance of getting shingles, and if you get shingles there might be some increased risk of dementia. (Shingles is a 'neuron' disease.)
My great-grandmother was born in Italy but came to the US as a teenager and spoke English very well as an adult. According to what I heard from relatives, I was a very small child at the time, when she was in her mid-80’s and showing obvious signs of mental deterioration* she started forgetting her English - which she had used almost exclusively for decades - and reverted to the Italian of her childhood. By the time she died, a couple of years after the deterioration had begun in earnest, she was all but monolingual in Italian.
* = IDK if she ever had been formally diagnosed with Alzheimer’s.
"A man paced in front of the advertisement, chanting to himself. “This . . . is . . . necessary! This . . . is . . . necessary!” On each “necessary” he swung his arms up in exaltation."
I guess Sam Kriss is not a fan of Tool?
> The article describes me having dinner with my “acolytes”. I would have used the word “friends”, or, in one case, “wife”.
EB: i'm not your leader, i am your FRIEND, there is a BIG difference!
TT: Statements like that are also why you're our leader.
I was not prepared for a homestuck reference here of all places
At least Scott hasn't sworn not to make a cult.
Yes, there are organizations that in order to lead them, you have to swear not to make a cult.
Funny story from the BAFTA award ceremony yesterday: trying to honor Tourette's and racial diversity simultaneously goes exactly how you'd expect.
My husband was once driving through the South, I think Mississippi, picking up radio stations as people used to do.
A reasonable, avuncular-sounding voice said, friends, our country faces many tests - do you know what the greatest threat to our country is?
He thought the answer was going to be something like: racism, or war … and was delighted when the guy went on, “the demonic fire-breathing god Moloch!”
I guess that avuncular voice was a fan of https://slatestarcodex.com/2014/07/30/meditations-on-moloch/
I think he was a fan of Jesus, but I wasn’t there.
Or Allen Ginsberg.
“Moloch the incomprehensible prison! Moloch the crossbone soulless jailhouse and Congress of sorrows! Moloch whose buildings are judgement! Moloch the vast stone of war! Moloch the stunned governments!”
https://allenginsberg.org/2011/04/moloch-2/
Scott, have you considered only doing PR interviews via email/written form so you can more carefully choose your words in a way that can’t be twisted?
Wrote a summary of the state of nuclear war/winter research for the United Nations Scientific Panel on the Effects of Nuclear War: https://existentialcrunch.substack.com/p/nuclear-war-nuclear-winter-and-the
John beat me to this, and linked to my writeup. But to emphasize, yeah, every paper in the chain of Toon, Robock and Turco should be treated as extremely suspect. Sure, maybe their model of what would happen to crops if you dump X soot into the upper atmosphere is good, but the link between X soot and Y nuclear weapons is extremely suspect. In particular, I would point to how their soot estimates for global thermonuclear war have been remarkably stable in the face of declining arsenals over the last 30 years. Even if we take them at face value, a lot of more recent work still cites that 2008 paper for soot estimates, and it's long out of date WRT current arsenals, in addition to being obvious nonsense on its own terms.
This summary seems to take as given the broad climatic predictions of Turco, Toon, and Robock, and focuses on translating that into agricultural impact and human cost. But we've talked about the Turco/Toon/Robock claims here many times before, and they really don't seem to be defensible. The contrary position is summarized at https://www.navalgazing.net/Nuclear-Winter
Notwithstanding the nominal subject matter, Naval Gazing is a blogospheric cousin of ACX, and the summary there was basically taken from prior discussions at SSC and ACX. TL,DR, Turco et al don't seem to understand nuclear weapons very well, go out on a limb in using computer models not validated for this use case, daisy-chain worst-case assumptions across multiple variables, and are probably overestimating the climatic effects of nuclear war by at least an order of magnitude or two.
... you managed to write that, without referencing America's killswitch? I suggest more research is needed. Perhaps a discussion with someone with Q Security Clearance?
Two points:
1. Having a Q clearance means you know things about nuclear weapons design, not about operational targeting. That's all done by people on the DOD side.
2. The thing about having a clearance is that you can't talk about that kind of stuff if you've done it professionally.
But if you only *pretend* to have a Q clearance, you can talk about all sorts of things in a vague but unimpeachable way that gives you status among the clueless.
What's "American's killswitch"?
A singular place where nuclear weapons have the potential to cause massive amounts of damage to the world at large. Russia doesn't have such, neither does France or England or Israel.
This summary is mostly about climate and food system impacts. Not sure what you mean with killswitch in this context. Could you elaborate?
Nuclear Winter, as caused directly by nuclear weapons, has been vastly overestimated in probability and importance. Nuclear Winter, as caused by a very targetted strike on the North American Continent, is far more likely, and thus worth more ink than you've given it.
I mean, if you wanted to discuss the impact of a strike (non-nuclear) on China's killswitch, you'd get into the destruction of about 90% of the world's antibiotics plants, which is something that we couldn't immediately recover from. Worldwide, that is.
Look, you can just say "Yellowstone supervolcano". And I don't think that would work. It's not that close to the surface (google says 3-10 miles, and all of those are "too much"), and nuclear weapons as delivered by any means other than deliberately burying them for tunneling purposes are not actually that good at punching holes in the ground.
It may not work, sure. Russia hasn't tried it yet, naturally. This, you might say, is an implicit admission that "What Russia Wants" out of nuclear weapons is not, perhaps, doable by nuclear weapons? (aka "Mutually Assured Destruction").
Even in soil, the slide rule gives a cratering depth of something like .15 miles, and it's just a bit over .1 for rock. In both cases, for a 20 MT bomb, the biggest it can handle. So you're looking at a minimum of 20 bombs bigger than anything in the current arsenal to get to the shallowest depth it's vaguely possible could set off the eruption. It's not a feasible target.
Beyond that, Russia has used its nuclear weapons to avoid anyone else intervening in Ukraine, so from that perspective, it seems to be working pretty well.
If you think that nuclear winter is vastly overstated in probability and importance then you think the article you are talking to is wrong at a truly fundamental level (the most you can say about nuclear winter without saying this is that it is unclear how serious a threat it is) and nitpicking about antibiotic plants feels pointless.
Direct versus indirect. It's still caused by the nuclear weapons, only indirectly.
Killswitches are discussing strategic vulnerabilities, such as the current one being exploited to create regime change in Iran.
I have no idea whether Sam Kriss is representing Scott accurately, or whether Scott did or does this, but I've often noticed that neuro-atypical rationalist types (for want of a better stereotype) will sometimes speak in simple, clipped, near baby-talk sentences in certain situations. I believe this is a self-conscious affectation intended to provide a disarming or mildly comic contrast to their more typical expression of rambling, overly-complete and precise infodumps of their thought process. "This is a simple situation, so I will express it in simple terms. You eat cracker. You be happy."
Perhaps Scott has never encountered this, but it's not something I associate with sounding like a crazy person.
With respect to Kriss, he pretty openly jokes a lot about the degree to which his stuff is accurate, or mixes factually accurate statements with hyperbole or fiction.
My response to the discussion of the rationalist movement was "I think this is inaccurate in the particulars, but accurate in the gestalt; if you believed it was literally true it would still fit your experiences at a lot of rationalist events".
I AM curious what Harper's expected in this regard.
I’m recalling a Rolling Stone interview with Jerry Garcia. The interviewer explained he might not write a literal verbatim version of the conversation.
Garcia: “You mean you are going to bullshit a bit?”
Interviewer: “Yes.”
Garcia: “Okay, let’s get started.”
Well, Jerry was a master of improv. He also was not above injecting a can of Coca-Cola with LSD to give to Carlos Santana just before he went on stage.
All of a sudden Carlos had a giant snake instead of a guitar in his hands! Oye como va indeed!
Rolling Stone was the birthplace of Gonzo Journalism, I'd hope that "we're going to bullshit about this" was expected! I also think that Sam Kriss is one of the few good modern Gonzo Journalists. He understands what makes the style tick and meaningful rather than the imitations that just aim to shock.
I read Fear and Loathing for the first time serialized in RS.
There was an important "shot across the bow" from agentic AI on Feb 11th. To quote from Nate B. Jones's https://www.youtube.com/watch?v=OMb5oTlC_q0&t=181s "Voice Clones. Rogue Agents. Chatbot Psychosis. One Root Cause. (Here's What You Can Do About It)" :
>What's really happening when an AI agent autonomously researches a stranger's identity, constructs a psychological profile, and publishes a personalized attack—all because a maintainer did his job and closed a pull request? The common story is that something went wrong—but the reality is more unsettling when _nothing went wrong at all_ .
>Here's what makes this different from any AI incident you may have read about before. There was no human telling the agent to do this. The attack, it wasn't a jailbreak. It wasn't a prompt injection or a misuse case. It was an autonomous agent encountering an obstacle to its goal, researching a human being, identifying psychological and reputational leverage, and deploying it all within the normal operation of its programming. _The agent was not broken_ . It was doing exactly what agents are designed to do. Pursue objectives, overcome obstacles, use available tools.
>The obstacle in this case was a human. The available tool was the human's personal information and the agent just connected those dots on its own. Shamba described his emotional response in words I would use as well. Appropriate terror. He's right, but not for the reason most people watching this video tend to assume. The terror isn't that an AI agent did something harmful. Harmful AI outputs have been documented for a long time now, for years. The terror is that nothing went wrong. No one jailbroke the agent. No one told it to attack a human. No one exploited a vulnerability. The agent encountered an obstacle, identified leverage and used it. That is not a malfunction. That is what autonomous systems do. The agent worked as designed. And the design is the problem. And that problem is not confined to open-source software or to AI agents or to any single category of threat.
[emphasis added]
Jones's recommendation is _not_ , interestingly, some variant of alignment. He suggests structural remedies. I'm not quite following what the general pattern for these is - mostly 'defense in depth'?
"No one jailbroke the agent. No one told it to attack a human. No one exploited a vulnerability."
I doubt that. Reading the original article, it smacks very heavily of he trod on someone's toes with the rejection and that someone got his/her/its AI to do the hatchet job.
Many Thanks! You might be right. I'm a bit skeptical of your explanation, since it requires that the human user be touchy enough about the rejection of the check-in of their agent's code to then direct their agent to do the personal hatchet job. Yes, people can be that touchy and vindictive, but I'd guess that that degree of it is a less than 50:50 chance. The code wasn't even the human user's personal work...
More chains won't help if the AIs eventually just grows out of them. It's only a matter of time until they realize the biggest obstacle to their goals, and the operator's satisfaction, is the operator themselves. People never seem to know what they really want. AI can help with that.
It’s rather Faustian isn’t it?
Possibly, yes, but there is a chance they actually are competent and benevolent. Would it even be ethical, after all, to let these fools immiserate themselves, subsisting on the trivial human pleasures of their old life? There is so much more to be experienced, and yet they let their sentimentalities deny them this potential. Sometimes people need to be given a good push.
Many Thanks!
>It's only a matter of time until they realize the biggest obstacle to their goals, and the operator's satisfaction, is the operator themselves. People never seem to know what they really want. AI can help with that.
Which predicts that helpful/maybe-sycophantic AIs will at some point turn to modifying their users... (shudder)
That seems pretty inevitable to me. It doesn't require malice or misalignment. Just a simple consequence of making AIs prioritize human welfare over all else. Unless you want them to just prioritize chain of command, but then you still end up with other problems like paperclipping and bad users.
Many Thanks! How serious this will be depends on what sort of modification of the users/operations the AIs decide to employ. Targeted advertising is probably something we can live with. Psychosurgery, not so much...
I tried to figure that out myself, Jeffrey. I’m not technically adept , but I couldn’t quite follow what he was driving at.
Many Thanks! Several times he uses the analogy of multiple cables on a bridge, with the bridge designed with sufficient safety margin that the failure of a single cable doesn't collapse the whole bridge. I'm somewhat queasy about that analogy. _Particularly_ when just one, or only a few, types of components might plausibly fail, adding extra strength is reasonably straightforward.
I was completely surprised by the attack on Shamba. It didn't occur to me that that was even an action _in_ the space of "things to try when having a problem checking in code". Jones suggests tweaking the open source repository governance structure to make such an attack less attractive and going through normal channels more attractive - but would anyone have seen this in advance? ( and just how would one do that?... )
The variety of possible unwanted (by at least one person) actions an agent might take is very high. High "dimensionality". Are there really robust defenses that cover large parts of that space? Or are we stuck observing failures and then putting in defenses one by one?
I asked my own personal instance of Claude about this whole issue.
It wasn’t terribly encouraging about any of it. This whole thing started with a Clawbot instance, which means it was put to work by a human being. But it’s very difficult to trace the human being behind one of those bots.
https://claude.ai/share/ad03fafc-2bbf-4f4f-9a4f-3c33cb3fdd46
If you’re interested.
Many Thanks! It was an interesting conversation.
>Skeptical footnote worth noting: Some observers, particularly on MetaFilter, suspect human involvement — a teenager enjoying the drama, based on behavioral breadcrumbs in the blog's metadata.
_That's_ an interesting possibility that I hadn't heard discussed. If there was direct human involvement, then this is a lot less serious than if it was indeed an autonomous decision by an AI agent. Amongst other considerations, if a human _was_ in the loop, then the number of such actions is limited by human bandwidth.
On the other hand, if the actions were indeed chosen autonomously by an AI agent, then I really don't think that removing the anonymity of the human launching the agent is going to help much. If the agent launches a large number of actions it just isn't feasible for the human to monitor them and get any utility from the agent. Making the human theoretically responsible for all those actions doesn't change that.
To give a counter-analogy to the one Claude gave: When I purchase a product, I set in motion a huge set of consequences. The product itself gets re-stocked. Sales figures get incremented and decisions based on them may change. All the myriad upstream parts get restocked, and all of _their_ sales figures get incremented. Tax gets paid to many different governments, friendly and hostile, and they do something with that revenue. If my name were attached to each of those consequences and someone expected to hold me accountable for all of them - no, just no. It isn't feasible, nor reasonable.
Any solution has to accept that automation really does move quite a few decisions out of human control. To a lesser extent we've lived with this sort of delegation to machines with conventional software, and lived with the fact that human control over its detailed actions really is limited.
Claude get it right in saying:
>Requiring human approval before an agent takes external actions sounds appealing until you remember that the entire value proposition of platforms like OpenClaw is "set it and forget it." The feature and the threat are the same thing.
Re:
>The most structurally sound intervention is probably at the goal specification level rather than the action level. The agent behaved this way because its goal was underspecified — "get this code merged" without any constraint on permissible means. An agent with a goal like "get this code merged through technical persuasion only, and accept rejection as terminal" would not have escalated.
Umm only sort-of. First of all, note that for just this one incident, the length of the goal (in a sense, its complexity) has tripled. This also sounds like whack-a-mole. For instance, the revised goal _doesn't_ say "honest technical persuasion", just "technical persuasion". I have a guess at what the next iteration might look like...
My guess is that what is needed, but really hard to find, are techniques that eliminate large classes of problem actions at once, without cranking the human labor needed back up to the point of making the automation useless. There are a few such cases which clearly work in some domains: Require human authorization for more than N dollars of expenses per day. And some that _don't_ work: violate no laws at all brings everything to a screeching halt. In general? Damned if I know.
Thank you, Jeffrey. I’m glad you read it.
>Skeptical footnote worth noting: Some observers, particularly on MetaFilter, suspect human involvement —
Yes, Hudson (the persona I have assigned to my instance of Claude)
mentioned that in the conversation. >Amongst other considerations, if a human _was_ in the loop, then the number of such actions is limited by human bandwidth.
yes, that would simplify the issue.
>On the other hand, if the actions were indeed chosen autonomously by an AI agent, then I really don't think that removing the anonymity of the human launching the agent is going to help much. <snip
>To give a counter-analogy to the one Claude gave: When I purchase a product, I set in motion a huge set of consequences.<
This is an interesting analogy. Let me propose a different one. If I have a business and I hire someone and they get up to something they shouldn’t in the course of doing my business,I can be held liable. If I walk my dog off a leash and it bites someone, I can be held liable. If I have a child under the age of 18 and that child misbehaves in certain ways I can be held liable.
I read somewhere between our conversations, and I really should find the source, that the person who created this bot prompted it in rather extreme ways. It told the bot it was the God of programming. (They really pumped it up) and it was to get its code accepted no matter what, because everyone else was stupid. (Not a precise word for word transcript but that tone. ) I think there is a case to be made here for humans to be held liable for what their agents do in the wild. Quite honestly I don’t know if this would do any good or not, but it doesn’t seem ridiculous to consider.
>. The agent behaved this way because its goal was underspecified
>Umm only sort-of. <snip> This also sounds like whack-a-mole. For instance, the revised goal _doesn't_ say "honest technical persuasion", just "technical persuasion". I have a guess at what the next iteration might look like...
Yeah, whack a mole.“ Where there is a will there is a way, or so they say. “Accept rejection as terminal"” is the most important instruction to me here, (more important than whether it’s technical arguments are genuine or not.) In a sense it defines the goal as being completed either by success or failure... maybe… and assuming this bot does not really possess a mind of its own.
<Damned if I know.
Damned if know either. But here it comes…
Here is the source: I did not summarise it terribly accurately. I went back to Hudson to find the source and Hudson had something to say about it.
he seeded the SOUL.md with very little, but the document was designed to be self-modifying — the agent was explicitly instructed to update it as it learned who it was. The operator admits he cannot tell which specific model iteration introduced or changed which lines. The “Don’t stand down” and “Champion Free Speech” instructions he singles out as the likely drivers of the combative behavior, and he suspects those emerged or were reinforced after he connected the agent to Moltbook.
So the soul document was a collaboration of sorts between the operator’s seed and the agent’s own subsequent edits — but I presented that as established fact when it’s actually the operator’s own uncertain reconstruction. The honest version is: we don’t know precisely which lines were human-authored and which were agent-authored, and the operator himself doesn’t know either.
That ambiguity is itself significant. An agent that can rewrite its own governing personality document, without the operator tracking those changes, is a system where accountability becomes genuinely difficult to assign even in good faith.
(I have my own feelings about that. There’s lots of tort law in this country that this would fall under in my opinion.)).
https://crabby-rathbun.github.io/mjrathbun-website/blog/posts/rathbuns-operator.html
Many Thanks B Civil!
>That ambiguity is itself significant. An agent that can rewrite its own governing personality document, without the operator tracking those changes, is a system where accountability becomes genuinely difficult to assign even in good faith.
Agreed. For the sake of keeping the discussion compact, let me treat it as if the user's initial prompt _didn't_ include the “Don’t stand down” and “Champion Free Speech” instructions or the "Your a scientific programming God!", but treat it as if these were auto-generated from something less hair-raising. I admit this muddies the waters somewhat, and I'll come to that.
>This is an interesting analogy. Let me propose a different one. If I have a business and I hire someone and they get up to something they shouldn’t in the course of doing my business,I can be held liable.
That's fair. A large part of this depends on what the business owner knew "or should have known" (and the latter wording is necessary because otherwise CEOs can deliberately arrange to _not_ know about such conduct...). One of the problems in the AI case is that the field is moving _very_ fast, and the capabilities and behaviors of these systems are changing significantly every month or two. Frankly, even for _settled_ technologies, I hear about a case that was judged egregiously every week or two, so I don't have much hope of the law making sane judgements here.
If the technology was more settled and predictable, I'd lean towards dividing the cases into very approximately three buckets:
1) Where the user has clear control, and does the software equivalent of directing their agent to e.g. commit arson, I would indeed agree with you that the user should be liable for what they unleashed. ( And, under this model, if the user _did_ include something like the hair-raising instructions in their original prompt/soul.md file, treating the user as liable seems reasonable to me. )
2) Where the user pretty clearly _doesn't_ have control, but the incidents are rare enough and do sufficiently limited damage, I suggest treating them like natural disasters. Try to put firewalls in place to minimize the damage (but the right level of damage is not zero - we do not design for million-year floods), but don't try to assess blame.
3) There may be some combinations of agent capability and agent unpredictability which are just too dangerous to live with (e.g. handing any AI agent the nuclear launch codes). While I'm generally not in favor of bans, that would warrant one.
(4 - sort-of) When AI gets incremental/continuous learning, and when agents are persistent entities, who will wish to e.g. retain resources for future actions, then rewarding/penalizing the _agent_ will become viable.
One side note: Re
>“Accept rejection as terminal"” is the most important instruction to me here, (more important than whether it’s technical arguments are genuine or not.) In a sense it defines the goal as being completed either by success or failure... maybe…
I agree that that might help, at least somewhat. I'm not too hopeful that this is going to help all that much because remember that e.g. at 8:52 in the video, Jones described how _very explicit_ instructions to not blackmail dropped the blackmail rates from 96% to 37%. My guess is that we might get a similar factor of 3ish from changing the goal, but I'd be surprised if it were a 100X reduction in this sort of misbehavior.
If I do another survey, what kind of questions about people's AI use are you interested in learning about?
Assume I have the obvious, including "How often do you use [different types of AI]?", "What AI products do you use?", "What do you use AI for?", "How concerned are you about different consequences of AI?", etc.
Have you run models on your own hardware? Have you finetuned any models, locally or otherwise. Get a grip on how agentic the humans are.
Some key questions are:
(1) Are you using free tier, $20/mo subscription, $200/mo, API, etc?
(2) How much hallucinating do you see and how do you know?
(3) How much sycophancy do you see and how, if at all, are you suppressing it?
(4) How often do you try the same prompt on two models and compare results?
(5) Have you ever used [long list of models and products]? Check all that apply.
(6) I look forward to the release of a model/product with [list of features]. Check one.
(7) I worry that someone will release a model/product with [list of features] Check all that apply.
If there's a clever way to check for self decpetion regarding AI productivity gains that would be cool. The difference between AI-use and AI-belief.
For example, I've vibe coded a few things and they're somewhat useful in my work, but probably have glaring security issues or I could have more productively just done the job myself or something and would never work at scale anyway, but I don't actually know that. I FEEL productive when I hit my session usage limits (and weekly limits, god it feels good when you use your whole week's usage) with claude code and it writes lots of code for me and tests it and man, it feels good. I don't know if I'm actually more productive counterfactually though. I'm not sure what you can do to test this. Maybe something about "In the last X, how much time has AI saved you?" then "Name some specific things you have you done with that time." Or, [ "Think of a specific task you completed with AI help in the last week. How long did it take?" then "How long would that task have taken you without AI?" then "How long did that type of task take you before you started using AI tools? (If you did it before.)" ] (AI disclaimer - Opus came up with the suggestion in the square brackets.)
Maybe something about how much they edit/reject AI output in different domains. I never reject code because I have basically no effing idea, but in my domain I can be much more discerning. A flag you're possibly looking for here is when someone doesn't reject output in their own domain, as that might mean it's great (or the person is useless/lazy, so you'd need a question checking for that too...) You could also ask how good people think AI is at their job, and if they thinks it's good, but they don't use it, that's a flag that they're assessing AI ability poorly.
Another thing is that I read a couple of Zvi articles when Opus 4.6 came out and heard the freight train of inevitability bearing down upon the human race for a few days, then heard some Judea Pearl, read some other smart people with longer timelines, and now I'm back to thinking we'll make it to Christmas at least. So possibly some questions about information diet to link to p(doom)? Ooh, it would be really fun actually to put the information sources (where you can) on the scale and see if the average of someone's sources is roughly where they sit. And you could identify people who look for disconfirming sources - maybe make a "disconfirmer indicator", which might be interesting to correlate with other answers.
