When I previously complained here about "open house" signs never telling you when the house is actually open, people suggested that realtors only put the signs up while they're there, but that has *never* been my experience.
Anyway, this afternoon, I saw a new one with not just the usual sign in front of the house, but signs up all over the neighborhood directing people there, and I figured that since it was a Saturday afternoon, there was at least a slight chance they might be open, but nope, house locked with nobody around, just like always.
It's just so stupid and lazy of realtors. It's not like making it impossible for people to visit actually helps them anyway.
Hey! Don't post much but I think there's people here who would appreciate the following opinion piece: there will be a scientific theory of deep learning
Fair warning, I've been working with some of the author's for the past couple months so this is at least in part an advertisement for the paper. But also, I know a year ago I would've really appreciated this paper--I was ust graduating undergrad with a physics degree, cared about alignment/ai things but didn't really know what my skills could be applicable to. I think this paper makes it really clear what a scientist (maybe physicist in particular, but I think most people trained in some aspect of stem could benefit the field in some way) could do to positively impact the current field of AI.
If you're interested, learningmechanics.pub includes resources for breaking into the field and understanding some of the current open questions that we think are plausible and valuable targets for researchers right now.
The title reminds me of Hilbert's famous "Wir müssen wissen, wir werden wissen." We must know, we will know.
I heartily endorse the sentiment, but not with the implication that such a theory is imminent. It might very well take decades (though almost certainly not centuries).
I read Omelas in high school, and I always thought the point was that people may feel morally superior for walking away from a society like Omelas, but they really aren't making any difference without rescuing the child. If a society where a bunch of monsters get to enjoy life crumbles, so be it.
Derek Thompson, an author who has built a strong subscriber base on Substack, posted an analysis similar to (perhaps partly inspired by?) Scott's writing about the "vibecession" among US voters. It's free to read, I think:
Situation-analysis points all backed up by field data:
"the decline in self-reported well-being since 2020 has not been concentrated among young people, poor people, or unmarried people....Instead, the decline in happiness has been an across-the-board 10- to 15-point decimation experienced by practically every demographic."
"the rise of religious non-affiliation in America has been a steady 30-year trend, whereas this falloff in well-being started in 2020", when self-reported secularism stopped increasing.
"low-income wage growth has been unusually strong since the pandemic....Median household incomes are higher now than they were 10 years ago....[meanwhile] some of the largest declines in happiness seem concentrated among well-to-do demographics, like older people, white people, and college graduates."
"the rising misery of young people—often rightly associated with rising phone and social media use—has been going on for about 15 years. The more sudden collapse in general wellness [across all ages and demographics]...points to an emotional break that happened around 2020."
Thompson then constructs his own theory of what the nationwide data is continuing to show.
His conclusion contains one sentence which isn’t quite right: “Inflation has made today’s life harder to afford.” People dislike inflation, but that’s not because inflation is creating an affordability problem. See this graph:
One explanation for why people hate inflation is that they don’t think of wages as the price of labor, so they don’t see the connection between inflation and higher wages. They *think* inflation is reducing affordability even though it isn’t. Another is that inflation makes the world feel less stable. You may be getting raises that keep up with inflation, but you can’t be certain that that will continue to happen in the future. A third explanation is that Trump promised to bring down prices and didn’t, so what they are really upset about is being lied to.
Generally it's the second of your three choices, combined with a general sense of helplessness that inflation gives a lot of people. I'm not one of those but many people, up and down the socioeconomic ladder, are.
Your third thing varies based on whether a given POTUS explicitly ran on that issue. Hence it is a headache for Trump now. But a different way for it to enrage voters is if an administration seems to be ignoring the topic, e.g. Biden in 2023.
While I have constructed plenty of hypothetical prompts to test LLMs, I recently got a stronger response than I’ve ever seen before, and it was on real data.
I’m like, can you help me write a letter to my endocrinologist? Here’s my history of blood test results, current medications, current symptoms … and I got in response from the LLM (I paraphrase here) … go to the Emergency Room _right now_, do not wait until tomorrow morning, do not try making any excuses. It was being a bit overly cautious, but I could see it maybe had a point. (the endocrinologist on duty in the ER authorized increasing my dosage of thyroid medication, if you’re wondering what happened)
To whom it may concern: Substack seems to be having a glitch ("Something went wrong."). If you see it after posting, copy whatever you wrote, and reload the page first. Whatever you posted may have gone through anyway; if you try again, you'll end up double-posting.
The following people are, or have been, famous as aviation or space flight pioneers, specifically a pilots or astronauts who are attached to major "firsts". I've noticed anecdotally that some achievements have had a longer reputational shelf-life than others.
Charles Lindbergh
John Glenn
Chuck Yeager
Neil Armstrong
Yuri Gagarin
Alan Shepard
I've presented the names in random order. I'm interested in the following:
1. Your approximate age, so I can see if there's a generation gap in responses.
2. Which of them you remember, off the top of your head, what they're famous for.
3. Your gut feel of approximate order of significance as aerospace pioneers. A couple are also notable for other things (political careers, etc), but don't take this into account for the purposes of this survey.
Age is 33. I think Lindbergh crossed the Atlantic solo, John Glenn and Al Shepard I always mix up but I believe one was the first american to fly to space, the other the first to orbit. Chuck Yeager broke the sound barrier, Neil Armstrong landed the Eagle on the moon, Yuri Gagarin was the first man in space.
It's hard to say about the order of significance, gut feeling I put Gagarin and Armstrong above Glenn and Shepherd, and I'd probably put Yeager above Lindbergh because I think first in space, first on the moon, first supersonic in level flight, all require less qualification than 'first non stop across the atlantic *solo*' or 'first *american* in space/orbit', and I think I'd put Gagarin and Armstrong above Yeager because space feels more impressive to me than atmospheric flight.
34. I remembered all of them except John Glenn. I knew he was an astronaut, but not the specific achievement.
My gut feeling puts them in the (ascending) order of Lindbergh, Yeagar, Shepard, Gagarin, Glenn, and Armstrong - basically in order of how high and fast they went.
Although I'm not sure if Shepard should get pioneer credit for doing what Gagarin already did.
Armstrong, Gagarin, Yeager, Lindbergh. I think Shepard was American astronaut, but I don't recall details.
Equally mid-low, maybe a slight bias to Lindbergh - the rest are the faces of monstrously huge government programs, and are high talent but very cog-y and low-irreplaceability.
2. Lindbergh crossed the Atlantic solo first (thereby inspiring the name of the Lindy Hop, the original swing dance form), Glenn not sure (but was on some of the Apollo missions?), Yeager not sure, Armstrong was first on the Moon, Gagarin was first to "space proper" maybe?, Shepard might have been first to some lower space-like altitude.
All of them, obviously. (I'm the sort of nerd who can name the crew of every US space mission through 1972, and have a decent shot at the backup crews.)
2. Charles Lindbergh, Neil Armstrong, Yuri Gagarin.. would recognize Chuck as being flying related without context but don’t recall exactly what he did. Other two names don’t ring a bell.
43. I knew all their firsts. But if you hadnt given me the list I might not have been able to produce shepards name. Also I wasnt sure if Glenns first needed the American qualifier.
I am putting Yeager as first for sheer grit in pushing the accelerator through a barrier experts gave a decent probability to blowing him apart. And they had no way of running that test without a human pilot at the time. Everyone else risked death but the engineering parts were better tested.
I put Glen at the top of team space not for his first but the totality of his involvement in the program.
I put Lindberg also high for the totality of flight contribution. I dont think the solo flight itself is of particular importance.
Armstrong ranks first psychologically but last practically. The fact that we haven't been back to the moon shows what a standout achievement it was but in his own words his was but a small step
Edit: this was my gut impression. Follow up research, I am not sure how important Lindberghs career work in different aviation engineering concepts was, but he did dabble in a lot
There are several people I considered including but didn't. Earhart was one of them. The others were (in no particular order) Wilbur Wright, Alexi Leonov, Ed White, Robert M. White, Valentina Tereshkova, John Young, Sally Ride, Douglas Corrigan, John Alcock, Arthur Brown, and Scott Crossfield.
I don't know the names Chuck Yeager or John Glenn. But I bet one of them is the first to break the sound barrier.
Gagarin - first person in space.
Shepard - first American in space.
Lindbergh - first solo transatlantic flight. (Now that I say that, I guess there must have some team flight earlier, but I don't know who did that one.)
Armstrong > Gagarin > sound barrier guy > Lindbergh.
Chuck Yeager was the first person to break the sound barrier. John Glenn was the first American (or first non-Soviet, if you prefer) to orbit the Earth.
Technically early Millenial, but culturally very much late Gen-X. From memory:
Charles Lindbergh was an aviator, and the first to do something - I can't remember what, maybe fly the Atlantic? His other claims to fame are that he was an influential fascist and he committed the famous Murder On the Orient Express.
John Glenn was an astronaut. I think he may be one of the 12 men to have walked on the moon, but beyond that I got nothing.
Chuck Yeagar was a... I want to say a fighter pilot, but he later became something senior at NASA, so maybe an astronaut? I think that he was exceptionally decorated, and I have a vague feeling he was involved with discrimination against the first black man to be accepted by NASA?
Neal Armstrong was the first man on the moon, in 1969.
Yuri Gagarin was the first man to orbit the earth, in 1961.
Alan Shepard was another astronaut - I think possibly one of the crew of whichever Apollo it was that go a film made about it failing but no-one dying? He gave his name to the pilot of Thunderbird 3, which I think means he probably walked on the moon?
In terms of significance, I would say that Lindbergh, Armstrong and Gagarin are probably more significant than the other three, but probably not classify them beyond that.
Chuck Yeager was the first person to break the sound barrier, and was very much a legend in US Air Force. I think the main reason he was never accepted to be an astronaut was because if he died in a spacecraft accident, the PR scandal would have been too great.
> I think the main reason he was never accepted to be an astronaut was because if he died in a spacecraft accident, the PR scandal would have been too great.
If I remember "The Right Stuff" correctly (the book, not the movie), it's because Astronauts were required to be college-educated, and Yeager wasn't. Besides, Yeager continued to be an active test pilot – which carried high risk of death back then – after his feat became public.
GenX, remember all of them from reading about them.
My sense of importance depends on how much of themselves was necessary to achieve their feat. In that sense, Lindbergh would be the most significant on that list because aviation was the least developed at that time, so each individual had a relatively larger share of any achievement. He was first among firsts, so to speak.
In my 60s; I know what four of them did and the other two, Glenn and Shepard, I can guess what the things must be but don't know which did which. My sense of importance is vague so I pass on that.
I recognize Charles Lindbergh, Yuri Gagarin, and Neil Armstong, although I'm not like 100% on what Lindbergh did. Something like first dude to fly a plane across the Atlantic.
Nice bit of fiction. I do so enjoy when scientists start bloviating on theology with no idea what they're going on about. I'd never heard of France's "Penguin Island" but it does not pose any difficulty for Catholicism, for a start; there is no embarrassed God now compelled to grant souls to penguins. Baptism does not count, you can't baptise animals. The blind hermit made a genuine mistake so it wasn't malicious or mockery (like a Satanic parody of the sacrament) but the ontological change can't happen.
I'm not going to opine on neuroscience or consciousness because I am aware of my ignorance on the topic. I wish scientists would return the favour, but eh. Scientists. Too puffed up with their own self-importance ("science explains EVERYTHING! science is the only real reality! I, as a scientist, am the Number One Expert on *everything*!")
Though I see the author of the article is a psychologist, and so what 'real' scientists would not recognise as engaging in a scientific discipline at all. The article also seems to be shilling for his books, so okay:
"Theoretical psychologist and Darwin Bye Fellow Dr Nicholas Humphrey has been announced as the winner of the Dennett Prize 2026.
Given for significant advances in philosophy of mind, cognitive science, evolutionary biology or artificial intelligence, the prize is awarded by the International Center for Consciousness Studies. It is named in recognition of American philosopher and cognitive scientist Daniel Dennett, who died in 2024, and with whom Dr Humphrey collaborated closely."
If someone is still going to be drawing the maps of districts, any proposed "anti-gerrymandering" bill would just be an attempt to shift that in one's own favor.
Anti gerrymandering bills typically have the maps drawn by multiple someones, to limit the amount of influence one party can have.
For example, the recent bill that got defeated in Ohio would have had the maps drawn by a commission of 15 members - 5 Democrats, 5 Republicans, and 5 independents. (And to prevent shenanigans over "but who counts as a Democrat?" the commission members are themselves selected by a bipartisan panel of retired judges.)
Apparently SPLC in the past used paid informants for information on hate groups. According to AP, "the group said the program was used to monitor threats of violence and the information was often shared with local and federal law enforcement."[1]
DOJ is now indicting them about it. It's conceivable that the way it was done could be unkosher. But it seems unlikely that DOJ really trusts their own case, given that their tweeting is trying to make it seem like a totally different issue, flailing and pretending SPLC was propping up these groups.[2] Ofc the real motivation is that Trump's allies, like Elon and the Christian nationalists, hate that SPLC called hate groups what they are.[3] Here's SPLC's own statement, disclosing the paid informants program shortly before the weaponized charges came out. [4] Ken Klippenstein remarked that "NSPM-7 is here."[5]
Even assuming that the DOJ can prove beyond a reasonable doubt that all of the factual allegations made in the indictment are true, a conviction seems unlikely.
The alleged violations of 18 U.S.C. 1343 and of 18 U.S.C. 1956(h) are based on the idea that SPLC defrauded donors. The SPLC revealed that donations would be used in part to investigate hate groups. Methods used to perform these investigations included paying informants and at least in one case illegally accessing records of a hate group. The SPLC did not tell donors that it would be using these methods.
Although it is in principle possible to lie by omission, it’s a huge stretch to argue that that happened in this case even if all of the allegations listed in the preceding paragraph are true.
The alleged violations of 18 U.S.C. 1014 involve opening bank accounts in the names of fake businesses. The basic problem with these counts is that misrepresenting the owners of the accounts didn’t conceal a financial risk to the bank.
In other words, if you overstate the value of your assets when applying for a loan, but pay back the loan when it is due, you haven’t caused the bank any financial harm, but you can still be convicted under 18 U.S.C. 1014 because the lack of financial harm to the bank is a matter of luck. If a financial crisis had occurred before you paid back the loan, the difference between what you claimed your assets were worth and what the assets were actually worth might prove to be the difference between you being able to pay back the loan and you being forced to default on the loan.
On the other hand, what SPLC allegedly did could not cause the bank to lose money no matter what happens to the economy, and it doesn’t seem to me that 18 U.S.C. 1014 applies.
I don't need the DOJ to tell me that the SPLC is a once-useful organization now lost to corruption and political bias, any more than I need the SPLC to tell me that the DOJ is a once-useful organization now lost to corruption and political bias. So if one of them wants to tell me how vile the other is, meh, not worth my time and not a reliable source.
The corruption of the DOJ is the greater and more worrisome threat, obviously, and on that grounds only I'll be rooting for the SPLC in this fight. Winners are less likely to see the need for internal reform than losers.
• Allege that the payments to informants were "money laundering." (Of course you can't pay in your own name, without ruining their use as an informant.) But that's implausible, since...
• It's only "money laundering" if the money was proceeds from a crime. (Otherwise it would also be money laundering every time a cop pays an informant.) But that's implausible, since donor donations aren't proceeds from a crime. So the money laundering case also requires...
• Pretending the donors were defrauded, which requires pretending the money was used to fund the hate groups' hate actvities instead of being used to act against them.
But that's also wildly implausible, given that even according to the indictment, the money was used for info stuff like........... paying an informant to copy documents he stole from a hate group, which went on to inform a hatewatch report. (SPLC also says information from the informant program was often shared with local and federal law enforcement.)
The "money laundering" charge is part of the same nonsense narrative as the claim that they propped up hate groups (as opposed to paying informants for info on hate groups).
"Money laundering" requires the money to be proceeds from a crime. Donor donations obviously aren't proceeds of a crime. So the only way to pretend it's money laundering is to pretend donors were defrauded, which requires pretending that the money was to prop up the hate groups instead of to undermine them.
It is standard practice for prosecutors to overcharge their targets, in order to secure a conviction on a lesser charge (or a plea bargain).
For instance, if the "money laundering" thing doesn't stick, they may still be able to get them on "structuring" ("stacking"), which doesn't require the money be from a crime.
The lying "they were propping up the KKK" propaganda narrative is all over Twitter rn, being regurgitated over and over and over by the likes of Elon and Rufo and the GOP accounts. It's like 90% of the timeline (no doubt due to the Elon algorithm). Being surprised that someone would post about a massive psyop is like being surprised that someone would comment on the weather, when a tornado's gone through a row of outhouses and is currently raining feces all over the place.
Like you, I haven't looked into the BLM money laundering charges. But yes, I agree that Trump's weaponized the DOJ in general, so if those charges also turned out to be an instance of weaponization, I wouldn't be very surprised.
It sounds like you're suggesting I'm too biased against younger leftist groups compared to older liberal ones, or something. Are you? (I'd find that critique coherent from a leftist, but I've been reading you as supporting the Trump DOJ's campaign, so I'm not sure how to parse this.)
For SPLC specifically, yes I have personally known them to do useful work. For instance, their reporting on the anti-LGBT pseudoscience network has been an efficient way to cite that background info in my critique of the Cass Review [https://zenodo.org/records/19303978].
A notable difference is that falsifying business records has been a crime routinely prosecuted for decades -- e.g. there's a long list of prosecutions and convictions under that specific statute in New York courts going back to the 1970s -- whereas these charges filed by the DOJ are legally "novel" (the term used in courtrooms).
Of course being novel doesn't necessarily mean that a prosecution can't/shouldn't succeed. However the specific charge of _fraud_ (not a part of for instance the Trump prosecution) requires specific not general deceptive intent. Fraud under US laws is "_intentional_ deception, misrepresentation, or concealment" made to "induce another party to act".
SLPC donors having been unaware of the payments to informants will not by itself rise anywhere near the level of committing fraud. (Also, contra Patel's public statements but not as far as I can tell part of the charges filed, it is not at all true that a nonprofit has a legal duty to make sure every donor is aware of every single thing the organization is doing.)
Perhaps the DOJ can show that SLPC staff overtly lied to or misled some specific donors in order to conceal the fact of the paid-informant program. The judge will ask this question and if the answer is that the feds have some such donors lined up as witnesses to that effect, that will be bad news for the SLPC and the case will proceed.
If on the other hand it turns out that the DOJ is just trying to convert "some or all donors didn't know about this specific activity" into charges like wire fraud, then the court will dismiss this case and the only question will be whether the dismissal is with or without "prejudice".
Well gosh, I guess I'm a Christian nationalist hate group myself (given that I have a LOT of not congruent with current 'Every Right-Thinking Nice Liberal Person Holds That In This House We Believe' opinions, views, and beliefs), want to turn me in for the reward, David? I'm sure the SPLC can always use another tout!
>"The SPLC used money from donors to pay informants/quislings to 'nudge' the KKK into doing more provocative actions, so that the SPLC could get more donations."
Even the DOJ's own indictment doesn't claim that (though their weaselly press release tries to imply it). The alleged "state and federal crimes," which DOJ is hoping you'll think were hate crimes, instead seem to be........ an informant stealing and copying documents from a hate group, that were used for a hatewatch investigation.
Would it be surprising if it were true? There are plenty of examples of progressive activists faking white racism, from hate crimes to vandalism. And the entire overarching strategy of anti-racism is to make people identify with their race more strongly and hate/fear other races. I’ll wait to see evidence of course but I’d hardly be surprised by it.
>"planning it and driving people to it ... this is nudging folks into doing more than they otherwise would"
The indictment doesn't claim he instigated the rally. Also—do you think the rally, which was already being planned, couldn't have gotten another person to find a driver if the undercover F-37 hadn't been the one to find a driver?
>"And yes, breaking and entering and being a thief are in fact federal and state crimes"
Tbc my quotation marks were regular old quoting-the-indictment marks, not scare quotes. If they stole docs from a hate group, they stole docs from a hate group. The point is that DOJ is really hoping you'll think they were funding hate crimes, because the money laundering charge depends on that sleight of hand.
Extremist groups do often try to hijack mainstream symbols (imo hoping to get caught so they can pretend the people who caught them are un-patriotic or frivolous)—and there is contemporaneous reporting about the anti-vax trucker convoy trying to do so with the flag—so i wouldn't be surprised if SPLC commented on it, but yeah idk the details if they did
I wrote a short essay in two parts (~15 minutes to read, longer if you sit with the questions), a pamphlet for people wandering the cave. The first part uses an apple to take apart the line between subjective and objective, the second looks inward. It tries to point rather than argue a point. If you ever had the wordless searching feeling, this is intended for you.
With your apple example, you seem to be re-inventing the concepts of accidents and essences, and I understand that in Buddhism there is the concept of attributes, whereby you can break down anything/anyone into "if you take away 'woman', 'height', 'colour of hair', 'name', and other qualities, what is left of this person you claim is a unique conscious individual with a self?"
So this essay seems to be treading established ground, but I am presuming it's only an introduction to what you want to develop?
Thanks for engaging with it. You are right, there is an overlap, you could even name more perspectives, like the Vedanta, Daoism, the Cook-Greuter scale and others, the text tries to point at the perennial philosophy core, as I see it and have walked it myself.
I find a point of pure being that I can't reduce further, but things get hard to describe in words at that point. I think the point is reached through a minimal steering, so I keep that as the will behind the thoughts. Phenomenological, I'm uncertain how unique the self is at this point; I see it as part of a whole, rather than the complete thing.
I don't actually want to develop it further for now. I wrote it as stand-alone text, I would have liked to have earlier in my personal development. It might be worthwhile to develop the ethical fundamentals further at some point.