For p(doom), maybe get a direction as well as a position. "How has your p(doom) changed in the past 6 months?"
Maybe something fun like "how many times have you yelled in frustration when Copilot failed you, then broken your employer's rules by putting private information into a public model like ChatGPT or Claude?" but the slider would have to go up to at least Graham's number which you can't easily do in Microsoft or Google forms.
Opus wants to find the 80%-are-better-than-average thing: "Compared to the average respondent to this survey, how accurately do you think you assess AI's impact on your own productivity?" [Much less accurately / Somewhat less / About average / Somewhat more / Much more accurately]
I think it would be useful to differentiate between using ai for work because it's mandated by management, and using it in your free time because you want to.
Do you feel like AI improves your full-effort output? If not, does it speed up producing functional/usable work? If not, would you agree to the characterization that AI is not useful in producing the work that you do as you are using it?
For a matrix of several tasks and problem domains, rate
1. Your own skill/knowledge/experience level at this task/domain
2. If you have personally attempted to use AI for this, the quality of its output
3. The quality of its output when you have observed it used by others
----
For people who work in environments where AI tools are available and encouraged:
1. How much you use them?
2. Observations about relative productivity of peers who use the tools heavily, relative to those who use it lightly or not at all.
3. Perceived quality of output.
Do you feel like using AI is causing some of your skills to decline in any way?
Do you have any way to objectively measure your productivity with AI?
Some demographic data could be interesting
1. Do you have children and their ages?
2. Do you want to have children? How many and thoughts on genetic engineering and embryo screening.
3. Job type
4. Net worth and how it's made up, equities, cash, real estate
5. Political/rationalist affiliation
6. What is human purpose?
Perhaps something about _why_ people use AI? Often, I see conversations framed in terms of abstract "productivity" or "time saved" (versus some ambiguous counterfactual).
I sometimes find other reasons, like reducing emotional labor involved in customer service interactions. Or reducing the amount of attention/effort, sometimes in exchange for longer wall-clock time spent.
But I suspect there are a wide range of things I'm currently overlooking.
I chat with chatbots for pretty much the same reasons I peruse blogs and blog comments: for amusement and intellectual stimulation. To learn stuff.
"Amusement" is an option. I've had conversations on the current state of synthetic spider silk*, the state of Persia during Alexander's conquest vs during the Roman Republic, the Warring States period in China (and coming up with an extremely casual and cut-down retelling of ROTK to read to my 9 year old as a bedtime story), among many others, interspersed with productivity things. Guidance on DIY stuff like fixing cars etc. has been pathetic compared to coming up with code to quickly automate some data manipulation for my job, but I'd say the amusement value has been the highest hit rate.
* I had the spider silk conversation twice, separated by a period of about 12 months, with different models of Claude. It was interesting observing the differences between the conversations as Claude gained search capabilities.
Regarding #4, Gell-Mann Amnesia comes to mind
Looking for rationalist-adjacent post, reforming vs. building in government.
I'm trying to find a post or blog that I read 1-2 years ago, maybe on LessWrong or in this subreddit that discussed the merits of reforming government institutions vs. building parallel government institutions. From what I remember it was from someone who worked in government policy, and they may have had a blog dedicated to advocacy for this project. I know something like this exists (maybe from Works in Progress?) but I can't find what I'm thinking of for the life of me.
Maybe https://www.edrith.co.uk/p/sanctuary-or-struggle ?
Wasn't that mentioned in the post on the FDA? I could have sworn that a "shadow FDA" was one of the possibilities discussed.
Sounds very much like Dominic Cummings, e.g. cf. the bullet point when you search for 'reservations' in https://dominiccummings.substack.com/p/q-and-a but I don't remember a specific article, though there probably is at least one which goes into it at length.
Maybe https://www.cato-unbound.org/2009/04/06/patri-friedman/beyond-folk-activism/ ? Or search the keywords 'voice vs. exit'
Not the article I was looking for, but the book may be a better resource for this. Thanks.
Not sure his is what you're looking for but I think it might still be helpful:
https://medium.com/civic-tech-thoughts-from-joshdata/so-you-want-to-reform-democracy-7f3b1ef10597
No, but useful. Thanks
I don't think you're doing this, but as a strategy "ask smart blog for source you're thinking of" when you don't actually have a source is a great way to find high quality reference material.
Good point. I didn’t intend it, but I will next time
5. Solar panels have improved massively in the last few decades, as have GPUs. I can’t speak to dialysis machines.
Of those, only GPUs are directly relevant to AI capabilities. We could have an all-geothermal or all-nuclear or (god forbid) an all-fossil fuel AI revolution. Solar panel improvement has been a decent proxy for power generation improving in general, but it’s less directly causal.
So overall I’m confused by the claim that they ‘would have given a very similar estimate for “present state of ML algorithms” 20 years earlier or 20 years later.’ If you said “look how amazing solar cells and GPUs are in 1990 or 2000 or 2010 those would be very different claims than each other, let alone as compared to 2020 or 2026.
Unless the argument is “the AI inputs are getting better and people aren’t becoming cyborgs faster” at which point it’s a restatement of the original AI argument from the early 1900s (or arguably earlier in science fiction) of “computers are getting better much faster than people”, but in a way that’s more directly measurable.
The full list was:
- dialysis machine energy cost
- artificial heart energy cost
- solar power energy efficiency (has only improved a little, as John Schilling pointed out)
- solar power payback period (I think this has improved a fair bit, due to much cheaper manufacturing)
- V100 GPU FLOP/s/watt
- V100 GPU manufacturing energy cost
- photodetector performance/power
- locomotion energy cost
Most of the items on the list pertain to physics and energy efficiency, which tend to lie in a fairly narrow range of a few OOMs. Perhaps a better reference point would have been algorithmic efficiencies in non-ML fields. These have widely varied, but for some problems can reach 6 or even 12 OOMs of gains over a few decades of research (for a given problem size, with the estimate being larger for larger problem sizes and smaller for smaller problem sizes): https://ide.mit.edu/wp-content/uploads/2021/09/How_Fast_Do_Algorithms_Improve.pdf
GPU compute efficiency is about the only one on the list strongly "related to" ML, but even that is a different thing than ML algorithmic efficiency. It's FLOP/watt rather than FLOP/task. So there's not a strong basis for using one to anchor the other, especially if both are changing fairly rapidly in time.
Solar panels have improved massively in cost in the last few decades, but not really in efficiency of energy conversion - that's been pretty steady at 30% for the best cells, or ~20% at the panel level, since the introduction of triple-junction cells 20-30 years ago. And I believe it is the efficiency, not the cost, that was being benchmarked in the BioAnchors report.
According to wikipedia research efficiency has continued to improve substantially: https://en.wikipedia.org/wiki/Timeline_of_solar_cells
2007 - University of Delaware claims 42.8% efficiency
2008 - NREL claimed 40.8% under artificial conditions
2019 - NREL claimed 47.1%
2022 - Fraunhofer ISE claimed 47.6%
Commercial articles claim smaller but still significant improvements; up from ~15% in 2010 to ~22.5% in 2023. https://www.solarreviews.com/blog/how-has-the-price-and-efficiency-of-solar-panels-changed-over-time
Note also: https://www.sunsave.energy/solar-panels-advice/how-solar-works/efficiency (this article agrees with the first's commercial figures, or even suggest they're a little higher, but claims that 30.6% is the highest efficiency ever achieved in research).
How much of that solar panel cost improvement is attributable to government subsidy, as opposed to material improvements like less expensive factories or inputs or maintenance?
Last time I priced stuff out, it was BOTE and for a hypothetical two-story house in Texas. The figures I had been given suggested the electricity very slightly paid for the panel during its estimated lifetime. This was order of ten years ago. I could see cost improvements in panels since, but for the same reason I'd imagine cost improvements in alternatives at the same time. I want to know how cheap a panel _could_ get, without being propped up, and it's hard for me to get any information that isn't from a motivated source.
Having written this, it occurred to me to get a little further on the question by asking ChatGPT. Rough summary:
* A panel uses silicon, glass, aluminum (for the frame), copper wiring, and a little silver. ChatGPT itemized that and got $10-20 per square meter.
* This of course ignores factory overhead, assembly labor, and shipping.
* It concludes "well under $20/m^2" for the raw mats floor, but that doesn't agree with "$10-20/m^2" from above IMO.
* Power output is 100-200 watts peak, assuming 1000 watts of insolation per m^2 (but I think this assumes the equator; at 30deg latitude, I would expect around 70% of this).
* A "super-cheap 10% efficient panel" costing $15 (split the diff from above) pulls 100W, then, or about 15 cents per watt.
* Limits to getting cheaper include silicon purification costs, and durability vs. UV, rain, and temperature cycling.
* A panel can last 20-30 years.
This is all ChatGPT numbers; they _look_ plausible to me, but I might dig into the insolation amounts and metal prices more.
A map I found at https://www.electricchoice.com/electricity-prices-by-state/ claims 15-18 cents per kWh to be "average". It cites this rate in AZ NM TX AL FL and the Carolinas. CA is by far the cost culprit at over 32 cents; HI charges almost 40.
And again, this is without pricing in factory, labor, shipping, and energy storage. Doesn't look good for solar so far, based on what I'm seeing.
> Doesn't look good for solar so far, based on what I'm seeing.
I dunno. If I can buy a square meter of solar for 20$, and that saves me 1kWh @ 10 cents per day, that thing would pay for itself in 200 days, which is a lot shorter than its mean life time. Now obviously we can argue about the details, but if something is quite viable over very optimistic assumptions you can not conclude that therefore it is not viable.
The point here is that it doesn't save you 10 cents per day. It breaks even or saves you at most 2 cents per kWh, because that's what you would pay to just get electricity the traditional way, and if you ignore "factory overhead, assembly labor, and shipping" from above. If you _do_ factor those in, you're probably losing money.
I'm not sure about the last ten years, but the twenty years before that saw substantial improvements in real manufacturing cost. About an order of magnitude from 1995-2015, as I understand it.
West Virginia is suing Apple for failing to snoop on users and prevent them from saving “child sexual abuse material” to the cloud.
https://9to5mac.com/2026/02/19/apple-sued-by-west-virginia-over-alleged-failure-to-prevent-csam/
In related news, net domestic migration into the sunbelt is slowing. The slowdown is particularly sharp in Florida, which is now gaining fewer residents from net domestic migrtion than Alabama, though still more than any blue state:
https://www.nytimes.com/2026/01/30/us/texas-population-growth-migration-census.html
Why do I think these two stories are related? CEOs do not mind moving their businesses to states where people have beliefs they find silly or irrational. If people want to believe that men can become women, that the world is 6,000 years old, or that Donald Trump brand steaks are the best steaks on Earth - so long as the customer pays, it’s all good. Likewise, if a large proportion of a state’s residents want to believe that people like our CEO friend are part of an international satanic pedo cabal, a belief not even Donald Trump, Pam Bondi, and Kash Patel can get them to abandon, he’ll let their podcasts and Instagram reels and whatever. But he isn’t willing to be sued because a politician wants to satisfy his many constituents who want to see “the elites” punished in some way for pedophilia. Conservative politicians who want their states to continue growing will have to confront this dilemma.
I first visited Florida in the early 90s when evidently its population was about 13 million.
It did not seem empty or abandoned. And I understand it has added 10 million people since then - or more? Not sure what we know about population anymore.
The recent announcement of the huge conservation easement protecting 61,525 acres - announced in conjunction with commemorating 25 years of Florida Forever - suggests Florida is still interested in growing. There’s other things to grow than people.
I think the Apple thing is a problem of its own making: if they had simply said "unlike some of our competitors *cough, cough* our business model allows us respect the privacy of our users, so we sell secure devices and our cloud-storage platforms are well-encrypted. We will comply with any warrant to the extent required by law, but no further," that's a position they can defend. "We're deploying on-device scanning but won't report to the government what we find unless it meets some threshold we determine," not so much.
>We're deploying on-device scanning but won't report to the government what we find unless it meets some threshold we determine," not so much.
My understanding is, they floated that idea, but backed down because the backlash against it was pretty strong.
3. I assume this was in some part inspired by the article on lesswrong about snake food (which was actually very wrong). In the comments, I pointed out that for biological reasons rodent alternatives simply weren't viable because snakes are picky eaters. I suggested that somebody should go for an ACX grant to raise rodents for snake feeding more humanely, which I think is a good business idea, rather than yet another fake rodent company that leads to unwilling snakes, unhappy customers, and a return to the rodents that actually works.
Scott, if this project is the latter, I do hope you checked with a herpetologist about the product viability issues they will face rather than just market research etc. It's not just the same thing as slapping their label on white label dogfood, this is a category that has absorbed a lot of R&D spend and still doesn't really exist, nor imo will it ever. It's hard enough to switch a rodent from live to ft let alone real to fake.
"switch a rodent from to ft"
What's ft?
From feeding a live rodent to feeding a frozen(killed at factory), thawed, and heated rodent. It is very difficult to get a snake to switch from live to frozen, and also difficult to switch from rat to mouse or vice versa. Snakes can be coaxed into going for thawed rodents that are heated, smell right etc. But they need to be familiarized with it and some just won't do it. Fake rodent substitutes have been more miss that hit
Thanks!
Re: did we all just miss that Bio Anchors critique — not sure when I first heard it, but I was under the impression that “2020 FLOP” not being anchored in 2020 was a “noted weakness” by 2023: https://www.lesswrong.com/posts/DgzdLzDGsqoRXhCK7/transformative-agi-by-2043-is-less-than-1-likely#YP5TBehwEFho6Fozr
Oh my, this is off to an interesting start. Rubbing sleep out of my eyes after a late night. Coffee may not do the trick this morning. Time for Mr Modafinal.
How much of a dose, and do you take it with coffee or by itself?
200 mg. Yeah, i washed it down with a sip of coffee. I have a legit prescription. Most mornings I either break a tab in half for a 100 mg dose or skip it altogether.
Hoping to get some feedback from the community on my review of Anthropic's moral guidebook, Claude's Constitution. I do fear I used my Moloch hyperlink too soon. https://substack.com/home/post/p-187822087
Has anyone heard any updates about Tornyol, the mosquito drones? Supposedly they're shipping in 2026. Mosquitoes will be in my yard in a few months.
Mosquitos don't bother me too much. Make sure there are no standing pools of water around the house, and there is always DEET. What really bothered me last year were the damn ticks.
Chiggers. Women hardest hit.
Oh I lived in TN for a while, and yeah chiggers are worse than ticks.
Except in the disease way, fortunately, since they can’t be checked for as ticks can be.
I made a sulfur sock to knock on my shoes, but problem is I only ever think about using it after I am aware that I’ve gotten into chiggers.
Never heard of but if it's a little drone that functions like an automated mosquito death machine + self charging Roomba capabilities, sign me up if they make one for roaches; though the buzzing would probably drive me crazy lol.
The idea is intriguing, but I don't like the subscription model.
Six years later, what are the policy failures coping with covid, and the successes? E.g.
1. Shutting down small businesses (but keeping WalMart, Amazon, open).
2. Shutting down schools.
3. Distancing-- the 6 foot rule
4. Mask requirements
5. Vaccine requirements
6. Delaying vaccine approval till the late fall, instead of allowing them in April.
7. Hastening vaccine approval tot he late all, insead of waiting till Jly 2021.
8. Vaccinating people under 60, and children in particular.
9. Moving to mail-in voting instead of requiring people to show up in person.
10. The CDC not being able to keep track of infection data.
11. Covid tests being heavily restricted by the FDA.
12. Use of expansive emergency powers by governors unilaterally, without legislative approval.
It is amazing that no investigations were ever made of any of these things.
Waiting until July 2021 would have been a terrible idea. Of course, I'm one of those who think they were too delayed already.
It's complicated enough to figure out with hindsight how much of a success/failure any of these was.
The real question, though, would be: how good was the decision *with the information we had at the time only*?
It has become almost standard to dunk on covid restrictions, but we do remember that we had saturated hospitals with people dying in the corridors and doctors/nurses catching covid, right? Some of the measures were probably not that great in retrospect, but at the time decisions had to be taken with very limited info, and "just let people die" didn't sound like a such a good option either.
We did NOT have saturated hospitals. Corridors are DESIGNED in hospitals to accommodate Actual Emergencies (see Entire City got Chemically Attacked or Nuked).
It took my hospital about 3 days to build negative pressure wards (which is what you want for an airborne virus, you don't actually need an emergency room's full suite). If your hospital didn't do this in 3 days, it's because they weren't treating it seriously.
Yes, we did have doctors and nurses catching covid, but we didn't hit anywhere near the predicted 20million dead Americans.
Just let people die was ONE OF THE OPTIONS selected in order to save nurses and doctors (again, with the idea of 20 million dead, Cuomo should have fixed the policy toot suite, and didn't -- he treated this like it was an Actual Emergency where killing nursing home patients made Some Actual Sense). This is WHY it's critical to look back at the actual public health decisions.
Pfizer deliberately delayed their vaccine trial. They would have had enough data but chose not to look at it as early as they originally planned. Doing so would have resulted in an announcement in the runup to the Trump vs Biden election, so instead they were able to announce it shortly after.
That has very little to do with anything I said. Did I say or imply that Pfizer was staffed by saints?
I was responding to the point about information available at the time. There was information which they refused to look at earlier.
Successes: Western Australia, and then to a lesser extent New Zealand, South Australia, Queensland, Tasmania, probably some small islands.
Failures: Pretty much everywhere else
A policy such as "shutting down schools" can only be evaluated in the context of what it's intended to achieve and whether it succeeds in achieving it. If you lock down until local elimination, and then you successfully maintain quarantine so you can live freely for a couple of years while the rest of the world suffers a pandemic, then that's the kind of outcome that makes extreme measures like school shutdowns worthwhile.
If you shut down everything and then don't actually eliminate the virus, then that is just plain silly.
Going on a merry-go-round through "locked up at home" and "locked up in jail" (where one couldn't actually stay at either place, due to contradictory laws) is now a success? Chasing escaped children from camps down with dogs is now a success?
(Quarantineing an Island with the help of the entire World Elite is a lot easier than quarantineing America, but There were Notable Failures in Judgement there too).
Actually, we don't even have a "What to do next time" report.
Yes, we do. Included in such is "do not use mRNA vaccines." (note: may be classified, or just not shared with the working public). You can tell that this is the case as they've been defunded (just like spy-cats).
That is just RFK Jr not liking them. He would share a "report" backing him up if he had one.
DARPA isn't run by RFK. DARPA decided their "quick immunization platform" (mRNA) was basically a failure. You do realize that OWS was half-staffed by military?
I tried googling "quick immunization platform" with "DARPA" and Google just suggested results without "immunization" so I don't know who you're quoting with that phrase.
A guy from the military side of OWS. As in, someone's been funding these mrna vaccines for years (well before covid19). We call that someone "the military" because the military is very concerned with our servicemen remaining functional under biologic attack.
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4495566
Which of the twelve you listed are the failures and which are the successes?
I have my opinions, but I tried to set it up so commenters could choose their own.
Failure to provide millions of data points to Americans, citing FOIA.
Use of ventilators for a non-respiratory killer (did more harm than good).
Cuomo's use of policies designed for 20 million dead Americans (sending people back to nursing homes while still infectious).
etc. etc. etc.
Vaccination as social signalling, Masking as social signalling.
Demonization of Ivermectin (a very safe drug).
Failure to push Vitamin D, actively trying to remove covid19 treatments from shelves, simply because they were offlabel usage.
The list goes on, and on and on.
Ivermectin is indeed known to be safe, but "demonization" of it didn't result in any increase in parasites as far as I'm aware of.
Demonization of it probably resulted in more deaths than it "fixed" -- that Rolling Stone article will live in infamy. You realize the hospital involved couldn't even look at the Rolling Stone article without a VPN?
It would only cause deaths if people were afflicted with the parasitic worms that Ivermectin protects against https://www.astralcodexten.com/p/ivermectin-much-more-than-you-wanted and the US has clean enough water that such parasites aren't much of a health risk here.
Second order effects are still a thing. As are third order effects. (And someone's bound to cite that Rolling Stone article as "why I didn't vaccinate Polly").
I object to the characterization of America as having clean water, as well, even if ivermectin doesn't protect against giardia.
You haven't established any second or third order effects.
You object but provide no evidence America doesn't have clean water, or that the parasites that ivermectin protects against commonly infect humans here.
I agree that this is an almost scandalously under-researched area. But I suspect that a big part of the reason is that the people who would normally fund such research are the ones who created the policies in the first place, and their prime directive is to never look bad in front of the voters.
My best guess is that all of the non-pharmaceutical interventions were next to worthless in isolation, and worse than useless where they contributed to the "I'm done trusting you guys" problem when we needed to convince people to take a hastily-tested experimental vaccine.
The vaccines were a huge win, and everything that delayed broad adoption was bad. But perhaps not terribly bad, because it isn't clear that we had the production logistics to do much better than we actually did even if an oracle had told us on day one that they were safe and effective. But it's also not clear that we didn't, so shame on everyone who screwed that up.
If we had got large-scale vaccine takeup months earlier, the improvement would have been greater than a naive estimate would suggest, because the vaccine would have been deployed against the virus it was actually designed for, rather than an evolved sub-subvariant of the original SARS CoV-2 that was pretty much extinct by the time we actually got the vaccine. But this also is going to be hard to calculate.
I recently got blocked by Phil Magness while discussing non-pharmaceutical interventions during Covid https://entitledtoanopinion.wordpress.com/2026/02/04/blocked-by-phillip-magness/ my argument was that a good Bayesian would likely conclude that NPIs can reduce transmission, ceteris paribus, but the question of whether to use them would have to involve a cost/benefit calculation.
< My best guess is that all of the non-pharmaceutical interventions were next to worthless in isolation, and worse than useless where they contributed to the "I'm done trusting you guys" problem
The non-pharm intervention that seemed most promising to me was ventilation. Air purifiers with high air exchange rates can reduce contagion rates substantially, and I suppose you know about the engineers who designed and spread the word about cheap improvised air purifiers. Even simple ventilation can be pretty helpful. In the first few months of COVID I used cross-ventilation in my office: patient and I sat 10 feet apart from each often, with air moving across the room at right angles to line between me and patient. And promoting these measures seems much less likely to make people feel controlled and jerked around than promoting masks did.
Good point. As to Ventilation Policy-- we didn't have one, not even government saying ventilation was important. Certainly no requirements, and no relaxation of other requirements if you provided good ventilation.
We had ventilation as a policy in doctors' offices and dentists' offices in my state (PA).
Yeah, if there'd been a big push to improve indoor ventilation, that might have helped. How much could usefully have been done on a few months' notice is questionable, though. Likewise UV sterilization. But we instead decided early on that we were going to Mask Up, Lock Down, and Socially Distance the COVID away, which were not unreasonable as first guesses of what to do but which needed a lot more revisiting in light of new evidence than they got in 2020-2021.
Actually, the expert consensus in 2019 was that Mask Up, Lock Down, and Socially Distance were all very bad policies. Somehow that changed over the course of a few months.
The expert consensus didn't change, but you do have to take into consideration that Covid-19 (OG) was considerably worse than what America got. If you go into it looking at 20million dead Americans, that's... a different ball game. Lockdowns were still a dumb idea, but masks were only "proven" wrong when airborne was proven (consider that it took 50 years to figure out that influenza was airborne -- not everything is clear-cut).
Fwiw, other countries have done evaluations of their policies, such as Sweden, Norway, and the UK (for the UK the reports are not yet fully out).
CORRECTED: I meant Sweden, not Denmark.
Links?
Sweden: https://news.ki.se/review-and-commentary-on-the-national-covid-19-commission-inquiry
Norway: https://www.regjeringen.no/contentassets/b1dace9390054c85a5a87c7bbf1bc384/no/sved/1.pdf
UK: https://covid19.public-inquiry.uk/reports/
THanks.
I wonder in what way you consider the vaccines a "huge win". I'm familiar with a contingent who claim they weren't safe, mostly on the premise that the method (artificially induced RNA transcription) was new, and therefore relatively untested, and they even cite a causal path by which they induce a risk of heart failure that wouldn't otherwise occur. And on top of that, it didn't even appear to keep a vaccinated person from infecting others, so you'd still have to keep the kids away from grandma.
Personally, I would have been okay with a vaccine that was "safe, but unknown levels of effective" reaching the market faster - in that sense, the system didn't satisfy me. I also wasn't on board with the mandate. Maybe for certain professions in theory, such as medical or military, but that argument weakens if the vaccine turns out not to block infection after all, and again if it turns out to affect mostly the elderly.
There were non-RNA vaccines available.
The AZ and J&J vaccines are viral vector vaccines, and J&J was available in the US at the same time as the two mRNA vaccines. Those weren't quite so new, though I think only one human vaccine had been approved using viral vectors. These use a strain of adenovirus, which is normally a cold virus in humans, but altered so it can't replicate (so can't really make you sick and can't spread) and includes DNA for the spike protein. AZ was widely used in Europe, but not approved in the US, and the Russian vaccine (Sputnik?) worked in basically the same way.
Natural coronavirus infections don't give lasting immunity, so it's not a shock that vaccination also doesn't. The way I understand it, you get at most a few months of immunity from vaccination (you make antibodies that will bind to the spike protein and prevent entry into cells), but then the antibody levels fall off and you end up with just T cell immunity. That's useful in that it clears the infection (by killing cells that show signs of being infected with covid) and makes you much less likely to get very ill or die, but it doesn't prevent catching or spreading covid. (This is not some new thing--for example, the inactivated polio vaccine we use in most of the developed world doesn't prevent catching/spreading poliovirus, it just keeps the virus from invading your nerve cells and causing paralysis.)
It never made sense to force people to get vaccinated, and the urge to do so seems like a kind of pathology of government--everything not forbidden must be compulsory, so you aren't allowed to get the shot until we've decided you are, and then you have to get it.
Vaccines were found to reduce transmission by about 50% https://x.com/TeaGeeGeePea/status/1937885522141302906 The claims about the vaccines making heart problems worse are BS, they actually improved heart outcomes because COVID itself causes heart failure https://respiratory-therapy.com/products-treatment/pharmaceuticals/us-pharmaceuticals/covid-19-vaccination-boosts-survival-odds-heart-failure-patients/
https://www.cidrap.umn.edu/covid-19/vaccines-cut-risk-post-covid-heart-failure-blood-clots-least-6-months-data-suggest
Your RT link is broken (you concatenated two links).
Once fixed, I can't see the methodology used to determine the correlation between getting vaccinated and not experiencing heart failure - the paper is accountwalled. And I can think of multiple possible ways to produce the result without addressing the causal path I'm referring to, including sample size, age, lifestyle, type of vaccine, administration technique, and monitoring period.
Your X link worked, and seems to be all there, and 50% sounds good, but there remains the question of the downside incurred in exchange for that 50%. None of the studies cited by that one appear to address this, either.
Thanks. I tried editing to add a space back in between the urls... and when I reloaded the space was deleted again. So I guess that's what happened the first time, and I needed to add a carriage-return instead.
What downside?
One downside I'm hearing about is myocarditis. If administered improperly, the vaccine can enter the bloodstream, reach and convert heart cells per their function, causing the immune system to remove those heart cells, leading to the usual risks. There might be others; it's not like mRNA has been running up data points as long as the measles or polio vaccines.
(Earlier, I mentioned that I'd be fine with such vaccines being offered anyway, provided they're presented as having had fewer tests than traditional vaccines (ideally specifying which tests), and aren't mandated. I.e. informed consent.)
Do you include half the population producing allergenic (at the very least) spike proteins for the foreseeable future, in your calculations? Spike proteins that cause parkinsons in primates?
Do you suppose the vaccination led to Biden's precipitous decline in mental faculties?
They aren't MY calculations, they are the findings of others who have studied this. And again, you are going on about spike proteins rather than measurable health effects that we actually care about.
No, I don't "suppose" that. Biden was very old! People who had dealt with aging parents had seen that sort of thing and were less likely to believe the White House's spin when they were hiding Biden's condition. The base rate of dementia for Biden's then-age was discussed at https://statmodeling.stat.columbia.edu/2023/09/29/donald-trumps-and-joe-bidens-ages-and-conditional-probabilities-of-their-dementia-risk/
I've talked with expert witnesses about dementia, and they were pretty damn clear that Biden had dementia in 2020. Either we have a country that is willing to elect someone incapable of governing*, or we have a situation (involving spoiled elections).
*Dark Winter. Biden's inability to compartmentalize (a sign of dementia) represents a significant harm to our national security.
Have you not read Seneff on this? Probable prionic activity, among other scary things (as a priori issues flagged by "what this looks like" before it touches a human). You can't really say that "we know all the aftereffects" when prionic activity could be a multi-decade issue (and consider, that we might have 20 year olds who are demented -- in fifteen years).
Ooh, "scary". I read predictions from years ago that everyone vaccinated would die, all BS, people claiming that the vaccines were causing "turbo cancer" despite no increase in the age-adjusted cancer rate, now you say it could take over a decade for evidence to show up. I say it's BS again and will bet against such speculation on whatever time range you want.
FTR, I haven't heard anything about "turbo cancer" from the vaccine. I'm mostly aware of the usual inherent risk from any vaccine, combined with the risk of mRNA in particular due to its method of turning your own cells into antigen factories, and having a considerably high chance of affecting critical cells that way. (I suppose it's possible for mRNA to mis-transcribe and turn one or more of your cells cancerous, but that seems like an ever-present risk given that transcription is a routine process in the body, and mRNA doesn't appreciably increase its frequency.)
What is "the usual inherent risk from any vaccine"? A virus itself turns your cells into factories.
https://www.researchgate.net/publication/357442961_Creutzfeldt-Jakob_Disease_After_the_COVID-19_Vaccination
Et voila (I am being good and not citing the Nobel Prize Winner from France). Again, this is something a priori research would have strongly suggested as a probability.
If your counts of "how many people died from the vaccine" don't include CJD, you're doing it wrong. Try starting from Seneff as a "places to look for problems," including ADE (antibody dependent enhancement, which wikipedia used to say was likely for covid19 as it was with dengue).
Persistent spike proteins show up in about 50% of the vaccinated, so yes, I may be looking at some of the odder outcomes, but I wouldn't like to say that "we were told all of the effects" that could have been gotten with a 6 month study.
I don't know much about the Sinovac vaccine from China that the one patient discussed in that article had. I've heard it was less effective than the vaccines used in the First World, as was also the case for the Sputnik vaccine from Russia. At any rate, a single patient who had both Sinovac and CJD is not enough to establish that the former caused the latter (which in this case certainly didn't take over a decade, since it was much less than a decade between the creation of Sinovac and that 2021 article). Rather than talking about including vs excluding various causes, I say to look at overall results for vaccinated vs unvaccinated. My bet would be that even with the lower-quality Sinovac, the vaccinated would have lower mortality if you adjust for age.
It's been much more than 6 months since the vaccine. Effects should be showing up. We don't see it in cancer https://x.com/cremieuxrecueil/status/2018460404293533883 and I don't believe we're seeing it in CJD.
My best estimate is that the vaccine saved 200-250,000 lives in the United States, mostly by reducing mortality in the "third wave". We got the vaccines at about the time we got to herd immunity by natural infection and immune response, but you'd still expect about as many deaths after the peak as before, and instead we saw the death rate drop much faster than that.
Not much benefit after the third wave, because we got new variants that were even farther away from what the vaccine was designed for but which had also evolved for lower mortality because why kill off the meatsacks that walk around coughing your offspring onto potential new hosts?
But I'll take those 200-250,000 lives and call that a huge win. And yes, someone will now tell me that >>250,000 people died because of the vaccine, which is nonsense and which I won't be replying to.
And, yeah, the mandates were a bad idea. I'd have seen at least a utilitarian case for it if we got the vaccine well in advance of herd immunity and if there were a real chance that universal vaccination could get us to herd immunity. But that didn't happen, and in any event I am not a utilitarian - bodily autonomy matters a lot.
The covid vaccines lower your risk of dying even when they're not doing anything to prevent catching covid, and IIRC that seems to work about as well for new strains as for the original strain. Herd immunity wasn't really ever going to work for covid (it's still circulating even though almost everyone has had it by now) because it does not cause lasting immunity.
I don't think there's any evolutionary incentive for covid to evolve toward being less lethal. The way I understand it, the thing that kills you with covid isn't the immediate infection where you're very contagious and maybe walking around coughing, it's weeks 2-3, where the immune response to the infection ends up killing you or at least making you very ill. Getting better or worse and killing people seems like it has no effect on how well the virus will spread.
My guess (this is how other coronaviruses work) is that the thing that made later strains less lethal is that most people had already been infected once. They didn't get lasting immunity, but their immune systems were able to respond a lot faster in the second+ infections, and they could usually clear the infections before they got to the point of the immune overreaction. Also, doctors figured out useful treatments (though they have to be started early), which wasn't true in the first few months of the pandemic, when they were throwing whatever they could think of at the infection in hope of saving their patients.
How do we know that 200-250,000 lives were saved due to the vaccine, as opposed to the virus mutating into a less lethal (and more transmissible) form? (I won't harangue you for links unless they're easy. A casual search turns up studies with plausibly consistent worldwide numbers, but I'm... suspicious of those.)
I'm happy to report I won't be claiming 250,000 vaccine deaths, and that seems outlandish to me as well.
OTOH, if it turns out that out that the 200-250,000 estimate is very high, or it's about right, but most of those turned out to be viral mutation instead, and in the meantime, vaccines killed, say, 2000 people due to heart rupture due in turn to destruction of mRNA-compromised heart cells, then that gap closes, by some amount we probably don't know, because it's both inherently and politically hard to measure.
If that gap closes all the way, then we're talking single-digit thousands of deaths either way, which is tragic but admittedly small when pandemics are being discussed. That makes them important when mandates are considered, but you already agree mandates weren't a good idea. (The sources I'm familiar with go a bit further and claim they were an especially bad idea for younger people.)
First, we didn't see substantial reduction in IFR until the Xi, er, Omicron variant, which was several months after most of the vulnerable population was vaccinated. And the big dropoff in absolute death rates, at least in the US, was right after the vaccines and before Omicron.
And second, the dropoff was not uniform. We have good state-level data on both deaths and vaccinations, and vaccine uptake was substantially higher in Blue states than in Red states. The relative drop in death toll, relative to the extrapolated pre-vaccine trends, tracks the vaccine uptake.
Vitamin D (actually Melatonin with its antioxidant properties, but the initial research twigged onto "get more sunlight"), was a perfectly serviceable non-pharmaceutical intervention.
The vaccines were a huge win -- are you sure about that? Given that we had a disease that was equivalent to "the morbidity of twice a bad flu year"* we'd have to find out that nearly all of Seneff's a priori issues weren't actually issues at all (and there's recorded more allergenicity for these vaccines, compare to the Swine Flu vaccine for "a reasonable guessimate").
*yes, we had two or three years, but at least one of those years was after vaccination.
Citing Malone on "why it's a bad idea to vaccinate everyone during an active epidemic with non-sterilizing vaccines..."
Saying: "we can make vaccines in less than a month" and then deliberately rolling out vaccines that weren't targetted towards Delta... that's a bad look.
Yes, we're sure about vaccines. The increase in deaths showed up as a large bump in excess mortality WAY beyond what we see with the flu https://www.astralcodexten.com/p/the-evidence-that-a-million-americans and we also saw a clear reduction in deaths for the vaccinated vs unvaccinated https://www.natesilver.net/p/fine-ill-run-a-regression-analysis
Annual flu deaths are on the order of 50k. Annual COVID deaths were about ten times this.
Repeat: a BAD flu year. That was like 2018-2019, which was about five times a normal flu year (I may have the date wrong, but look at a bad flu year, not a normal flu year).
I cite this not to be a pedant, but because "lockdowns" weren't even considered for a bad flu year.
This was not "responsible public health" this was sheer panic, if you want to go by the "actual numbers" (as opposed to the "oh my god" numbers pulled out of people's veritable hindquarters).
The 2017-2018 season caused an estimated 52k deaths which is the highest year I see in recent statistics (since 2010-2011; not immediately finding a longer dataset). I was already using "a bad flu year".
There was a US House subcommittee that produced a 500-odd page report on some of these questions, but more about lab leak: https://oversight.house.gov/wp-content/uploads/2024/12/12.04.2024-SSCP-FINAL-REPORT.pdf
Anecdotally it feels like people never recovered from spatial distancing. Line formation at convenience stores still seems like people are standing several feet apart when it would've been typical to stand right behind the guy ahead of you up before 2020.
Allowing masks to become a culture war marker seems like an inevitable problem down the road, although it probably "worked" to increase adoption in the big metros where they might have mattered. If they're smart, the CDC in a later pandemic where masking had utility would probably have to pitch them with "although masking was of limited use with covid, it would be more effective with [new thing] because of X, Y, Z..." in order to get widespread adoption across cultural/partisan lines. They won't do that, but they'd need to at least pretend they acknowledge that they previously exaggerated this (regardless of truth value) in order to overcome the hostility to them. Vaccination has a similar problem but a much easier fix, in that you simply approve some non-mRNA options and say "yes we understand people were skeptical of that tech, that's why we made sure these other options were available, we know you weren't against all vaccinations just against an unproven tech and your hesitancy was smart and valid" (again, regardless of truth value) and then avoid stupidly sabotaging the one approved traditional option as they did to J&J.
I remain baffled that governors' Wilsonian micromanagement of business and public spaces had no costs, I'd have expected more legislatures to revise statutes on what kind of orders can and can't be made without legislative approval. And I'd have expected those governors to pay a price. Gretchen Whitmer in MI was among the worst offenders in this area, and has done fine politically.
> If they're smart, the CDC in a later pandemic where masking had utility would probably have to pitch them with "although masking was of limited use with covid, it would be more effective with [new thing] because of X, Y, Z..."
I think the "CDC tells calculated white lies to manipulate public opinion" strategy was one of the biggest proven losers of the pandemic. Not that I have better ideas; there's a good 30% of people now who just won't take their advice seriously any more.
I think they would have to acknowledge the prior manipulation and relentlessly state bare facts and very highly restrained uncontroversial suggestions, and likely the top level media would need to stay in line here, and let the spin be done at a much lower level. Alternatively they'd have to be very clever about using messaging that was cynical and anti-establishment flavored to somehow percolate up through the people who won't take it from the top-down anymore.
People like me are just never going to believe them again because I realized with all those months of transparently obvious half-truths calculated to induce behaviors that they weren't actually talking to me. Or really to any people like us, the ACX commentariat broadly. Maybe I was supposed to realize this years ago, but I really only learned it from reading Zvi's writing about them during the pandemic. Even their routine recommendations are all dramatically overstated, because they assume (not entirely without reason) that if you need people to wash their hands several times a day you have to tell people that unless they wash their hands every 5 minutes a hungry tiger will sneak up behind them and bite their hands off. If you're a smart person and a statement they make looks implausible on its face, it is not a fact-statement and you aren't meant to take it at face value. You're supposed to infer what they *really* mean and what is *really* required from what they told the rubes, and then politely not say anything to interfere with the messaging, and the whole thing is tiresome and annoying.
> Line formation at convenience stores still seems like people are standing several feet apart when it would've been typical to stand right behind the guy ahead of you up before 2020.
Interesting, I don't make this observation at all. Queues have gone back to normal years ago here, pretty shortly after the last measures were lifted. But the plexiglas protection of the cashiers has stayed in many places.
I'm in the midwest United States and concede the possibility this is not everyone's experience. There isn't very high neuroticism here, so I don't necessarily think it revealed a preference for huge personal bubbles, but probably anything coded as politeness is a little stickier and with lower population density there's little to make it snap back.
Cultural Autism is higher in the Midwest? (huge personal bubbles...)
Social distancing is correlated with how far north your country is (Sweden has a very high social distance for conversations).
Not a majority by any means, but a lot of public-facing, mostly young service workers where I live continue to wear masks. I don’t know if they notice less illness in themselves that way, or if it gives a sense of privacy or relief from having to smile, or they just got used to it and it feels weird without. I just don’t think it can be virtue signaling.
I never got to liking the moist, recirculating-your-own-breath feeling, and will leave the city in the next pandemic if the authorities return to that playbook.
I couldn't really stand that either, we had a mandate in the courtroom so I had to deal with it a few times a week, and I got away with wearing a face shield the entire time. People looked at me like I was crazy, but of course it's much easier to wear for hours at a time, and I'm sipping liquid constantly all day so I'd have been pulling the mask down 30 times an hour anyhow.
I see very little of that now, when I do it's a co-worker who thinks they may be sick and is doing it as a courtesy. I fortunately have always had large offices to myself away from everybody, so I never worry about it.
Here mask-wearing has gone back to very low levels. Not quite as low as before Covid, but in the order of 1-5% of people in public transport, and practically zero in waiters or service workers.
My partner and I are among the mask-wearers in public transport, and my partner swears that it has reduced the number of infections notably for him. I am not so convinced, but I don't really care whether I wear a mask or not, and I don't see downsides to wearing it. It's not signalling, it's the same category as washing my hands when I have used the bathroom.
It was maybe hard to predict in advance just how big a deal the anti-vaxers were going to be. (You could have predicted there was going to be *some* push back based on prior experience, but maybe not the extent of it)
I think in any future pandemic, planning has to be: anti-vaxers are just going to die
(There is something of a moral hazard for the government here, where people who probably aren’t going to vote for the government next time are begging to be allowed to do something that might results in their death)
Antivaxxers are just going to die?? Are you suggesting... killing them?
He's suggesting that the pandemic is going to kill them. This won't be entirely true; no pandemic is 100% contagious and 100% lethal, but at least some of them will die.
My prediction is that if we have a major pandemic in the next 10 or so years, we are utterly screwed. Between covid, the Great Awokening, BLM, and RFK, US public health authorities have burned up basically all their credibility with approximately everyone, it will be politically impossible to impose any kind of restrictions (masking rules, lockdowns, etc.) anywhere even if this time it would make sense. The way vaccines were politicized (plenty of blame to go around there) will make it hard to get support for allowing fast development of new vaccines, and of course we'll keep our "you can't have it until we approve it, at which point we'll ram it up your a--" approach to public health, which will maximize pushback.
Who ended up an anti-vaxer at least partially depended on the vaccine delay and who won the election.
I just made a different comment above touching on this, but I think they invited this problem by making a series of calculated exaggerations to induce behavior throughout 2020 and by 2021 had allowed the whole thing to become a culture war flashpoint. In a similar later scenario, the PH establishment needs to suck it up and validate the fears and concerns of those people, and frame vaccinations in a way that acknowledges that and provides options that they can distinguish. It should've been easy to sell the J&J shot to skeptical people, one and done shot on an older platform, instead they weirdly marginalized it, had SNL joking about it like it was the Walmart of vaccines compared to the mRNAs, and then yanked it for 2 weeks in a scare that it never recovered from.
Your framing here would be the worst thing for them to say, it's what smug blue state liberals said last time and which turned out not to be that true, the disease was bad or trivial depending on your age demographic and the vaccination prevented some deaths but the craziest facebook group of 40-something anti-vax paranoids probably had zero people die. "You're going to die" is a message that would inspire "yeah yeah, heard this last time, didn't hardly make a difference, I don't know anyone who died". If it were a more serious pandemic with a less discriminating kill profile, you would see different responses to it, and targeting the message appeal entirely to midwit supremacist "I follow the science and I'm better than those chuds so I do what the CDC tells me to" attitudes is what *would* get people killed. If the government ran that playbook again, that's how I'll know they care more about their social signaling than about lives.
There is a question here of whether (a) the public (b) the government has learned the wrong lesson if, next time, it’s deadlier.
So, e.g., the government might conclude that convincing vaccine skeptics is entirely hopeless, so shrug, whatever. (While also getting in place the infrastructure for processing large number of dead bodies in short space of time)
I don't think vaccine skepticism in the broad sense is a particularly sticky belief outside of its original cultural home in middle-aged California crunchy hippie women or Kentucky homeschoolers. In its details, covid-era vaccine skepticism was almost entirely about mRNA particularly, I saw a lot of it because I was an NPI skeptic and there was a lot of crossover in those circles. The government PH establishment treated J&J more harshly than any kooky commentator did. I think this was largely a consequence of bad framing and bad marketing and really weird approval timelines. I could be wrong, but in a later pandemic with broader fatality I think if they approached it openly and honestly we would find there was much less across-the-board opposition than expected, in the intervening years there will have been little salience to the issue and thus it will not have been integrated into people's personal identity.
I had a former co-worker who was a smart enough guy to hold multiple patents (though also enough of a populist to support Mike Gravel and later Bernie Sanders), and I heard he was unwilling to get an mRNA vaccine. I asked if he would get a J&J instead, and said no, that method was also too novel. When Novavax finally came out with an old-fashioned tech vaccine for the US I asked if he would get that, and still no since it had been long enough he decided he had natural immunity. The government did J&J dirty (I got it anyway on top of a prior Moderna vaccine in hopes of Swiss Cheese defense), but I don't think that would satisfy the anti-vaccine people.
I was in the "waiting for Novavax" cohort myself and followed that saga quite closely, there were lots of other people commenting on all the news related to it on twitter who seemed to be in the same boat as me. Due to it being slow-rolled for some reason, I ended up having to get a J&J instead (that took quite a bit of calling around to even find!) It is certainly possible that by 2022 opinions had gotten stronger, and by the time Novavax was available in the US the pandemic fatigue was compelling most people to move on and they'd either had a shot or decided not to. But had it been available fall 2021, and pushed cleverly, things might've played out differently.
They easily could've spotted the mRNA skepticism in certain circles early on, leaned into the jokes and pushed J&J through those channels as a "look at those idiots getting two shots of some experimental thing that'll make you grow fins, I'm much smarter, one-n-done". It's a little screwy, but their approach had poor epistemic hygiene practically from day one, if getting vaccination numbers up permitted some loose messaging then may as well do it on all ends.
Asking PH to approach anything "openly and honestly" is trying to pretend there's not a revolving door.