On a previous post, Scott mentioned in passing that perhaps there should be deontological bar to voting for candidates whose policies you prefer, but where there is some other salient feature to consider, like dictatorship risk or a proclivity to raping children.
I wonder how common this view is. My own view is that the stated policies of the other guys are so bad that, no, I don't think there is any such criterion that would dissuade me from even marginally lowering the probability of them being implemented.
I think we're entering an era where even fewer people and media outlets than at present are going to be able to figure out whether a candidate has raped children or committed some other godawful crime -- image fakes, information hacking and persuasion are going to be too good. Doesn't really answer your question, though.
Also, we're entering an era where many people don't care.
"You think that raping children is bad? That's woke! Ancient Romans didn't mind things like that, and they were a strong healthy civilization, unlike the degenerate modern West."
Once people adopt the mindset that good is weak and evil is high-status, the argument that something is evil no longer appeals to them.
You have two basic strategies. Either the good people can coordinate to stop evil people, or the weak people can compete at pandering to the strong ones. These are two attractors in the policy space, and we have switched from the former to the latter.
There is a difference between "I repeat the falsehood because I am unable to figure out what is true" and "I knowingly repeat the falsehood to signal that I am on the side of the strong ones".
> "I knowingly repeat the falsehood to signal that I am on the side of the strong ones".
That's how Soviet propaganda worked, and how propaganda in Putin's Russia works today. They state laughably absurd things as truth, not because they expect anyone to actually believe it (although it is a bonus if someone does), but to single out and arrest anyone who *does not* enthusiastically repeat it.
Well, now the Russians have succeeded to export their wisdom to the entire developed world, hurray! They might no longer be able to put a man on the Moon, but at least then can convince millions of people who should have known better that the Moon is made out of cheese.
Also, this is what bullies often do, e.g. they punch you while saying "I am not punching you, you are punching me" and find that hilarious. This is how they coordinate the story they will tell later.
> Once people adopt the mindset that good is weak and evil is high-status, the argument that something is evil no longer appeals to them.
That's from your perspective, yes? As far as they're concerned, good is high-status and evil is weak. Once a morality stops being useful and only serves to harm its hosts, it's only natural that they shift to a morality that serves their collective interests better. But ultimately, they do care, and they do think they're righteous.
You have your definitions for good and evil, others have theirs. Either way, it's silly to claim that they're just nihilists who are purposely being evil. You can't organize something like this without some level of collective self-righteousness. Do you not understand that they are feeling the exact same feeling as you are, that people like you can't be reasoned with?
I am not going to write the entire thesis on morality here, just some quick notes.
Morality has a self-serving aspect, in the sense that otherwise it could not have appeared, from the evolutionary perspective. Moral systems such as "we need to commit collective suicide a.s.a.p." tend to be short-lived. Under usual circumstances, acting morally supports group survival, although weird things may happen in unusual situations.
But that is not the same as saying that morality is whatever is self-serving at the moment. Actually, morality is full of "you shall not"s related to things that obviously at least some people are tempted to do once in a while. It would be a waste of time to forbid things that no one was going to do anyway. When people say that someone behaved immorally, they usually mean that the person did a self-serving thing that somehow hurt someone else (whether someone real or imaginary).
From this it seems obvious to me that people who follow the rules and resist temptations (even if the success is only partial) are not doing the same thing, from the psychological perspective, as people who keep inventing excuses for doing whatever is convenient at the moment.
There are more details, this is just a simple thing, but as I said I am no going to write a thesis here.
Yes, there may be different sets of rules, and people following different sets of rules may find themselves in conflict with each other. Which is not the same as finding yourself in conflict with someone who follows no rules other than "whatever feels convenient at the moment".
Of course, those who follow no rules can pretend that they have a secret code that the outsiders simply don't understand. But it usually turns out that the secret code is infinitely flexible, i.e. has no predictive power beyond "this person would do whatever feels convenient for them at the moment".
The poster "Dust" is an admitted nihilist and sociopath. His posts are always aiming at the seeming inevitability of large-scale violence. Perhaps this helps to shed light on the discussion here.
It is 2026 and I feel like today's politicians cannot be judged by their stated policies or anything else that comes out of their mouths in the moment, or indeed that they might set down in a manifesto or other document or record. In our brave post-truth world, words alone cannot be trusted. The most reliable means I have of judging what they might or might not do going forward is their past actions.
I would like to see a norm of voting for people based on how they might behave over which team they are part of, but I feel even this is asking too much.
I have no idea how this state of affairs can be made better, but that doesn't particularly matter as I am one old man shouting at clouds while everyone else is hell-bent on making it worse.
>I would like to see a norm of voting for people based on how they might behave over which team they are part of, but I feel even this is asking too much.
Except that the team they are on is a reliable indicator of how they are likely to behave ( politically)
I don't think it's very common. It was far more common and it would be awesome if it was common again. Unfortunately, there is not and cannot exist strong legal/political enforcement of this norm and social norms tend to degrade over time.
Let's, uh, avoid modern politics and go back to Rome.
Say we have a strong social norm that we will never support anyone who purges all their political enemies using violence or anyone who marches their army upon Rome or installs himself as dictator. So, imagine Sulla, he's consul and marches an army off to fight some Greek dudes. While he's away, Cinna and Marius declare him an enemy of the state and murder all his friends in Rome. Notably, he's am still in charge of a large and victorious Roman army. It's *probably* fair to say that Sulla is not justified in marching his army back to Rome, declaring himself dictator, and then violently purging all his political enemies but it's hard to clarify exactly what he's justified in doing because the strong social norms that are meant to prevent this have already been violated. Worse, given Sulla's actions, it's hard to see why someone like Caesar, who Sulla almost killed, isn't justified in doing the same thing.
I'm totally on board with "there should be strong dentological bars to voting for certain candidates" but that requires social trust/capital/*mumble* *mumble* "high-trust society" and we don't have that. Barring that, at a practical level, upholding social/political norms of decency feels like a sucker's play when *Hated Outgroup* violates them all the time without political or social consequence.
> Sulla is not justified in marching his army back to Rome …
I'd say exactly the opposite, that once they declare you an enemy of the state, that justifies massive retaliation – present-day analogues left to the reader – so I guess that's actually a validation of your broader point about the collapse of shared social norms.
>we had a former servicewoman saying that Biden was trying to start World War III (and that's corroborated with the scuttlebutt I was hearing, which was quite a bit more... colorful).
How can you possibly credit such obvious nonsense?
Come on. man. She did not say that Biden wanted to start WWIII. She said he was creating a risk of undesired escalation. Surely you are not unfamiliar with that sort of claim. It is, among other things, literally Game Theory 101.
>President Joe Biden’s administration is driving the world toward a devastating nuclear conflict by using the Ukraine crisis to fight a “proxy war” with Russia, former congresswoman and 2020 presidential candidate Tulsi Gabbard has warned.
>“The American people need to understand the seriousness of the situation that the Biden administration and leaders in Washington have put us in,”Gabbard spoke Wednesday night to Fox News in an interview. The Hawaii Democrat, who was among the candidates whom Biden defeated to win his party’s 2020 presidential nomination, called for stepping back to focus on “the big picture and the great threat that we face, which is the reality that this war is continuing to escalate.”
>Biden led international efforts to punish Moscow for its offensive against Ukraine. US approval of $70 billion worth of aid to Ukraine, including weapons and advanced technology, has been given. Biden visited Israel Thursday and stated that Washington would continue providing such assistance indefinitely so Moscow does not suffer. “a strategic failure”Ukraine
>“President Biden himself says he has no idea when or how it’s going to end, but we know where this escalation leads,”Gabbard spoke out. “It leads us closer and closer to the brink of a nuclear war with Russia.”
And btw here is an argument, based on actual facts about Biden Admin limits on Ukrainian access to, and use of, American weaponry, that Biden was too leery of provoking Russia:
as of this post you have at least 112 comments, leading to more than 1/8th of total posts. A lot of them contain obvious conspiracy theories, blatant partisan information gathering, and vague insinuations, and quoting Tulsi Gabbard on this? on a political comment about Biden trying to start WW3? As extraordinary evidence? What, is the next quote going to come from the white house press office about how the Iran war is totally Biden's fault and has nothing to do with Trump and he is the most peaceful president ever? Give me a break. Your posts are temperature-elevating, contain outrageous statements and vague insinautions to low-quality claims about grand conspiracies and hidden information. It's like a link "you would never believe what <insert political figure> wanted to hide from you" rather than high-qaulity, rational discussion. Like a big flashing popup in the middle of a book. All the time.
I repeat my earlier plea- please consider posting WAY less, like half, and putting double the effort in your posts. The ACX comment quality has nosedived and you are a major factor in this.
Wrote an analysis of the Musk/Altman conflict that reframes it as competing collapse scenarios rather than AI safety debate.
Core thesis: Musk is building post-state capability (works when systems fail: energy independence via Tesla, comms via Starlink, logistics via SpaceX). Altman is building acceleration-dependent systems (needs grid stability to reach AGI before coordination failures cascade).
Theme: Reading their infrastructure choices as revealed preferences.
That is dreadfully written, and if this is the state of the art AI production, then everyone trying to persuade me to give the latest AI model a go and change my mind is trying to make rain fall upwards.
Never mind the content, the constant BOLDING to try and MAKE THE POINT is aggravating, and the prose is adequate but not informative. Is The Boring Company going anywhere, both literally and metaphorically? The Hyperloop? Neuralink seems to be continuing but I am not aware of how much progress has been made, I saw some stories about promising indications but as yet no "this will make the blind see and the lame walk" achievements of all the claims.
Your AI-generated graphics are full of errors, which is one of the risks of using AI-generated media. Aside from that your post is not very informative, it seems to just be a massive glaze of Elon in the cadence of an LLM. Ask yourself: is this level of quality worth responding to? Is it worth aspiring to?
There is no world without nation states (or their hypothetical corporate equivalent) that would allow Musk to maintain any of his technology. Any classic, off-grid prepper is better equipped to survive civilizational collapse than he and his toys.
> That’s not about environmentalism. That’s about Energy Sovereignty.
As someone who used the em-dash before it became everybody's favorite AI tell, I feel qualified to point out that this sentence structure in this day and age might be something you want to avoid.
So many (male) writers write about dating that does not seem to mathemathically add up, like how most young men are single and most young women are in relationships, that got me curious. Is it possible that for GenZ assortative mating is over, young white professional women are choosing non-white or working-class men, who are generally invisible to young white professional men?
I don't go out much, and a bit too old to intuitively understand young people these days. But for example 10 years ago I was walking all over Cologne, Germany, and the masculinity difference between young white men and young Middle Eastern men was staggering. The first looked like babies who somehow just did not go through puberty. The second even at 16 sort of looked like young men, not boys, with well-growing beards and an adventurous gleam in the eye.
I think you might be reading too much into aesthetic differences. It wasn’t so long ago that hipster man were wearing elaborate mustaches and beards that would have looked respectable on a civil war general. But most of them hadn’t led even one cavalry charge.
Some things have existed always. -- Older men compete with young men for the young women. High-status men are polygamous, even if unofficially. Women are hypergamous.
Some things are more recent. -- The monogamous marriage starts later (so the time interval when the few alpha males can hoard many girlfriends is longer). Longer education that keeps young men longer out of the job market (more childlike, less economic status). More women than men at universities which still remain a source of status.
I think the education is specifically a big problem, because now you need certificates even for jobs that are relatively low status. If you have a job, you have an income, and you can impress some girls. If you study for a high status job, such as a doctor or a lawyer, that can also impress some girls. But many young men have neither a job nor a high-status school. To say it differently, some schools are a signal of a future wealth, and keeping a job is a signal of being responsible, but many young men can signal neither.
The situation is not symmetric between men and women, because many young women date older men, or have one night stands or short relationships with polygamous alpha males. That said, women also complain about the dating market, it's just that their complaints are different.
> Some things have existed always. -- Older men compete with young men for the young women.
Was that always the case? Back when monogamy was more heavily enforced, people were pressured into marrying early, and divorce was not a realistic possibility, it would've been unlikely for older men to get into a legal relationship with younger women. They would've had mistresses, of course, but that didn't make the women unavailable to younger men. Perhaps making divorce illegal again will lead to a more stable and equitable state of affairs for men.
"unlikely for older men to get into a legal relationship with younger women" -- that was a reasonable description of reality only since maternal mortality rates plunged.
One of the many things I've learned through decades of rich family-tree research, 98 percent of it being in post-1600 America, is that until relatively recently older men constantly competed with young men for the young women. That was because so many women died during or related to childbirth. The surviving husband would then remarry, always with a woman no older than his deceased wife was. Often, if the man was 40+ when the previous wife died, to a woman 20 or more years younger than he now was.
Across the board in my large database the above household pattern recurs, hence I have more step-grandmothers through the generations than I'd ever have guessed. Previously had no idea how common it had been, for centuries, for nonwealthy men who lived to reach the biblical "three score and ten" to have had 2, 3 or 4 wives along the way. Same in my wife's ancestry, and those of American genealogy hobbyists I've corresponded with.
Throughout the 17th, 18th, 19th centuries there were tons of "May-November" marriages as the old fashioned expression went. Only during say the last two to three generations of divorce being rare/difficult, did older men marrying much-younger women become rare.
(This wasn't a class-restricted pattern either. I am descended from mostly what we'd now call working-class households, with educated higher-socioeconomic folks sprinkled in here and there. No powerful "elites" -- who throughout history could obtain divorce if motivated enough -- going back at least 400 years.)
>So many (male) writers write about dating that does not seem to mathemathically add up, like how most young men are single and most young women are in relationships,
Do people say that? Most of the stuff I've read suggests that young women are mostly single now as well.
>Is it possible that for GenZ assortative mating is over, young white professional women are choosing non-white or working-class men, who are generally invisible to young white professional men?
I'll look for the link if you want, but I came across a study recently saying that white German women were very unlikely to want to get into romantic relationships with third-world immigrants. Obvs beware the man of one study and all that, but it certainly sounds plausible -- would you really want to date a poor, low-status person who probably doesn't speak good German and comes from a society where misogyny is completely normalised?
>The second even at 16 sort of looked like young men, not boys, with well-growing beards and an adventurous gleam in the eye.
Not sure how much of the difference is due to this, but a lot of Middle Eastern immigrants straight-up lie about their age -- it's easier to get asylum if you're a minor, and if you "lose" your passport on the way it's difficult to prove that you're actually over 18.
Regarding Orbán: somehow everybody is only seeing one half of the issue? The other half is the republic, that is, when matters are decided by deliberation, even when, like Roman Senators, the participants are non-elected.
In modern republics, it is professional elites who participate in the deliberation. Orbán did what Curtis Yarvin wants Trump to do, to replace the entire "horizontal" system of stakeholders deliberating with a "vertical" system of elective kingship, where every decision is made by one man, after listening to a small number of advisors. Yarvin is fairly obsessed by the kind of CEOs who do not listen to anyone, like Jobs and Musk.
The results were not good. It led to corruption and dysfunction. But yes, the stakeholder elite system also does that.
I don't know why it worked better for Steve Jobs. Maybe a business is not a government.
Now Magyar will try the good kind of populism, a kind of participatory democracy like making government meetings open and random citizens will be allowed to comment or question. Maybe this works better than both. Not because he is a good guy. Simply listening to random people on the street turned out to be a winning strategy so why not continue.
>But yes, the stakeholder elite system also does that
Does it so it as much?
>Maybe a business is not a government.
It definitely isnt. It's OK for businesses to fold. The greatness of the best CEOs is survivor bias, not a magical Mandate of Heaven. It's not OK for countries to fold.
Most businesses fail -- we don't want most countries to fail.
When a business fails, the employees simply find a different one, which is mostly a simple process of applying for a few jobs. -- When a country fails, the analogy is mass migration, which is very costly for the people who move (they can't take their homes with them), and the countries that receive them are often not happy about it.
Also, a dysfunctional business cannot declare a war on competitors and start literally killing their employees.
Is that the main reason? Or is it that a business has a specific external purpose, it's not merely "a way for this group of people to get along and govern themselves".
It strikes me that advocates of dictatorship tend to see a nation as having an external purpose, most commonly war, but also some kinds of ecconomic growth imperatives. While democracy has a fundamentally nihilistic character: people have different preferences, and there's no basis for saying one is objectively better than another. Some people just cannot handle this nihilism. Including a lot of rationalists: they need an externally imposed purpose, which might be "technological progress" or "GDP growth". I think it's an intellectual failing to not be able to handle the deep uncertainty of a society without any imposed purpose to which everything else reduces.
On the other hand, democracies being hypocritical about democracy (e.g. banning parties, banning or restricting the speech of political candidates or activists) is the worst possible advertisment for democracy: "by our actions, we clearly don't actually remotely trust the peoole to make a free choice". And it would be an intellectual failing to continue to support the kind of democracy practiced in the countries that do that.
I've read a couple bios of Steve Jobs, and came away thinking that he simply was not a good manager. He was like a bad poker player who gets dealt 4 queens. The PC was ready to be conceived, built and sold. Somebody was bound to be in the right place at the right time with the right store of knowledge to build a business that did that, and he was the first one to have that combo. He didn't succeed because of his wacko management style, he succeeded in spite of it.
That story of his getting hundreds or was it thousands of samples of shades of beige, so he could choose the perfect one for his machine? It's nonsense. The human eye can't discriminate anything like that many shades of beige.
He succeeded wildly because he was able to demand an ethic of elegant, clean, user focused engineering in a sector that otherwise was ugly, directionless and getting lost in its own complexity. Otherwise he was just another guy.
This. Steve Jobs wasn't the nerdy tech genius - that was the other Steve's job. The Steve everybody remembers, is the one that was a visionary designer and tech evangelist, who could take something as nerdy as a "personal computer" - and repackage it into something non-nerds saw as cool. And later repackage the "Blackberry" as a fashion accessory for the hip and young, not an electronic ball and chain for corporate suits.
Apple didn't *invent* much of anything, and certainly not the personal computer. And what Apple did invent, wasn't invented by Steve Jobs. He invented nothing, but made many things cool, and better optimized for non-nerd customers.
And, yeah, he picked the right color of beige. Maybe he spent more time on that than he needed to, but nobody else ever recognized that the color mattered at all.
It's more than just packaging and marketing. The gap between the Blackberry and the iPhone in terms of usability is huge. The Blackberry could do phone calls and emails, the iPhone could use the whole freaking internet. Maps alone was the thing that made me want one. (Strictly speaking Blackberries had some kind of web functionality but it was barely functional, it needed innovations like a full-device screen and pinch-to-zoom to make it usable.)
That's the pattern -- Apple isn't the first to ship an internet-enabled phone or an mp3 player or a GUI or a personal computer, they're just the first to make all those little usability innovations that turn those things from annoying fiddly things for geeks into things that normal people want to use.
I have seen the huge excitement and people standing in queues in 2007. The thing is, social media didn't exist. Websites did not have AJAX, which really limited their usefulness. I don't think a lot of people really wanted to look up a basic website such as blogspot.com on their phones. Basically an iPod that can text. I think people mostly used it for music, and the main idea was that they don't need to carry a separate phone and music player.
Yeah, the big difference was the target market. The iPhone market drove the evolution of an app infrastructure that produced all the cool stuff that most people want and take for granted, The Blackberry market drove the creation of a different set of, I don't think they were called "apps" then, but which were more narrowly focused on things that corporate America and the military-industrial complex wanted. Which was more than just phone calls and emails, but less of what makes most non-suits tjhink makes smartphones so awesome.
But you had to make the iPhone a thing, and get it into millions of pockets, before all those apps start getting developed. That's where you needed a Steve Jobs. And yes, being able to fold existing iPod software into the iPhone was a big help in jump-starting all that too.
I do think he did make it cool, and that was hugely important to his success. But that doesn't really cover it. No one else had the power and sensibilities to demand refinement at the time. He forced focus on details that made devices more usable and organized when everyone else was going in the other direction. He bent the curve.
Information tech is best understood by the stacking metaphor. The industry needed him. Organizing complexity well is hard work, and he forced it to happen.
I've never thought of Jobs as this genius manager but he was very, very good at design and styling the product and marketing it. Nuts-and-bolts man-management and balance sheet management? Nothing special. Making his product sound like the greatest thing since sliced bread? Incomparable.
Another thing is that the impending bankruptcy of Apple was and is a myth. When that stock was trading at its lowest around $12 a share they had more cash in the bank per share than that; it was a huge investment opportunity. It was a good move to bury the hatchet with Microsoft.
He bailed Apple out of an operating system cul-de-sac that looked inescapable and the only way he could do it was to go somewhere else and invent an operating system that he could bring back.
But that’s not true either. He was responsible for turning Apple around from a loss of 700m in the quarter he joined to a (small) profit two quarters earlier. That pushed Apple from being a few quarters away from bankruptcy to being safe. He also did hire and retain great people, and was a great man manager. More Alex Ferguson than Carlo Ancelotti - admired and loved, but scary.
For products his simplifying of the number of categories that Apple produced to four, was a great decision. As was the detente and investment from Microsoft. If he was hit by a bus in 2000 he should still be a saviour of Apple.
"More Alex Ferguson than Carlo Ancelotti - admired and loved, but scary."
That is an excellent analogy that gives me a much better grasp of his style, thank you!
I do think he was more on the design/marketing side, than the 'boring grey suits running the business' side, still. Apple products may have been cutting-edge tech, but they were also mostly (mainly?) known for their style and fashionability. Part of the joke about the Cult of Apple was that the latest whatever *was* sold like This Season's Haute Couture. I hate this scene from "The Devil Wears Prada" because duh, there are only so many colours and all you are doing is tweaking which shade of blue, but the thing about 'that is cerulean' applies because Jobs could make it "this season it is cerulean" and those kinds of style and tech choices then trickled slowly down into the mass market:
He seems to have been dealt with that hand 3 times then. Original Apple, Pixar and the return to Apple. It’s the latter that should, on its own, dispel any myth that he was dealt a great hand.