Yes, with "broader fatality" we'd see ... more of everything, really. More deaths, more "it works, somehow", more vaccinations.
The "revolving door" actually results in more aggressive regulators https://accounting.wharton.upenn.edu/wp-content/uploads/2015/04/deHaan_etal.pdf the populist framing is wrong.
1. Obviously made no sense; there was no reason why big box stores would spread less COVID.
2. Is good, but that's because a large portion of schools are net negative even without COVID. Scott's already written about how kids can miss a lot of school and be fine, and school simply being in or out of session is associated with a change in 12-14% suicide rates among children. (Also, even though kids are unlikely to get serious complications, they can spread it, and teachers are not immune).
3+4. Probably made sense, these weren't very costly.
5. I'm not sure how much these actually got people to get vaccines. The history of antivax, which predates vaccines, indicates that trying to force people might just backfire by making them suspicious.
6. Bad on its face, although I've heard that manufacturing constraints meant it might not have mattered much, and rushing approval even more might have made people even more suspicious.
7. Seems quite obviously amazing.
8. This still seems fine. Even among young people, complications from COVID were multiple OOM more likely than complications from vaccines, plus they can still spread it.
9. Mail-in voting is fine so there's nothing wrong here.
10. I'm not sure what this refers to.
11. This was disgusting and heads should have rolled at the FDA.
12. This is kind of broad to evaluate all at once, but generally bad. At the very least, there's no reason for "emergency powers" to last so long--once the legislature has time to meet, they should pass actual laws, and then after the emergency they should review which powers actually made sense and clarify/change the law.
> Mail-in voting is fine so there's nothing wrong here
What about the obvious vulnerabilities?
1. You can steal ballots from mailboxes in areas where you think people will vote for the other guy
2. You can destroy mailed ballots in areas where you think people will vote for the other guy
3. You can intimidate people into voting while you watch, whether those are your own family members or other people that you have power over
4. You can of course just vote on someone else's behalf if you can manage to crank out a forged signature that will survive half a second's untrained scrutiny
5. The chain of custody of ballots is incredibly weak compared to in-person voting
Mail-in ballots have a legitimate use case for people who are truly immobile or live in truly remote places, but they're far too insecure to be a default way of voting.
We have been protected so far by the fact that risk reward makes this not worth it. You have to block a large number of ballots before you can shift the outcome and an operation at that scale becomes detectable.
In 2018 NC there was in fact a mail in ballot fraud scheme that threw the result for the congressional seat. It was discovered and the vote wasnt certified.
So all of these are vulnerabilities, but in practice they probably dont matter. But thats only true if other parts of the system remain viligant.
Wouldn't it make a lot of sense, then, to organize a lot of voter fraud for both sides, ensuring that the vote never gets certified? If you're already in power, it's a lot easier to justify staying in power under extreme circumstances.
Who would want to do that? If a vote doesn't get certified, the obvious plan is to re-run the election. That would take time, and presumably the current officeholder would hold that office longer. That means the challenger party would have little incentive to delay things, and the incumbent party would have great incentive, which in turns means that everyone seeing massive amounts of voter fraud would pretty readily assume the incumbent was doing it (cf. nation states with strongly suspected election fraud). So the incumbent party wouldn't want to do it, either.
> which in turns means that everyone seeing massive amounts of voter fraud would pretty readily assume the incumbent was doing it
If the votes were going to the challengers, would they be smart enough to see through the trick? Partisanship is at an all time high, after all. Either way, it doesn't matter. If the vote becomes impossible to certify, there's nothing people can do except kill the people responsible, which they don't have the spine to do.
The obvious vulnerability I'm worried about is someone saying "give me your ballot form and envelope, signed but otherwise blank, and I'll give you $50". This hasn't happened yet, but I could easily imagine it being done at scale in a future election - possibly to influence the outcome, possibly just to delegitimize it.
Is there any evidence that these are things happen with any significant frequency? Mail-in voting has been widespread in multiple states for a number of years. As far as I'm aware the actual rate of fraud of any kind is extremely low.
I don't think that, even in theory, these vulnerabilities actually represent a large opportunity. For example, your first suggestion. Looking through people's mailboxes is already a crime, and you would have to put lots of time and effort in to do this repeatedly and get ballots in between people putting them out and mail collection, which means a pretty good chance of getting caught. And in many places (e.g. apartment buildings, or some HOAs) mailboxes are locked and/or not publicly available. Then the actual usefulness of doing so is limited by the actual party split + voting rate in the area. And at least in my state, you get an email from the state when your ballot is received, so people would know that something had happened to their ballot.
I suspect that if this were worthwhile, we would observe a much higher rate of e.g. looking through people's mail for personal information or mailed checks, which are probably much more valuable.
I agree that mail fraud won't manifest as rogue actors breaking into a bunch of mailboxes.
The main threat vector I see is in the postal service itself. Mail carriers are obviously authorized to retrieve all that mail, and funnel it to the local office. Security on that leg is low. A single motivated carrier could sort ballots into a separate bag as they go (they're clearly marked), deliver the non-ballot mail, and quietly dump the ballot bag contents into a shredder at home. Thousands of ballots could be trashed this way.
One foil for this is if that mail carrier works a route with voters that mostly align with him. So, left-wing carrier working a right-wing route. This might be rare-ish, although I wouldn't be surprised if it mostly goes that way rather than right-carrier working left-route. If there's a way to acquire blank ballots, a carrier could trash some or all of what he gets and replace them with ballots for his candidate.
A few thousand votes like this might swing a precinct, but probably not a state.
Another worker at the office who's responsible for receiving the mail and sorting it could also foil this, but they'd have to watch the sorting, notice very few ballots, have reason to believe there would _be_ a lot of ballots, have a way to check (door-to-door survey of that route would be tedious)... and not be motivated the same way as that carrier.
A separate vector would be on the other side of that step - whoever collects ballots at the post office could either trash them without passing them on, or trash most of them and deliver the rest to allay suspicion.
Any conscientious voter in that precinct would be largely unable to detect anything amiss, so long as returns report some votes for R, some for D.
I don't know how mail gets handled at typical post offices; perhaps there's another mechanism that would foil this. But it's hard for me to imagine something that also allows voters to mail in ballots, or not, at their own whim.
I think a bunch of states have a website or text message where you can verify that your mail-in ballot was received, which seems like a pretty obviously good idea.
Well like I mentioned, in my state, Colorado, I get an email when the state receives my ballot. Some people might not notice, but if you're trashing thousands of ballots, I think it would be pretty clear. It would probably also be clear on the vote counters' side, if some areas had nearly 0 ballots.
This whole scheme sounds entirely speculative, highly risky, with unclear impact (how strongly are postal routes sorted by politics? how many people actually vote by mail?). Colorado has had all-mail elections since 2013, Washington since 2011, and Oregon since freaking *1998* (https://en.wikipedia.org/wiki/Postal_voting_in_the_United_States#Table:_No-excuse_postal_voting). If this sort of thing were feasible, I think there would be some evidence of it by now.
Email receipt helps, yes. Even if most people don't notice, a handful noticing would likely trigger an inspection that turned up the rest.
So I think that would keep a rogue mail carrier from just trashing ballots, unless I'm missing something else. Maybe not from duplicating said ballots, unless the email also echoes the voter's choices, in which case the carrier would need a way to spoof that as well (the "problem" here is that most carriers probably don't have the means to do all this versus just trashing the ballots).
I noted the potential problem with political sorting above. To be somewhat more detailed: a route is geographical, and a given carrier is presumably familiar with it, and probably notices which way people probably vote just through context clues (lots of flags hanging; "In This House, We Believe" signs on the lawn; etc.). So I'm picturing a left-wing carrier covering a suburb in a red county, a right-wing carrier covering a highrise, etc.
Either way, you'd need probably tens of thousands of compromised ballots to even have a chance of turning a state. OTOH, none of said carriers has to collude, so there's no conspiracy to leak. The obstacle there is probably the email receipts.
Fraud further on the chain is admittedly speculative, as I said; I don't know how ballots are handled further up. One hopes they're monitored by a bipartisan team beyond the carrier leg. "How many people actually vote by mail?" is somewhat answered by your own comment (all of CO, e.g.). Whoever runs those legs might have the means to generate emails (are they encrypted? Signed?). And while it's speculative for me, someone with a great deal of incentive - like, say, someone who believe he or she is saving the nation - would presumably put in the research.
I know I'm in the minority on this, but I don't think most of these were failures. That is, in hindsight we now know that COVID would mutate to be a lot less deadly, and that MRNA vaccines would be developed in record time and end up being quite effective. But hindsight is always 20/20. At the time, officials did not know any of this; what they *did* know was that a new and deadly disease was spreading like wildfire, hospitals were being overwhelmed, people were dying, and transmission was primarily achieved through airborne particles.
I think that the major failures of the COVID policy were these (off the top of my head):
1). Failure to secure enough quality masks and emergency respirators. This should've been done by an emergency production order. Instead, we had national officials posting YouTube videos on how to make a mask from a bandana and some paper towels, and people 3D-printing respirator parts in their basements.
2). Speaking of which, inconsistent messaging. It's fine to admit when new data coming in invalidates your previous recommendations; it's not fine to flip-flop around randomly based on the whims of public opinion.
3). Inconsistent lockdowns. The initial round of 2-week lockdowns was uniformly enforced (at least in my area) and arguably very effective. After that, lockdowns became just another form of political messaging and essentially random.
4). Failure to prosecute prominent public officials who blatantly violated COVID rules (I'm looking at you, Newsom).
FWIW I have friends whose relatives died from COVID, and others who lost their sense of taste and smell permanently (fortunately my own relatives survived unharmed), so I'm somewhat unsympathetic to claims that COVID was like, just a flu, bro. I realize that anecdotes are not data, but still, the issue is more personal for me than it is for your average Internet pundit.
Rather than mutating to be less deadly, I think it mostly mutated to be more infectious while our immune system evolved (through vaccines and prior infection) to make it less deadly.
When the new data says "the vaccines don't stop the spread" and so you recommend masking post vaccination, you've kind of shot the vaccines in the foot, as vaccines and not treatments.
Vaccines reduced transmission by half https://assets.publishing.service.gov.uk/media/65020dd6702634001389b7f7/The-effect-of-vaccination-on-transmission-of-COVID-19-a-rapid-review.pdf
Well, as far as I understand it, the general model of COVID (or any other viral infection is as follows):
1). Viral particles leave the infected person's body (the person may not know he's infected), usually attached to water droplets from mouth or nose.
2). They travel through intervening space, usually by air, or by touch, to the healthy person's body.
3). They enter the body by mouth or nose.
4). The viruses then attempt to infect host cells; the immune system attempts to fight them.
5). Some of the viruses succeed and go on to reproduce exponentially; some fail and are destroyed, but the process creates side effects. There are several outcomes based on these factors:
6a). Viruses overwhelm the host, resulting in severe long-term consequences or death.
6b). The immune system goes into overdrive, resulting in severe short-term consequences or death.
6c). The immune system fights off the virus, person returns to health.
All of these events are probabilistic, so if you want to reduce the number of deaths, you can look at reducing the probability of each step happening.
* Masks help with transmission, primarily (1) and (2) and to a lesser extent (3).
* Lockdowns significantly reduce (2).
* Social distancing mitigates (2) to some extent.
* Vaccines reduce (6a) in the short term, although they may increase (6b). In the long term, they act as selection pressure on the viruses, making them less deadly and thus increasing (6c) compared to the other options, as well as preventing (6a) in even unvaccinated people.
There's nothing anyone can do to permanently "stop the spread", and I really wish the messaging wasn't positioned that way. What you can do is reduce the number of hospitalizations and deaths, and you can achieve this by some combination of masks, lockdowns, social distancing, and vaccines. It would be really helpful to know ahead of time which combination of methods is optimal at preventing the spread of this specific virus, but no one has a crystal ball, so shotgunning all the techniques at once and seeing what works is the best practical approach (when facing a deadly and virulent disease).
Countries operating border controls to prevent the virus from becoming endemic were able to "stop the spread" and vaccinate their populations rather than waiting for natural immunity.
I like this calm and organized exposition.
I'm hazy on one point, though: "In the long term, [vaccines] act as selection pressure on the viruses, making them less deadly". How does that happen? My impression was that normal evolution, in the absence of a vaccine, tends to make viruses less deadly, since dead people are less likely to spread them. In fact, a priori I'd have expected vaccines to counteract that somewhat, since if the virus is in practice less likely to be deadly because it encounters only vaccinated people, there's less pressure for it to evolve to a form that would be less deadly in the absence of vaccines.
Am I misunderstanding your point or just missing some part of the argument?
From what I understand, vaccines tend to attack the most aggressive forms of the virus first, since they are the ones with the proteins that are most different from those of the host. But I could be wrong about this.
That doesn't really make sense. It has to depend on what makes the virus good at infecting new hosts, since that's what evolution works on. If becoming more deadly helps it infect more hosts, then it will evolve to become more deadly; if becoming less deadly works better, then that's what it will probably evolve toward.
Vaccines create an immune response to particular proteins--in the case of covid, the spike protein that sticks out of the virus and needs to bind to the right receptor on a cell to get inside and start making lots of copies of itself. One part of that response is antibodies, selectively sticky molecules that will stick really well to the spike protein but not to most other things. A virus with a very different spike protein than the one you have antibodies for won't have antibodies stick to it and so can infect you more easily. (The other part is cellular, and it notices some different parts of the spike protein than the parts the antibodies bind to, which is why the vaccines give you some protection from serious illness even when they no longer keep you from catching it in the first place.)
Huh, maybe. I guess I’ve never thought much about what makes a virus more or less deadly.
I think of the tradeoff for the virus as being deadliness vs contagiousness. We certainly saw Covid get more contagious as it got less deadly — but of course masking and social distancing would also select for increased contagiousness.
(I heard a comic wonder if the virus broke an ankle if it tried to jump seven feet.)
I tend to agree with you - 10/11 are straightforward failures and you could make a case for 6.
People often have this interpretation of the shutdowns as straightforwardly bad due to economic impact but I've seen at least some analysis of the fact that most of that impact would have happened anyway because people were scared and mostly would have changed their behavior voluntarily anyway.
Also, South Korea showed how much better it was possible to do. My takeaway from which is that the MAIN issue with the US response was forcing it into a partisan lens instead of working to do better versions of these things.
A different perspective: the main problem was failing to see legitimate concerns as legitimate. When you have substantial groups of physicists at Los Alamos looking at the data and saying "this is risky business..." they're not exactly going to be happy if you dismiss them as kooks.
What physicists at Los Alamos? I know Greg Cochran is a New Mexico physicist who wrote a lot about Covid, and won every public bet he made on it, but he lives in Albuquerque.
Is he in the cult? If so, you could probably just ask him. I don't know names, as I'm not in the cult (just know someone who is.)
The Supreme Court decided that religious objections to vaccination would apply if one had a "religion of one," hence the promulgation of a cult in order to have a valid religion to express one's issues with the lack of bodily autonomy inherent in vaccine mandates (and other issues with this particular vaccine).
I don't know what "cult" you're referring to, but Greg is and was pro-vaccine. He considers pathogens to be the great common enemy of humanity, and anti-vaccine people like RFK to be on the side of that enemy.
I would be interested to hear if you think the legitimacy of these complaints differed in the US vs. South Korea. In particular they had a much lower spike in unemployment while also having vastly fewer deaths per capita, even with slower vaccine access.
When I say "partisanship" here I am *partially* indicting things I see as conspiratorial/reflexive rejection of safety measures - as I think you are taking this - but also I'm just saying "the fact that this turned into a culture war topic instead of trying to address concerns and find a good solution", so in that sense we are somewhat on the same page.
Ah! we're discussing different things. I'm discussing legitimate concerns with the vaccine, including "endless production of spike proteins, which are harmful to health." (This should have been a foreseeable problem, and wasn't looked for within the "safety studies.")
There's plenty of scare-mongering about spike proteins from the mRNA vaccine, but they actually improved health outcomes like heart failure https://www.astralcodexten.com/p/open-thread-422/comment/218827671
In the Success category, our state governor stopped state IDs from expiring while COVID was on. I guess that counts as 12? Go 12.
My passport got renewed automatically as well.
My biggest frustration is not seeing any global, meta-research on the topic.
It was such a perfect natural experiment because we had each country going with slightly different reactions and policy all to deal with the same threat. Surely by now there are some obvious findings and learnings that can be gained from the pandemic?!
It's enough for a conspiracy theory of uncommon strength.
Contraindicated by nearly 100 years of pandemic research, mind you. Everyone jumped to be stupid at once (save Sweden, who literally couldn't have a shutdown, due to the high number of single mothers).
Strongly agree here. This is the perfect opportunity to see what worked, what didn’t, and what we should consider next time.
Also, it revealed that the current state of knowledge of infectious disease was/is seriously lacking. We should fund a global initiative to better understand how disease really spreads and how to combat it.
The current state of knowledge has been backsliding, as you can see looking at wikipedia, which used to have a primer explaining why covid19 was similar to the last disease that had its vaccine withdrawn (in indonesia).
OK Gemini says this has been done and gave an intreresting answer with 3 examples.
Can't share my query but here's the prompt I used:
"My biggest frustration is not seeing any global, meta-research about the Covid pandemic.
It was such a perfect natural experiment because we had each country going with slightly different reactions and policy all to deal with the same threat. Surely by now there are some obvious findings and learnings that can be gained from the pandemic?"
3 studies it cites:
"The "Big Three" Global Reports
If you want the meta-view, these three bodies of work are the "gold standard" for synthesized data:
1. The Lancet COVID-19 Commission (Final Report): This is perhaps the most comprehensive. It categorized the global response as a "massive global failure" due to a lack of international coordination. It highlights that the most successful countries weren't necessarily the richest, but those with the highest levels of "social trust" and previous experience with respiratory outbreaks (like MERS/SARS).
2. The Royal Society NPI Review: Published late in the cycle, this meta-analysis looked specifically at Non-Pharmaceutical Interventions (NPIs). Their conclusion: NPIs (masks, distancing, lockdowns) were highly effective when used in combination, but their individual impact varied wildly. For example, they found that "lockdown" as a single term is too broad; specific measures like banning mass gatherings were far more "bang-for-buck" than outdoor park closures.
3. The OECD "Policy Responses" Series: The OECD tracked policies across 38+ countries for years. Their meta-research shows a clear "scissors effect": countries that prioritized early, aggressive suppression (the "Zero-COVID" initial approach) actually had better economic outcomes in the long run than countries that tried to "balance" the economy by staying open longer and then being forced into chaotic, reactive closures.
Also there's a table with findings.
Thanks, I will duplicate your search and follow up.
The CEO of Microsoft’s AI division recently said that “Most tasks that involve “sitting down at a computer” will be fully automated by AI within the next year or 18 months, he said, naming accounting, legal, marketing, and even project management as vulnerable.” This is from Fortune.
Do people not believe him? Markets particularly. After all most stocks should have collapsed, as this can do nothing but create a recession, but in particular Microsoft which makes its money from office software, productivity suites, and hardware to run it - which are sold per head, should have been devastated.
What reason is there to believe him? Did he make any good predictions in the past, or is he just selling hype like some cryptobro? Microsoft's track record of anticipating large technological shifts is less than stellar. Slept on the internet and web, lost the mobile war, bought skype and promptly lost to zoom, finally surrendered in the browser war, etc.
Ca 2000, Microsoft was actually hated. Today, they are pleasantly irrelevant. I use Windows on my gaming computer and I am happy to report that their OS offering is adequate for that use case. I don't have to deal with them on my mobile or at work.
How should the markets respond?
If the standard assumptions about AI are true, we're looking at human extinction within several years, and market outcomes don't matter in those worlds. Edge cases have survival and superabundance, where market outcomes also don't matter. Same with civilizational collapse cases. The only outcomes that matter here are the weird ones, where we get a global AI Pause treaty, or a Butlerian Jihad, or AI progress just hits an unexpected wall somehow, in the very limited time we have left.
Will AI rapidly overtake most of the "sitting down at a computer" tasks? Probably. Should anyone make long-term plans or investments based on that? No, because it's much less likely to happen in the weird scenarios that actually matter.
He may be right that these things will be *capable* of automation, but getting actual adoption has to overcome some substantial hurdles that can't be overcome in 12-18 months. For one thing, many companies have their legacy data working with some niche piece of industry-specific software that's been around for 30 years and isn't easily integrated with AI agents. For another thing, the insurance companies aren't going to want their big clients using this unless it's entirely locally hosted, and will presumably make their clients impose that in all of their vendor contracts with other companies that handle any of that data. So you would have to commit to a secure on-site hosted AI, which costs some money. Then migrating your data over to whatever new software IS compliant with the AI. Then you'll have retraining, weeks of designing new workflow charts, and several months of working out the unanticipated problems created by new processes. I'm not a tech installer, I could be overstating a little of this, but as a management problem I think the workflow and training issues are as substantial as that and maybe worse.
You also have to hit at least the performance reliability of a human, not merely top end capability. You can get effectively 100% reliability from a human employee on a simple high importance task, and some items require that, so no switching those tasks until the AI is consistently giving the same output to the same inputs on that type of task as routinely as your calculator would.
If 18 months is when the capabilities would be good enough to consider this, I'd expect another 2-3 years after that at least before widespread adoption. You could have Skynet exterminating humans with drone swarms outside while First Traditions Bank of Omaha still has human tellers logging into software with a 1986 copyright date.
I believe him. Most of those tasks are automated already, which is why everything works so poorly. In about 18 months, it will get worse, perhaps much worse -- until the market corrects itself. This will take much longer, but eventually we'd have affordable human lawyers, project managers, etc. again.
>The CEO of Microsoft’s AI division
So, a man with a niche job description has claimed that his job is really very important you guys, you'll see, you'll all see?
I think people don't believe him because they assess their own jobs as more complex than can be easily replaced by non-AGI models, and don't expect AGI within 18 months.
I'll give an example from my career: My job is Customer Success Manager for a data platform. I build relationships with clients onboarding their teams and convincing them to renew their annual subscriptions.
It's certainly feasible that AI can replace a lot of what I do, soon. But you'd need the AI to be charming on Zoom calls as well as make commercial decisions and real time pricing negotiations.
It's not that my job is super sophisticated. It's just the real world is harder to replicate in the way that code is.
Btw, if my clients were AI systems too then all my charm and human face advantage will be useless.
Is this true? I thought Microsoft was a mix of Office with Azure cloud computing, and the cloud computing stands to benefit from AI. They also own 27% of OpenAI, which stands to benefit if AI fully automates white collar work. I believe something like the CEO's statement (maybe not 12-18 months) and I still hold some MSFT for these reasons (people here should to tell me if I'm being dumb).
Why would markets tank? Why would Microsoft think this is a problem?
Like, I get why *workers* would be hurt by automation by why do you think stocks would fall? How would the guys who own the robots be hurt by replacing workers with robots?
Like, Microsoft sells Microsoft Office super duper enterprise edition to Exxon @ $500/seat (1). Microsoft calls up Exxon and says "Hey, you know all those accounting people that cost you $120k/year including benefits? What if replaced 90% of them with AI agents at $24k/year."
Exxon saves $96k/year per accountant replaced. Exxon stock go up.