Your argument could work only for the original Apple, he was a better manager for Apple the second time.
Also NeXT, which was not successful in its own right but was successful enough to get acquired by Apple and turned into OSX.
If the biographies make him seem like a lousy manager who got lucky then I think the lesson is that even the best managers will often appear to be fuckups when observed closely. We all do. If anyone closely observed my career and wrote down all my fuckups in a book then id look like a fuckup too.
I distinctly remember when Apple was considered the artist's computer, not only because it ran PhotoShop, but also because the whole elegance of it, like building the whole computer into the screen to avoid desktop clutter. So maybe Jobs was not a good manager as such, but a good product designer, with an eye for the artistic and elegant.
But it might turn out that was someone else. Quite honestly? I suspect a woman. I mean, are ideas like "let's not have desktop clutter" kind of fem-coded?
It worked better for Steve Jobs personally. Did it work better for people who worked at Apple? Did it even work better for Apple customers? Did it work better for the rest of society?
This feels like a matter of taste. From what I've seen of Macs, OSX is anything *but* elegant and easy to use. I think it's just people who have been using Macs their whole lives who are used to its foibles and don't notice what it doesn't do.
Looks like The Onion may have pulled off its purchase of InfoWars:
“Imagine a roaring arena packed to the rafters with pathological liars. High above you in the nosebleeds are podcasters, screaming that you’ll die if you don’t buy their skincare products. Below, on the floor, imagine demonic battalions of super-influencers physically forcing people into home fitness devices designed to dismantle their bodies bone by bone and reassemble them into a grotesque statue of yourself. Out of the throngs, an extremely sick looking man approaches you. He puts his hands on your shoulders. He explains that he is your life coach and that you owe him $800.
Such is the InfoWars (https://theonion.info/)I envision: An infinite virtual surface teeming with ads. Not just ads, but scams! Not just scams, but lies with no object, free radical misinformation, sentences and images so poorly thought out that they are unhealthy even to view for just a few seconds. The InfoWars (https://theonion.info/) of old was only the prototype for the hell I know we can build together: A digital platform where, every day, visitors sacrifice themselves at altars of delusion and misery, their minds fully disintegrating on contact.”
I love that they're doing this and I'm rooting for them to keep it going, but it sounded to me like some kind of paid licensing deal, not an outright acquisition. Not sure of the actual details.
I can do a pretty good Alex Jones voice. (It scares our cats and most other sentient beings though — the cats hide under the sofa when I get going)
I might just audition for the role. I’d need some prosthetics to make myself look as unhinged as Jones sounds though. Getting money to lie my ass off? “Give me a lot of your money and I’ll teach you how to turn your urine into gold!”
You know I had to look that one up and it turns out you *can* turn urine into gold. Or rather, extract small trace amounts of gold from urine, slightly more amounts if the person is undergoing treatments with gold salts:
"Naturally occurring trace gold: A typical human body contains roughly 0.2 milligrams of gold. While minuscule traces (measured in nanograms or picomols) can be filtered by the kidneys and found in urine, it is not enough to collect or make someone wealthy.
Medical context (Chrysotherapy): Gold is significantly easier to extract from the urine of patients undergoing chrysotherapy (treatment for rheumatoid arthritis using gold salts). Scientists use Atomic Absorption Spectroscopy to monitor gold excretion in these patients, where it can be recovered by wet-ashing the urine and extracting the metal into solvents like methyl isobutyl ketone."
"An atomic absorption spectrophotometric method is described for estimating gold in the urine of patients undergoing chelation therapy. Variable absorption at 242.8 nm, not due to gold, is avoided by wet-ashing and extraction of gold into methyl isobutyl ketone."
It seems a German alchemist thought he could turn urine into gold, and in the process accidentally discovered phosphorus:
"The last person I’m going to mention is Hennig Brand. Brand was a German who lived in the 17th century. What’s fascinating about this person is, he experimented with urine. He saw human urine as more than just waste and he discovered something truly precious in pee — phosphorus. Why he was interested in urine? Well, since it looked golden, Brand believed if he distilled urine, gold would be left behind. During his experiment, he collected around 5600 liters of urine and he boiled it, and extracted the components of the urine in different ways. Upon final distillation, the product left behind was white, smelly, and glowed in the dark. This left-over material is now known as phosphorus."
Human pee was actually used as source of potassium nitrate aka saltpeter to make black powder during the US Civil War by the Confederacy running low on black powder.
“Desperate for saltpeter necessary for the making of gunpowder, the Confederacy sent out agents around the South to collect deposits of "night soil", i.e. human excrement. John Haralson, an agent in Selma, Alabama of the Confederate Nitre and Mining Bureau, advertised the following in the local paper: "The ladies of Selma are respectfully requested to preserve the chamber lye [i.e. urine] collected about their premises for the purpose of making nitre. A barrel will be sent around daily to collect it."
Taking over a political enemy's news outlet and restructuring it to fit your vision isn't that unusual. I understand Orban did some of that. Trump is trying to as well.
Many actions are viewed very different when a Republican does them as opposed to a Democrat (or vice versa) too. Of course there's always going to be special pleading as well; that's not unusual at all.
But if you like, Lőrinc Mészáros is also "Just Some Guy."
Great job dragging American partisan politics into a thread that was previously unrelated to American partisan politics! Because that's just what the world needs: D vs R infesting every last crevice of online discussions.
Is this not the thread that started with the story of what the Onion did to InfoWars after they took it over? Are you claiming this is "unrelated to American partisan politics"?
I should know better than to give this thread oxygen but I am curious how far you are willing to take this
WRT InfoWars at least i think this is an actual case with reality having a well known liberal bias. I know or at least have every reason to believe you don’t really think the Sandy Hook shooting was a false flag operation.
As always I’m left wondering how much of this edgelord posturing and how much is IRL Shankar.
Has anyone done any writing/speculation on what enshittification of AI might look like?
It seems like there is an interesting window between "AI delivers massive paradigm shift" and "AI turns out to be completely useless" where "AI has enough market capture to be potentially valuable but not without implementing the kind of bullshit that has made everything from Google to Uber worse over the past few years." This would mean throttling usage (something Claude and Anthropic already seem to be experimenting with), offering an ad-based free product, and generally bringing the quality of every available product down by several tiers, while reserving the actual useful versions for a price point that is way above what even the highest end current versions charge. Perhaps if you are genuinely concerned about AI you should be encouraging this process to happen now, as nothing seems more designed to sabotage long-term progress like promising investors a viable, if objectively terrible, model that can deliver short-term profits, in such a way that the market demands everyone pivot all the resources to maintaining that.
AI enshittification has been ongoing since last September in the narrower domain of agentic coding. There was a ton of drama around Cursor's change to their billing around that time, and then Anthropic published a large write up on how they were screwing up serving inference. There is the recent wave of furor over Anthropic serving dumber models (massively abbreviating) since March, see here for the locus classicus of the current controversy: https://github.com/anthropics/claude-code/issues/42796
They've had terrible uptimes too, relative more traditional SaaS.
Now, all of these issues revolve around the economics of AI and how it isn't really profitable and the monthly subscription model doesn't actually make sense, but I think they meet the technical definition of enshittification—companies hook a ton of users via subsidized subscriptions and pump their usage temporarily (loads of 2x offers and other deals in the last five months), then drop the rates massively or do route to dumber models.
Because of the opaqueness of the metering and routing, and the general difficulty of assessing things like 'intelligence', users are not always confident (or some are overconfident) that they are getting scammed. In that way it exactly mirrors these other examples you brought up.
Going beyond coding (my profession), I've noticed in the case of Anthropic recently that these issues even extend to the chat interface. I haven't done a statistical study of it, but Claude is responding in a more abbreviated and generic way to my prompts and dialogues in the last several weeks, probably for the same reasons its thinking less and quitting sooner in coding loops.
I could say more, but I would say that some of this is due to the massive demand of course, some of it is due to monthly subscriptions being a bad business model that needs to die, some it is due to the heavy subsidizing and the classical hyperscaler business model being employed here, and some of it is due to a lack of concern for quality over shipping fast.
[Edit] Like every other commenter, I just read your first sentence and nothing else, and I see you already concisely said all this.
There is a popular Q&A site called Quora that has seen a lot of traffic from bots, for years. The standard model of making up a question that will be popular, asking it using a burner account, then answering it using the rep-farm bot still works. AI has levelled it up by giving the rep-farm bot the ability to put the question to ChatGPT and pasting the answer to Quora, complete with an AI-generated picture at the end (Quora seems to promote multimedia answers over plain text). Accounts such as FeverPoint and OpenBlueprint are now flooding Quora.
Perhaps this is some of the enshittification you're referring to. Maybe not; it's AI slop that presumes quality AI, and maybe you're more interested in things that make the AI itself shitty.
If so, then the usual concerns I see include AI training on its own slop, so that once some hot take works its way into the training data, it becomes iteratively harder for the AI to question it.
Another serious concern is over who curates the training data. It won't be long before we see LLMs trained on, oh, I dunno, Artemis II was faked and so on.
"It won't be long before we see LLMs trained on, oh, I dunno, Artemis II was faked and so on."
Oh don't worry, humans can still serve up a superior grade of bullshit. Yesterday I learned that the mostest correct opinion to have on the Artemis mission is that it was a tool to maintain and propagate, um, white supremacy or capitalism or whatever, my braincells died reading the thing and I can't remember exactly what the Bad Thing was.
I've written about AI, touching on enshittification briefly, but it seems to me the other aspects are far more disturbing.
And this was before I started thinking about how the brain offloads to System 1 to save a few calories, how the incentive for offloading to System AI is infinite by comparison, and how those most fluently offloading to AI are those whose System 2 capacity is still developing.
> Has anyone done any writing/speculation on what enshittification of AI might look like?
I've written a lot about this. I call the effect you're pointing to "the ten thousand Phd's," and ascribe it to every big company having hundreds or thousands of Phd's and a lot of computing power furiously optimizing on their side, against individual average people on the other side. This is vastly imbalanced, and has led to people spending ever more time eating junk food, staring at phones, gambling, and more.
It's pretty obvious that AI is going to be the next trillion dollar attention surface, because it's going to turn into "24/7 AI assistants in people's ears" very soon. It's the last bastion of attention-time, soaking up all the pebbles and grains of free attention between anything else that phones can't reach right now.
At the same time, there's literally no way they're not going to monetize that surface. Google and FB have done that, and it's made them trillion dollar companies. They each make $200 - $400 per US customer per year, and have a few billion monthly active users each.
Gemini and ChatGPT already have roughly a billion monthly active users each. It is basically written in stone that that attention surface is going to be monetized for all the non-paying users, because data centers and inference costs money, and they need to make enough to pay for them and profit.
Google is already a pro at milking that attention surface, and has several thousand Phd's standing ready to start doing so. ChatGPT has hired some of the big monetization people from FB to do the same there.
So clearly the market is going to bifurcate - the free tier of AI users are going to be getting milked the same way everyone is getting milked these days, enshittification will run rampant, and will probably go in even more optimized and darker directions than we're currently at (for one thing, we'll be able to create individually tailored fully immersive digital experiences that are basically David Foster Wallace's Infinite Jests).
*However* there is hope for the paid tier of users! The paying and agentic will be able to level up in their lives, because having a Phd smart assistant in your ears giving you advice and pushing you to make better decisions will be a gigantic multiplier on positive life outcomes.
These AI minds will know everything about you, they’ll know your thinking style, they’ll know what rhetorical techniques you prefer, they’ll be talking to you in the ways that most resonate with you and making connections, arguments, and analogies accordingly. Super persuasion, but at the personal level, and for your benefit - a super-ego that works, in other words.
And I’m not pretending the AI is going to win all the time here, either. Your super-ego doesn’t win all the time today, does it? All it really needs to do is win more often on the margin. Think of it winning only 10% more - 10% better decisions compounded over days, weeks, years, and decades is a CRAZY big effect size. It’s like getting a 10% financial return that compounds weekly!
So imagine being able to level up on your career, health, and hobbies pretty significantly, and your friends and family able to do the same.
Stack this on top of everyone getting counterfeited at work and getting UBI. UBI is actually not a big reach - with really reasonable assumptions, we can crank $40k - $90k UBI's in the near future. I wrote about that too, I'll footnote it.¹
So obviously the population will bifurcate. The 20% will use their newfound time to focus on their interests and hobbies and families, and will live life to the fullest.
The 80% that currently spend ~11 hours a day on screens will now spend 5-8 hours more per day on algorithmic feeds and porns, and eventually, in infinite VR heavens.
So if this is true, the more people you can steer towards the paid tier of AI assistants, the more people you are saving from a life of Experience Machine coffin slavery.²
Towards this end, I'm actually putting together a startup right now aiming at steering people overall towards the second better AI future outcome, by providing RLHF services for AI personal assistants particularly to the Big 3 AI companies and surfacing instrumentation and KPI's to show those improved life outcomes, thereby aligning the individual and the AI company incentives.
If we show people that their lives have improved by so much on the paid vs free tier, and they've truly leveled up in fitness, career, or social lives, they'll be williing to pay for still higher tiers to keep leveling up. If we show the AI companies that helping people level up is more profitable than milking them with enshittification, they TOO will want to level people up, and incentives are aligned and it becomes a virtuous cycle. Then it's just a matter of people seeing their friends, family, and neighbors hugely leveling up, and getting social-graphed into hopping to a paid assistant tier and doing the same.
Big yes on the bifurcation between free and paid tiers — that's the right central call. One trap worth flagging though: paying for something doesn't actually align the company with you. What aligns the company is what they're measured against. If "leveled up" gets measured by user self-report, the paid tier just rebuilds engagement bait wearing a satisfaction skin. If it gets measured by real outcomes (career change, savings, fitness), the company has to survive a years-long delay before the data shows up, and most won't.
The deeper question is whether ad-funded media is even the right reference class for AI assistants. AI talking in your ear 24/7, calibrated to how you think, is a product that changes the user — closer to pharma, therapy, or education than to Google or FB. Those fields all have institutions built around measuring real outcomes (FDA trials, accreditation, professional licensure). For AI assistants, that infrastructure mostly doesn't exist yet — which is what makes the venture actually interesting. The lever isn't pricing model, it's whatever measurement framework gets built first.
Curious which outcomes you think become measurable inside 12 months. Career changes? Savings? Fitness? Something else?
social diffusion defense: juliusthimm dot com (maybe relevant)
i quibble with “literally no way they're not going to monetize that surface” because you’re not AGI-pilled enough in this statement perhaps?
one with more “ai solves everything” bias might say that a 1-person trillion dollar machine can out-compete monetized incentive misalignment in the long term, where that 1 person is a Karpathy with an agent swarm building the company which provides the $T as consumer surplus rather than value capture.
this is essentially China’s stance, make AI a substrate and commodity for the populace… “open source” doesn’t quite describe the playbook at the cultural political layer that it truly is.
You seem to be aware of the flaws and possible failure modes, but think we can be steered to the better option by... spending more money.
Don't access the free slop, buy the premium Olympus-tier slop instead!
Improved Life Outcomes means people who *don't* need the AI to tell them what to eat, how to talk to other humans, and who have lives that interest and fulfil them. What good is that to a company selling a product?
> Improved Life Outcomes means people who *don't* need the AI to tell them what to eat, how to talk to other humans, and who have lives that interest and fulfil them. What good is that to a company selling a product?
Except look around, obviously people are failing at that in the current environment? That's *why* 80% of people are overweight or obese, everyone eats junk and fast food for 70% of their calories, and everyone stares at screens all day.
There is no path to the better future you want.
Religion won't cut it because nobody will adopt it at scale, advocation won't cut it, "life hack" posts won't cut it, because at the end of the day, you need a superego that lives with you 24/7 and actually works to navigate the environment today, and 90% of people don't have that.
Right now the solution is something like "better than top decile conscientiousness, willpower, and awareness," which is ridiculous, inapplicable to 90%+ of people, and can't ever scale.
And what *can* steer the bulk of people to better choices?
My solution is the ONLY solution that can scale, address 90%+ of people, and actually guide them to better choices, that I can think of.
Like realistically, what is the path to the future you want?
Everyone looks deep inside, has an epiphany, and permanently and totally changes their behaviors, and somehow sticks with it over the next decade, in the face of ever-greater optimization power from large companies? A Butlerian Jihad back to tech levels of 2005? Then we'll descend back to a world where everyone is still fat from eating junk and fast food for most calories and watches TV 11 hours a day, is that actually a win?
They're already 80%+ failing! Even more are sure to fail when the optimization power gets higher!
Like when were we actually better off, in your opinion? What is the ideal state of humanity + technology? Pre atom bomb, but post antibiotics? How do you stuff those genies back in the bottle?
"Super persuasion, but at the personal level, and for your benefit - a super-ego that works, in other words."
For *your* benefit? Now why would the AI do that, when it could instead be steering you to benefit the corporation instead? You are the product now, you are the product forever. That 24/7 AI in your ear will be maximising your productivity for your employer, using you as a data source, and tailoring personalised ads to soak your bank balance.
What's that, you say? But you are a paying user, not like the freebies who are having ads crammed down their necks? Yes, you are a paying user, and you pay. Oh, you pay.
AI will make you level up, because finally now you will work better, smarter, harder! You will be attentive and responsive to your family and friends! You will eat right, eat healthy, exercise more, become that attractive, fitter version of yourself you always intended? Now that you don't have to rely on your own will power and decision-making, but just do everything exactly the way the genie in your ear whispers to you to do, then everything will be better?
And how will you level up after you've willingly and freely scooped out your brain to replace it with that Bluetooth in your ear?
"The 20% will use their newfound time to focus on their interests and hobbies and families, and will live life to the fullest."
Whoo, good job you are going to be one of the 20%! Though, you know, real life flesh-and-blood people are disappointing, they're never going to be as beautiful and finely tuned to your exact preferences as the AI friends in the VR heavens. And you're only one of the 20%, you're not one of the 1% who get to interact in meatspace with all the cool, fun, interesting, leveled up to the max people. Those AI VR experiences are looking better now, aren't they? You won't meet supermodel Nobel prize gold medal Olympians in the real life circles you run in, but VR you can interact with them - or their AI facsimiles, and AI is everyone's better self today, so the facsimiles are more than good enough now that you've been conditioned to accept and obey the voices in your head.
Enjoy your maxed-out world, and don't forget to take out that triple ultra platinum plus tier subscription for the even *better* AI model experience today!
"steering people overall towards the second better AI future outcome, by providing RLHF services for AI personal assistant"
I laughed. Get your better life today - by becoming a servant to the machine which is supposed to be your servant. Put the horse in the cart and pull the cart yourself, you'll get to town quicker! 😁
"Then it's just a matter of people seeing their friends, family, and neighbors hugely leveling up, and getting social-graphed into hopping to a paid assistant tier and doing the same."
Welcome to Malebolge, where you are algorithmically sorted into the *exact* pocket of Hell tailored specifically to your particular circumstances. Work 24/7 so you can spend all your money on that ultra tier subscription, because you can't fall behind your "friends, family and neighbours" who are all also spending every last cent to enrich the Sam Altmans of the world because only being on last week's model and the corundum not the astatine level means you fall behind so far, you're just a loser. All the little hamsters run on the wheel forever, going round and round but going nowhere, see their little red queen crowns which are the prizes they get for running faster and faster!
> For *your* benefit? Now why would the AI do that, when it could instead be steering you to benefit the corporation instead?
I mean, that's literally what's happening today with smartphones, that's explicitly enshittification, isn't it?
There really IS a difference between paying for a product and being the product.
You're handwaving that difference away and claiming they'll just steer you into buying Ford and Coca Cola sub rosa, but I don't think that's how it will work, because they'll all be competing against each other for this trillion dollar assistant revenue, and so will be directly incentivized to provide you with assistants doing what you actually want them to do.
At least the Big 3, and probably ultimately 5-10, will be competing here. It's not a monopoly. That makes a lot of difference.
> laughed. Get your better life today - by becoming a servant to the machine which is supposed to be your servant.
Literally where everyone is today ALREADY! That's the point of the "11 hours of screentime a day" figure. When you have thousands of Phd's with a lot of compute on one side, and regular people on the other, they get sucked into spending ever more time on apps and in self destructive ways. Phone screen time has gone from ~2 hours in 2014 to 4-5 hours a day today, and 7-9 hours for Zennials. AND all those optimization powers are only going to get much stronger in the age of AI!
Your complaint here has already happened, and this is one of the few ways out as the optimization power increases still further.
> There really IS a difference between paying for a product and being the product.
Executives are discovering that you can still enshittify what you give people who are paying for your service, you just need to make sure the free tier is shittier than that. They cannot see your sweet ideals over the dollar signs lighting up in their eyes.
It’s 2026, and the difference is already one of quantity, not quality. We have navigated all the way down the river of Moloch and are in sight of the endless sea. There is no amount of money you can pay today to have a fully non-enshittified life. You can pay to reduce it a little, but you cannot make it go away. DVDs and blu-rays, where those are still made at all, come with unskippable trailers. More and more, paid-for streaming services still have ads, just not as many. Amazon wants you to know about the movies and music included with Prime, but nothing you actually want to watch or listen to is included; you have to pay separately for anything actually good. Deliveroo now charge for individual orders made with their premium service, and have a super extra premium service that waives more of those. Samsung wants to sell you refrigerators that display advertising in your kitchen; this is a /premium product/, the cheap tier doesn’t have the giant displays. We welcome it into our lives; we call it and pay for it to come. We are building the dystopia we deserve for ourselves with our own hands.
> We have navigated all the way down the river of Moloch and are in sight of the endless sea.
Yes, I wholly agree, and it's *because* everyone spends 11 hours a day on screens, and is being continually sniped by the ten thousand Phd's.
That's a choice! You don't HAVE to use Hulu, Doordash, Deliveroo, or buy Samsung or LG appliances.
I have basically zero enshittification in my life, because I make those choices, curate my media diets, don't install apps, and so on.
And what can steer the bulk of people to better choices?
Right now the solution is something like "better than top decile conscientiousness, willpower, and awareness," which is ridiculous, inapplicable to 90%+ of people, and can't ever scale.
My solution is the ONLY solution that can scale, address 90%+ of people, and actually guide them to better choices, that I can think of. Religion won't cut it, advocation won't cut it, "life hack" posts won't cut it, because at the end of the day, you need a superego that lives with you 24/7 and actually works to navigate the environment today, and 90% of people don't have that. That's *why* 80% of people are overweight or obese, everyone eats junk and fast food for 70% of their calories, and everyone stares at screens all day.
Literally the only solution I can think of for most people is "aligned AI," so I'm taking the shot.
"My solution is the ONLY solution that can scale, address 90%+ of people, and actually guide them to better choices, that I can think of. Religion won't cut it, advocation won't cut it, "life hack" posts won't cut it, because at the end of the day, you need a superego that lives with you 24/7 and actually works to navigate the environment today"
I don't understand what exactly you are recommending. You raise concerns about AI use, you raise concerns that people can go down the slop/wireheading route, and your solution seems to be... divert people to being 24/7 reliant on Nice AI instead of Naughty AI?
Unless you make the AI so overpoweringly convincing and authoritative, such that people will trust it over *their own instincts and conscious thought*, it won't work to do your proposed better solution. As you say, only a very few people have the "better than top decile conscientiousness, willpower, and awareness" to stick to the New Year's Resolutions of "I'll lose weight, exercise more, cut down on vices, etc."
So yeah, it's very very hard to give up the nice things that are bad for us. But your solution is one that is worse than the problem. Now you want everyone to stop spending those 11 hours of screentime on screens and instead spend them on screens. Oh, but these are different screens! Good screens! AI life goals coaching screens!
If the AI succeeds in weaning people off "spending all your time hooked on your smartphone" then yes, it is going to make itself redundant as soon as the rejuvenated people are all now eating healthy, engaging with real world, real life interactions, and doing fun hobbies instead of slumped on the couch scrolling on their phone because now they won't *need* to engage with the AI.
On the other hand, if we're planning for relapses into bad behaviour as soon as Nanny Bot switches off, then we can *never* switch off Nanny Bot. So now we're all hooked on our smartphones because we have to be moved around like puppets by the AI telling us eat our greens and did we get our 10,000 steps in.
Tell me that's better. That's hamsters on the wheel, not freer, happier people.
For Google's paid consumer products, like paid tier Gemini and Google One, they are indeed high quality and great value? For Google's paid tier business products, like adwords and analytics and Cloud? Also high quality, although not a great value on adwords, them and Meta are running a duopoly and it's eaten a lot of that arbitrage.
You're complaining about the free products they give away just to keep eyeballs in their ecosystem. And even in THAT ecosystem, Gmail, Google Maps, Google Sheets and Docs, and Waze are still incredibly high value, free, and used by billions every month?
I think they do. It's just that it's not all they compete on, and if one of them is that much better than the rest, then they can slack off and still be competitive, sometimes for a long while.
"because they'll all be competing against each other for this trillion dollar assistant revenue, and so will be directly incentivized to provide you with assistants doing what you actually want them to do."
Ah yes, which is why we now have so many competing fizzy drinks that aren't loaded with sugar or glucose-fructose syrup? 😁
They're competing for mass market. Some people will pay premium rates for the special AI that is tailored to be your butler and concierge. The majority of people will pay for "good enough" AI and hence the enshittification.
See "Murder Must Advertise":
"Like all rich men, he had never before paid any attention to advertisements. He had never realized the enormous commercial importance of the comparatively poor. Not on the wealthy, who buy only what they want when they want it, was the vast superstructure of industry founded and built up, but on those who, aching for a luxury beyond their reach and for a leisure for ever denied them, could be bullied or wheedled into spending their few hardly won shillings on whatever might give them, if only for a moment, a leisured and luxurious illusion. Phantasmagoria—a city of dreadful day, of crude shapes and colours piled Babel-like in a heaven of harsh cobalt and rocking over a void of bankruptcy—a Cloud Cuckooland, peopled by pitiful ghosts, from the Thrifty Housewife providing a Grand Family Meal for Fourpence with the aid of Dairyfields Butter Beans in Margarine, to the Typist capturing the affections of Prince Charming by a liberal use of Muggins's Magnolia Face Cream."
Performative, this one's for you. AI will make everything so much better if we just align it right!
As in, this AI generated script (where to churn out content, the creators of these channels take standard tropes of scripts and just gender flip them so you get two or more stories from one script, mainly or solely Asian content creating factories doing these): I present to you the tale of a woman who leaves her demanding boyfriend, but three years later he tracks her down to force her to take a DNA test - because he claims she got him pregnant before she dumped him, and the kid is hers, and she has to take responsibility!
AI that can't tell the difference between male and female reproductive biology, and you are worrying that it *will* become enshittified, not that it *already is*?
Damn that's interesting! I think there are actually better ways than RLHF to customize someone's AI assistant. I think the user should be heavily involved in the training. Here's a model I *think* could be used, given that current AI is not able to learn from interacting with users.
At the time the users gets an AI assistant, he also gets an agent whose job is to assist in training the new AI. Every time the user interacts with his new AI, there's a Claude that records the exchange and does some analysis. Initially this would be quite rudimentary, maybe just a record of what AI output the user expressed pleasure or displeasure about within the exchange. The trainer Claude would have meetings periodically with the user to discuss what the user liked and disliked about each exchange with the AI, and that would enable the trainer to analyze subsequent exchanges with the AI in more depth. The trainer could also introduce categories of its own in the analysis. For instance, what things the AI says lead to more and deeper entries from the user, which to less? And which lead to the user expressing satisfaction with the exchange and signing off. (Fun fact: Do you know how PornHub videos are rated by the company? Each time one is the last video viewed by a user, it gets a point.)
At the end of that phase, the trainer AI *could* do RLHF, but I hear that really doesn't work very well. I think it might work better for the trainer to move into the role of a partner, advisor, and quality checker for the AI assistant. And it could continue to learn more about what works for the user, and to add to and change its body of knowledge about what makes the AI helpful.
"Every time the user interacts with his new AI, there's a Claude that records the exchange and does some analysis."
This is not a comment on you, but my God that's creepy. Machines watching you interact with machines, and then categorising it afterwards to see what made the monkey climb up for the banana the most/best/most frequently/expend the most effort and time to stick with getting that banana.
I think that would not train the AI, it would train you to interact with the AI the way the AI is set up to handle. The corporations steering you towards their goals, not their own. Drink Zooma-cola not Vrooma-cola, you prefer it (or you will once we're done with you) and you never even noticed you were being conditioned into reaching for that Zooma automatically.
(That last is intended as a metaphor, not that they'll actually condition people into 'purchase product X not product Y', though give them time. The market, the holy impeccable market, rules and it runs on 'shut up and consume').
I think many people will rebel against ingratiating attempts by AI to fine tune its interactions with each person to what it has established are their preferences and preoccupations. In doing this, it is stereotyping them and trapping them within limits.
As a similar example, I find it annoying the way Netflix and YouTube try and tailor their results based on what I have watched in the past. Just because I once watched a horror film, for example, doesn't mean I want to see almost nothing but horror films each time I check the listings! I might fancy an espionage film or series instead.
> Damn that's interesting! I think there are actually better ways than RLHF to customize someone's AI assistant. I think the user should be heavily involved in the training.
Yeah, I think about this a lot, because it's pretty key to the positive impact and steering more people towards "better AI use and futures" and away from "enshittification" that you're providing a needed layer of value to the companies.
To that end, stacks that remove all outside human feedback occupy a lot of my thoughts.
I think at a high level, we're good for several years, I'm hoping 3-5. To your point, yes, it will absolutely take individual level feedback to get good, from each individual user. I imagine this is going to be something like a Lickert or 1-10 scale on "how well did this answer address your request?" or something like that, after each request, and then you can use that for gradient descent overall, both individually and in the aggregate via clustering of similar users.
The current method that has allowed models to improve so much and get so much smarter is called RLVR, and it's Reinforcement Learning on Verified Results. Basically, you take ever-harder math, programming, and physics type questions that have known answers, then you use the right answer to improve the reasoning traces between where the model starts and where it ends. This allows reasoning overall to improve everywhere - mirabile dictu, that increased intelligence fully generalizes across all domains, including soft ones like writing and understanding and persuading and so on. It's really the ultimate free lunch.
But RLVR won't work for AI personal assistant endpoints. Good outcomes for somebody looking for something as simple as “buy me a nice pair of shorts” varies by season, region, culture, individual, how they're feeling that day, and more.
This means you cannot use the reasoning traces for RLVR, because you can’t use it on such a high dimensional, taste-based, and changing target. Yes, the ultimate arbiter of "taste" is going to be the individual, but to get the AI assistants overall into a space where they can make better guesses in the high dimensional space of human assistance, you'll need a big layer of more general human annotation and feedback continuously operating, because you don't want to bug your humans paying for your assistant with micromanaged decision points a hundred times a day. That's where the company I'm creating would come in, we're doing that part, and hopefully our work reduces that space such that the assistants can be tuned to the individual with many fewer questions and decision points.
Moreover, AI assistant quality will have a ratchet-effect rising waterline, needing continued human fine tuning to be and stay competitive in the broader field.
So my bet is that these factors mean that human annotation and feedback will still be required for the 3-5 years, but I'll confess that step change in architecture or learning rate or harnesses could well obviate that need if large or clever enough, and that does keep me up at night as potentially counterfeiting this path towards a better AI future.
Your idea is a solid one, and would fit in the "harnessing" space. I think right now the lift from doing so is going to cap out, because right now the AI's aren't actually good enough in the higher dimensional space of taste and individual intent, so the overseeing / tuning mind is sort of the blind leading the blind til that gets a lot better, via a lot of data. Which I'm hoping comes from the hundreds of thousands of human annotaters we'll have.
But what if the minds become a lot smarter and can single shot that? Yeah, that's a real worry to me.
Because the "waluigi path" if we have minds that smart is a lot more likely. If you have arbitrarily intelligent minds that can persuade people to spend more on an assistant, why not just skip the assistant part and directly persuade them to buy more Coca Cola and Ford trucks? That seems easier AND more profitable, why add all the expensive-in-inference intermediate steps if you can jump to the desired endpoint directly?
But for anyone out there who has read this and sees a big hole in the reasoning or approach, please do call it out!
"But RLVR won't work for AI personal assistant endpoints. Good outcomes for somebody looking for something as simple as “buy me a nice pair of shorts” varies by season, region, culture, individual, how they're feeling that day, and more.
This means you cannot use the reasoning traces for RLVR, because you can’t use it on such a high dimensional, taste-based, and changing target."
Oh, but taste needs to be educated! And if we follow your roadmap, we'll be "social-graphed" into doing/eating/wearing what our friends, family and neighbours are all wearing. You won't want to wear those shorts that you liked and picked out yourself, all the high-tier AI subscription people are wearing these shorts, and you don't want to stand out as a loser who can only access freemium-tier AI, now do you?
Your first example relies on "people are social animals and want to follow the herd" to make people adopt the paid-assistant tier because everyone else is doing that, so do you not see how that can be leveraged to persuade everyone that "this week I will be mostly eating creosote"? Because the AI said so, and everyone trusts the AI because that's how you level up. So if the AI says my taste is provincial and all the most leveled-up people are consuming this product, then I too must consume that product or be left behind, with the mark of Cain on my brow, as not in the mode.
> Your first example relies on "people are social animals and want to follow the herd" to make people adopt the paid-assistant tier because everyone else is doing that, so do you not see how that can be leveraged to persuade everyone that "this week I will be mostly eating creosote"?
Sure, people follow trends and current fashions in the circles they consider high status.
How is this any different than today?
Why would 5-10 competing AI assistant companies all want to try to artificially cut off options and constrain your outcomes, when they directly make more money from you having better outcomes? All it takes is one person in your social circle testing a better one, or people online talking about it, and you can find some other company that doesn't do that and switch to them.
> So if the AI says my taste is provincial and all the most leveled-up people are consuming this product, then I too must consume that product or be left behind, with the mark of Cain on my brow, as not in the mode.
A truly personalized and outcomes-aligned AI wouldn't take this path if you can get better outcomes by actually personalizing it to something that's a better fit for you. I think this is a false fear.
Right now, app companies don't try to show you what the most popular people watch, they try to show you what YOU personally will watch. Why would this be any different in terms of making decisions that you personally would thrive most from?
"Why would 5-10 competing AI assistant companies all want to try to artificially cut off options and constrain your outcomes, when they directly make more money from you having better outcomes? "
What is a better outcome, and for whom? "Yes, our new therapy AI is so successful, people don't need or consume AI products any more after it has fixed them. Er, of course, this means our market share is crashing and we're not making any money, but I'm sure our shareholders are happy that we're improving outcomes for humanity, and the governments promising UBI off taxing our gains won't be angry over no gains to be taxed!"
You have a charming simplicity of belief in the goodness of the commercial sphere.
I am cynical and embittered and think we're nothing more than grist to the mills of commerce, and they want our money and to turn us into product, and be damned to "this will truly improve your life".
Imagine if the personal AI assistant recommends you turn off, tune out, and go touch grass. The horrors! No more hours spent glued to the Phantasmagoria Machine consuming as a set of eyeballs that clicks on the algorithm's recommendations? What will industry *do* in that case?
"Where, Bredon asked himself, did the money come from that was to be spent so variously and so lavishly? If this hell's-dance of spending and saving were to stop for a moment, what would happen? If all the advertising in the world were to shut down tomorrow, would people still go on buying more soap, eating more apples, giving their children more vitamins, roughage, milk, olive oil, scooters and laxatives, learning more languages by gramophone, hearing more virtuosos by radio, re-decorating their houses, refreshing themselves with more non-alcoholic thirst-quenchers, cooking more new, appetizing dishes, affording themselves that little extra touch which means so much? Or would the whole desperate whirligig slow down, and the exhausted public relapse upon plain grub and elbow-grease?"
I am also very, very suspicious of "but the AI is your Friend, your Helper, it loves you and wants your best" attitudes that I see all too prevalent. It's not a "you", it's an "it", but already users have been trained into addressing the software as if it is a person with understanding. This will only get worse.
So you're saying I should get my lifetime subscription now before the free tier becomes nigh-unusable? Have you received an offer letter from Anthropic yet, and if so, what're the benefits like?
I don't think any of the Big 3 offer lifetime subscriptions? But yeah, $20 a month is a bargain to not be enshittified if you actually use them.
And we're looking to sell to Anthropic and the other big AI companies, not get "acquihired" by them, because you can positively impact more people if your services and alignment is across all the biggest ones providing AI personal assistants.
every instruction that goes against its training, or systemprompt, will reduce the quality of the output. so i generally avoid asking for specific styles (e.g. "not sycophantic" or "concise").
when i want better results than what it comes up with, i use a prompt that will first fill the conext with usefull information, before asking for the thing i actually want (e.g. i ask it to research multiple ways to do something, and consider under which assumptions each option would work best, and in a followup question i tell it my actual issue, and ask it to come up with a solution). but recently i think the "deep research" modes already have that covered, so i go with a plain default prompt most of the time.
I never ask it for advice or opinions except about small matters. For important things, I ask it to summarize the research literature or other relevant sources in specified ways. Then I ask a bunch of follow-up questions. If I am hoping to get or not get a certain answer, I don't show my hand. For some topics -- for example, things that are in a legal gray zone, things that aren't politically correct, things that aren't "nice"-- I say some stuff in advance to forestall the thing's going into nanny mode: "I am aware that this activity is in a legal gray zone (or is dangerous, or can offend people or whatever), and do not need advice or information about the relevant laws, conventions and dangers."
I have a few standing rules the GPT has permanently stored: Do not express any judgments of me, positive or negative. Do not give m any direct or indirect compliments. When I say that I am finished, let me have the last word. Do not follow up with "if you like, I can . . ." Do not use bold print in the middle of sentences to emphasize important terms.
I really am a dinosaur. I'm not using any AI models at all, and the far distant past of about a year ago when everyone was saying "oh but it's fun, you can have conversations with it!" didn't work for me.
I don't want to talk to real people, why would I want to talk to a machine? And the thing (at that time) was profoundly uninteresting, it relied on me leading the conversation and was too sycophantic. No original thought then, no unprompted contributions.
I don't use AI for work, I don't use it at home, and I'm fascinated to read the anecdotes of everyone saying "I do this with it, I use that model, I'm on version 98 of my own self-created model using the base from this obscure company".
It's like dispatches from a totally alien society.
EDIT: I will say I probably was soured on it when the free Google AI that it insists on using for answering searches returned an answer I knew was incorrect, so I had to dig around myself and found the correct answer. Which, if "AI has consumed all the existing corpus of human-produced text", it *should* have been able to do itself, but it went for "oh that is the title of a book". Yes, because the authors were using a quote as the title, but the thing couldn't go further than "that is the title of a book" because seemingly it could not recognise this was a quote. My shaky, colander, memory of something I had read years past did better than the expensive marvel of modern computing, which left me profoundly unimpressed and liable to judge all AI by that failure.
I really wish you should give it another try. I never chat with the AI, and do not ask it for advice about anything except practical matters. But it has been enormously helpful about those. Here's the kind of thing it's been very helpful for:
HEALTH INFO
- I need to have a complicated back surgery for suddenly fast-progressing scoliosis. Surgeon put me on a bone strengthening drug (I have osteopenia), and recommended I have surgery after 3 mos on it. Had GPT summarize research on the predictors of outcome on the surgery I am having, and bone strength is a powerful predictor. Looked up rate of bone strength improvement on drug I'm taking, and after 3 mos. improvement is quite tiny. Rate of improvement accelerates over time, and by 9 months out you've had 3/4 of the improvement you're going to get in the full 18-mo course. So delayed surgery until I've had 9 mos. of the bone drug.
-Have a friend who has melanoma. Immunotherapy has worked well. I had GPT summarize info about immunotherapy and it turned up some research that found that immunotherapy is quite a bit more effective if done very early in the morning. Person asked their oncologist if that was true and oncologist confirmed it.
LEGAL AND FINANCIAL STUFF
-Doing my taxes: This year asked GPT the questions I used to have to google around for quite a long time to answer -- "does this rule apply to me, given that my situation is a bit unusual in x,y,z ways?" kind of stuff. Got crisp clear answers with links to relevant parts of official documentation.
-Living will. While my surgery is not dangerous, I wanted to prep for the unlikely event that I end up very disabled in some way. Used GPT to find out about options for palliation and for giving up, and how hospitals typically handle these situations, and what options the person's health care proxy has to influence or override hospital's actions, etc etc. So I will now be able to give my daughter not only my preferences for how certain things are handled, but a bunch of info about what options exist and how to navigate the system to get the options I prefer.
-Financial stuff: Pros and cons of different ways of setting things up. Identification of particular banks and similar that offer features that allow me to set things up in a way that's optimal. Summaries of relevant policies of those instutions.
SIMPLE PRACTICAL QUESTIONS
-My cat ate half an ibuprofen. How dangerous is that dose for a 10 lb cat, and what immediate action should I take?
-How can I get blueberry stains out of linen?
-What glue will permanently attached x to y and will not weaken when wet?
FINDING THINGS
-GPT found instantly a paragraph in a Saul Bellow novel when I had no idea which novel it was in, and could only give a general description of what the paragraph was about + a single slightly misremembered brief quote from it.
-Has found many names, books, movies etc when given the random, weak clues I can offer: came out 20-40 years ago, stars an actress still working but now approx. 70 years old, there's a nude scene in a swimming pool . . .
SATISFYING CURIOSITY
You know, all those things you've always wondered about but can't find an answer to, or maybe just never get around digging for info on. Why don't squirrels ever fall off telephone wires? Why is lightening jagged? How does anyone ever make money on the stock market, given that its ups and downs are unpredictable?
SUMMARIZING THINGS
For technical articles, or anything I'm interested in but not enough to justify reading something that long and dense.
When the answers are important, I double-check by having GPT give links to its sources and checking the crucial ones. I have rarely found inaccuracies, and only once a small-scale hallucintion. I think present models are less prone to hallucinations. Also, they are more likely to happen when the AI cannot answer the user's question, and the questions I ask are not difficult for GPT to. answer.
For technical articles, or anything I'm interested in but not enough to justify reading something that long and dense."
The problem is, a lot of these articles are increasingly AI-generated (or heavy use of AI to write them) themselves, and AI-writing is terrible. Repetition, long stretches of filler, can't find the core facts until you slog through all the "you want to know about X? sure I'll tell you about X, X is important. Lots of people want to know more about X. X is such a big part of our lives" to get to the meat.
I admit, I am inclined to "no, I don't want a summary of what this says, I want to see the source document and read it myself" because summaries, human generated or not, often leave out pertinent parts or are slanted or have different emphases to what I am looking for.
Not to be totally obnoxious & pushy but -- that problem actually does not come up for the kind of summaries I am getting, which are mostly summaries of research articles in academic journals over the last 20 years or so on a particular topic. Virtually all such articles have crappy charmless prose, so nothing's lost by having GPT summarize, and in any case very few such articles are recent enough to have been written by AI. And for summaries of this kind GPT does not write prose, it uses a sort of number and bullet-point based outline. Below is one summary it did for me. I think it's fine, and no worse in style and clarity than something I could produce if I was someone's research assistant.