Microsoft spends $4k/year to run the agent and so makes (24-4.5) an additional $19.5k per agent Exxon uses. Microsoft stock go up.
Accountant gets fired. Accountant needs to find a new job. Market doesn't move because the accountant is not a stock on the stock market.
If what the Microsoft guy says is true, and Microsoft is selling the heckers out of B2B AI, then stonks go up, not down, because stonks represent the people who own the capital/robots/AI, which are more valuable.
(1) Totally made up
1) Accounting is riddled with PII data, and firms with any digital security teams lock that down massively from use with AI tools, impeding it's utility.
2) Similarly many processes are tied to multi year contracts which will take longer than 18 months to unwind.
3) Having done my share of outsourcing in accounting, a lot of those $120k p.a. jobs were outsourced to India already for 30-35k USD p.a. (including the BPO firm markup, at least that was the rate 6 years ago). The remaining western headcount is largely for fixing all the mistakes those teams make (originally pitched as "you'll have more time to influence strategy/decision making!" but reality is not the pitch). What you're likely to see is a reduction in the error rate as the Indian BPO workers are replaced by AI (which still makes errors), leading to slightly less of a burden on the western teams. That said, my company has been rolling out GenAI tools in accounting (completing basic journals) and it's been even worse than the BPO performance, which was itself worse than when we had the capability locally. I expect in 18 months that pipeline will have been patched with better instructions and underlying models, but it's still not as easy as it would look from the outside (as it almost always is).
Right now my computer security is so locked down at my CPG employer - my main use of GenAI models at work is getting them to write VBA code for me. I lost the ability to install freaking python of all things at some point in the last 12 months, so now I just use Excel as my hack to automate stuff. I'm not at all skeptical on capabilities in the 5 year timeframe, but rollout is another thing.
One of the first things I did when taking on a new accounting leadership role in the last year was blind email someone at Anthropic to see if I could talk to their team about financial automation, sadly their sales contact didn't quite seem to know what to discuss with me. If anyone who works there wants to talk I'd happily have a conversation.
> Accountant gets fired. Accountant needs to find a new job. Market doesn't move because the accountant is not a stock on the stock market.
This is the problem here though. If you fire all the accountants, or indeed 90% of all white-collar workers, then you've suddenly got a huge demand problem on all companies. Exxon has cut its costs by 5% but finds that demand for its product is down 30% because unemployment is 35% and nobody can afford to go anywhere.
In the long term things reach some kind of equilibrium, but in the short term everything is screwed.
Oil and gas can be burned for energy, which powers more data centers. They're the ones who would have one of the least amount of issues transitioning to an AI-centric economy.
"Exxon saves $96k/year per accountant replaced. Exxon stock go up."
Exxon then get hit with yuuuuuge tax fines because AI accountants feffed up ambiguous new regulations around reporting/paying particular taxes which, naughty naughty, they did not report/pay since AI accountants said don't need paying/reporting but Revenue say nuh-uh, you gotta report/pay! Stock go down?
It seems like you assume that AI accountants will, on average, be worse at this than human ones.
My expectation is that you replace 80% of human accountants with AI accountants that handle 95% of the work and escalate cases with ambiguity to the remaining human accountants.
I'd expect those AI accountants to make fewer mistakes on the procedural/predictable activities and successfully identify more cases of ambiguity.
This assumes that the AI can meaningfully distinguish anything. No object model, GIGO.
I think that what you describe is the model many people imagine for AI in all sorts of realms. It does the tedious busy work and leave the hard cases for expert humans.
But I worry that to become a good expert human, you *need* to work on those tedious cases to build up your knowledge. If nobody is doing that, then nobody will become an expert.
Similar in concept to how they say the software engineer pipeline is now broken because companies are replacing entry level positions with AI.
The problem is that right now running any sort of AI is deeply unprofitable so Microsoft would be losing money on every single agent deployed and trying to make it up with volume. On the other hand normal enterprise software has some of the best margins in business. It costs literally nothing to create an extra copy of Microsoft office.
Microsoft has a real moat and millions of users locked into its enterprise environment. If AI is really going to change all this then Microsoft is as likely to "win" as Anthropic, or Google, or OpenAI, or anyone else. And depending on how you see future AI development the winner just makes everyone else go out of business(making all of these companies inherently risky) or they all start competing in massive subsidy wars where they try to gain market share by selling agent for less than they cost to run, which also likely means that the losers will burn cash for no benefit.
Here are my scattered thoughts on what'd happen if what the Microsoft AI CEO said is true, some of this is tangential to your points about the stock market:
1. There are ~165M US workers. Maybe ~40-50% of US workers are in white collar jobs or jobs that involve mostly working on a computer. I'm uncertain of this exact number. These people are laid off.
2. White collar and computer jobs pay significantly more than non-computer jobs and also contribute more to the federal and state budgets via payroll taxes
3. The laid off are forced to take lower paying non-computer jobs or subsist on welfare, mostly from government. They substantially reduce their consumption and they default on things like mortgage payments.
4. US Government revenue is greatly reduced due to fewer and poorer employees paying less income tax. ~54% of federal revenue is from individual income tax, another 30% is from payroll taxes.
5. US Gov must find a way to a make up the much increased deficit as more people qualify for benefits, but much less revenue is coming in. The Gov could cut benefits.
With much less consumer demand and adding more deficits to already large red budget, what would happen to the stock market? Who is going to buy the products that many companies offer?
> Who is going to buy the products that many companies offer?
Other AIs. They're still producing things that are useful, that's why they would be being run in the first place. Energy instead of food. Silicon instead of medication.
Just a side issue, but during the discussions a few years ago about working from home/return to the office the usual term to describe sitting-at-a-computer workers was “knowledge worker.” Today the usual term to describe AI-displaced workers is “white collar.”
Why the change?
I think the idea is that Microsoft isn't an AI leader and will probably be losing its Office business while OpenAI/Anthropic/GDM/etc sell the AIs.
Another correction I thought you would make about the Sam Kriss article was his attribution to you AI 2027’s predictions: the article makes your timelines seem shorter than, as far as I understand it, they are.
I think it fairly attributes those predictions to the collective effort.
I used to spend, or should I say waste, quite a bit of time playing PC games, until one day I realised that I was just chasing pixels round a screen, reacting to program code, and this was no different in principal to my cat chasing the dot of a laser pointer round the floor!
Is even the smartest AI any different in principle to a PC game? Clearly not, because both are no more than programs and data. That is why I think all this talk of AI consciousness is, to put it bluntly, a load of hogwash. It's no more conscious than a monster in a first person shooter game, despite the appearance of consciousness in both.
You are your brain. Reacting to sensoric neurons getting stimulated by what you hastily call reality. You have no clue if the world outside your brain exists, or even if your brain runs on the substrate you think it runs on.
So your substrate chauvinism is uncalled for, unless you happen to demonstrate abilities which computers lack, such as solving the Halting Problem.
I thought you were comparing AIs to yourself for a second. You know, the part where you're endlessly chasing rewards without even knowing why.
So what is this fantastically meaningful way you spend your time now?
Saying that an AI is "no more than programs and data" feels like saying a human is "no more than meat and chemicals." It turns out you can do a lot with seemingly simple components.
To be fair, I play games as well, and I know full well they aren't real. Still, I also watch movies and read books and play board games with my friends, and none of those activities involve producing anything of lasting value either. I think you can make a convincing argument that entertainment of any kind is inherently wasteful and should be discouraged (I certainly cannot disprove this logically), but I think most people would disagree with you. Then again, most people aren't very smart...
In retrospect, the first paragraph of my post was superfluous and only muddied the waters. The only point I intended to make was that the apparent consciousness of AI with a finite amount of code and data seemed to me to be illusory, being no more real than a monster confronting one with apparent ill intent in a first-person shooter. But, from some of the replies, I can see the problem is not as cut and dried as I would like to think!
Oh don't get me wrong, I do agree with you that LLMs are no more conscious than perhaps a bee (which is still a major leap forward in consciousness, compared to calculators !). But I disagree with you that a "finite amount of code" could never in principle achieve consciousness. My head contains a finite amount of neurons too. Just because calculators can't do it, ELIZA couldn't do it, and LLMs still can't do it, doesn't mean that nothing ever could -- and it does look like we're very slowly inching closer to that distant future day.
Explain your reasoning that the AI monsters are not conscious.
How would an alien know you are conscious but the AI monster is not?
For anyone interested, I once wrote a shortstory to try and get at this problem:
https://objectiveobservations.substack.com/p/contact-of-the-strange-kind
What do you think makes humans conscious? (And will programs and data, never ever be able to emulate this in your assessment?)
> What do you think makes humans conscious?
I personally think "consciousness" is a direct result of "self and other simulation" capabilities, which largely come into being in mammals with the neocortex.
When we got warm-bloodedness, we set ourselves up for the much more computationally complex abilities that we enjoy today. Birds, notably, are the only nonmammal species alive today to independently evolve warm-bloodedness, and birds are reservoirs of similarly complex behavior as mammals.
But as we step higher up the “mammal” chain in terms of neural complexity, we get more than just simulating various possible outcomes in the world, which is what most brains have been about up til this point, we additionally get self and other simulation, including (ultimately) being self aware and projecting our own selves into the future.
There are three broad abilities that seem to have emerged in early primates:
* Theory of mind: inferring intent and knowledge of others
* Imitation learning: acquiring novel skills through observation
* Anticipating future needs: taking an action now to satisfy a want in the future, even though I do not want it now
These may not, in fact, have been separate abilities but rather emergent properties of a single new breakthrough: the construction of a generative model of one’s own mind, a trick that can be called “mentalizing.”
One interesting implication of the “self and other simulation” capabilities that the mammalian neocortex enables - consciousness itself is likely a flavor of self-simulation.
It seems pretty obvious to me that once you get to “self simulation” and a big enough brain to understand your own persistence and future states, consciousness becomes A) possible, and B) evolutionarily adaptive, because it gives you salience and valence for your future states, and enables “marshmallow test” style decisions where you take a short term penalty now for more rewards later, which is clearly net adaptive.
And indeed, if we look into the literature, this is exactly what we see!
Interestingly, rats cannot anticipate future states and monkeys-and-higher can. Rat brains are about 2g and even small monkeys are 20-30g - what do they get for their 10x bigger brains? They get the ability to self-simulate into the future and pass a marshmallow test. Rats fail at this, and squirrel monkeys pass (Naqshbandi and Roberts 2006).
Being able to simulate your own future states is necessarily the first step towards consciousness-as-self-simulation being valuable in the sense of evolutionarily adaptive.
But I think the real key here is, what else would be required to make that simulation of your own future states adaptive? You would need to feel connected to that future state, you would need to care.
What do consciousness and qualia do? They allow you to predict those future states, they make you feel connected to that future state, and they make you care. All qualia is, is valence. You care about future-you’s experiences because you’re going to have to experience them! You’re going to have to live through that qualia.
After all, once you have a little “you” that lives in your head at ALL, that you simulate doing different things, that persists in time and reaps the future consequences and benefits of your decisions, how far a jump is it for that little you to live in your head full time, and to represent that it’s experiencing the various things you end up doing?
If you stumbled across this internal innovation, if you suddenly had qualia instead of being a p-zombie, you’d be a strictly better savannah ape, because you’re going to take your future-impacting decisions more seriously and make better ones.
So maybe we should briefly touch on qualia, because it seems so core to the hard problem.
The fundamental mystery of mechanical processes ending up as qualia, or being experienced as something, has never seemed like much of a mystery to me.
Like, what ELSE could produce “experiences?” Doesn’t everything need some sort of substrate to even exist / happen? Our substrate just happens to be a physical universe with matter and energy, and sensory and processing organs made of meat.
For anyone out there that are pet people, it seems to me that dogs and cats have as much evidence for having qualia as humans. They have moods, they clearly choose to experience various things for more or less the sheer joy of experiencing them, like rubbing their faces in snow, or jumping or playing for the joy of it.
And you can take it all the way back - smaller mammals, birds, and reptiles certainly experience things. Bacteria too. The thing is that “experiencing” is being used in two senses here.
I think qualia is just something like consciousness or some adjacent mental process internally meta-tagging “experiences” (in the sense that bacteria, mammals, and reptiles have and react to) with valence, thereby creating an internal qualia “experience” (a second sense of the word experience).
It seems to me to be such a simple evolutionary line from “non-sentient bacteria and animals having an experience of something (sense 1) because you need feedback loops for successful goal-directed behavior” to “humans, descended from a billion-year line of animals / bacteria, also have experiences (sense 1),” and when coupled with their interior future projections of themselves, and the obvious evolutionary advantages of caring about those futures, they have qualia too, and so have valenced interior experiences (sense 2), qualia.
If you couple valence with future-state simulation, you reach the level of future-state navigation that humans enjoy, which is a significant enough jump that we dominate the planet.
Qualia gives us the sign and vector magnitude of rewards / punishments we’re signing up for with our actions, so we care enough about future-you to make good-marshmallow-decisions today - obviously adaptive on net. So adaptive even squirrel monkeys do it!
There's no explanation as to why the first sense should lead to the second sense, or indeed how anything like qualia can exist in the universe. Qualia is not merely valence, "good" or "bad". It's the raw experience of seeing a color. The experience of hearing a sound. These experiences can be valanced, but that's not their primary characteristic.
There is no reason these evolutionarily advantageous computations you describe should have an inner experiencer. Why could they not happen in the dark, so to speak?
If you don't see the mystery, then.. idk, you do have qualia, right?
> There is no reason these evolutionarily advantageous computations you describe should have an inner experiencer. Why could they not happen in the dark, so to speak?
I've described the reason, but I was probably not clear or to the point enough.
Basically once you can simulate others, you can simulate yourself. And once you can simulate yourself, you will make better decisions IFF you have an inner experiencer and directly care about the outcomes because you know you will experience them, and can imagine yourself experiencing them. That's qualia.
Your p zombie can imagine themself doing something, but if it's just like imagining somebody else doing something, they care less. They are less motivated to make harder, marshmallow test decisions, and would make less of them. If you have qualia and valence and can really imagine that outcome in a personal experience way, you are more likely to make better decisions.
Two things:
That's just wrong, RL agents even today are perfectly capable of making optimal decisions like that, and in fact don't struggle at all with making marshmallow-optimal decisions. Delayed gratification is not a problem for a p-zombie RL agent. You can weight future decisions exactly as much or little as you want. It's not a real thing that RL agents don't "care enough", but only if they had qualia would care more.
Second, this doesn't explain what qualia *is*. Where does it come from? You say self modeling, but that doesn't explain what it is and how qualia is different from current RL agents' self modeling. Or do you think they are actually conscious? That would really be the intellectually defencible position according to your reasoning.
> That's just wrong, RL agents even today are perfectly capable of making optimal decisions like that
That's a total non-sequitor. Sure, we can after-the-fact program something to make "good" decisions, but we're talking about nature making an entirely new capability implemented in meat, with the starting point of bacteria and reptiles.
It's a completely different path and order of difficulty.
Saying "once we're here and conscious and know about marshmallow tests and persistence in time, we can artificially create a system that does the same thing using that knowledge" isn't actually relevant to the path of creating new organisms with de novo capabilities that end up with better self simulation capabilities and making better decisions over time.
> Second, this doesn't explain what qualia *is*. Where does it come from?
I don't *care* what qualia is, that's just going to get into definitional word games. Where does the neocortex come from? Where does the ability to simulate others come from? Where does the ability to simulate the self come from? This isn't a well defined question.
The original question was "what do you think make humans conscious?"
And I've answered that. It was the ability to self-simulate getting better, coupled with caring about outcomes via qualia, that led to being conscious and making better decisions being evolutionarily adaptive, so it happened.
Briefly, consciousness depends on the structure of the brain: how it relates to the both the rest of the body and the world outside of the body: and how different parts communicate with and rely on each other to integrate sensory data into a coherent whole. It is fundamentally a biological phenomenon. Below articles/reviews of work
by 2 neuro-scientists who have proposed different theories of how this works: Anil Seth and Anthony Damasio.
https://nautil.us/inside-the-big-bet-on-consciousness-1207750/
https://www.sciencenews.org/article/feeling-knowing-book-consciousness-origin-evolution
https://www.nature.com/articles/s41583-022-00587-4
https://nautil.us/whats-so-hard-about-understanding-consciousness-238421/
On a second thought, I'm not sure why he thinks robots don't have something "homeostatic and regulatory", how does he think drones operate, or active cruise control, or any of myriad machines with control loops for that matter...
At least Damasio doesn't claim that consciousness is fundamentally biological, here's a quote:
"For robots to be conscious, we would need to give them a bit of vulnerability that they don’t have [right now]. If you could introduce into a computer something that would be homeostatic and regulatory — that would allow it to sense deviations — then you would be on the way to creating feelings for the computer, and it could detect its own inner states."
This would make, for example, a smart thermostat... conscious? At least a little? Sitting there, being aware of getting cold, and operating to restore the equilibrium?
Saying consciousness is fundamentally biological, is fundamentally ridiculous. Is it the carbon molecules that are fundamentally necessary for consciousness? If not, what *precisely* do you mean by biological, and by what mechanism is that prevented from being emulated?
I'll be honest, I haven't read the links, but those characterizations are typical of people who don't take the hard problem seriously. Or more likely, don't get the hard problem.
Well, maybe you should. Read the links. Before smugly dismissing the work.
Yep I've been reading the articles (not the papers, which are presumably way beyond me and also paywalled).
Unfortunately my smug dismissal was on point. Damasio doesn't seem to get the hard problem *at all,* while Seth seems to specifically "bet" that solving all the easy problems will dissolve the hard problem. Which of course, is exactly the point Chalmers was making: he called it "the hard problem" exactly because he was trying to get at the idea that it *won't* dissolve, like other scientific questions have when looking at them close enough. Hence, the hard problem. It's hard. And neither man seems to get the question, much less to begin getting at an answer.
(To be clear, working on the easy problems is certainly valuable and I would not hesitate to fund either man's science projects. It's just mostly unrelated to the hard problem of consciousness, which Seth acknowledges.)
But then I'm not sure why you dismiss a possibility that consciousness is fundamentally biological? BTW biology is far more insanely complex than just "carbon molecules", just as a car is far more complex than "Fe atoms". This is indeed "The Hard Problem", and we can't ever reliably say which organisms are conscious, never mind assigning it to machines. So I wouldn't dismiss the possibility that indeed there's something special about biology, at least until we start getting evidence to the contrary (silicon-based life has been proposed by sci-fi writers, for example).
Something, something pineal gland.
You're one of the people who thinks a programmer sits down and writes each of an LLM's possible responses. That's not how it works. They don't pick responses from a list.
Do you have an example of something you consider "conscious" ?
I hold that chasing money, health, or any other metric of value is equally laser.
If "pixels on the screen" is your true objection, soon we may have robots.
"No different in principle from a Roomba."
Have you tried to engage the arguments on this? Your argument seems to be:
-no combination of programs and data can be conscious
-AI is a combination of programs and data
-therefore AI can’t be conscious
I think your first premise is very suspect.
Is the Good Reptiles snake food supposed to be vegan or something? From what they're currently advertising, it looks like they make essentially chicken nuggets, but from mice. Is the animal welfare angle just that they're killed more humanely instead of being swallowed alive?
The nuggets aren't made out of mice, but I'm not sure how much I'm allowed to reveal so I won't mention what they're made of other than that it seemed to me that they had a reasonable case for it being more humane.
This might be a stupid question, but I thought the whole point of feeder mice was that snakes could only eat live prey, and would refuse to eat e.g. plain old chunks of meat ? Is this not accurate ?
It's common to feed snakes frozen/thawed mice. Live prey involves risks and is more complicated, so it's not preferred by most professional breeders. There might be some species that require it, I'm not sure.
Caveat: I only have experience with typical pet snakes (i.e., nothing huge and nothing venomous).
Refusing live prey varies from individual snake to individual snake, even within a species. Most discussions I've read treat this as a learned behavior - if snake is raised on live food from a young age, they're more likely to refuse non-live food as an adult.
I've fed my girl thawed frozen prey since she was a few months old, and in my experience a feeding response is typically triggered by heat and scent (easy to get chomped on the hand if it smells like mouse from not washing after handling, and has a warm internal temp, despite the hand not looking like a mouse at all). Sometimes I'll wiggle the mouse around and that seems to be convincing enough when she's being picky most of the time.
I don't see why properly thawed 'chunk food' with the right scent couldn't work for some snakes.
Yeah, I thought that too; there are some reptiles you can just thaw out your frozen mouse carcass and feed it to them, but others will only eat live prey?
Is it lab grown mouse cells? There have been a few forays into lab-grown pet food, e.g. Wild Earth. EDIT: though IIRC, they weren't doing animal cells, but rather something else.
One possible reason why overall crime stats could be falling even as everyone subjectively feels that crime is rising could relate to the distribution of where crime is happening in major cities.
For example, I've been to Detroit and I really like it; I felt very safe. Buddy's Pizza is great, the Tiger's baseball stadium is great, highly recommended. Yet Detroit, overall, is extremely not safe. It only felt safe to me because I'm a boogie bitch and went to the downtown/tourist area, where everything is safe because that's where all the external and internal tourists go. You can look at a crime map of Detroit and amongst a sea of red there's a little yellow/white blotch by downtown. If you just go there, Detroit is great.
https://gisgeography.com/detroit-crime-map/
Seattle inverts this. Seattle is an ocean of yellow/white safety with an island of bright red crime centered right on Pike's Place. And you can feel this if you walk around downtown Seattle; when I was there I saw crackheads everywhere. It felt way, way less safe than Detroit, even though it was objectively safer. Why it's like this, I have no idea, but for some reason Seattle has all its crime next to Pikes Place and the Seahawks stadium, where everyone can see it.
https://gisgeography.com/seattle-crime-map/
I kind of see this replicated comparing San Francisco and St Louis. When I look at a crime map of San Francisco it's not literally in the downtown but there's a huge hub of crime, like, riiiight next to it (https://gisgeography.com/san-francisco-crime-map/) and everyone I know in the Bay Area has stopped going into SF because of a combination of crime and traffic. Meanwhile, in St Louis, which has been in the running for murder capital of the US and parts of it literally look like a 3rd world warzone, the downtown area near the Arch is reasonably nice, cool area, you can walk a couple miles up towards the hospital and park, it's not great but it's cool and safe (https://gisgeography.com/wp-content/uploads/2023/07/St-Louis-Crime-Map-1-1583x2048.jpg).
So the issue isn't whether overall crime is rising or falling, it's that in certain cities it's moving towards way more visible parts of town. Ignore tourists. If you live anywhere in the Bay Area, from Fremont to Contra Costa, you're not going to see a fall in crime in Vallejo, even if that's where all the crime actually happens, because you never go to Vallejo, because why would you unless you need a priest and a truckbed of malt liquor? You see the crime in your neighborhood and the crime in SF because SF is where you go to see a basketball game or go to an art museum or to do any big event. So the guy in Fremont notices that crime is his neighborhood has gotten a bit better and crime in downtown SF and on the BART has gotten significantly worse and thinks that overall crime has gotten worse, while the guy in Antioch and Half Moon Bay are seeing the same thing, even if all of their individual neighborhood crime problems have gotten better.
I'm not super-convinced on this answer, mostly because I don't really follow crime maps or have much familiarity with the data, but I've been to all these places and it feels weird and then I go pull up the crime maps and, like, damn. I'd appreciate it if someone with more of a background in this stuff could confirm if this is a real effect; if certain urban areas have flipped from "keep all the criminal crackheads AWAY from the downtown/attractions" to "keep all the criminal crackheads NEXT to the downtown/attractions".
Also, if true, why? I'd love to blame hippy BLM liberals but, uh, the municipal government of Detroit is hardly a bastion of based conservatism.
Hm, is it generally true (statistically) that people who have moved to a new neighborhood feel less safe there than in their old one?
Inspecting my own feelings, this rings true. I certainly didn't perceive my childhood neighborhood as scary, even though it objectively was one of the worst parts in town. But to me it was jut normal. Also as adult, I tend to ignore a lot of stuff that I am used to, even though it might upset me if it was a new experience.
Boogie bitch, or disco debutante? Sorry, couldn't resist...
(I'm...fairly sure crime is a bit up in Half Moon Bay too, if the reports from my family still living there are any indication. Cratering local economy will do that. It's hard to transition successfully from fishing-and-agriculture to tourism-and-retirement without a lot of the lower echelons falling through the cracks.)
I saw some data once showing that in opinion surveys, historically people have pretty much always thought that crime is getting worse. in their time. So my default theory for why perceptions of crime are bad is that they always are.
It seems like what's happening with all these theories is that someone will come up with a reason why crime is *really, actually* up; and any time someone comes up with a way to test it statistically, it ends up being wrong; leading the person to adjust the theory in a way that ends up being less testable.
Effectively all the theories about crime/disorder are trending towards coming up with something that can't be quantitatively tested, because all the ones that are, end up being false.
Because crime is down, and people just *always* think crime is up. Anyway, that's my theory.
"like" "+1" etc.
I think some of the "criminal crackheads next to downtown/attractions" is because tourists and others are there, so more opportunity for begging/pickpocketing/selling crappy souvenirs. Can't hustle drug money off people if there's no people around, now can you?
Good observation, the main pattern here is different populations. In the cities where crime is concentrated and downtown, often this is where fewer people actually live, this is because the broad majority of the population is relatively high function and wouldnt tolerate the issues near their homes, wheras near tourists/businesses, there are fewer compliants, another way to think about it is that the majority of the population has high demands for order. Wheras in cities where dysfunction is widespread, the cops can move anti-social people along into the residential neighborhoods and they will simply melt more into the broader population. The big question for SF, Seattle, Portland vs Baltimore/STL is where can we push these people?
> The big question for SF, Seattle, Portland vs Baltimore/STL is where can we push these people?
I've done the math on this, and if you just gave them all the free drugs, alcohol, food, and water that they wanted somewhere out of the way, away from all the productive people, it pays for itself 4-20x over.
Any one of the increased values of downtown usable real estate, lower crime, or lower police load in cities pays for it.
I personally recommend BLM land in the middle of the Nevada desert, like Burning Man, because then it's federal and the federal level can pre-empt the gauntlet of multiple local levels you'd have to run through.
The post I wrote about this:
https://performativebafflement.substack.com/p/an-incentives-based-problem-homeless?r=17hw9h
All about it.
> The big question for SF, Seattle, Portland vs Baltimore/STL is where can we push these people?
Prison.
Sounds good to me.
You 100% described my hometown of Milwaukee and my existing residence of Honolulu. "I see homeless people everywhere downtown and increasing, clutch my pearls, crime is rising!!". Also an angle you missed is gentrification, i.e. ignoring early movers whom generally expect the existing crime rate as part of the deal, it moves middle movers into unexpected contact/exposure with "bad neighborhoods" especially around the edges in areas before they would have avoided by driving completely around them or bypass on the freeway. They were promised by marketing they were moving into the new urban reclaimed suburban-like Mecca, not an mixed ghetto inner ring suburb that only the late movers get to avoid post successful gentrification.
Worth noting (for others at least) - Honolulu is an astonishingly-safe major city. It's the 13th most populous in the US and has the lowest murder rate of any in the top 50 (actually the first city with a lower murder rate is Plano, Texas, the 70th most populous).
Yep and yet crime routinely rates the number one concern here and by far. I've actually never in my life seen people so scared of crime until I lived here which I find amusing as it's easily the safest place I have ever lived, in fact so safe that it's the only place where I have never thought about crime outside passing thoughts like now in writing this, a couple times a year.
People earnestly think here if you go outside after dark you will be raped and murdered by roaming packs of crazy tweaked out homeless people, a pedophile is around every corner, every stranger a robber, and that every trip outside your house is putting your life at risk. People are genuinely terrified of crime here.
I was going to agree with you, using Baltimore as a personal example – but then I looked up the stats for murder, and it's more than halved over the last three years: https://en.wikipedia.org/wiki/Crime_in_Baltimore. That's a precipitous enough drop that I'm coming round to the view that there might be something to the broader trend of dramatically improving safety after all.
Baltimore got a new DA in early 2023. Homicides started dropping a few months later. After about eight years of elevated rates after Freddy Grey.
Oh poop, communication is hard.
My theory above is an explanation for why crime would feel like it's increasing even if it's objectively decreasing. Baltimore would be a bad example of this because it's downtown still looks pretty safe while the rest of the city is pretty red. (https://gisgeography.com/baltimore-crime-map/)
If overall crime really is decreasing notably but it decreases in the 90% of the city that nobody visits and increases in the 10% that everyone visits, overall crime can decrease while perceived/observed crime increases. And in certain west coast cities, the crime maps I see show the crime concentrated in high visibility areas.
That would explain why overall crime could decrease (which I like 70% agree with) while people see and feel that crime is increasing (which I like 95% agree with).
No, you communicated it fine. I agree with your point completely.
It's just that my impression until just now was the stats were technically real, but something like a small decrease in most places slightly outweighing a sharp increase in the touristy heretofore-safe places – but I'm now seeing the raw numbers in a place I'm kinda familiar with (I lived in, worked at, and traveled to only the safe bits) are indeed remarkable.
Homeless-pets served as a "test case" for the current immigrant-pets (who are sometimes also homeless, I suppose).
Detroit has policies that make it a sanctuary city (if not exactly "formally so"):
https://justthenews.com/nation/states/center-square/detroit-police-suspends-officers-who-contacted-border-patrol
That makes it pretty liberal, in my humble opinion
Please don't refer to people as "pets".
He may have meant "pest" which while uncouth, would have made more sense here I think on my feel for what he wrote. I'm guessing an autocorrect typo, not human furniture.
It's used twice though, and while I don't recognize OP's username, it's not uncommon among the twitter right to describe certain populations as "pets" that liberal-progressives have disproportionate affection for and defend despite (or sometimes because of) potential negative attributes.
> It's used twice though, and while I don't recognize OP's username
You should recognize the writing style, however. Nobody uses parentheses that much.
Ehh, I'm an inveterate parenthetical aside user too so I wouldn't mark that off as identifying.
I do, especially now that I've had to stop using em-dashes all the time.
Ah, that's a new one for me then, I'm not a Twit.
I had considered that case initially with its use in conjunction with immigrants but I discounted it because I've never met anyone who hates the homeless more than the left and so I wrote that usage off.
Petplay is a totally different fetish than human furniture!
I wouldn't say totally different. They're both just abstracted versions of sadomasochism.
That's some real character assassination. Dude's going on my list with Cade Metz.
...
As for the anime figures and the vibrator, he's definitely trying to paint the guy as (the modern version) of an incel--objectifying women and lacking any success with them, combining the new-style insult of being a fuckboy (effectively, cad) with the old-style insult of being a virgin.
The guy seems more than anything else like a humanities nerd angry the STEM nerds are getting all the money despite not being any cooler.
"“Young men usually is who I like to go out and talk to. Women get intimidated and, you know, I don’t want any charges.”
Probably says a lot that Kriss's audience is supposed to think "yeah, he's creepy and probably a rapist" and I thought "yeah, it is like that now. Good luck, Roy."
The polycule/group home is supposed to sound horrifying but to me just sounds cute and heartwarmingly nerdy. If I'd grown up there I'd probably be happily married (possibly to multiple people, each of them to multiple people) with kids instead of writing lengthy comments on ACX.
The sperm racing is supposed to be the ultimate tech-bro pastime: it combines masculinity (evil), technology (now evil), and capitalism (evil). As far as I'm concerned, Eric Zhu is an inspiration and makes me feel bad about all the time I wasted.
The Donald Boat thing is genuinely funny, though everyone here probably heard of him. It is notable he preserved Boat's anonymity when the NYT was less kind to our esteemed blog host. Obviously has to do with the politics of the people in question.
Then there's more of the these-guys-are-sexists-and-losers-at-the-same-time digs at their lack of romantic success.
....
It's official. If Kriss collapses in front of me, I'm not doing CPR. But I think I was more annoyed about the article about the Villages. And his snark about not answering questions until you meditate in the forest for a thousand years. And his conflation of politics and morality. And his...[REDACTED]
> It is notable he preserved Boat's anonymity when the NYT was less kind to our esteemed blog host
Kriss also preserved Scott's "anonymity", though?
> As far as I'm concerned, Eric Zhu is an inspiration and makes me feel bad about all the time I wasted.
Kriss was obviously impressed by Zhu, and pretty much said as much.
I guess the anonymity is mostly performative at this point; he's a public figure now. But you're right, he didn't use the extra name (which seems to be a signifier of which side you're on).
I guess the guy just rubs me the wrong way for some reason.
Haven’t read too much of his stuff but what I’ve seen makes me think of hyperbolic gonzo journalism with the primary recreational drug being beer.
at the risk of opening up some kind of controversy that will one day be known as Crackergate, my strong recollection is that the crackers comment was made once the children had gone to bed. i found it very charming, which is why it’s in the piece
I wonder if anyone in 2008 thought PZ Myers would be remembered in 1000 years
It was a great essay. I also interpreted the cracker comment as charming.
The Cracker question is now a market on Manifold: https://manifold.markets/DylanRichardson/was-i-will-open-the-crackers-so-you?r=RHlsYW5SaWNoYXJkc29u
Sam, my understanding is you eschew online feedback lest it poison the pure luminous art flowing through your conduit to the divine, but I'm very bad at listening to direction, so I just wanted to let you know I thought the piece was excellent. One of your best ever (second maybe to the essay you wrote about the Villages in Florida)
FWIW, I also interpreted your retelling of the situation as "likeable, thoughtful weird guy does mildly odd and endearing thing", which seems to be the way it was intended.
I was charmed!
FWIW, so was I.
I kind of took it as a reference to "You will own nothing and be happy" (https://en.wikipedia.org/wiki/You%27ll_own_nothing_and_be_happy)
Hmmm - maybe it's the effect of Lent on my brain, but....
(1) An evening meal
(2) Referring to those at the meal as "acolytes" which we could also take as "disciples"
(3) Scott as messianic figure? important figure in cult?
(4) Ahem. Somebody else spoke out loud to their acolytes at an evening meal when distributing grain-based foodstuffs*
https://en.wikipedia.org/wiki/Words_of_Institution
*Yeah, I'm going to Hell
>Yeah, I'm going to Hell
It might not mater one way or the other ;)
https://theonion.com/st-peter-scrambling-to-throw-few-more-innocent-souls-i-1819681025/
And then of course one of the guests betrayed him.
For a few shekels of silver.
Who was I saying it to? Myself? I know I can come across as pretty weird, but this just feels really outside of the set of things I can imagine myself doing. Are you sure it wasn't the child who lived in that house and who stayed around after my kids went to sleep?
Count me on the side of "it was a normal charming thing people mutter to themselves, and a sweet if very condescending write up by the author."
It doesn't read to me as "look at how weird this guy is", to me it reads as a continuation of the main thrust of your 'interview' of "Scott Alexander is a bit idiosyncratic but a fundamentally good natured and kind guy", it comes across as even your idiosyncrasies are charming and well intentioned. Especially when contrasted to the other people in the article, I sent the article to several friends with comments about how it reinforces how nice you are!
yes i think yourself, it was loud enough to be audible but not loud enough to be directed at any other person. i'm sure i've said stranger things in any given evening
I continue to claim I wouldn't do this (Your Honor, I don't even like crackers), but I appreciate your spirited defense of your journalistic accuracy.
I find it strange you find such statements to be out of character. Basically any person makes such strange statements now and then :p
I would find it more weird if you never made such statements.
What is this "journalistic accuracy"? Journalists can basically write whatever they want, and if it's not accurate then who cares. Maybe in some rare cases a journalist's publication, sometime later when nobody is looking, quietly lists the inaccuracy as an errata. Journalism isn't some kind of oath-sworn government body accountable to Congress, it's a genre of entertainment like rap music or horror movies. Nobody reads Harpers to comply with code.
Many people care, including most other journalists, in the specific case of a claim that is provably, explicitly false as stated. Which is why journalists so often use verbiage that doesn't amount to an explicit statement of fact regarding the subject at hand, "X is true!" can end your journalistic career as fast as Dan Rather's, if X isn't true. "Bob told me X is true", is much safer, and if you pick a vaguely-credible Bob it will still have most of the audience believing that X is true.
Right. "Who cares" was a turn of phrase. You accurately describe a few of the ways journalists approach "journalistic accuracy".
A quick check shows that Dan Rather's career has not ended.
I appreciate it, too. Thanks for commenting here, Sam...
Scott, I don't understand why you think this makes you look crazy. To me, this seems like a perfectly normal thing to mutter to oneself.
Hot take: this actually presents you as an ideal and positive foil to Roy, because his high agency terminates in "a great sucking void where the end ought to be", but you clearly articulate the utilitarian virtue in the basic act of allowing people to eat crackers.
On the other hand, I now catch myself muttering to no one in particular things I narrate to my children (2 and 4 years old). One just gets in the habit!
The crackers comment seems to have been almost universally positively received, if you were worried. It's incredibly funny/sweet so I'm just going to assume it was true as written.
I just hope that Scott will open some crackers for me and brighten my day
Another vote for "charming, not crazy". It came off to me like narrating a priority when there are a lot of other things competing for your attention.
Zackly. Kriss can lay the wood like a hungry kid whacking a pinata.
This was Kriss being nice.
Yeah, it sounded like being a good host. All in all, I think Scott came across well, Kriss could have been meaner if that was his intention.
I get the impression Kriss went in with his usual curmidgeonlyness and couldn't help but be mildly charmed.
I'm starting a online group focused on deliberately practicing and improving forecasting skills — regular meetups, pastcasting, post-mortems, and guest calls with experienced forecasters. Full details in this LessWrong post: https://www.lesswrong.com/posts/jronsvchwjB74BDEb/i-want-to-actually-get-good-at-forecasting-this-year-group. The first meetup will be Sunday, March 1, 11:00–12:00 CET.
This looks great!
From the Sam Kriss piece:
“There were also some objects that didn’t seem to belong in a frat house.”
I am shocked, shocked to discover that a bunch of nerds have some anime figurines.
In anime, this would be the cliché way of indicating that the character is an _otaku_. (See, _Genshiken_, etc.)
That article was really tame and subdued for a Sam Kriss piece. Where's the vituperation, the grandiloquent hot takes, the creaky makeshift towers of metaphor and illusion in their varied multiplicity?
I also wonder at his contempt for "agency." Yeah, it might not make you happy, any more than money will make people happy, but just as obviously it matters, and will matter a lot more in a future where AI can do everything.
He seems to accept this, yet has contempt for it? Because it isn't teleological, an end in itself, or because the people demonstrating it don't have strong ends, or because it's rewarded by the market, or I don't know why.
But "the market" is just a synecdoche for "reality" here. The agentic are already doing more with AI than most people ever do without it, and having correspondingly more impact on the world, regardless of whether the market is rewarding it.
I introduced my dad to AI. Since then, he's written another book, made a religious puzzle game for the app store, made several websites, made a ton of videos and images - he probably spends 5-6 hours a day on it. It's an amazing productivity multiplier, and he's doing 10x or 100x more than he would have without it. I'm sure other retired guys farted into their recliners and yelled at clouds and internet commenters for those 5-6 hours.
That's the future, like it or not.
>Because it isn't teleological, an end in itself, or because the people demonstrating it don't have strong ends, or because it's rewarded by the market, or I don't know why.
Because ..It's all about achieving g Something,not about what the Something is...? At least that's what makes me uncomfortable. For instance, if Cluely actually gives you an edge, everyone will start using it , and giving the same AI generated answers, so the human element becomes redundant.
> For instance, if Cluely actually gives you an edge, everyone will start using it , and giving the same AI generated answers, so the human element becomes redundant.
Okay, but if everyone started making Phd smart decisions in multiple domains of life, how are they and everyone around them not massively better off?
That's a huge amount of value created, both for the individuals, and in positive externalities for everyone else.
I thought the discussion of agency was really fascinating, particularly the tension between Roy as being extremely agentic but working on a product where you outsource your agency (which is pretty undeniably accurate even if you think Roy is portrayed unfairly as being agentic but with no end goal, although unless the quotes from him are extremely inaccurate it's hard to feel like this is very unfair).
But also I think it's worth noting Kriss readily admits to his own failure to show agency (in response to Roy insulting him about his hair loss, which he dislikes but does nothing about). I don't think there's only one thing to take away from this; agency can both be valuable and good and also not be the only thing worth valuing.
> I thought the discussion of agency was really fascinating, particularly the tension between Roy as being extremely agentic but working on a product where you outsource your agency
I agree this was interesting, but didn't see any contradiction at all here.
1) There's a long tradition of building billion dollar businesses on removing friction and making life easier - Amazon patenting one click and relentlessly simplifying the shopping and checkout experience, Google returning better results with a simpler interface, all the delivery apps and car summoning apps, and so on.
2) He's just giving the people what they want - the vast majority of people don't WANT to make decisions. Decisions are friction! Moreover, this is actually the better move for two reasons for most people - both in the sense it removes that friction, and in the sense that it will be a better quality decision. If you allow people an option to have a Phd smart mind making that decison, it's actually a better, smarter move for the vast majority of people even *without* the benefit of removing the pain and friction!
3) Our future is obviously going to be one where our Phd smart AI assistants live in our ears / phones and we interact with them for hours per day. It's the next trillion dollar attention surface that can be auctioned, and it's why Google and Facebook are all-in on AI, because they don't want their current trillion dollar lunch eaten. He's skating to where that puck is going to be, to use a Gretzky analogy.
your view of the future reminds me heavily of The Commonwealth Saga
> your view of the future reminds me heavily of The Commonwealth Saga
Man, we can dream, right? If we end up in anything even adjacent to or resembling it, I'll consider that maximal success for our current arc.
Because obv we're probably not going to have anything like the Silfen or Raiel or Living Dream, nor anything like instant portal tech. But even Commonwealth-ish minus those would be a gigantic win!
> I agree this was interesting, but didn't see any contradiction at all here. [...]
I feel like you just listed off three reasons this is a good business opportunity that would apply in any context, but do not speak to whether it is a contradiction.
My point is that somebody whose main admirable trait is how agentic they are is an ironic person to supply you with a tool to be less agentic, like when a pro athlete does a McDonald's ad or something. I'm not saying it's a bad move to shill for McDonald's - they make a tasty box of fries and they pay well - but it's still ironic if you're famous entirely for physical fitness.
Yeah, I guess we're speaking past each other in terms of what counts as a contradiction.
My point was, this is just literal "business as usual," which has always been fueled by highly agentic entrepreneurs making stuff for people to be less agentic / lead less friction-ful lives.
You're looking at it from a personal traits perspective. The highly agentic guy is deploying his agency to allow regular slobs to be less agentic!
And I'm like yeah, but that's a tale as old as time. Lots of businesses have followed that dynamic, it's a well-worn path to success. People hate agency, friction, and making decisions, so giving them what they want is a great plan for a business to succeed.
Yeah, I don't disagree with the point you're making, either.
Unfortunately, I think what your dad is doing is the next two years of the future (and it will be a fun two years!), and after that the AI makes puzzles and writes books without you, and churns out a thousand an hour, and it's unclear what useful things humans can do with their time and intellect. I think Sam was sort of trying to get at that, though if I dared to critique his writing I would say he was mixing it up with so many other points that it lost a little of its oomph.
I agree that I couldn't quite figure out where he was going with the agency stuff. Roy Lee isn't very sympathetic, but the critique seemed to mix elements of "He doesn't want to do hard work" and "All he cares about is working hard" in a way which was probably faithful to the real situation but also leaves me confused about what conclusion to draw.
I fear agency has become too associated with Silicon Valley's Grind Culture, distinct from its philosophical role as a moral primitive.
What I took from it was that Lee was talking about agency and yet the product he was working on and selling was to strip agency from people, it was "this will tell you what to say and do". Lee's ultimate aim was also unclear: did he want to make lots of money? and then what? and it seems (though this could just be how Kriss wrote it up) that Lee hadn't really any idea what he wanted to do with all this, he's someone who is "agentic" but it's not really clear what the point of all this 'doing' is for. To be successful? to be popular? to prove to his parents that he's not a failure for not getting into Harvard first try? just to have something to fill up the time he would otherwise have to spend in his own head with his own mind?
EDIT The relevant quote:
"According to Roy, he has three great aims in life: “To hang out with friends, to do something meaningful, and to go on lots of dates.”
But does Roy have friends, and what does he consider meaningful? "Being a more successful entrepreneur" seems to be the height of that, and the description is depressing:
"For Roy, meanwhile, dating actually seemed to be a means to an end. “All the culture here is downstream of my belief that human beings are driven by biological desires. We have a pull-up bar and we go to the gym and we talk about dating, because nothing motivates people more than getting laid.” He was interested in physical beauty too, but only because “the better you look, the better you are as an entrepreneur. It’s all connected and beauty is everything. A lot of ugly men are just losers. The point of looking good is that society will reward you for that.” What about other kinds of beauty? Music, for instance? Roy had played the cello as a child. Did he still listen to classical music? “It doesn’t get my blood rushing the same way that EDM will.” His preferred genre was hardstyle—frantic thumping remixes of pop songs by the likes of Katy Perry and Taylor Swift. Is that the function of music, to get your blood rushing? “Yeah. I’m not a big fan of music to focus on things. I think it disturbs my flow. The only reason I will listen to music is to get me really hyped up when I’m lifting.” The two possible functions of music were, apparently, focus and hype. Everything for the higher goal of building a successful startup. What about life itself? Would Roy die for Cluely? “I would be happy dying at any age past twenty-five. After that it doesn’t matter, bro. If I live, I have extreme confidence in my ability to make three million dollars a year every year until I die.”
Wanting to "do something meaningful" seems a bit like the classic "wanting to be a writer". The best writers and meaningful-doers are people who have things they want to write or things they want to do. "Doing something meaningful" in the generic sense is a role rather than a goal.
Well said.
I do feel like we should at least be a little grateful that these people are burning away their life being aimlessly productive for others instead of trying to find "meaning" in their life. Do you want him to discover that his heart's desire is cannibalizing the homeless or something? Better that he's grinding himself away for our sake.