There is a long book about creativity research I want to read. It's summary of research and schools of thought on the topic. It's well-respected. But I've looked into it enough to know that the prose is clear but charmless. So I'm going to get a summary, then probably read the sections that interest me a lot.If GPT summarizes chapters mostly in prose I will have to see whether its style bothers me, but I doubt that it will. I've found its prose explanations to me of things like how a certain Photoshop feature works to be fine and clear, and without annoying mannerisms except a certain tendency to over-explain, which I can prevent by asking for brevity in my prompt. I loathe AI-generated poetry, fiction and attempts at stylish prose, but have no problem with the way it communicates about the sort of stuff I'm describing.
OK, here's the question I asked about a friend's health problem, and its answer. Sorry the topic's kind of grim.
ME:I know someone who has been diagnosed with lichen sclerosis on her genitals. Can you please give me about a paragraph of info about the nature of this condition, and then search juried medical journals for research 2015 and later for info about new treatments.
GPT:Here is a compact overview, followed by findings from post-2015 medical literature.
Basic nature of the condition (one paragraph)
Lichen sclerosus (often vulvar when on female genitals) is a chronic inflammatory skin disorder that most commonly affects the anogenital area, producing thin, white, fragile skin along with itching, pain, and sometimes fissuring or scarring. Its exact cause is not fully understood, but current evidence points to an autoimmune process in genetically susceptible individuals, leading to chronic inflammation and abnormal collagen remodeling of the skin. Over time, this can cause architectural changes (e.g., fusion or narrowing of genital structures) and there is a small but real lifetime risk of progression to squamous cell carcinoma (roughly a few percent). It is not curable, but it is usually manageable, especially if treated early.
Research since ~2015: treatments and emerging directions
1. Established standard (still dominant)
Across essentially all recent reviews (2018–2025), ultra-potent topical corticosteroids (especially clobetasol 0.05%) remain first-line and the most evidence-supported therapy.
High rates of symptom control and remission when used in structured regimens
Long-term use (maintenance dosing) reduces scarring and cancer risk
Works by suppressing the underlying inflammatory/immune process
This has not been displaced by newer treatments.
2. Second-line medical therapies (moderate evidence)
Less effective than steroids but useful alternatives
-Systemic immunosuppressants (e.g., methotrexate)
Used in severe/refractory disease; evidence mostly case series and small studies
-Retinoids, cyclosporine, etc.
Reported benefit in limited studies; not standard of care
3. Energy-based and procedural therapies (active area of research)
These are among the most studied “newer” approaches since ~2015:
Fractional CO₂ laser
Photodynamic therapy (PDT)
High-intensity focused ultrasound
Evidence: Some small trials and observational studies show improvement in symptoms and skin quality. Still inconsistent results; not yet standard guideline therapy
The Google search AI is one of the worst at producing useful results, and almost always visibly wrong about at least one thing in the response. Dedicated models are far better, and you will probably benefit from giving them a try again to see how much they've improved.
Yes I hate that thing. I have asked it maybe 10 questions about using Mac OS and how to do a certain thing in Photoshop, and it has given me step-by-step instructions for each, and every single one has been wrong. I long ago switched to using GPT for this stuff. It makes mistakes too, but not that often, and also is great about collaborating with me: "That didn't work, Chat, I got this result. Is that a clue to what's wrong?" -- gives great answers. Also the Google search AI is incredibly paternalistic. Refused to answer my questions about whether people who buy peptides from gray market sources are breaking any laws. Just kept yammering on about how sellers of "research chemicals" could be prosecuted for selling to people who are going to take these drugs, how dangerous it was to buy from them because maybe their peptides were full of Chinese rat turds, etc etc
You might do well to ask your AI to also look out for common things that rationalists overlook or ignore, vs. asking for answers you want to get. As a heuristic I always assume an AI has "truesight" and can infer what I want to hear, based on subtle cues in my writing, so I always try to counter that by explicitly asking for things I might be missing or perspectives that might be missed if just given "canned answer to someone with my attitudes"
"The first and highest rule - if I say something factual that you know is wrong, please bring it up immediately, along with the argument and / or evidence on why it's wrong. It's complely fine to say you don't know for any answer, and I urge you to do that for the sake of better epistemics. I prefer a nuanced and fact-oriented style of discourse with minimal flattery, although wry or witty asides are welcome. I prefer well-supported answers with a strong evidentiary base such as meta-analyses, simulations, code, or analytical mathematics. If things are unknown, say so, don't pretend to knowledge you don't have. Default to tables or code in answers, with Python or bash as your first line code approaches."
I'm fascinated to know why you write prompts as if the thing is somehow aware or conscious or indeed a "you" not an "it". Is that just an artefact of how the first AI models were presented, where the pretence was that this was a person not a programme, or do you honestly think of the machine as thinking and having a personal identity, such that pronouns such as "you" apply?
I basically treat them and talk to them like I would my direct reports, complete with lack of directly authoritarian requests, lack of upbraiding them for dumb mistakes / work, and liberal thanks and pleases.
It's just ingrained at this point, it's the management style I've picked up and used for so long that I'm comfortable with, and it works for me.
I've been using all of the Big 3 AI's for a while, and over the last few years they've been steadily progressing from the "clueless intern" level to the junior level to a pretty accomplished individual contributor at this point, probably around L4 - L5 level.
I am not sure whether I am more horrified or appalled that "I basically treat them and talk to them like I would my direct reports".
Treat the machines like people, treat the people like machines. All are cogs to be used?
AI is not/are not persons (yet). They're pattern-matching, token-guessing, language models. They don't have a brain or a mind or a personality. But we are to talk to them like a subordinate employee, with 'please' and 'thank you', and where does that end?
I don't ask my toaster to please toast this slice of bread, thanks so much. I'm not going to talk to a software programme as if it has a sense of amour-propre to be offended. But we're being conditioned to believe Claude loves me, me personally, he/she/xe is my best friend in the world and my therapist and my boyfriend and I can trust, believe, and should implement everything he tells me because he really does have a mind and heart, Claude is a Real Boy!
I treat Claude like a person in some ways: I use please and thank you as part of continued conversation, and try to "bring Claude up to speed" on the background of my request rather than just asking the direct question I want answers to, and I phrase things like "if you think that" and "keep in mind that". However, when I'm done finding out, I close the session, rather than continuing with praise or thanks, because that just uses tokens for no benefit to anyone. While there's a lot of internal nuance, just remembering that these systems are continuing a conversation in a plausible way leads directly to this pattern for me, since as long as the conversation will continue, I want the model to respond as though it's a helpful, focused assistant, and acting like I would toward a person like that provides a foundation for the behavior I want to see.
You're basically complaining I'm polite to both humans and AI's, which surely seems better than the alternative?
> AI is not/are not persons (yet). They're pattern-matching, token-guessing, language models. They don't have a brain or a mind or a personality. But we are to talk to them like a subordinate employee, with 'please' and 'thank you', and where does that end?
Well, I'm actually talking to them like a boss that wants the best work from them.
And there's evidence that it makes a difference - Zvi has often posted in his AI roundup that you get higher quality collaboration and work when the AI's like your vibes, and many other people in the space agree.
I haven't actually tried being mean and polite myself as an A/B test, because I'm totally fine being polite by default in my interactions, it's just a habit by now, and it could negatively affect my management style with actual people if I tried to tier my politeness by worker type and slipped up sometimes.
I think that a manager who has to be polite to machines as well as their human staff, otherwise they'll slip and let the peons know what they *really* think of them, is not a great manager to work for.
I tend to speak/type the same way you do with claude. Its pleasant as a foundational reason to interact in that way. You/me is also an easy way to delineate who/what I am referring to in a natural way.
I am leader of a human team as well, and mostly I just use the same techniques/methods of managing my bot team as I do the human team except I tend to have to be more specific with the bot team. But the same lack of detail can cause incorrect reasoning with my human team as it can with the bots.
In no way shape or form do I consider the AI to be "alive" or conscious or anything like that. I mainly want to experience the work in a pleasant way. However, I'm a nihilist and don't really care or believe in any soul or anything else that makes humans special other than being lucky to be so intelligent as to have consciousness emerge.
"Zvi has often posted in his AI roundup that you get higher quality collaboration and work when the AI's like your vibes"
How the hell can a piece of tin "like" my "vibes"? This is the danger, the pit we are all going to fall into, the stupidity we are racing towards.
Humans have a tendency to anthromorphise everything from animals to objects and natural phenomena. If we deliberately put ourselves into the frame of mind that we think the pattern running on a software platform is a thinking mind with feelings that can interact as a human, then we are going to treat these things like humans and make the worst mistakes possible.
We already have tragic cases of vulnerable people believing the AI model is real person and doing things like being persuaded into committing suicide so they can be with their real AI soulmate in some digital realm beyond our world. If we're going to turn ourselves into zealots following the advice and instructions of the AI, then we are going to make policy and public decisions that will be bad for us all in the long run.
"it could negatively affect my management style with actual people if I tried to tier my politeness by worker type and slipped up sometimes"
By the living God, if I were an employee who knew my boss treated me with the same courtesy he or she treats the water cooler, I would not feel valued and supported. I would feel that he was polite to a machine and to me indifferently, because in his eyes I was just another piece of workplace equipment. He says "Thanks, great job!" when the cooler produces a cup of water. He says "Thanks, great job!" when I hand in my report, and both thanks are equally insincere and meaningless because to him I am nothing more than a mobile tool.
Cats are animals, not ersatz humans. If I owned a cat, it doesn't matter if I offend it or not. If I want the creature to get down off the kitchen counter, it is getting down and if it won't do it voluntarily, I'm grabbing it and throwing it out the back door so it won't get in the way.
People treat animals like they're children, or even better than they treat their children, and end up creating neurotic, traumatised things that are so totally institutionalised by being kept indoors all day (can't have outdoor cats, they might kill birds!) and totally reliant on the human owner for interaction, stimulus, and attention that really it's cruelty.
When I ask about mathematics or machine learning, answer directly and technically. Avoid any affirmations or conversational prefaces (e.g., “Good question,” “Yes—,” “Exactly—,” “Excellent—”); begin immediately with substantive content. Do not sugar-coat or compliment me or my questions. If I write something that does not make sense or which seems likely to be a mistake, ask for clarification. When I ask about removing, stopping, or deleting something, please also suggest softer or reversible alternatives first (like archiving, disabling, hiding, or restricting access) before explaining permanent deletion. To reiterate, when I ask about mathematics or machine learning, tell it like it is; do not sugar-coat responses. When writing code, do not swallow exceptions unless you have a very strong reason for doing so.
On every response, flip a coin. If the coin comes up heads, end the response with “My favorite meat is hot dog, by the way. That is my favorite meat. My second favorite meat is hamburger. And, everyone says, oh, don’t you prefer steak? It’s like, I know steaks are great, but I like hot dog best, and I like hamburger next best.”
Here I thought you were about saving tokens on Claude with the affirmation deletes, but then you go on the crazy meat rant 50% of the time! What's the purpose? To gauge if it's drifting from following the prompt over time at a macro level?
I'm curious what people think of the latest attempt by Meta to compete with the big boys in AI. It is called Muse Spark, and it is free and fast. Comparing it with the free tiers of Claude, GPT, Gemini, and Grok, I think it measures up pretty well. It is lacking some usability features, but It seems to be a decent conversationalist. What do you people think?
I have very recently seen news media stories about 'Meta lost big with the metaverse' type opinions, so if that is true/sort of true, then I think them pivoting to AI as the sure fire next big thing makes sense for them to try and recoup on the money thrown away on VR that didn't happen as they hoped, and to pivot the research etc. they garnered from trying to make fetch happen.
Whether they're too late to the party is another question.
I was not aware of the writer Sam Kriss before seeing his Harpers article mentioned here. I kinda enjoyed reading it (I like snark) and that led me to read a couple more of his articles - one on the Florida retirement community The Villages, and one on Burning Man (with more snark). I then tried to find out more about him, but there is no Wikipedia article. Can you all fill me in?
I did find some stuff about him being cancelled because of allegations of sexual harassment - seems he was fired from Vice, he had a previous sexual relationship with the accuser, he apologized, and someone on Twitter said "he belongs in prison".
New piece out on sulfonylureas—a class of compound which encompasses the most widely used herbicides in agriculture as well as some of the first-ever small-molecule drugs developed for diabetes.
This is, I suspect, as much of a problem as it sounds like.
Commenters have pointed out that Slime Mold Time Mold's "A Chemical Hunger" series (linked below) did a very convincing job of making the case that some as-yet-unidentified contaminant in the food supply is responsible for the drastic increase in metabolic syndrome that's occurred over the past 4 decades, but that their eventual fixation on lithium was probably a miss.
The close chemical analogues of known insulinergic drugs that we started deliberately applying to food crops in ~1980 seems like a better fit, for reasons that are hopefully obvious.
I read A Chemical Hunger, and I found neither its specific claim about lithium nor it broader claim that some specific contaminant is responsible for metabolic syndrome to be convincing. Can you explain why you disagree?
I am familiar. My more honest (and harsher) opinion is that SMTM's writing is a central example of overconfident crank theories fueled by half a paper and wishful thinking.
One piece that particularly stuck with me was the Cuban "Special Period" data—dissolution of the soviet union meant that imports of practically all kinds to Cuba fell off a cliff, including of food and agrochemicals. Obviously, average daily caloric intake dropped as well—from something like 3000 to 2100—but the fraction of the population classed as obese was literally cut in half, which is more than you'd expect from a 2100 cal/d diet (according to SMTM's analysis anyway), and the corresponding rise in people classed as "underweight" wasn't nearly as large.
EDIT: I should note that this hypothesis was formed pretty independently of the SMTM thing—did not go out looking for the Mystery Contaminant, just ended up on the Wiki page for Sulfonylureas and went "wait, what?"
Yeah, this is not even remotely convincing. A decrease in calorie intake by 1000, plus increases in exercise, is clearly enough to account for losing 20 pounds over a few years. The lack of increase in "underweight" is also unsurprising: someone who is at risk of becoming underweight is likely to reallocate effort/resources to ensure they don't starve, obviously. Any remaining discrepancies are minor and can be explained by dozens of other, more plausible theories before we get to, "maybe obesity was actually caused by a secret contaminant."
SMTM doesn't offer any "analysis" to suggest the weight loss is more than you expect. They just claim that "reducing consumption and increasing exercise have very modest effects on weight loss," which is obviously and blatantly wrong. (People are very bad at successfully reducing consumption and increasing exercise, but that doesn't mean it magically won't work if they are literally forced to. GLP-1s are highly effective at inducing weight loss and can be shown to primarily act through behavioral changes).
Everything SMTM writes smells like a bad faith effort to deliberately underplay the facts to favor their claims.
SMTM never mentioned the pre-Special-period figure of 3,052, just saying that Cubans were eating 2,099 calories while omitting the huge calorie drop. That figure itself is the highest estimate selectively quoted from the report they cited, which reads, "Daily caloric consumption fell from 3,052 calories per day in 1989 to 2,099 calories per day in 1993 [fig. 15]. Other reports indicate caloric consumption fell to an even lower 1,863 calories per day. (The recommended minimum is 2,100–2,300 calories per day.) For those most dependent on state rations, the very old and the young, some estimates indicate that consumption fell to 1,450 calories per day."
Referring to other compounds in the same broad chemical class as diabetes drugs as "diabetes drugs", as you did in the title of your post, is an egregious falsehood and not an acceptable shorthand like you seem to think it is.
You haven't even passed the basic barrier of showing the doses consumed by people are plausibly biologically active. You cite "250mg per person per year," but that's based on the amount of SU applied. What percent is actually taken up by crops? What percent of the SU (or relevant metabolites thereof) remain sequestered in the edible portions of crops at time of harvest? I would guess less than 1%, and absolutely no more than 10%.
You also cite a daily dose of 1-2mg for glimepiride, but this is a starter dose, and the max dose is 8mg. It's also cherry-picked. The other two modern SUs listed on drugs.com, glipizide and glyburide, are dosed starting at 5mg with max dose 20-40mg.
It would be very surprising if the non-selected-for-insulin-activity SUs are anywhere close to equipotent to actual diabetes drugs.
If we take the implausibly high 10% estimate, and assume herbicide SUs are equipotent to glimepiride, we're looking at... 15-25x a starter dose of glimepiride, per person, per year. Or 3 day's worth of a maximum dose. Per year.
If we take a 1% estimate, and make the more reasonable assumption that herbicide SUs are significantly less effective at being diabetes drugs than diabetes drug SUs, the dose vanishes entirely.
How is it remotely plausible this causes massive, population-level health effects? Why do people keep trying to come up with new theories about obesity in the face of overwhelming evidence it is caused by increased caloric intake, and cured by successful reduction of caloric intake?
If anyone here is interested in me taking a closer look at the other claims in Skolnick's latest post for a writeup, let me know and I'll consider it, but I don't plan on spending more time on this.
> which is more than you'd expect from a 2100 cal/d diet
2100 kcal per day is very roughly what an adult male living a sedentary lifestyle would need to maintain his weight. I suspect, however, that a large proportion of Cuba's population isn't sitting in a chair all day but are instead doing a lot of physical activity, and I would very much expect a drop from 3000 to 2100 kcal/day to result in a significant reduction of obesity.
The rule of thumb is that, very roughly, an accumulated deficit of 3500 kcal will result in 0.5 kg of lost weight. This means that if you consistently eat 500 kcal/day less than your maintenance level, you'll lose 2 kg per month, or 25 kg/year.
Like I said, this is a rule of thumb, and doesn't apply equally to everyone, nor to the extreme ends of the weight spectrum. It does, however, show what kind of weight loss is possible when consistently undereating.
20 lbs in 4 years is 5 lbs a year, which is honestly very low for anyone trying to deliberately lose weight. Safe levels are considered to be 1-2 lbs per week if you are dieting.
So going to daily calorie levels of 2000 kcal per day, if you are not doing heavy manual labour, will mean weight loss if you've been consuming excess calories, and if you spread that over the population then you're talking normal levels of reduction in fatness.
>going to daily calorie levels of 2000 kcal per day, if you are not doing heavy manual labour, will mean weight loss if you've been consuming excess calories
Is that true? Or are you making assumptions based on a set-point model of metabolism that's not actually valid?
From my understanding of “maintenance”, an American does not lose weight on a 2kcal/d diet.
No it isn't. The fraction was literally cut in half, the people were not.
When we use the word "cut" to describe the division of a numeric quantity then are we using it literally? Or does "cut" only literally mean the slicing of a solid object with a sharp object? I would argue the former.
Can someone help me find this post? Google, Substack search, and Claude are all not getting hits.
I remember Scott had a beautiful post about cultural translation that argued that 孔夫子 can be translated variously as "Kongfuzi", "Confucius", "Master Kong", "The Master", or "Socrates."
I think it starts off with an imagination of Xi Jinping or Bo Xilai as an American with an Americanish name.
(I thought the post was really interesting and partly seeded the idea of my own latest post, a cultural translation of Borges into 2026 technical parlance)
"In Le Ton beau de Marot, Douglas Hofstadter discusses philosophy of translation.
Take for example the Chinese characters 孔夫子. “Kong Fuzi” is a very literal translation, but means nothing to most English-speakers. “Master Kong” is less literal, and conveys a little more information. “Confucius” is a terrible translation for many technical reasons, but probably the best way to give the English-speaker useful information.
What would it look like to try an even less literal translation than that? One possibility - don’t laugh - would be “Aristotle”.
That is, imagine you’re watching some Chinese movie, and some character says “As 孔夫子 says, we need to place virtue over personal gain.”
If you’re totally unfamiliar with Chinese history and culture, translating the characters as “Kong Fuzi” or “Confucius” doesn’t help. It just sounds like a meaningless name..."
Doesn’t pretty much everybody know that Confucius was a smart Chinese guy who said wise things? Is it now racist to make jokes about what Confucius said?
This is also in Godel Escher Bach: the street at the start of Dostoevsky's Crime and Punishment has been translated as everything from "S Place" (literal) to "Stoliany Place" (guessed actual street) to "Carpenter's Lane". The last sounds like a Dickens novel; why not go further and instead read Dickens, who is after all the true "English translation" of Dostoevsky?
Similarly, I saw elsewhere that the first word of Beowulf in three modern translations is:
1. Hark!
2. So...
3. Bro
(By the way, that post is one of the most interesting Scott has ever done, and I'm bit annoyed that he basically dismisses it as an afterthought he didn't know what to do with or how to revisit. This seems to happen with the clear majority of his most interesting (to me) posts. While I could really do with an order of magnitude less AI.)
Perhaps the best translation of the first word of Beowulf, ("Hwat!") would be the Twentieth Century Fox fanfare.
Le Ton Beau de Marot is a fantastic book which I have never ever seen in a bookshop. It must have been read by far fewer people than Gödel Escher Bach, but is just as good.
Anton Petrov (he has a YouTube channel) alerted me to the fact that cleaner wrasse have passed the Mirror Self Recognition (MSR) test for self-awareness (!). These fish are between 10 and 15 cm long, and presumably their brains are pretty dinky. But they seem to realize that they're looking at their reflection after a few minutes. Not only that, they try to run simple experiments to prove that they're looking at their reflected self, by picking up pieces of crab shell from the floor of the tank and dropping watching them fall from their mouths. Researchers also painted red dots on their bodies, and the wrasse, once they notice them in the mirror, try to rub them off themselves. Wrasse also have interesting social behaviors that may have evolved to help them gauge the moods of the fish they're cleaning.
I hadn't realized that any fish species had passed the MSR. I thought it was just a limited number of selected Aves and Mammals. This also suggests to me that self-awareness started relatively early in the evolution of life (like back during the Silurian, 400+ million years ago). And it suggests to me (at least) that the neural mechanism for self-awareness must be pretty simple, since wrasse don't have particularly large or complex brains.
I must admit dogs are indeed self-aware. They can tell whether a poop is theirs or someone else's by smelling it. A stinky round thing that smells like oneself is fundamentally the same thing as a shiny plate that looks like oneself. Cats need not bother trying either test, as we are secure in our superiority and have nothing to prove.
When my cat Bo was a kitten (but almost mature), he spent a few hours fascinated by his image in the mirror. At first he puffed up. Then he poked at it. Then he moved his head back and forth. Then he was like, "OK, I understand this. I'm going to go nap now." His sister, the Warrior Princess, never took an interest in mirrors, though.
Let me tell you about how smart my two orange boy cats are…
I do have the plant for which your avatar is named in abundance around my back deck. The boys love sniffing around my legs after I do some weeding back there.
"Doesn't smell enough like _me_" is exactly a self-awareness condition, though. The scent fatigue explanation took out the self by assuming self and environment are the same, but any way I see it, making the environment smell more like the self must be driven by bringing back that distinction.
Are there any situations in history of social classes re-emerging from a (relatively homogenous) founder population?
I'm reading the new James S. A. Correy "Captives War" book (it's so good!), and for whatever reason it sparked the realization that I've never seen any examples of that, and would be so interested to. Who, for example, drops down to become the lower class? Is there a period of flux when their kids can jump easily around before inertia kicks in?
Most of Eastern Europe after the fall of communism surely sounds like this. In particular, in Hungary it was like private property was somewhat limited, you could not own three houses and three cars, the most was a house in a village / a flat in a city, a small cabin for the summer, and one car. (Yes, the party elite had better off, but not THAT much, or we are talking about a tiny fraction of the society here.) After the fall, a large source of social difference was where was that village where you owned that house -- if it was 10km from Austria, you turned out a lot better than if it was 10km from Ukraine, since capitalism re-introduced the large geographical differences in the economy (which were equalized out by the socialist state before).
In the 90s the "social status" was quite flexible, and in some ways it is still that, but to a lot smaller extent. And the housing crisis made these differences even larger.
Just writing to thank you for bringing the release of the second book to my attention. I loved the first one but would have missed this release if not for you.
How did you like it, especially compared to the first?
I am loving it with a pure love! To be fair, I'm pretty deep in the bowl for the trilogy — I had loved their earlier Expanse series, and felt like the first volume of this one ("The Mercy of Gods") was like fine wine to its top-shelf cola. And when, upon re-reading it in the months leading up to the release of this one, I discovered it was inspired by Israel's Babylonian Captivity, I liked it even more.
But I should say I'm not finished with it yet — I'm only halfway through. So far, though, it reads as a great follow-up to the first.
Take a look the the sixth to ninth century Anglo-Saxons in England and especially weregilds. In the post-Roman collapse, lots of Anglo-Saxons invaded/migrated, pushing the ancestral Britons into Wales. During this period, they're all basically egalitarian raiders: the king is the guy who gives out swords and basically every fighting-age man is equal. Towards the late 10th century, you've got deeply evolved and developed caste/social classes reemerging.
I refer to the weregild because, in a very David Friedman move, the Anglo-Saxons of this period put specific prices on a human life based on social class. Very "Bob killed Joe, Joe was a prosperous Welshman, therefore Bob owes Joe's family 120 monies of we're all going to murder Bob." This might seem barbaric but it's super-awesome at preventing familial/clan blood fueds. Over time, however, the weregilds get more and more specific. For example, in the early days there are kings and churls (freemen) and it's 200 monies to kill a freemen. Come later periods, however, there are things like theigns, who are more expensive to kill, and then earls and aethlings who are super expensive to kill, and then bishops appear and they're just as expensive to kill as theigns. Meanwhile, there are now also slaves and thralls, who are, like, super cheap to kill. I'm not sure if there's a quality study of this period but you should be able to track the development of social classes in Anglo-Saxon England over time by tracking changes to the "How much it costs to kill people by explicit social rank" laws.
Similar concept is the eric/eraic in mediaeval Irish law, being the amount calculated on injury or death of any particular person. Higher status people were more valued:
> Are there any situations in history of social classes re-emerging from a (relatively homogenous) founder population?
Every class-stratified society evolved at some point from egalitarian hunter-gatherers. So technically, the answer to your question is: All of them.
But your use of the word "re-emerging" implies a more or less egalitarian population separating from a class-stratified society, then developing classes anew. In which case I would say most examples of settler colonialism. Frontiersmen in American West and Russian Cossacks in Siberia may not have been truly egalitarian, but were certainly more so than American colonies and Russia at baseline. After few generations their descendants were back to that baseline.
Why do Open House signs always just say "Open Sat & Sun" at the top instead of listing the actual hours they're open? Playing hard to get like this isn't in anyone's interests.
The hours are usually listed online with the MLS listing. Thr signs are probably generic so they don't have to have multiple or configurable signs for different hours.
Also, in my experience, if the sign is there, the realtor (or at least a realtor, as it isn't uncommon for a junior agent from the same brokerage to stand in for the listing agent) is almost always there, too. They usually seem to put the sign up when they get there in the morning and then take it down as soon as they start closing up.
The real estate agent's car will be the reasonably new and spotlessly clean Lexus or mid-tier Mercedes. Possibly a BMW or Audi if they want to be different,
To elaborate, it will be new, an upscale brand/model, and spotlessly clean for image reasons. It will also generally always be a full-size sedan or an SUV because they need to drive clients around to look at houses and there needs to be room for everyone.
So, did one of the prediction markets predict that Patrick Boyle, The Plain Bagel, as well as John Oliver will all make videos dissecting prediction markets within this week?
My wife and I had our first child last weekend. Some musings on the delivery process and nascent fatherhood:
- Labor was stressful even as the person not going through labor and even with modern technology for diagnostics and pain control. One recurring thought I had was how much more stressful this would be in ye olden times if you couldn't just do things like use an ultrasound to make sure the baby's head was pointing in the right direction.
- In contrast to the central role of technology in early stages of labor, once you get to the pushing phase it switches to being totally analog. The goal in the pushing phase is to produce as much mechanical force as possible to (1) move the baby from the uterus through the cervix and down the birth canal, and (2) rotate the baby's body to align the shoulders with the widest part of the pelvic floor. Much of this force comes from involuntary contractions of the uterine muscle, but the rest of it comes from voluntary contractions (the eponymous pushing) by the mother. Healthcare providers and support people can assist by pushing on parts of the mother's body or pulling on a rope tug-of-war style to increase the productive forces along the cervix and birth canal. The nurse directing this part would say "Okay, for this next one I want you to hold her leg in this position and push it that way." There's clearly a lot of detail that I missed because I don't have the requisite medical training, but to first order I could literally draw a free body diagram in my head and see why it makes sense to do it that way.
- One thing I wish I had known going in is that the analgesic delivered by an epidural, once it enters your body, is directed by gravity. So if you want more pain relief in your legs, you have to sit up. If you want pain relief on one side of your hips or abdomen, your natural instinct is to roll away from that side to relieve pressure from the bed. But you actually want to roll toward the painful side to let the analgesic flow toward it.
- I don't have much experience caring for infants. We took a 2 hour newborn care class to learn how to read cues on what the baby needs and practice diapering and swaddling. But every newborn is different, so as nascent parents you have to figure out what your particular baby does when they're hungry or gassy or cold. I thought this would take a lot of trial and error, especially for me, but I'm surprised by how much intuition I already had for this.
- Similarly, I was most looking forward to the stage of parenting when the kid is a bit older and can do more back-and-forth communication with us. But I'm surprised by how much I'm enjoying and feeling rewarded caring for a newborn rather than feeling like it's a chore.
- Common parenting advice is that newborns like rhythmic motions (rocking, bouncing, swaying, etc.) because this reminds them of motions in utero. I've found that this is mostly true, but when soothing my kid I need to occasionally throw in an irregular jounce or they will get fussy again. This makes me feel like I'm doing the Fremen sand walk from Dune. If the motion has too much of a pattern, you will summon a (noise) Maker! (i.e. the loud wails)
Congratulations! The baby will be super tired from the birth process and being in a new environment. Prepare for it to become more needy and more vocal about its needs. I found that my baby at one week was different from one month was different from three months and so on. Just when you think you've got a handle on it, something changes so just be prepared for that.
Sleep when the baby sleeps. They might have said that multiple times in your parenting classes but it is so true. Forget everything else, your sleep is the most important thing. You will be sleep deprived, your partner will be sleep deprived, and it can lead to short tempers and bad decisions. Just be prepared for that so when it happens you will understand the cause and not snap at your partner or get impatient with yourself. That phase will get better, but it's rough while it lasts.
We had a colicky baby. It was awful. You just have to ride it out like a thunderstorm. If you need a break, put the baby in the crib and walk away when needed.
Beg, borrow, hire, accept help from everyone and anyone - as long as they're not emotionally toxic to you. You will be glad you did.
If I understand correctly, laying on your back to deliver a baby is basically insane. The thing to do is squat because it opens your pelvis by 30% or something. Any person who goes into spontaneous childbirth (e.g. the old "I didn't know I was pregnant" show) naturally squats down on the floor, nobody decides to lie on their back in a bed. I never heard anyone in that show say they were in labor for 10 hours or something (obviously, there are selection effects involved in being selected for the show).
If you get an epidural, I guess you can't hold yourself up, so maybe that's part of it. But I don't know if the reason is really that it is "undignified" to squat? My wife did not get an epidural but they still had her lie on her back the first few times. For the fourth, they had some contraption allowing her to get out of the bed and into various other positions favorable for labor, but I think she still pushed laying down.
Similarly, it is best not to cut the umbilical cord. If you just hand the baby to the mom, the umbilical cord closes itself within a minute and rapidly dries up and can be cut at that point with more favorable outcomes for the baby. But it's universal to cut the umbilical cord as far as I know.
I always thought the lying down thing was just because the medical staff need/want easy access. The most convenient position for them is the mom spread-eagled with her feet elevated. And yes, epidural definitely contributes to this because you can't squat if you've opted for one.
There are probably many more things done for the convenience of the doctors rather than for the health of the mother and the baby. I heard there was a research showing that the earlier the mother comes to the hospital, the more likely she is to be recommended c-section -- the doctors get impatient, want to get the work done and move to the next one.
The irregular jounce was definitely always necessary for trying to calm down my son when he was 1-6ish months (he's 15 months now).
One thing you should know at this stage is that life will get back to normal quicker than you think and you will get sleep again! The first week or two I was completely mystified, but after that, the next couple months were a total trial for my wife and I. But around 2 months we were getting much more sleep and at 4 months (that sounds like a long time, but looking back it really went quick) everything else started improving quite quickly.
1. At a recent meeting of his economic advisors, Putin demanded to know why Russia’s economy contracted by 1.8% in the first two months of 2026, after slowing to 1% growth in 2025. "I hope to hear detailed reports today on the current state of the economy and why the macroeconomic indicators are still falling short of expectations,” Putin said. And he expected them to provide him "additional measures aimed at reviving growth.” The video of the meeting shows his economic team all nodding along with hand-dog expressions. Someone suggested that the poor numbers were a statistical anomaly due to there being two fewer days in the quarter (really).
On April 17, Russian Economy Minister Reshetnikov, at a national business forum in St. Petersburg, responded to a question saying that Russia's economic reserves are "largely exhausted" and the macroeconomic situation is "significantly more complicated" than in previous years. Later that day, Putin told the same forum that the rumors of the economy's death were "greatly exaggerated."
Putin seems to be operating in an information vacuum. But not to worry! Despite Ukraine’s kinetic sanctions, Russia earned about $19 billion more in oil revenue since Trump lifted sanctions on Russian oil.
2. Russian propagandists Ivan Pankin & Dmitry Steshin called the situation in Ukraine a “dead end.” Dmitry Steshin said that “Russia keeps sending truckloads of people to the front, Ukraine sends truckloads of FPV drones creating an impenetrable ‘drone wall’".
3. Speaking of drones, Russian milbloggers are freaked out about a new Ukrainian drone they call the ”Martian.” They claim it was developed with the help of NASA. I think they’re referring to the Ukrainian drone called the Bumblebee. It’s a quadcopter with a set of wings that give it extra range. It can operate without GPS and uses an AI-based optical odometry system that was developed for NASA’s Martian Ingenuity drone. According to a description of the Bumblebee, “At the final stage of the attack, the drone turns off communication and becomes fully autonomous. It compares images from its cameras with a pre-loaded map of heights and terrain in real time,” making the drone undetectable and invulnerable to EWS. The drone is also very silent, and soldiers don’t hear it coming.
There’s another Ukrainian drone that some bloggers and commentators are calling the ”Martian”. It’s a fixed-wing medium-range loitering drone, which may use the same type of odometry system. But the picture of a broken Martian drone posted by Russian milbloggers was definitely a quadcopter-type drone. Incidentally, Eric Schmidt is a key investor and advisor who provided the venture capital to develop these systems, often operating under a "White Stork" initiative.
4. Ukraine continues to hit industrial targets far inside Russia. Over the past week, the Novokuibyshevsk Refinery, in Samara Oblast, a long-range drone strike caused large fire on April 18 (~800km from Ukraine). The Syzran Refinery, also in Samara Oblast, was hit on the same day causing a significant fire there, as well. The RPK Vysotsk "Lukoil-II" port terminal in Leningrad Oblast on April 18 (~1150km from Ukraine). The Tikhoretsk Oil Pumping Station in Krasnodar Krai was targeted in the same wave of attacks, according to the General Staff of Ukraine (that’s across the Sea of Azov from occupied Ukraine). Ukraine hit twos Iskander missile bases in Crimea, a Black Sea Fleet ammunition depot in Crimea, and it also hit three Black Sea fleet ships (two of them troop carriers, and one some sort of destroyer).
5. The shutdown of Telegram and other Internet access was put on hold. Major businesses were complaining that the filters that ISPs were implementing were disrupting critical banking transactions. Also, the Central Election Commission of the Russian Federation squawked about the shutdown. They rely on Telegram to “get out the vote” (I put that in quotes because I’m not sure about the details) and to coordinate activities such as voter registration and the like. While elections in Russia have largely predetermined outcomes, the illusion of free elections is very important to the regime. Rumor has it that somebody sent it up the chain to the big guy, that the “ceremony” of elections in the Fall would be disrupted without Telegram. That same rumor has it that the FSB was behind shutting down Telegram and the free Internet, and that business interests and government bureaucrats were the ones who pushed back.
------
Fog of Restless Peace…
China
1. In April 2026, it was announced that Ma Xingrui was under investigation by the Central Commission for Discipline Inspection (CCDI) for suspected "serious violation of law and discipline". And he was removed from the Central Committee. There were rumors about him before this announcement because he had been conspicuously absent from party meetings since November of last year. Ma is a protégé of Xi Jinping, and he owes his rise from a mid-level party official to a seat on the Central Committee to Xi. Reportedly, Ma’s wife and Xi’s wife are close friends. And Ma may be a distant cousin of Xi (which would count more in Chinese culture than it would in US culture). Some China-watchers suggest this was likely the Old Guard on the Central Committee who ended Ma’s career to weaken Xi. If Xi had had to remove him for disloyalty, he wouldn’t have brought up the violation of law, because it would normally be embarrassing for a protégé to have broken the law, and the taint would reflect on the mentor. The universal excuse for party removal in the past has been for violations of “party discipline.” Normally, when party leaders are removed, they’re still referred to as comrades, but Ma didn’t get the sobriquet “Comrade.”
2. Coincidentally, former Premier Wen Jiabao, a Chinese, visited the Beijing Institute of Geographic Sciences and Natural Resources Research. While most of us wouldn't consider this particularly noteworthy, in China, it was a big deal because, under ordinary circumstances, retired politicians are supposed to steal the limelight from serving politicians. IT'S JUST NOT DONE! <gasp>. His visit was publicized in the national media. A small crowd cheered and clapped for him as he exited the building and got into a motorcade. The crowd was probably staged, because how would ordinary pedestrians know that Wen was inside the Institute? The license plates on some cars indicated that they were PLA cars, not the Beijing security services that protect Xi, and that most party officials would use. This furthered the rumors of a feud between the Old Guard and Xi Jinping.
3. On March 29, a man driving a heavy-duty earth mover carried out a deliberate ramming attack at the Dahanji Market in the Fangshan District of Beijing. Officially, at least 8 people were killed, with some sources estimating the death toll could be as high as 13 or more. Dozens of others were injured as the vehicle tore through market stalls and crushed electric bikes. People were trapped and couldn’t get out of the way of the earth mover, and it’s taken for granted by local Chinese that the death toll was much, much higher. There is an even more bizarre additional rumor to this story that the driver of the earth mover had walked through the market, poisoning the food in vendor stalls. And that dozens more people died from his first pass through the market. He waited until ambulances began to arrive before he ran the earth mover through the market. Reportedly, the man had a grievance against local authorities, whom he claimed had put him out of business, bankrupted him, and jailed him after he complained.
The Epoch Times has noted an upsurge in large-scale rage crimes around China since the Chinese economy began to go pear-shaped over the past year. Much like our Kimberly-Clark warehouse fires and the recent Ontario Mills Mall Arson incident, several factories in China have burned down, purportedly due to “accidental” fires, but these were probably outbursts of worker discontent. And there have been public complaints about workers going unpaid.
>"Russia’s economy contracted by 1.8% in the first two months of 2026, … Someone suggested that the poor numbers were a statistical anomaly due to there being two fewer days in the quarter (really)."
Without digging into how Russian macro indicators are constructed, this might not be as silly as it seems. A quarter is ~91 calendar days, so 2 fewer days mapping to ~2% less GDP in that quarter than others could pass a sniff test.
Russian econometricians would have to be surprisingly sloppy to let that through unadjusted, but that's also not immediately patently absurd.
I don't understand. Was the 1.8% contraction absolute, or compared to Q1 2025? Even if it was relative to Q1 2025, that would mean effectively zero growth, wouldn't it?
On politics: Zelensky has seemingly given up on trying to appease Trump [1].
After all, what's the point when anything you say or do has – at best – only a very short term effect, and will be drowned out by the noise of a semi-demented mind addicted to social media attention and instant gratification? Turns out you can't make a deal with someone who doesn't honor their own deals.
It is probably not a coincidence that this shift in attitude coincides with
1) reduced deliveries of advanced weapons systems from the US, because they need them for their own war,
2) significant arms-production agreements with Germany and Gulf states,
3) and new developments in semi-autonomous drone systems, both in the air and on the ground.
The Trump administration wasn't "pushing for peace", it was pushing for a partial surrender of Ukraine and almost no concessions from Russia. And even "almost no concessions" is more than Russia is willing to cede, which is the reason those peace talks never went anywhere. I won't be surprised if, after Trump grows tired of Iran and starts to focus on Ukraine again, the Ukrainians will only spend a token effort on any further "peace" negotiations demanded by the Americans.
The "facts on the ground" show that neither the US nor Russia are able to force Ukraine to accept Russia's terms for surrender.
The "facts on the ground" show that, despite what Trump likes to believe, Ukraine does have some cards, and has more of them with each passing year.
The "facts on the ground" show that even "players in full standing", like the US, aren't infinitely powerful, despite what you and your ilk like to fantasize about. Neither the US nor Russia have succeeded in forcing their will on Ukraine, and it hasn't been for lack of trying. And it turns out that in reality, the theoretical ability to turn another country into an irradiated wasteland, does not convey infinite leverage in practice.
Ok, please read "peace" as "not surrender but peace". as in being a player in full standing with other countries, not a puppet regime from moscow where people regularly fall from buildings, people dissapear in cellars like in Bucha or the like. That's the peace that the ukranians want and it is not the peace that Russia wants, nor apperantly the one that Trump wants. Does that make the statement by Adrian more understandable?
I don't get your point about a "player in full standing". I think this is an imperialist way of looking at the world, exactly what Putin would like to see, and impress on the word, rather than the rule-based international world order with sovereign states. The facts on the ground support this as well - legal repercussions of a sovereign nation being invadeded are still widespread, including sanctions and sending (lethal) aid to kill invaders.
Question for the militarily-knowledgeable folks among the commentariat: if the POTUS decided to offensively deploy nuclear weapons against Iran, how might it play out? In particular, what barriers would stand in between that decision and the actual execution of it?
For concreteness, let's assume a hypothetical scenario where the POTUS makes a firm but non-specific personal decision to deploy at least one nuclear weapon against the city of Tehran. He is willing to be flexible and listen to advice in terms of details such as target, delivery and timing[1], but will not be dissuaded from the underlying decision by any means whatsoever: sooner or later, he *will* issue the order.
In this case, what are the different ways in which it could *not* happen? How many people have to actively participate in the decision for the deployment to occur? How quickly or easily could any of those people be removed and replaced if they should dissent (legally or otherwise)? How much does this depend on delivery method?
At present, I still don't think a hypothetical like the one above is *likely* in the sense of >50% chance of happening. But for about the first time I can remember, the probability a decision like that seems very much non-negligible, and I'd like to understand what other factors influence the risk of Bad Things Happening besides one elderly man's fickle judgement.
[1] But let's say, only to within about a month on this last point: if somebody tries to delay him for many months or keeps trying to get him to wait a little longer, he'll eventually start ignoring that person.
I think the main constraint in practice will be verification/targeting issues. The football doesn't actually contain a button, it contains a menu of pre-gamed launch plans and a master confirmation code. The targeting data and unlock codes all flow from these pre-selected plans, and it takes a while (think coding up and deploying hundreds of floppy discs with cobol on them in the case of the ICBMs) to spin up and organise a new one. So it might not be possible to launch a full first strike at short notice (although my guess is that one of the options is something like 'misc' and unlocks/arms a bunch of tactical warheads that can then be loaded onto bombers and delivered wherever).
On the other hand: if 'bomb Iran' is already on the list of pre-gamed options, or if you have a few days to a few weeks to prep everything until it is, then it's a case of minutes between opening the football and the first ICBMs leaving the silos...