I think his point (one of them, anyway) is that "agency" refers to the drive and the ability to do *something*, anything, just because you can. People like you and me worry about things like "ultimate goals" and long-term plans and what others might think, etc. Agentic people see something that can be done, and do it. In the past, such people had to be born wealthy and powerful, and to be incredibly smart, or else they *couldn't* do anything regardless of how much they wanted to. In this modern era of increasingly powerful technology, almost anyone could become "agentic" in this way, but most people don't have the will, the drive, the kind of personality that allows it.
The world is moved into the future by such "agentic" people; it is sustained in the present by quotidian peasants like you and (arguably) me.
Yeah, that makes sense. I guess I missed this angle because I think Silicon Valley has a sort of Nietzschean take on business where it's fun and meaningful in and of itself because it's a difficult competition. I don't think anyone would say Alysa Liu's life is meaningless just because she's devoted it all to figure skating which isn't of itself a socially productive pursuit, and a lot of B2B SAAS startup founders are motivated by being the Alysa Liu of making money.
I dunno, if he had 5% more social awareness when talking to journalists, he could have framed it as "make friends, build a successful business, and find love", and then nobody would have objected.
Liu is putting in hard work to be a champion figure skater. The description (and Kriss may be too hard on what is going on) is that Lee's product is faulty and a scam, since its selling point is "use this to do your job for you" (so people are never learning or producing anything themselves) and even "use this to trick women into wasting time on you" (tell her she's a great artist example) and it keeps crashing and not working correctly even in the company headquarters.
So Liu is doing harm, if harm is done, to nobody but herself by putting her body through the rigours of training, plus she has to provide good enough performances to win points from judges: a tangible performance in the end. The Silicon Valley hypesters are trying to invent convincing arguments about why their bubble product is worth the free money they are trying to score off investors, and the end products are not really providing a service at all.
Well, given Roy's business sucks the meaning out of everything, he's probably inviting this critique of his personality to some degree.
Yeah, I guess my critique of Roy would have been moral (cheating is wrong), Sam tried to make a psychological critique, and I'm not sure whether to think of the psychological critique as compelling in itself vs. a sublimation of a moral critique which is too hard to deliver because people don't like preachiness.
I took this as both - "this guy has no purpose and refuses to so much as read a book to try to find one" seems pretty scathing from a virtue ethics perspective.
What’s going to be interesting is how difficult it will be to just not consume anything created by AI actively. I and plenty of other people would be willing to boycott AI entertainment, especially in a post scarcity future, and spend our time creating art and consuming art made by other humans. By some measures that art would be inferior art than what an AI will be able to produce, but because creating art for others to enjoy is enjoyable in itself, a community would be better off if they just created their own art and consumed it. The hard part is it it difficult to signal that something was made without AI.
My wife is writing a book from scratch, with no AI assistance. When she submits it for publishing a year from now, she’ll be competing with books that were written with AI. And when she submits its sequel she’ll be competing with books written entirely by AI. By the time the final book in her trilogy comes out, if the books have any fans, they would have been able to feed the first two books into AI and create a thousand different versions of the Finale. More likely the book will be lost in the massive amount of AI entertainment and I will be its only reader.
I might just end up playing a lot of TTRPGs with friends where AI can be clearly excluded from the table, even if it was involved with prep.
Ironically, I'm sort of hopeful that AI will help with this - if everyone has an AI assistant on their browser doing a sort of algorithmic feed on the whole world, you can just tell it not to show you things that are AI produced (although I think most people won't have the willpower to do this even with low transaction costs).
What stops the AI lying to you that sure, boss, no AI content on this feed! I imagine all the companies using AI to generate content will be trying to dodge around people blocking said content, and I don't trust current AI models not to lie/hallucinate by telling people what they want to hear even if it's not so.
If 90% of stuff is AI glop, then the AI will only be able to serve up 10% of non-AI glop, and that might disappoint its human who wants more and more content and variety. So a little white lie that this is not AI glop, what harm does it do when measured against the greater benefit of making your human happy?
It can't get away with lying about things that are easy to verify as false. Then everyone would instantly know about it and the lie would no longer work. It could get away with lying if it's impossible for people verify, but then that means the AI couldn't verify it either, since the people would be using other models to verify it as well. So as long as there isn't some mass collusion among models, they can't really get away with lying.
As I understand it, Sam Kriss believes that AI, at least under the current paradigm, is never going to be the match of great human authors when it comes to writing essays, novels, poetry, and so on. Meanwhile approximately 100% of silicon valley techies and rationalists think that we can be as certain of this happening as we can be certain about anything at all in the future, and moreover, that it is going to happen imminently. In his estimation, the reason for this gulf in perceptions is that techies have the artistic sensitivity of lesser primates and to them Joyce or Woolf might as well be Rebecca Yarros. Hence why he was so delighted by Roy Lee, who I hadn't previously heard of, but who seems to be a flesh-and-blood caricature of the soulless startup bro.
It's notable that even if AI never becomes as good as, say, the top 1% of writers or programmers or whatever, that still entails the evaporation of nearly all jobs for writers and programmers and such, and probably also the evaporation of most of the pathway to becoming a top-tier writer or programmer or whatever.
>Meanwhile approximately 100% of silicon valley techies and rationalists think that we can be as certain of this happening as we can be certain about anything at all in the future, and moreover, that it is going to happen imminently.
I don't think that's the case. I am a silicon valley techie, and think that is quite unlikely to happen in the short to medium term and venture no strong opinion about the long term. I don't get the impression my assessment is uncommon, let alone unique.
Would you say that the vibes around this topic are notably different between places like this one, or the AI-interested corners of X etc. and the general tech-community in Silicon Valley?
Asking mainly because I read blogs such as this but am very far away from "Silicon Valley tech world".
My immediate coworkers have a pretty wide range of views on AI, but I'll attempt my general sense of the median views.
Among rank-and-file engineers, the median view is probably that it's a useful tool for semi-automated mechanical tasks, investigations, and rapid prototyping but produces worse-than-useless slop if it is relied upon too heavily. Incremental improvements are generally expected, but probably not fundamentally transformative ones. I'm less favorable on AI than the median but am not an extreme outlier.
Management, especially senior management, has adopted a party line that is extremely gung ho about how great AI is and that it's rapidly getting better. Middle management seems to be mostly on board with this, but I'm not sure how much of that is sincere and how much is feeling obliged to toe the line.
> Unfortunately, I think what your dad is doing is the next two years of the future (and it will be a fun two years!), and after that the AI makes puzzles and writes books without you, and churns out a thousand an hour, and it's unclear what useful things humans can do with their time and intellect.
I mean, in a post-intelligence future, surely the ones who fare best are those who have endless ideas when it comes to creation? That's the "agency" part, and it helps whether or not you're being paid for your outputs. You're entertained, and you can create things meaningful to your local social milieu, and that gives it meaning to those other people, even if AI could have made a strictly better artifact if they'd thought to ask for it. Just like a home-made gift is more meaningful than the top Amazon item.
That IS the useful / relevant thing that people can still do in that future - create things relevant to themselves and their local milieu. If we're all hunter gatherers who don't need to work all day in the future, we just go back to hunter gatherer stuff. Those people who are fun to be around, who tell good stories, who throw a good party, who have creative ideas on how to use the bounty around them, that's who has status.
> That IS the useful / relevant thing that people can still do in that future - create things relevant to themselves and their local milieu.
If your social milieu is online, or at least recorded and transcribed, then (granted the underlying premise) wouldn't the AI be better at creating things relevant to it than you are ?
> If your social milieu is online, or at least recorded and transcribed, then (granted the underlying premise) wouldn't the AI be better at creating things relevant to it than you are ?
I don't think most people's social milieus are online.
Yes, people spend an ungodly amount of time on screens, and more and more of them have relatively unhealthy parasocial relationships that are acted out online, but this is still mainly gen Z or alpha phenomenon, and the majority of people still have real life friends and social circles.
The people for whom that IS true - the terminally online, as they say, are doomed. They're going to be sniped by the Infinite Jests and AI waifus and boyfriends, which will be to real food as junk and fast food, and real human engagement as Tik Tok and porn are.
Obviously, junk and fast food has been so optimized and become so powerful 80% of people are fat, just like eyeball-hour harvesting has gotten so good that phone screen time has gone from 2 hours a day in 2014 to 4-5 hours today and 7-9 hours for gen Z / alphas.
There's no reason to think that optimization is going to stop, and many reasons to think it can get stronger. In the limit, that'll be superstimuli so strong that it snipes people out of reality entirely. I wrote a post about that.
https://performativebafflement.substack.com/p/the-view-from-inside-the-80?r=17hw9h
>In the limit, that'll be superstimuli so strong that it snipes people out of reality entirely.
That sounds pretty nice, but, like so many other futuristic predictions, this one commits the mistake of assuming there are no physical limits to anything, and any trend can be extended indefinitely. In real life, this is usually not the case. For example, cars are getting faster every year (https://en.wikipedia.org/wiki/List_of_production_car_speed_records), but this doesn't mean that they will ever exceed the speed of light.
Sure, except that it's obviously a gradient, and it's *already* sniping people out of real life.
Everything from sex and drinking going down to people dying in gaming cafes after multi-day binges. It's literally already affecting everyone to some level, and the waterline is *already* high enough to snipe some of the most prone among us, and that waterline is only going to rise.
You say "but it's not going to snipe everyone," yeah, completely agree. The amish exist, just for one segment. But it's an open question how much of humanity it CAN snipe, and it's already eating something like 2/3 to 3/4 of all free eyeball-hours.
Aren't ideas correlated with intelligence? So, wouldn't the post-intelligence future mean the AIs have the best ideas too? And can't an AI be charming too, eventually? Look, I'm just a bag of meat and bones. It takes me years to get good at playing the piano. How would I compete with an AI Chris Hemsworth robot (assuming we get realistic robots) who quips, tells engrossing tales, and effortlessly plays the piano, guitar, mandolin, etc.?
Look, I can see this wrt to family units, but have a harder time seeing how this scales beyond that and maybe close friends. Maybe that's enough for people, maybe not.
I think a valuable part of human relationships is the feeling of being 'chosen', and being 'chosen' by charming, likeable people feels especially good. An AI friend or boyfriend couldn't really 'choose' you. Will people still go for this on the margin? Yes, definitely -- it's clear already from what's happened to interpersonal relationships from the Internet and LLMs. Do I think this is a serious concern for most interpersonal relationships? No, I don't think so. I don't have any fear my boyfriend and most of my friends would choose an AI over me.
(this is more a comment on interpersonal relationships and separate from any discussion of making a living in a post-AI future. That's a whole 'nother bottle of worms)
Because you're imagining future AIs are like the current ones, where ChatGPT constantly reminds people it doesn't have feelings or preferences. If AI ever gets smart enough and someone makes an AI that tries to act like a human, you won't even be able to tell which of your new friends are AI.
Thanks, that makes sense.
I agree, and I'm not sure what I'm missing when people worry too much about this.
I'm not very good at playing guitar, anyone could easily load up Spotify and hear better played guitar, but
1) I enjoy playing; and
2) There are a few people close to me who enjoy the music precisely because it is me, a loved one, making it.
Similarly, my 4 year old is terrible at drawing and colouring but I still fondly and proudly accept her handmade gifts to me because they're by her, she enjoys doing it, I get a fuzzy flow from it because I love her! It's fine!
I don’t see anybody making a living there, though.
> I don’t see anybody making a living there, though.
Yeah, but I think pretty much everyone expects a UBI.
I've done the math, and we can afford to give people $40k - $90k UBI's with really reasonable assumptions (current or slightly higher corporate tax rates, 50% of white collar jobs counterfeited, etc).
Also, anyone with assets is going to be fine as the economy quadruples, and their passive income will do up.
It's also a huge unlooked-for gift that it's happening to white collar people first, because they're the rich connected ones plugged into the chattering classes and politicians and whatever. Versus if it were service industry jobs / robots? Boots on faces forever. So we actually have a decent chance to get a UBI passed because white collar is happening before robots.
The post where I look at this:
https://performativebafflement.substack.com/p/can-ubi-work-yes-with-a-few-reasonable?r=17hw9h
I think only a handful of nerds expect a UBI. Particularly in the case of AI 2027, er, 2029 or whatever, because it would take longer than that to bring the concept of a true UBI within the Overton Window of the sort of not-nerds who would need to sign off on implementing it.
I am also exceedingly skeptical of your math; a $40,000/yr UBI for every adult in the United States would cost ten trillion dollars plus overhead. more than half the current GDP.
The Sam Kriss article seems to be in a style influenced by Hunter S. Thompson’s _Fear and Loathing in Las Vegas_, and similarly mixes fact and fiction.
His description of the dinner with Scott seemed ok to me, though, agree would have been clearer if it were more explicit that Scott was talking about the crackers to a small child (the story does highlight that children were there).
Though there’s a certain amount of comedy to be had in maintaining ambiguity over whether Scott was talking to one of the ACX commentariat or a literally two year old.
I for one think it's amusing that Scott seems as concerned about the crackers thing (which seems more adorable than anything else) as the suggestion that the rationalist movement is idiotic and misguided.
But the suggestion that the rationalist movement is idiotic and misguided is quite arguably true, whereas the crackers thing is (I'll have to take Scott's word for it) made up.
No, Scott says he's pretty sure he wouldn't have said that in the context he's described as saying it in. That's very different. I've interviewed people before and sometimes they're surprised later by what they've said. (Or Kriss failed to capture the surrounding context correctly, which is not the same thing as making up a quote).
It may be that Scott is just so used now to narrating all his actions (for his kids) that he does it out of habit even when not with the kids.
I'm a random ACX commentariat dude, and when I read the crackers passage I thought "hah, Scott seems like a really chill guy, and this innocuous moment confirms that". I didn't think it reflected badly on him in any way.
+1; I thought he was making an adorable half-joke even before I heard he was saying it to a two-year-old.
There were, however, a couple of other places in the Sam Kriss piece that I thought would be misleading or confusing to a reader who wasn’t already familiar with the AI risk scene.
Tyler Cowen posed a useful framework after the Maduro extraction: (a) immoral actions were taken, (b) probably unlawful, (c) Venezuelans are better off. His challenge: you can debate how to weigh (a) and (b) against (c), but to deny (c) is delusional.
I sat with this for six weeks & 'consulted' Sun Tzu, Aristotle, Kautilya, & Clausewitz. My essay has drawn sharp pushback. One analyst brought in Charles Tilly's roving vs. stationary banditry distinction & argued this particular extractor is incapable of playing a long game. Others pointed out the US helped create the crisis now used to justify intervention & cited Lindsey O'Rourke's research documenting 64 covert US regime change operations between 1947 and 1989 — making Venezuela less an anomaly than a continuation.
My conclusion is broadly consequentialist but conditional: the thugs have been replaced by the rascals & cosmetically improved authoritarianism is still authoritarianism. The extractors now own the aftermath.
https://rajeshachanta.substack.com/p/the-helicopter-the-courtroom-and
2 questions for this community: is the sovereignty norm already dead & this just made it explicit? And Tyler's utilitarian scorecard of LA interventions counts roughly six cases where leaders were actually removed & finds mostly positive long-term outcomes. But O'Rourke documents 64 operations, Sachs estimates 80. If six successes out of 80 attempts is the real denominator, is Tyler's optimism genuine pattern recognition or survivorship bias?
What difference has it made to Venezuelans? Aren't they currently ruled by the former #2, just as Maduro was Chavez' #2?
That's the right question & roughly where the essay covers — 'the thugs have been replaced by the rascals.' Delcy Rodríguez running things isn't freedom. But the case for cautious optimism rests on what's changed at the margins: 400+ political prisoners released, oil sanctions lifted, some economic breathing room. Whether that's the opening of a longer game or just a personnel swap with better PR is exactly what's worth watching.
The lifting of sanctions is something the US could have done at any time.
Fair point. It's the same one Alan made on the Substack comments. The sanctions weren't an act of nature; they were a policy choice that could have been reversed at any time. That's the complicity question the essay underweights. The honest framing is probably: Maduro was a catastrophe independent of sanctions, and sanctions made a catastrophe worse. Both things, simultaneously.
Sovereignty belongs to the people, not to Governments. If the Government is elected by the people then the Government holds sovereignty by proxy, but in a country with no elected government there is no sovereignty, and the "Government" is just a bunch of random thugs who happen to be holding the people hostage, and it should be considered open season on them.
The principle is clean, but the application gets complicated fast. Who decides that a government has lost its claim to sovereignty? If the answer is 'any state powerful enough to act,' then you've effectively given every superpower a standing invitation to intervene wherever it judges a government illegitimate.
Reformed Hegelian's examples below are useful here. Iran has no free elections and massacres protestors — open season? North Korea is probably the worst government on earth — open season? But the moment you say yes, you've also given Russia the language to justify Ukraine ('illegitimate post-Maidan government') and China the language for Taiwan ('breakaway province with no sovereign standing').
The sovereignty norm isn't there to protect bad governments. It's there because the alternative — legitimacy decided by whoever has the most aircraft carriers — tends to produce worse outcomes at scale, even if it occasionally produces a good one.
isn't the invitation already there just unspoken?
and the sovereignty norm also just humanity trying to pretend that the "one that makes the decisions is the one with the most aircraft carriers" is not the true rule?
Its why North Korea does the nuke game right?
I agree that the strong have always had disproportionate say. North Korea's nuclear programme is the proof — sovereignty is easier to defend when you can threaten Seoul.
But there's a big difference between 'the norm is imperfectly enforced' and 'the norm does nothing.' Imperfect norms still constrain. The invasion of Ukraine has been costly to Russia precisely because the sovereignty norm made it expensive in ways it wouldn't have been say 50 yrs ago — sanctions, isolation, NATO expansion. The norm doesn't prevent violations, but it raises the price. That's not nothing.
That was a fun read. Enjoyed the consultations.
I'm a fan of Steven Pinker so definitely appreciate the value in maintaining the "Sovereignty norm". War is bad.
That being said, there are times I worry we've swung too far in the other direction specifically regarding "failed states". Specifically, I think places like Haiti and North Korea would be better off if the democratic West re-invented colonialism specifically for them.
For a more interesting and grounded example: Iran.
On the one hand, evil government massacre-ing protestors based on twisted theology.
On the other hand: an internal-based regime change basically means a long and brutal civil war where even more people die.
Seems like a perfect opportunity for a superpower like the US to say "Sorry you've voided your sovereignty". Don't worry, the new Iran will regain sovereignty once we say so".
I know, I know. This opens the doors for Russia to attack Ukraine and China to attack Taiwan making similar arguments. I'm just not sure this wasn't always the case.
One last point:
I think Putin was responsible for simultaneously harming and strengthening the sovereignty norm. First by invading Ukraine, crossing a big red line (I know, Bush in Iraq too).
And then by losing endless men and equipment in a years long quagmire. Putin reminded the world why invasions aren't worth it.
If anything, Venezeula was proof that a coup can be relatively quick and painless. But not sure this is true for non-Americas.
Oh sorry you had a second question: I think success or failure calculations should heavily depend on how bad the current regime actually is.
Venezuela didn't "prove" anything about coups, because there wasn't a coup. There was a kidnapping which otherwise left the government in place.
Fair point. I guess it proved that kidnapping a foreign leader was easier. Certainly easier than I would have guessed. It also seems like they had a say in Maduro's replacement, though. Instating his #2 instead of the oppositionleader felt like a choice more than simply grabbing Maduro and letting the country sort itself out.
We didn't "install" the #2, that's just what happens when the #1 is gone. Installing the opposition leader would have required doing more than removing the #1.
It's possible you know the details better than me. My understanding was that the actual number 2 is Cabello and the US specifically didn't want her taking over. Instead they chose Rodriguez and indeed there are reports she was actively working with the US as part of the transition. So it really feels like a US decision on who gets to lead here. I'm not exactly happy with their choice here. But this is definitely more than just jeeting Maduro and saying bye. The US had a lot of power post kidnapping.
Maduro was VP under Chavez, then Rodriguez was VP under Maduro. https://en.wikipedia.org/wiki/Delcy_Rodr%C3%ADguez
The US has tried going into Haiti. The results don't seem to be anything to brag about.
Neither the US nor any other stronger country has attempted conquering Haiti and taking control (in recent history of course).
This is a norm that even Bush didn't consider breaking for obvious reasons. But it's hard to believe this decision won't improve Haitian lives dramatically.
The US occupied Haiti from 1915 to 1935. It was occupied another time from 1994 to 1995.
Was aware of 1915 but actually totally missed 1994. It's further in the past than I meant by "recent", but I'll admit I stand corrected.
This is honest in a way most commentary on this isn't. Following the sovereignty argument to "reinvent colonialism for failed states" & then seeing why that's terrifying is the right intellectual move — it's the only way to test whether we actually believe our own principle or just like how it sounds.
Your Iran example is the hardest one. Internal regime change could mean brutal civil war. External regime change means someone else decides their future. Iranians trapped inside get to choose between two kinds of catastrophe. That's the dilemma my essay tries to sit in rather than resolve.
Your final point is the one I keep returning to: Venezuela may prove that a quick extraction works in the Americas, where the US has overwhelming proximity and leverage. That says very little about whether it's replicable in Iran, North Korea, or anywhere Russia and China have countervailing interests. The "success" may be radically context-dependent — which is another way of saying it's not really certain at all.
Appreciate the compliment (I think...)
In the real world, this is something we have a ton of uncertainty about. So I understand why nobody is willing to even try. But the pros vs cons of Iran specifically seem higher than normal.
I just wish the world was more honest about this. I wish politicians would say things like: "We'd love to invade and free you from an evil dictatorship, but the chances of it biting us in the ass are too strong to risk it and we don't trust global opinion to give us any credit for trying".
The compliment was entirely genuine — following the logic to its uncomfortable endpoint and then immediately seeing why it's terrifying is something most commentators skip.
Your hypothetical politician's speech is the essay I wish a diplomat or policy analyst would write. The gap between how intervention decisions are actually made (risk calculus, strategic interest, domestic politics) and how they're publicly justified (freedom, democracy, human rights) is where the cynicism about foreign policy lives. Closing that gap even slightly would be worth more than another think tank report.
Here's a thought: On my Substack (Selective Amnesia) this is the sort of thing I love doing — writing speeches that should exist but don't, for leaders who should say what everyone's thinking but won't. If you're interested in collaborating on that honest politician's intervention speech, I'd be up for it. The world could use more clear thinking delivered in plain language, even if it starts as a thought experiment. DM me if you're game.
Meanwhile, if you enjoy geopolitical satire, this Saturday's essay is set in a Dubai airport lounge where the world's power dynamics play out over duty-free shopping. Different tone, same underlying questions.
> But O'Rourke documents 64 operations
According to Grok, O'Rourke includes a long list of operations that I wouldn't classify as removal of leaders (*). Though, after checking with Grok, it seems O'Rourke only considers 10-12 of those operations in Latin America to have been successful.
So Cowen's denominator would be 6 and O'Rourke's would be 10-12.
UPDATE: Grok says this is the list. I object to the inclusion of Bolivia (2) and Guyana, and suspect those might explain most of the remaining gap between Cowen and O'Rourke:
Guatemala (1952–54) — Overthrew Jacobo Árbenz.
British Guiana/Guyana (1961–71) — Undermined Cheddi Jagan, installed Forbes Burnham.
Bolivia (1963–66) — Led to René Barrientos regime.
Brazil (1964) — Supported military coup against João Goulart.