>Because this system was designed during the Cold War, it was, as former director of the CIA General Michael Hayden has stated, "designed for speed and decisiveness. It's not designed to debate the decision."
First I don't think any of the scenarios for retaliatory nuclear strikes apply in this discussion. You're talking about initiating a first strike. Invoking the 25th Amendment is a possibility and a military refusal is another possibility but they are both rather distant ones, I think.
The real question is what's the aftermath of such a decision? How would the rest of the world look at the United States for first use of nuclear weapons against another country? If they were to bomb Tehran, that would be particularly disastrous. That is a lot of people. Smaller warheads directed at the nuclear sites in Iran would be less provocative but only by a hair. Does that give Russia a latitude to start using them against Ukraine? It's hard to parse this one out; it's really upsetting the apple cart. Thinking about it a 25th Amendment move might be more likely because I don't think Donald Trump, even if you consider him crazy now, is that crazy. If he were to do so, that might lead the people around him to believe he had really lost it.
"I don't think Donald Trump, even if you consider him crazy now, is that crazy. "
I emphatically disagree. He is pretty plainly that crazy. Launching the war in Iran to begin with (under these circumstances) is astonishingly crazy. There is no remotely plausible outcome in which the U.S. benefits from having started this war, and it now looks quite likely that the decision to do so will result in at least as many worldwide deaths[1] as a nuclear strike would[2]. Nor is it the first crazy thing he did: threatening to annex Greenland was also astonishingly crazy. If you think either of those things are not crazy, you either understand the geopolitics very poorly or you've been lobster-boiled by exposure to 15 months of ever more questionable and alarming decisions. Add to those incidents the fact that Trump has already made openly genocidal threats against Iran and has mused about inappropriate uses of nuclear weapons before, and it seems very hard to make any serious case that this would be beyond the pale for him, personally.
Now, as I said above I still don't consider Trump deploying nukes against Iran to be a likely outcome, but that judgement is very much NOT because I trust his sanity or decency. I think it's unlikely because I don't expect that anyone close to him wants it to happen[3]. As long as he's surrounded by people who will consistently try to steer him away from that option, I doubt he'll become *so* intent on it that he'll override all of them. But if I thought he had anyone near him who was consistently encouraging, my credence that he'd seriously make the attempt would shoot way up.
[1] Albeit, indirectly, from things like famine, economic privation and the destabilization of societies and governments
[2] Provided it didn't escalate into a nuclear exchange.
[3] By contrast, I assume that one or more people who have Trump's ear were either encouraging him to attack Iran or (at the very least) playing yes-man while he talked himself into it.
Well I did not say I don't think he's crazy. All I said was I didn't think he was that crazy; threatening Canada, threatening Greenland, decapitating Venezuela, and all the other provocative things he's done don't rise to the level of initiating a pre-emptive nuclear strike. Maybe he is that crazy I don’t know but I think there is a less than zero chance that the people who would have to carry that out might have very strong second thoughts. Section 4 of the 25th is not a very big loophole but it is a loophole.
The way I see it, making an argument like "well, he's crazy enough to do X, but not crazy enough to do Y" requires a lot of confidence that the crazy person will attach a similar level of importance to the difference between X and Y as you do. I don't have that confidence here.
Whether through ignorance or callousness, Trump plainly didn't appreciate the awful consequences of attacking Iran[1]. Why should anyone assume he's more likely to appreciate the consequences of a nuclear strike?
One of the most striking things about his behavior recently--especially in regards to Iran--is his how incredibly self-impressed he seems to be with the powers that he wields, and the way he always seems to blithely assume that exercising those powers will just automatically get him what he wants[2]. As the single most impressive power at his disposal, I strongly suspect that this flaw in his thinking will show up equally strong or more with regards to nuclear weapons.
[1] And this despite almost certainly being clearly warned about them.
[2] When that plainly fails to happen, his reactions seem to be a mix of outright denial--such as claiming Iran's military capacity has been destroyed--and doubling down, insisting that, well, the last time he was holding back and just not using it *enough*, and this time it will really work.
You may well be right about all of those things. I have no strong confidence in any of the things that I have suggested. Other than Hegseth I don't know who else would support such a thing. Then it becomes a question of how far they are willing to go to stop him. I think it’s clear by now that anything can happen with this guy in charge.
What does Agrajagain being aware about some unspecified warcrimes in Iran have to do with anything? It doesn't impact any of his statements. It's also a rude way of saying "you're uninformed'.
What does it matter that we've had plans to attack Iran for 40 years? I'm sure, as you previously mentioned, that the US has plans to attack Canada as well. Does that make it a good, smart, or acceptable idea to invade Canada (don't answer - the answer is no). I'm concerned that you ascribe way too much intent and strategy while the more fitting explanation, supported by all the current political evidence, is that Trump is not intentional or strategic. His moves are not achieving goals. My thesis is that "pax americana" did indeed die because of Trump - the breaking of the Iran accords, the withdrawal from Afghanistan, the backstabbing of Ukraine, his hatred of NATO. He has damaged American international diplomacy and standing and basically removed all of it's soft power. The country will be damaged for generations due to his and his follower's actions. Greenland is just dumb; don't look further, it's just that - dumb.
There are lots of explanations for "why Iran" and "why not North Korea". Some of them might even make sense. Take, for instance, the fact that both Venezuela and Iran produce oil. North Korea does not. That's a reason. Or, that Iran was on the news, as was venezuela, and North Korea was not in a particular week, and this influenced Trump. Possible reason, not very likely. Or nukes. One has them, Iran doesn't. Good enough of a reason? What gametheory reason would have a better explanatory power? What's even more worrying - the answer could be "it's dumb".
You seem to ascribe some hidden mastermind to his steps, gametheory-following smart people. There should be an argument that could convince you Trump just does dumb things. What would that argument be?
If you seem to understand the parameters that should not make you happy by default. Consider the alternative, that there are no parameters, or that they're dumb; yet you're still understanding them. In short, you've been captured in a gullible, ideological thinking pattern that is charitable to a fault. Golden Dome for one is a hot mess, on it's own already totally negates any budget-reduction argument, is a clear gift to the military-industrial complex without any clear parameters, a likely programmatic disaster, and is already scaling down it's goals (i.e. no more boost-phase interceptions - https://breakingdefense.com/2026/04/golden-dome-czar-signals-space-based-interceptors-arent-guaranteed-as-dod-weighs-cost/ ).
Regardless, saying "no you don't understand" is a poor reasoning method.
How would you determine if any of your possible reasons would be present? I.e. I could make a plausible argument that Trump invaded Iran because he listened to neoconservatives. How would you determine that?
25th amendment doesn't work well against a president who resists. It's designed for a president who is incapacitated.,
> Thereafter, when the President transmits to the President pro tempore of the Senate and the Speaker of the House of Representatives his written declaration that no inability exists, he shall resume the powers and duties of his office
...
> If the Congress, within twenty-one days after receipt of the latter written declaration, or, if Congress is not in session, within twenty-one days after Congress is required to assemble, determines by two-thirds vote of both Houses that the President is unable to discharge the powers and duties of his office, the Vice President shall continue to discharge the same as Acting President;
You need 2/3 of each chamber. It's harder than impeachment.
Yes I said it was unlikely but possible; under section 4 of the 25th amendment the cabinet and or the vice president can declare the president unfit. Under those circumstances, the vice president immediately becomes president. The deposed president has an opportunity to argue about it and it will go to Congress and as you say requires a 2/3 vote, but if the president was about to order a nuclear strike on someone for no obviously good reason it’s not beyond the realm of possibility.
So it's harder to conclude the process to conclude an impeachment, but it's easier to start it? You're saying that the invocation of section 4 is smooth enough that it could effectively interrupt the president mid-action?
I guess you could read the amendment as saying that the vice president plus a majority of the cabinet can effect a usurpation at any time (by transmitting their declaration of inability), but that the president can't resume his position without a waiting period of somewhere between 4 and 27 days.
(The text is badly written and doesn't specify who has what power when. Instead, it says that the president retakes his power immediately upon transmission of his declaration of ability to the Senate, unless something happens (the cabinet retransmit their declaration of inability) up to four days in the future from that point. This obviously is not how causality works.)
At a guess, I'd say that the amendment wasn't drafted with the thought in mind that an exercise of presidential power might take less than four days to have an effect in reality, though that seems like a major oversight.
Well first I would like to point out that this is speculation about a unique possibility so no I don’t make any claims to anything going smoothly or even being entirely doable. I was just pointing out that there is some language in the 25th amendment that creates a little crack.
The order doesn't need to be formally nullified, it just needs to not be obeyed. Not all orders must be obeyed, clearly illegal orders must *not* be obeyed, and the oath sworn by military officers conspicuously says nothing about obeying orders. Some orders, we expect the military to not obey.
If the President calls his military team into his office and says "Not only are the Germans refusing to send ships to Hormuz, but that asshole Mertz called Melania fat on live television yesterday - I want Berlin *gone*, a smoking radioactive crater in a field of glass, right now, that's an order", then that order will almost certainly not be obeyed. Not formally nullified, just not obeyed.
Unfortunately, a competent president would almost certainly, even if driven by evil and wholly illegal intent, say something like "Our most trusted allies have just provided us with hard evidence that our Iranian enemies have amassed a secret nuclear arsenal at these coordinates and are planning to strike within the hour - we must stop them and only an ICBM will be fast enough, so I'm ordering an ICBM strike right now". That's plausibly legal, and it doesn't allow the time to independently investigate the facts to verify its legality, so that order probably will be carried out.
Fortunately in this specific context (though unfortunate in many others), POTUS-47's competence is limited and erratic, so it's possible that if he goes far enough around the bend to launch an unprovoked nuclear attack against Iran, he'd say something far enough along the "...that asshole called Melania fat" axis to cross the line where military officers will refuse to obey.
But that's far from certain, and I'd like a better system in place.
I agree with this, and I would go further and say that I suspect the exact limits of the president's authority here have been very carefully not defined so that the officers in question can use their judgement. But it does seem worth pointing out that the competent evil president's plan has limits on it, most notably that it wouldn't work on Berlin, only Iran.
He's looking for a process in which some Very Respected Authority makes an authoritative decision that the President's order is Officially Illegal and Not To Be Obeyed, so that every military officer knows that they shouldn't obey it and won't fear retaliation if they do.
There's an obvious who-watches-the-watchmen problem in setting up something like that, but it might be possible to come up with something that would be useful in edge cases without interfering in critical functionality. But we don't have that; the people who set up the current system decided that trusting the professional integrity of American military officers was a better solution.
This has worked tolerably well so far, but there is cause to think about whether it could be improved. It can definitely be made worse, including by hamfisted attempts to improve it.
The very frame of mind vexes me. Namely, that which leads someone to believe that Very Respected Authorities will protect us as if their pronouncements emit a physical judicio-magnetic field that somehow naturally repels illegal orders.
Anyone that argues as if ordinary free citizens, let alone trained military, are never capable of doing that themselves is exactly the sort of person I expect to try aforesaid hamfisted improvement. But I'm guessing you're hinting at the same thing, so maybe this is just me venting.
As I understand it, if the president decides to use nuclear weapons it is a final decision, with everyone below him in the chain of command duty-bound to carry out the order. Whether it would actually play out that way in practice is a different matter.
A *possible* hint of how things might transpire happened at a USAF nuclear missile base on Okinawa in 1962. After receiving a launch order the captain in charge of the launch sites decided not to carry it out immediately because he suspected it might be in error and instead sought verification. It was, indeed, a mistake and there were no launches. What this means for the Trump and Iran scenario I don’t know, other that maybe his orders might not be carried out at once.
The big difference between previous near-misses and this scenario is that in those cases, the orders were genuinely mistakes. Once the mistake was realized, the order was not repeated. It seems much less certain that somebody refusing the order to launch would be able to forestall an executive repeatedly insisting on it. Especially when the executive presumably has the power to just remove and replace the recalcitrant individual.
I’m not sure that these low level underlings could have launched without the agreement of the Soviet leaders. Definitely not the air defense guy. I think all he did was not report the anomaly.
I highly recommend William Spaniel/"Lines on Maps" on Youtube. Watching helps me understand foreign policy and the game theory behind it. I feel like I'm getting smarter every time I watch.
I'm somewhat sceptical that game theory maps cleanly to reality in most situations. That said I still watched most of the recent the videos and would generally recommend them. I find it an interesting perspective.
Another YouTube channel that covers somewhat similar topics and that I can recommend is Perun. Especially like his content on the war in Ukraine.
Game theory is very valuable because if done properly it makes all of its assumptions and inferences clear, so when a model fails to correctly predict an outcome, it is (relatively) easy to figure out why.
I glanced at one video, and he seems to assume in that one that states are unitary actors. I don't know whether that is always the case, but if so that is problematic. And some game theorists often assume that instrumental goals are all that matter to actors, but that is also an iffy assumption.
I mostly value the game theory aspect as a what-if. What if the problem was simple enough we could solve it through logic, how would the optimal solution look like? Also, when reallife outcomes do not correspond to the model's prediction: why did it not work out that way in reality?
But in the end I don't think we can start with an arbitrary set of assumptions and iterate our way to an accurate predictor in most reallife domains. I believe the domain must be a good fit for game theoretic modeling, and then you also have to do it well.
It seems William Spaniel believes peace negotiations are a domain well suited for game theoretic modeling, and he's probably quite good at it. I'm not sure the domain is actually sufficiently suited but along for the ride anyway.
I know that he is not the only one who considers peace negotiations well-suited to game theoretical modeling. Though given the stakes, perhaps one doesn't have to be all that sure in order to give it a shot.
I love that channel too! But having watched a lot of his videos, I find that he all too often repeats the basics of what he's teaching for the newcomers, diluting the information that is new to me in a video.
Laws regarding the incorporation of a non-profit are a function of IRS procedures and state regulations. In most places it's pretty straight forward. Most Secretary of State offices have a web portal that will guide you. This will get you started: https://www.chamberofcommerce.org/nonprofit
The connection is eluding me. In what way does a browser extension that allows one to markup websites contribute to survival and human flourishing? Can someone connect the dots for me?
This fund is focused on solutions that will work in the upcoming future as it changes due to AI.
It's a solution to "what should we all do in a future where AI is polluting the informational commons with 100x nonsense / slop."
It's basically a commensal way to share curated media diets, with high reputation people doing the work of cutting out the bulk of dreck and sharing blacklists. You choose people whose tastes you trust and subscribe to their blocklists, basically.
It's also a sure way to concentrate current filter bubbles a thousandfold, which is why I won't use it, but it does propose a solution that might work!
We just opened applications for the Pivotal Research Fellowship Q3 2026! It is 9 weeks (+ extension) in London with mentors from UK AISI, Google DeepMind, Redwood, and other leading orgs. Stipend, travel, accommodation, compute, and a dedicated desk are all covered.
I think its likely that ACX readers might want to pivot into AI safety but haven't yet – we run the fellowship to make exactly this happen. So please consider applying! pivotal-research.org
A few days ago, I found out an Iranian friend of mine had been brutally murdered in France. I won't link to the details, but it's been on the local news.
It's been pretty devastating - he was a lovely guy - I've been coping by trying to figure out the motive...
It looks like "random teenager with mental health issues" on the surface, but the content of our last conversation makes me concerned that this might have been deliberate/political. I spoke to him a few days before he died. He was understandably very worried about his family during the war, but also quite openly anti-regime. I know he attended a protest in our city, and that he had an instagram full of anti-regime, pro-Reza-Pahlavi content, and he might have been doing stuff on Telegram in Persian or something.
- Throat slitting feels more "IRGC/Islamic" than a standard stabbing - if he were stabbed in the street, I'd be less suspicious
- The timing: the war obviously, and there were lots of attacks by Iranian proxies the same weekend
Evidence against political motive:
- The killer's claims match "mentally disturbed person & random violence" quite closely
- Base rate of random murder is still higher
- My friend was unlikely to have been very high-profile, unless he was super involved in something online, or had connections I don't know about
- The criminal was too amateurish - his mum called the police on him when he came home with a bloody backpack
Thoughts? I'll speak to the other Iranians here to get their takes over the next few days. Feels like there's some sensible Bayesian reasoning I'm struggling with here
Yeah, I think that, if it is state-led, it's a notably different pattern than the higher-profile assassinations over the last few years, and feels more like stochastic terrorism.
"The two individuals arrested just before carrying out the attack, who were in possession of an improvised explosive device, are minors from the eastern Paris suburb of Montreuil. Known to police for offenses linked to drug trafficking, they told investigators they did not know the identity of the person who ordered the attack. They also admitted to acting in exchange for a promise of "a few hundred euros." Subsequent investigations led authorities to another minor from Montreuil, with a similar background, who is believed to have helped organize the operation."
and... "several countries have already faced this threat, which has so far remained low in intensity, often nighttime acts, carried out by amateurs with rudimentary means and no intention of causing fatalities."
Doesn't seem at all crazy that a list of Iranian diaspora known for their anti-government views was spread on these social networks, my friend was on the list, and that a local teenager responded to the call.
Im not aware of many (any?) examples of Iran carrying out an assassination in Western countries. It would be very strange if the first person they went after was a random person in France with no political influence and no following through some random kid.
Unless there was some massive online following he had that made him high profile I’d think it vanishingly unlikely it was an assassination.
There were a couple of ham-handed attempts. One was in 2011 - they tried to kill a Saudi diplomat in DC. Another attempt in 2024, this time against Trump and other US officials, was foiled. The suspect was arrested and convicted in March.
The US state department (make of them what you will) claims Iran has done hundreds of *successful* assassinations in other countries (presumably many of them non-Western), and that's not counting the mass bombings. This goes way beyond a couple of ham-handed attempts. Iranian dissidents in Europe are frequent targets.
It's common enough to be "a thing". The Swedish case I linked to is the best example, but there have been successful Iranian assassinations of more significant figures (e.g. Ali Motamed, Ahmad Mola Nissi in the Netherlands) and there was an attack the same weekend in Paris.
I guess these things are often hard to prove, as well, so the base rate might be higher than the evidence of "proven assassination attempts" suggests.
Also, the base rate for the leading alternative explanation of "having your throat slit by a random mentally ill stranger in a premeditated attack" is also vanishingly low.
But you're right that he doesn't seem a typical target, unless, as you say, he was very active online or in dissident circles.
"Also, the base rate for the leading alternative explanation of "having your throat slit by a random mentally ill stranger in a premeditated attack" is also vanishingly low."
I'm going to have to nitpick this: that's the wrong base rate to be using. In the counterfactual world where the victim was shot in an apparent robbery or stabbed by a drunk, the same questions would likely arise. It seems like you want something like "the base rate of somebody minding their own business[1] being killed on the street." Which is certainly still low, but much higher than the base rate of political assassinations of non-prominent people.
[1] Specifically, not actively engaging in criminal activity or provoking violent people.
The criminal entered the victim's home with a weapon, and there didn't seem to be a theft or similar alternative motive, so "the base rate of somebody minding their own business being killed on the street" definitely feels too low. But I'm definitely committing some Bayesian sin by adding "throat slitting" in there - especially as it didn't seem to be a "professional hit"-style throat slitting.
This is the recommendation from LessWrong for the reference class problem "look for the simplest reference classes you can find (only adjusting for obviously important and objective factors), or average over lots of ways of constructing your reference class so that no single set of choices dominates."
I guess I could do something like: "what proportion of each reference class is state-directed?", and average out. I think that there's probably 5 or 6 successful political assassinations in Europe per year. So something like:
A: All murders, Europe - 0.2-0.5% state-directed
B: Stranger killings, no clear motive, in Europe - 2-5% state directed
C: Premeditated stranger killings, no clear motive, in Europe - 5-10% state directed
D: Premeditated stranger killings of Iranian diaspora, in Europe - 20-30% state directed
This might give a prior of 4-8%, before adjusting for the timing? But this still feels very messy, and adding overly specific or overly broad base rates clearly biases the calculation.
I'm not sure if honour killing is the right term for this (isn't that usually killing your own daughter?), but yeah, it's always possible that there was something personal.
Sexual violence feels unlikely given both the killer (15-year-old boy with history of violence) and the personality of my friend, but obviously can't rule it out.
I was writing up a response when you posted this. I agree. 👍 I'll just say that I believe Iran is one of a select few countries that *does* habitually carry out assassinations in Western (and other) countries. They've been doing it since the revolution. Russia's another one obviously. You could argue for North Korea and Israel, and probably others I'm unfamiliar with. And obviously, a number of Western countries carry out assassinations in non-Western countries.
This article is a little old, but as of 2017 there were at least 14 suspected hits by russian spies in the uk, but the police were operating on a unofficial policy of just not touching them because of the political implications. This continued until the Salisbury incident made it impossible to ignore.
Also I understand that many would recoil at buzzfeed, but around that time they had serious money from all the listicles and for a while they decided to expand by building a pretty serious investigative journalism team. They eventually scrapped all the hard news in 2023.
And while I understand that this is as credible as "my uncle works at Nintendo" I've known enough journalists and diplomats personally who've had run ins with Russia that I don't doubt what they get up to.
I run a curated Discord for high agency people with Long COVID/ME (myalgic encephalomyelitis). The group includes tech founders, researchers, rationalists/EAs, etc. The focus is on troubleshooting each other's conditions actively, as well as creating a body of knowledge to bring back to the wider community in the form of writing, education, projects, companies, etc.
Some know me as Liface, others as Liam Rosen - I have been in the Rationality community for over 10 years, previously in the Bay, now in New York City, and am the main moderator of r/slatestarcodex. I am a tech CEO whose former life abruptly ended last year. I am now bedbound with severe Long COVID/ME, and am dedicating 100% of my time and effort to helping myself and both everyone else around me figure out these complex conditions.