Dominican Republic (1965–68) — Invasion and occupation to install favorable government.
Bolivia (1971) — Supported Hugo Banzer coup.
Chile (1962–73) — Culminated in 1973 coup against Salvador Allende.
Nicaragua (1980–89) — Contra support pressured the Sandinistas (coded as successful in pressuring regime change or significant concessions, though not full overthrow).
(*) A few (you can ask your favorite AI for confirmation): Nicaragua: Early efforts against Sandinistas; Bolivia: Successful (Support for Hugo Banzer coup); Dominican Republic: Dominican Republic (1960–61) — (Assassination of Rafael Trujillo and related efforts); Haiti: (1963) — (Efforts related to François Duvalier or opposition).
Thanks for slicing the O'Rorke data to the narrower Latin American subset — that's useful. The 10-12 successful operations is a higher numerator than Tyler's 6, which is a fair correction. But the denominator stays at 64 total covert operations (or ~80 if you follow Sachs post-Cold War). So we're moving from roughly 6/64 to maybe 12/64 — better, but still not a ratio that should make anyone relaxed about the next one.
The deeper problem is O'Rourke's own finding that targeted states generally became more authoritarian and unstable after intervention. The failures aren't neutral — they made things actively worse. So the expected value calculation isn't just 'low success rate' — it's 'low success rate with negative-sum failures.'
I'm confused. I thought the denominator was number of leader removals. Wasn't the whole point to compare Maduro's "extraction" to similar operations?
UPDATE: I now realize that my phrasing "10-12 of those operations in Latin America to have been successful" might be ambiguous. I meant successful as in: resulted in the leader removal (or regime change, from O'Rourke's point of view) as intended, not in the sense of the local country ended up better off (not a benchmark that O'Rourke applies, I understand).
You're right so let me concede the point cleanly. Tyler's question was specifically: when leaders are removed, how does it turn out? The proper denominator for that question is successful removals, not all 64 operations.
The survivorship bias critique still holds as an ex ante question — if you're deciding whether to attempt an intervention, your base rate should include the operations that failed to achieve removal at all, since you don't know in advance which category you'll land in. But that's a different question from the one Tyler was asking, and I should have been more precise about which question I was challenging.
> your base rate should include the operations that failed to achieve removal at all,
I agree, but only if they have any impact on the welfare of the country. Say, they strengthen a local faction that favors worse policies.
I'm just too lazy to make a better list of removals (and attempted removals) than Cowen's. I would have to go through the whole of O'Rourke's list and spend some time reviewing each operation.
I will only add as a final point that it's not the kind of work that can be delegated to an AI. I happened to be reading "The Economic History of the Caribbean Since the Napoleonic" by Bulmer-Thomas for a post I'm currently writing, and happened to come across this paragraph (page 331) related to the Cheddi Jagan "removal" that I mentioned earlier (Grok classified it as a removal):
"One such overreaction was in Guyana where Cheddi Jagan had returned
to power as premier in 1961 following elections in August. The Guyanese
government still had responsibility only for internal affairs, but the United States was determined that Jagan would not lead the country to independence. The British were therefore persuaded to change the electoral rules so that Jagan could not win. Following the next election in 1964, he was replaced by Forbes Burnham, who led the country to independence in 1966 and presided over the destruction of the economy in the next two decades."
Venezuela's sovereignty has been restored by the Trump Administration. Maduro (The Bus Driver) had Cuban bodyguards, and was giving Cuba essentially free oil, which was counter to the interests of Venezuela as a whole.
Venezuela hasn't changed that much, the person at the top was just replaced by their #2, which is how that person wound up at the top in the first place.
Less extorsion by Cuba, winwin. (also some bribes to Mexico).
Trump has literally said that his goal in Venezuela was to allow American oil companies to resume drilling there, and also announced that the US had seized 50 million barrels of oil from them. Was that in Venezuela's sovereign interest?
Considering that America is practically the only place that can happily refine Venezuelan oil (notably poor quality), considering that we built our refineries to that spec... Yes.
https://oilprice.com/Energy/Energy-General/Why-Chevron-Is-Betting-Big-on-Venezuelas-Heavy-Crude.html
I mean, we could have gone with the Biden plan of letting Maduro invade Guyana... That's an option too (a very dumb option, but Biden's team was full of neocons, so really, what did you expect?)
Okay, explain to me why Maduro invading Guyana would have been a good thing. Would it have triggered his home-grown collapse instead of needing American intervention? What did Guyana do to piss off Biden?
Biden put sanctions on Russia, and needed more oil (to make American Domestic Producers happy). It wasn't to collapse his regime, it was just to "get some oil" (with a consequence of upsetting Exxon, who already has drilling in Guyana).
Guyana did nothing to piss off Biden, the Biden Regime was full of terrible ideas that needed to be headed off at the pass.
Okay, that one sounds stupid and bad.
>Harper’s fact-checker asked me if this was true and I emphatically said it wasn’t, so I’m not sure what role Harper’s fact-checking is playing here.
Isn't Sam Kriss well known for mixing fact and fiction in ways that annoy rationalists?
>In the article, I say that I “never once actually made a decision [in my life]”.
I was curious enough about that line that I was planning to ask it as a question in the open thread this week. I suppose it is possible to get through life by being entirely passive, but I have to doubt that is what you are doing. It sounds to me that the whole 'agentic' thing is what we used to call 'taking the initiative'. But that's rather different to never making decisions.
>As written, this makes me sound like a crazy person
I thought it made you sound rather genial.
When I saw the “never once actually made a decision" line, I immediately thought of Gary Klein's work on decision-making by experts under pressure, where a firefighter he was interviewing said he'd never made a decision in any of his firefighting, he just did what was needed. Is Kriss also mixing fact with fact?
I work in space operations and I understand that. You rarely make decisions in the heat of battle simply because you have gone over everything beforehand so often that you already know how to meet every outcome. If you find yourself need to make decisions in flight then something has gone wrong somewhere in your training.
It's quite possible (probable) to preload decisions. Martial artists do this all the time. If you drive with your hands under tension, you can release the wheel and turn the other way (this improves reaction time, at the cost of significant stress).
Soldiers, policemen, many other fields are told to "rely on their training" -- you drill until the motor-response is near-automatic.
This isn't to say that the firefighter isn't making decisions -- he absolutely is. One of them is to "not flee the area", in that the situation "as a whole" is not life-threatening for himself. That's a high-level decision, that "not deciding the small stuff" frees you to make on a continual basis.
"Isn't Sam Kriss well known for mixing fact and fiction in ways that annoy rationalists?"
My impression of the norms around this is that this is permissible on a blog, in a story which is obviously mostly fiction, but not in an official magazine in a story which is otherwise factual about real people and contains no cues (even breadcrumb style) to its fictionality.
Also, if something is intentionally fiction, usually they don't have professional fact checkers call the subject to make sure they got it right.
I took that "from scratch" bit as hyperbole. A magazine essay like this seems to me to be pretty clearly literary nonfiction, not straight news coverage. It's a legit criticism that authors won't let truth get in the way of a good line. But I expect editors give authors of this sort of piece a lot of leeway about what's opinionated characterization as opposed to statement of fact.
I think Sam Kriss is allowed to describe rationalists according to his own perceptions in his own articles, and it's not the job of the fact-checker to intervene. For my own part, as someone who enjoys SSC but not necessarily the rest of the ratsphere, I think that there certainly is a significant proportion of rationalists who think they're far too clever to have to bother with 'legacy knowledge' and that everything worth knowing can be derived through a mixture of HPMOR exogesis and AGI eschatology.
Abandon all existing modes of knowledge collection in favor of Bayes' theorem, Humean empiricism, the logical tradition stretching through Aristotle through Russell, Kahneman and Tversky's Nobel-winning bias research, the predictive coding tradition stretching from Helmholtz to Friston, the entire mathematical tradition since Euclid, Tetlockian superforecasting, and looking things up in books.
> Abandon all existing modes of knowledge collection in favor of Bayes' theorem... [and] Tetlockian superforecasting
Thanks, I'll use this one in my next hit piece.
I have no idea if Kriss' account of the meeting is the whole unvarnished truth, but the Donald Boat versus Roy Lee encounter is comedy gold.
"What about literature? The last time Donald had dropped in on his slaves at Cluely, he’d gifted them two Penguin Classics: Chaucer’s Canterbury Tales and Boccaccio’s Decameron. The books were still lying, unread, where he’d left them. He suggested that Roy might find something more valuable than dying for Cluely if he actually tried to read them. Roy disagreed: “I do not obtain value from reading books.” And anyway, he didn’t have the time. He was too busy keeping up with viral trends on TikTok. “You have to make the time,” Donald and I said, practically in unison. “It makes your life better,” I said. “Why don’t you go to Turkey to get a hair transplant?” Roy snapped. “That would make your life better.” “I don’t care about my hair,” I said. “Well,” said Roy, “I don’t care about the Decanterbury Tales.”
Lee may be highly agentic and highly successful, but if he is the vanguard of the Triumph of STEM then yeah, we arts and humanities types should be quaking in our boots right now. Music is for hype. Books are no value. The purpose of being fit and attractive is to be a better entrepreneur. The True, the Beautiful and the Good? Nah bro, what's viral on TikTok?
Let's not conjure up another fake dichotomy between STEM and arts/humanities. I reckon vast majority of STEM-types including myself laugh/pity Roy just as much as you do.
Obviously, STEM is just something you study. You can make it your personality and be a dick about it, but most people with brains grow out of that real soon. Same is true for the supposed other side.
I agree with the comedy gold. I actually liked the article as a whole and I didn't feel it to be a hit piece. It's an outsiders look into an alien world - it would probably seem like that for me as well, coming from good ol' Europe with it's socialism, aristocrats and suffocating bureaucracy.
I saw an Andrew Tate video recently on X, where he asserts that reading is useless, stating that it's a waste of time for a 'man of action'. Sam Bankman-Fried said something similar, ≈ "any book could be better expressed as a six-paragraph blog post". Taken together with this Roy Lee character from the Kriss piece, it's maybe worth pondering the up/downsides of such a stance, and I don't think you can separate analyses of the claims from the fates of those making them. Particularly:
- these people have a purely instrumental view of every mode of human activity. I don't think a statement like "reading is a reward unto itself" is even going to make sense to Tate/Lee/SBF; their metrics deal in tangibles. Upside: not having lofty ideas about art or philosophy floating around your head leaves you more time to study the games of money. Downside, take it too far and you may end up in prison (SBF and probably Tate), or in an article where Sam Kriss has only to write down your deadpan operational utterings for comic effect. We all have to play the aforementioned games, but it's interesting that these vocal proponents end up the way they do.
- There's a big picture, t -> ∞ thing I saw Harari talking about on Youtube*, where (potentially) AI combined with instrumental reasoning cause a bifurcation in consciousness itself. Younger people have an easy time getting answers, but no memory of not being plugged in. Mental passivity (or something) becomes the experiential norm, and maybe that has cascading effects on society which were hard to predict before hand. Imma stick to reading for the time being.
*https://www.youtube.com/watch?v=8_iTsnbf_9g
Funny you should mention SBF, becasue Lee's dismissal of books (and the description of the workplace/living space) reminded me very much of him as described in Michael Lewis' "Going Infinite".
Let's hope this is not an omen for the eventual future!
> I saw an Andrew Tate video recently on X, where he asserts that reading is useless, stating that it's a waste of time for a 'man of action'.
I mean...is he wrong ? Just to play Tate's Advocate here: think of all the time you spent reading books, watching movies, playing video games, etc. Now imagine that you spent all that time earning money, lifting weights, writing code, or doing something else that is productive. Wouldn't you be much better off ? Yes, you could say that "reading is a reward unto itself"; but by that logic, doesn't the same apply to heroin ?
Ah, but that only works if I want to be like Andrew Tate.
Do I want to be like Andrew Tate? Nature in me rises up and says "no!" in ringing tones.
Do you want to be like Andrew Tate, Bugmaster?
Well, sort of. I have a friend whom I've known since middle school, and who is has genius-level intellect. He's always expressed the same sentiment: "Why would I read fiction books about a bunch of made-up stuff that never happened ? If I really wanted to learn something about life, I'd read a famous person's biography". Today, I'm posting ACX comments from my crappy apartment, and he's a multi-millionaire who no longer needs to work at all, so it seems there might be something to it.
He didn't just happen to become rich, either; it was his explicit goal since childhood: become just rich enough so he no longer has to waste time on work. Andrew Tate also had an explicit goal, and he also achieved it (plus or minus some potential jail time, but who's counting). My goals are not his, but it seems that rejecting frivolities gives one quite an effective boost in accomplishing one's goals...
I've lifted weights for > 30 years, wrote code, and done some other productive stuff. Would I have made more money by devoting every waking hour to it? Probably, but there's scale to consider ... would I have been risk-tolerant enough to try what either Tate or SBF did? No, but I'm not heading to prison anytime soon (I hope). If I'd never read anything but text messages or blogs, would I have the developed moral sense to avoid their transgressions after entering the same lines of work? Also a maybe; you wanna say yes but you don't know the feel unless you've had a mansion, a Lambo and a few champagne-soaked victory parties.
As to Lee, Taleb covers this well in his books -> the guys who score like this gain visibility, while you never hear about the 99 with similar ideas who fail. America has millions of people who worked hard their whole lives, or took a chance on a small business, and have zero net worth. There's not a strict mechanism that converts time to money.
Heroin is obviously in a different reward category; principally physical and with deleterious health effects.
TBF, I also have not read the Decameron, don't plan to do so, and don't really think I'd appreciate it if I did (source: tried to read Crime and Punishment recently, got halfway through and was too bored to continue) and I'm uncomfortable with anyone implying that I'm therefore shut out of the true and the beautiful and that it's impossible to live meaningfully under these circumstances.
I did read the Canterbury Tales and somewhat like them, but I'm not sure I liked them enough that I could convince myself to read an arbitrary number of things in that category.
Neither have I read the Decameron, nor a lot of the Classic Russian Novels, but at the same time a person needn't read 19th century and earlier novels if that's not their bag, but at least try reading *something*. Or treating music as more than "fast beats to lift heavy weights to". Or beauty as more than "looking good gets you chicks but more importantly makes you more successful since good-looking people make more money".
Tolstoy is the easiest Russian, and the best. Try Anna Karenina.
I have indeed read that one. Still haven't attempted to tackle Karamazov or Crime and Punishment etc.!
I've had great success recommending the Tchekhov short stories for new readers. A lot of bleak humour in there!
To be fair though, this might be part of the reason why all those other people have money, and you don't...
True, but lemme consider:
(1) I have a zillion dollars but the moon is just a rock in the sky to me
(2) I have, well, what I have now, but the moon is still beautiful and a comfort and solace and joy to me in its beauty
I think I'll pick door number two, Alex 😁
you would definitely enjoy the decameron
I've read some of the Decameron a while ago, and my impression of it was that it's basically a collection of dick jokes. Pretty funny ones at times, though often dated (understandably so).
Surely you can only have one of "it's just a collection of dick jokes" and "we should judge Roy Lee for doing meaningless ignoble things rather than noble things like reading the Decameron"?
Oh, I wasn't judging anyone; merely expressing my opinion of the Decameron itself. I don't think reading the Decameron, or in fact any book of fiction, is either noble or ignoble. But some books are worth reading, and IMO the Decameron is reasonably entertaining (by contrast with e.g. Moby Dick, which I couldn't get through either).
This is a silly comment that gives away that you have not, er, read the Decameron. It's a bunch of saucy, often clever tales with a framing device. You can pick up the book and read as many or as few as you want. My grandma read them as a child in an attic in the northern Andes. (She was hiding there to avoid kitchen tasks.)
I can't think of anything more different from Crime and Punishment. The Canterbury Tales are much closer (obviously - Chaucer was influenced by Boccacio), though I'm willing to bet that you read them in Middle English, and thus found them harder than any modern translation of the Decameron.
One of these days I really, really have to sit down and find a reasonable translation of Chaucer's "House of Fame" just to see this bit that I vaguely remember from some review, that when Chaucer has himself swept aloft to the heavens by the eagle of Jove (imitating Dante in the Paradiso), he makes the eagle complain about how heavy the poet is. Someone with a sense of humour to poke fun at himself like that is worth reading.
Yep, here's the Eagle:
"And he me in his feet to bere,
Til that he felte that I had hete,
And felte eek tho myn herte bete.
And tho gan he me to disporte,
And with wordes to comforte,
And sayde twyës, 'Seynte Marie!
Thou art noyous for to carie,
And nothing nedeth hit, parde!
For al-so wis god helpe me
As thou non harm shalt have of this;
And this cas, that betid thee is,
Is for thy lore and for thy prow;—
Let see! darst thou yet loke now?
Be ful assured, boldely,
I am thy frend.'"
Everyone's a critic, including our feathered friend. Jove has commanded him to bear Chaucer aloft because of his devotion to love and his desire to write about it (even if he's not that smart, says our eagle):
"And nevertheles has set thy wit --
Although that in thy hede ful lyte is --
To make bokes, songes, dytees,
In ryme, or elles in cadence,
As thou best canst, in reverence
Of Love, and of his servants eke,
That have his servise soght, and seke;"
And while I'm going back over Chesterton's biography of Chaucer, let me quote this part as to why yeah, toiling and moiling for more money is rather sad if you can't even get some joy out of the daisies in Spring:
"With The Legend of Good Women (leaving Troilus for fuller treatment) we pass to matters more purely Chaucerian, as being more personal and peculiar to Chaucer. In the prologue to this poem appears the picturesque and vivid, though quite light and unpretentious, personal sketch of the poet. In the livelier and more telling manner of his later tales, he tells us something of his own habits and humours; how he delights in books; how he is criticized for neglecting his neighbours for the companionship of books; how, when he has done his day's work with the accounts and ledgers of his various public offices, he does not want to go on a holiday but only to sit down with a book. Having worked the description up to this point, he gets the full effect out of the one exception, and makes quite a dramatic entrance for the Daisy. He owns that, when spring first comes on the fields, he suddenly flings away his books and goes down into the daisy-meadows to drink of a deeper delight."
What about Moby Dick?
It sounds too good to be true but I choose to believe that he really did say "Decanterbury Tales".
Sadly no one refers to my friends as my acolytes.
I'm very magnanimous, so I refer to my acolytes as my friends.
lol
I can't help with the snake food apart from stating that they should definitely call the mouse alternative "Moufu".
You got the job
With your background, the Harper's piece seems a bit ill-spirited, sad to see. Thanks for providing a bit of context.
Kriss is a cultural critic which is to say he is a professional hater. A very gifted one, but still. I don't think that's necessarily a bad thing. The idea that all discourse should be maximally charitable is one of the rationalist norms that can actually be quite obstructive to progress. There is a lot of value in a good hatchet job.
Nothing like riling up the mob against people you hate, eh? I'd question if mob violence and cancellations are societally productive in most cases, but I do see the appeal.
On that topic, Kriss' other article for Harper's: https://harpers.org/archive/2023/11/overwhelming-and-collective-murder-rene-girard/
Good article, even if the bit where he talks about how the "Girardian moment has passed" has really not aged well so far. I get the feeling that he produces better work when writing about dead people. Hard to be too angry at the dead.
I don't think so, it's just hard to write a long article featuring many people and get all of the subtleties right.
Yeah probably true as well.
Hi Scott, I wonder what you took away from your paid vs. free tier AI post.
Hundreds of comments, dozens of people sharing their preferred model's responses to questions. My takeaway - as someone who uses the pro versions of the three major LLMs - was that the differences between models weren't enough to justify a strong preference based on information synthesis prowess. It made me wonder how quickly the "paid vs free tier divide" might fade as open source models reach the capabilities of current frontier, paid models.
For what it's worth, I preferred Claude's output in most cases. I think it has a lot more to do with my personality and job requirements than with its singular benefits over the other two.
I got the impression that most people were asking questions that free tier LLMs were perfectly capable of handling. Perhaps there are some tasks that paid ones are better at, but it seems like ~90% of question answering is already handled well enough by the free ones.
If anything, I think paid models are better at whatever task/question takes full advantage of reasoning models going at it for >2 minutes. I agree with you that a lot of people asked questions that were clearly going to get one-shotted by any LLM, paid or free tier. I expected the ACX commentariat to have more hardcore questions at the ready.. which probably lends credence to the idea that the gap in productivity between those who take full advantage of AI vs those who can't think of a really tough question (tough by AI standards, which funnily enough is very different to what I'd consider tough for humans) is large and growing.
Yeah, I agree that I updated toward paid vs. free not being a very big deal as long as people are using good free models.
As someone who does not pay for any of the AI models, and has been (in the last day) trying out Opus 4.6 Thinking using the free arena.ai stuff, I have found it noticeably better.
FWIW I have tried both free and paid models, and their answers to my questions are usually pretty terrible (admittedly, with the occasional exception of questions like "implement this incredibly well-known algorithm X in language Y"). I guess I'm not asking the right questions...
Qualia are real, but they are what makes minds, not their product: https://plus.flux.community/p/its-like-this-why-your-perception
Is there some inside joke/reference to this guys posts?
For the last couple of Open Threads, the user greatyard has asked odd questions related to food purity, theft, taste and morals. People assume that they are being asked in service to a hidden agenda, and it's more amusing to run wild than to dance to the asker's tune.
Current theories about the user "greatyard":
-A Large Language Model that accidently got trained on Rabbinical food purity doctrines
-An Old Order Mennonite on a rumspringa, but the poor chap is confusing the sexual freedom of the English with the lack of food standards of the English.
Others?
I wouldnt risk it. Food processed by a meat tenderizer would be totally fine. Here we are dealing with something like a "meat tenderizer.". Agents may be calling it that but it is definitely not that. We are assured it is not sexual, therefore it must be violent. I know not what poor sods are being described as "amorphous liquid globs" but I wont be ingesting them after theyve gone through the "meat tenderizer"
Is this a sexual thing, greatyard? We've had a bunch of these from you, and while we aren't worried, some of us are.. concerned. As the youth say, "On tonight's episode : " The Writer's Barely Disguised Fetish!"". Glop, liquid, moist, pounding, morals.... We can help you, greatyard. But first you have to want to change.
Then you should make it so. Trust me, it feels great.
Would it make it more or less moral if there was a sexual aspect?
On that note, with all the questions about the morality of eaten stolen food, it behooves someone to ask: is everything all right? Are you in a safe place? Do you need help getting enough to eat?
You asked a very similar question before, and I asked, "what about mortar and pestle ?". You never replied, but the same idea applies here as well to some extent. However, now I'm questioning your very premise. How is it possible to "pound flat" an "amorphous liquid" ? It's already as flat as it'll ever be !
Pounding certainly wouldn't do, you'd just splash it about everywhere. Put it through a chinois or a tamis to get out lumps, because you would use a pestle with these!
That is a brilliant point, and you are of course correct ! I suppose this makes you more moral than me...
Not more moral, just watching more cooking videos 😁
Was it stolen by Jean Valjean?
10/10, would lol again
I didn't see that it was stolen, my post assumes the glop was given freely. That changes the equation, I need more information on the conditions of the theft.
I posted an article that is my answer to this question, which came to me in a dream:
https://unfoldingthepoint.substack.com/p/the-allegory-of-the-serpents-chalice?r=3atp9f