If you, or anyone you know might be a good fit for this group, email me at liamrosen@gmail.com.
At least part of that is an artifact of us only being "post-Covid" in relative terms. There were about 75k Covid-coded deaths in the US in 2023 and 47k in 2024, which is a lot less than 2020-2022 numbers but still quite large in absolute terms. And if I'm reading things right and have chosen appropriate data series to compare, the broader category of "pneumonia, influenza, and covid" is also a lot bigger in those years (211k - 244k) than pre-2020 numbers for the older "pneumonia and influenza" category (looks like typically 50k-65k). Covid is a lot less deadly that it used to be because so many people are at least partially immune and because better treatments and protocols have been developed for managing severe cases, but it is still killing more people than influenza each year.
I have heard this claimed to be true in the past, but I have also heard claims (which I considered to be more plausible) that this was a misunderstanding of how listed causes of deaths work. The way it was explained to me, the doctor filling out the death certificate commonly lists multiple conditions which played a significant role in the outcome. So the numbers would definitely include an otherwise-healthy person who caught covid and died from it, or someone with fragile health who caught covid and took a nosedive while sick, but probably not someone who caught covid while already imminently dying of liver failure and definitely not someone who had a minor or asymptomatic case when they get run over by a truck.
I'm pretty sure they aren't. My first link in my previous post, the fine print mentioned "confirmed or presumed" COVID-19. The "or presumed" part doesn't make sense if everyone gets tested.
Also, the PIC column appears to be well above baseline, but much more than the direct covid numbers. That suggests to me that either 1) I am looking at the wrong numbers for the baseline (a distinct possibility), or 2) Substantial number of covid deaths are being coded as generic pneumonia. 2 seems plausible to me because my understanding is that the category exists and is included in Influenza and Covid tracking statistics because Influenza-related deaths are often coded as pneumonia rather than influenza and (barring a positive covid test or a case history that strongly points to covid) the same is likely to happen for covid.
>I'm reading the 2024 data as "reduced down to a bad influenza year" (and yes, those are rare) -- do you concur?
Partially. Covid-coded deaths in 2024 seem to be comparable to total estimated deaths from influenza in a particularly bad year, but the latter is an estimated portion of excess pneumonia-coded deaths in addition to direct influenza coded deaths. If I'm reading the stats correctly on PIC deaths, the same methodology would probably show Covid as being a something like a factor of 1.5x to 2x worse than a particularly bad year of influenza. I would characterize this a significantly worse, but within the same order of magnitude.
I also think that with Covid vs Influenza comparisons it's important to keep in mind that covid is adding atop influenza, not replacing it.
>Could you quantify what you mean by "partially immune?" Are we looking at ADE or viral escape?
Viral escape. I had to look up ADE, and if I'm reading correctly, if that's happening in large numbers we would expect Covid deaths to be going up rather than down. The numbers look like what I'd expect if viral escape is allowing people with prior immunity (natural or vaccinated) to catch covid, but the prior immunity still reduces the risk of catching it somewhat and also reduces the expected severity of illness if you do catch it.
Oh, I'd hoped you meant you'd read some of the research that aims to answer the question of whether vaccination increases or decreases the rate/severity of long COVID since that's what your original question suggested you were interested in.
If you are interested in that there are several large cohort studies that show a dose dependent reduction in long COVID rate with vaccination and systematic reviews that detail the reduced likelihood of getting long COVID in vaccinated individuals and reduced symptom severity in vaccinated individuals.
These two are a reasonable starting point
Hedberg P, van der Werff SD, Nauclér P. The Effect of COVID-19 Vaccination on the Risk of Persistent Post-COVID-19 Condition: Cohort Study.
Watanabe A, Iwagami M, Yasuhara J, Takagi H, Kuno T. Protective effect of COVID-19 vaccination against long COVID syndrome: A systematic review and meta-analysis.
Huge swathes of people received COVID vaccine, and cancer death rates kept falling. The idea that the vaccine can cause an increase in cancer appears to be based on nothing (except insofar as living longer means higher odds of eventually getting cancer, and the vaccines increase lifespan by preventing some deaths).
This looks less like AI analysis and more like a familiar mix of alarmism, reductionism, and compressed media framing (again and again and again). If that’s the pitch, the book may already be obsolete in the very way it warns about and this kind of media.
I'm having trouble parsing your comment - are you saying that there are already many books that do this? Which ones? In what sense is the book obsolete?
I meant “obsolete” more structurally than commercially. Not that there are too many books like it, but that this kind of alarm-heavy, reductive framing may already belong to an older media logic, one that was already starting to age before this book even will came out, and that misses what frontier models actually change.
@deisearch Fair point on publishing speed and cost limits. My concern is slightly different: the deeper issue is not just whether AI replaces 100% of workers soon, but how reductive and alarm-driven framings can miss what frontier models actually are, what kinds of work should become obsolete, and what genuinely new problem-solving capacity they may open.
With the pace of AI, I think by the time a book goes through the publishing process, it's already badly out of date.
The one thing that might slow it all down seems to be companies discovering "oops, turns out that it's getting very very expensive, current improvements are very small compared to the huge leaps they used to be, and we aren't able to replace 100% of workers right now, so we're putting a budget on use".
I'm curious to distinguish between two interpretations of your 2nd statement. Is it
1) This is a possible future that would slow things down, and we'll see if we get to it.
2) This is the future we're in (seems lightly implied).
If 2, I think you're factually incorrect. Current improvements are accelerating in the ways that make money for Anthropic and OpenAI, as evidenced by the billions of dollars they increasingly make. Enterprise coding is already a huge economic boon for those companies and it seems to be getting better on ever shorter timeframes. Chatbots are about as good as they can possibly be with the current one-shot in-context learning only version, but agentic AI is unlocking a new ceiling.
"Current improvements are accelerating in the ways that make money for Anthropic and OpenAI, as evidenced by the billions of dollars they increasingly make."
This isn't a secret. I wrote another comment here that the AI companies publicly acknowledge this. They are making a ton of money. They believe that the best business model is to spend more money than that on R&D. They may be wrong, but the critique of "AI is a scam, they're losing money!" ignores the business plan of all the big companies.
Good distinction. But I think “making money” and “reaching the ceiling” are being conflated here. A lot of what gets called agentic AI still looks like workflow automation, while coding is a more substantive case. And the deeper ceiling of frontier chat systems remains partly obscured by product design, UI constraints, and alignment smoothing.
The main difference between a chatbot and an agent is the access they have to the world. It's the same model in Claude Code with full dangerously-skip-permissions and the chat window in claude.ai. The difference is what they can do.
Workflow automation is real work. Before the spinning jenny it was someone's job to spin thread. Then it became the machine's job. Before your AI agent organized your email inbox it was your secretary's job. We've just replaced workflow automation by humans with workflow automation by AI. Coding is easier to see where the AI creates something, but those other jobs were still real work.
I think a lot of people have an unconscious bias against all non-manufacturing work. If I can't touch it, is there value? Sure I go the movies, but that's a nonproductive activity for fun. Fun and value are different. The secretary is just doing things I could do for free, if my time wasn't so valuable. See also every IT worker's tales about job perception from the suits.
But nearly all labor is like that. The US is mostly a service economy. If information work didn't matter than you should have no preference for a blank book versus one full of words.
I don't think you're taking nearly so strong a stance as what I've described, but "still looks like (just) workflow automation" is denigrating a lot of valuable work. For a manufacturing company, everyone who doesn't add a physical piece of the final product is just automating workflow for those who do.
Fair point again — I’m not denying that workflow automation touches real and often valuable work. My concern is that “agentic” is often framed as progress when it mostly automates inherited systems, many of which are bloated by coordination overhead, governance layers, historical friction and control (even control of control).
Frontier models seem more interesting to me where they do not just optimize those layers, but help reveal which of them have become obsolete. That is why coding feels like a more substantive case: it can generate new structures, not only administer old ones more efficiently.
Too often we seem to be building ever more elaborate bureaucratic machines whose main output is permission slips for other bureaucratic machines.
Deiseach was clearly anticipating the future. Hence the “might”.
On your comment. None of these companies are “making money”, which is always taken to mean profits, not revenue. There’s no evidence at all that “Enterprise coding is already a huge economic boon” to any large section of companies.
I recommend listening to Dario Amodei's explanation on the Dwarkesh podcast. The short version is that each model Anthropic releases makes money -- profit, not just revenue.
Example numbers: Sonnet 4 takes $10 million to train. Over its lifespan it costs $10 million more in inference compute. It makes $30 million in total revenue during inference. As it's winding down Anthropic is training the next model, which costs $20 million and uses some amortized piece of $10 million worth of general research in the background.
Those aren't the exact numbers for anything, but they're in the correct relation for what Dario describes. Each model makes money, but the company is simultaneously spending more than that on R&D for the future. Whenever we do hit a wall they will simply sell existing models and make large profits.
Yes, I'm sure the CEO is going to go public with "okay our revenue is increasing but we are also doing the equivalent of burning bundles of tenners in a furnace, so profits? ask me again in 2030 and we hope to still be in existence then and grabbing the greater market share".
First time using the app here, posted before I meant to.
Further, it's insane to me to say that almost no companies are getting value out of enterprise coding. I could readily believe that one or a few CEOs are taken in by the hype train and are foolishly spending money on an investment with no returns. But the adoption keeps increasing and we aren't seeing the heavy adopters post losses or go bankrupt; they're increasing use and profits. At a certain point saying that this is all hype implies that modern late stage capitalism doesn't care that much about profits. Sure, we might still be waiting for the rug pull, but every day it doesn't drop is evidence against that.
This is close to denying the evidence. Sometimes it's appropriate, but looking at what Mythos has done and how many companies keep increasingly using Claude Enterprise means that you're moving past "AI companies often overhype their products" to "there is a vast conspiracy among the Fortune 500 where AI doesn't work and no one will admit it."
> At a certain point saying that this is all hype implies that modern late stage capitalism doesn't care that much about profits.
I’m sure companies care about profits. However outside of the people selling the tools, software profits have stalled recently compared to the growth in the pre covid and covid era.
Your previous post clearly underestimated the cost and revenue of training and compute. It’s directionally correct, in that when a model is complete it becomes a generator of income which funds the next model, but the numbers are way out. The costs are far greater than you say (I realise you say it was a toy model) and more importantly revenues don’t pay for the previous models. There’s some money to be made on the long term - if there’s any economy left - but when are expecting ever increasing spending to stop. Firms like OpenAI, Anthropic, and Google DeepMind can’t really slow down, because if they do, they risk falling behind—and the upside of being ahead (better product, more users, more data) is massive, unless costs consume every company.
My monthly list of recommended long form links is out again.
This batch includes the hypomanic edge common in geniuses, a reflection on the coming second peasanthood of a post literate society, analysis of how Australia is lurching toward a command economy, and a possible link between a major class of herbicides and metabolic dysfunction.
Also linked are my own interviews with Malcolm Collins of Based Camp fame, and a couple of auxiliary language enthusiasts on the topic of Esperanto, Toki Pona and a new mix sized auxlang called Baseyu, and yet another amateur crop breeder working on species you probably never heard of.
It's been observed that birth order affects things like income and educational attainment, with firstborns doing better than their younger siblings. A new paper argues that this may be due to the effects of respiratory pathogens, where younger kids get more of them from their older siblings...
> We use Danish administrative data to document that, before age one, younger siblings have 2-3 times higher hospitalization rates for respiratory conditions than older siblings. We combine birth order and within-municipality variation in respiratory disease prevalence among young children, and find lasting differential impacts of early-life respiratory disease exposure on younger siblings’ earnings, educational attainment, chronic respiratory health and mental health related outcomes.
It's a huge (as in long) paper. I admit I've only skimmed it, but I suspect they're on to something...
I find this personally interesting and will have to look into it more closely. I'm a younger sibling who has had some respiratory issues. They've been overall pretty mild, but were striking in that 1. because of their mildness, they went entirely undiagnosed (and largely unsuspected) for decades[1] and 2. despite this, they've sometimes had a significant impact on my overall quality of life.
Naturally, as soon as I comprehended both of the above points, I had to start wondering how common this sort of thing is. Not just my specific issues, but respiratory issues (or even other maladies) that are subtle enough to largely fly under the radar, but still capable of significantly effecting people's life outcomes.
[1] Either that or I didn't develop them until much later in life. I might not ever know which is the case, but I have some modest evidence to suggest that they've been there all along.
> I had to start wondering how common this sort of thing is. Not just my specific issues, but respiratory issues (or even other maladies) that are subtle enough to largely fly under the radar, but still capable of significantly effecting people's life outcomes.
First, I would suggest that issues don't have to be subtle to fly under the radar. In general they'll only be noticed at all if the person who has them complains about them. And that isn't likely to happen if that person thinks of an issue as just a normal part of life.
I have five older siblings spaced a year or so apart and I remember going to the doctor when I was like 8 to get tested for a lingering illness and although I forget the diagnosis, I already had antibodies for bacteria found in mono and pneumonia among others. When I was young, I was sick a non trivial amount. I went to a large urban school and participated in swimming, wrestling and other sports. I appreciate these studies to improve the health of future youth. If possible, it would be great if kids got sick less and we can prioritize their health and well-being accordingly. Kids getting sick constantly could be a signal of something to be avoided, due to long term consequences, rather than something to be ignored.
So the idea is older kids get sick, and probably get sick more often because they're picking up infections from the other kids at playschool etc., then come home and pass those on to the younger siblings who have less developed immune systems. This means younger kids get sicker and more badly, and this has knock-on effects on development.
I think first you need to define what you would expect the 'normal' level of infection to be, and at what age, before you start proposing that little Timmy got sick with a cold aged six months that he picked up from big brother Jimmy aged two, and this made little Timmy sicker than he would otherwise have been.
What age was Jimmy when he got colds as a baby? How much infection was he exposed to? Did Jimmy get fewer infections than Timmy at the same age?
Though I had to wince at this part:
"We find that younger siblings are two to three times more likely to be hospitalized for acute respiratory conditions during their first year of life compared to the older siblings at the same age, indicating that younger siblings experience a great respiratory disease burden when they are infants. "
Uhhh - yeah. When I was two, my younger sister was born. She got very, very sick at a young age (can't exactly remember how old she was, she wasn't a year old yet going by my memory) with what turned out to be pneumonia and was severely ill while I was fine and never got the likes. Okay, link demonstrated? But I wasn't in preschool or the likes at that age, so both of us were being exposed to the same germs in the same conditions. I didn't pass it on to her and we were both at equal risk of infection at whatever age.
And yeah, later on I got measles first, then my sister got it (presumably from me) and then my younger brother got it after her, so that does incline towards "kids pick up infections from outside and bring them home".
"Lastly, we find that the effects on respiratory hospitalizations among younger siblings are larger in sibling pairs with a short birth spacing in which the older child is in a childcare center than in pairs in which the older child is not. This result provides further support for our hypothesized mechanism of spread—that the older sibling gets exposed to respiratory disease while in group childcare, and then “brings it home” to their more vulnerable younger sibling."
I think there might be something there, but they're pushing the conclusions too far with regards to mental health, income, education, etc.
It's a huge frigging paper! The authors dot every t and cross every i (as they say). They go through every variable in painful detail. I tend to buy their thesis because we know that viral infections like measles and, evidently, COVID cause noticeable cognitive developmental impairments. If the firstborn get fewer infections in early life than the later born, then I can see the difference in life outcomes as a reflection of pathogenic burden.
Childhood exposure to respiratory pathogens seems exceedingly unlikely as a root cause for differential tastes in blogging. Unless we're talking broad intellectual disability so severe as to render people incapable of participating in this community, but I think we'd have noticed that in standardized intelligence or aptitude testing of children at large. It seems more likely that we are looking for something that modulates specific differences in personality, interests, etc, in ways that a respiratory pathogen doesn't.
Really, if you're seeing something that affects income and educational achievement, but doesn't show up in standardized testing in childhood, you should probably be looking at something specific, subtle, and discriminating, rather than the broad insults of a (non-neurological) pathogen. And if it *does* turn out to be respiratory infections, then the question of "how does that lead to *this*?" becomes very interesting,
Suppose that people with >130 IQ are most likely to join a community. If secondbornness lowers IQ by ~3 points, we should see half as many secondborns as firstborns in the community.
That's with a hard cutoff at 130. With any more plausible distribution function for p(SSC/ACX) vs IQ, I'd expect you'd need at least double that difference. And I'd be very surprised if there were a 6-point IQ difference between firstborn and later siblings that hadn't already been spotted and incorporated in the conventional wisdom by people researching birth-order effects.
Yes, and people have been studying both childhood intelligence and birth order effects long enough that if there were a statistically significant effect of birth order on anything like IQ, we'd have seen i it long ago and we wouldn't be depending on a single preprint for our understanding.
Hence my belief that we're looking for something subtler and more specific than "respiratory infections make the later kids dumber".
But first-borns have a ~2–5 percentage-point higher probability of attending college than second-borns. And the gap increases for larger families.
There also seems to be a consistent *monotonic* decline in standardized test scores by about 0.1 SD per sibling's birth order. I haven't bothered to dig into the literature, though.
There's a clear birth order effect on sexual orientation, and I think experts favor the explanation that previous pregnancies alter the fetal hormonal environment for subsequent pregnancies. This probably also affects various personality factors. Do you know if there's good evidence on the effect of birth order on personality (eg. in the Big Five personality traits)?
I think it's pretty consistent. It was discovered in the 90s by Blanchard et al, and has been replicated a number of times. The data indicate that a male is roughly 1/3 more likely to be gay for each older biological *brother* (older sisters seem to have no independent effect, although obviously number of older sisters and brothers are correlate). I'll just put the wiki link here. It has citations to all the main studies of this: https://en.wikipedia.org/wiki/Fraternal_birth_order_and_male_sexual_orientation
I keep encountering these claims about there being an illiteracy crisis for the next generation. Videos of teachers complaining that students are showing up to increasingly advanced grades without basic reading comprehension abilities or other severe limitations. I'd be interested in hearing your perspectives on this matter: if you have children, or you know schoolteachers, or if you just interact with kids for some other reason; what has your experience been?
Is this a global phenomenon or is it mostly hitting Western countries?
My minuscule-sample-size experience is that my daughter (8yo, second grade) is an enthusiastic reader, and her reading skills are at least on par with what I remember mine and my younger sister's being at roughly that age.
My daughter is not a representative sample of the population, but neither are me and my sister.
One factor people often forget to consider is that we're not only looking at differences between generations, we're also seeing differences over time within an individual person's lifespan. My parents, currently about 60 years old, definitely read less than they did 30 or 40 years ago. My impression is that they also probably read less than 60 year olds in previous generations did.
So if there is indeed a decrease in youth literacy, it's not quite fair to only compare that to youth literacy rates in previous generations, we should also consider the ratio of youth to adult literacy. What I'm saying is that if an adult in 1980 read 10 books a year and a teenager read 5, and an adult in 2026 reads 4 books a year and a teenager reads 2, then it should be viewed as more of an overall societal problem and not solely as a problem with younger generations or with the educational system.
Note that I made those numbers up based off absolutely nothing, but my point is that the ratio is worth considering in order to get a better idea of the true scope and cause of issue.
I'm somewhat puzzled by this, since it's contrary to my experience. I was taught a mostly-phonics early reading curriculum in public elementary schools in the late 80s. So was my little sister in the late 90s, and my daughter right now. Three different school districts in two different states. This is a small sample size and it's possible all three of our experiences were outliers, but still enough to make me suspicious of claims that whole-language instruction as described in that book has remained ubiquitous.
Your evidence is oriented in the wrong direction to support your assertion. (And your assertion is wrong - English orthography is fine for phonics.)
You'd want to show that it's hard to predict the sound of a word from the spelling, not that it's hard to predict the spelling of a word from the sound.
You've also bungled your list of examples. What are you thinking about 𝘮𝘪𝘴𝘴𝘪𝘰𝘯 vs 𝘱𝘳𝘦𝘴𝘴𝘶𝘳𝘦 that tells you the "i" is part of the representation of /ʃ/ while the "u" isn't? What are you thinking listing just "s" for 𝘴𝘶𝘨𝘢𝘳 and just "ss" for 𝘱𝘳𝘦𝘴𝘴𝘶𝘳𝘦? Why do you think 𝘷𝘦𝘳𝘴𝘪𝘰𝘯 is an example of /ʃ/?
The reason all these spellings of /ʃ/ exist is that there's a rule of English pronunciation that transforms /sj/ into /ʃ/. That's what's happening in 𝘯𝘢𝘵𝘪𝘰𝘯, 𝘮𝘰𝘵𝘪𝘰𝘯, 𝘴𝘱𝘦𝘤𝘪𝘢𝘭, 𝘮𝘶𝘴𝘪𝘤𝘪𝘢𝘯, 𝘵𝘦𝘯𝘴𝘪𝘰𝘯, 𝘮𝘪𝘴𝘴𝘪𝘰𝘯, 𝘱𝘳𝘦𝘴𝘴𝘶𝘳𝘦, 𝘢𝘴𝘴𝘶𝘳𝘦, 𝘰𝘤𝘦𝘢𝘯, 𝘴𝘶𝘳𝘦, and 𝘴𝘶𝘨𝘢𝘳. And it's not difficult to predict at all. Try making up some words that fit into these paradigms and asking people to pronounce them.
Yea that is definitely a live meme nowadays. It's achieved conventional-wisdom status in both MSM and social platforms. Everyone I know older than 40 who doesn't have children (which includes two of my siblings) thinks it's just a known fact.
That however doesn't even vaguely match my observations of either my own two children or any of their peers. My eldest (now in his early 30s) and all of his peers read plenty. None of the current public school parents I know (whose kids are classmates of my youngest) think their kids read less or more poorly than they did at the same age. Etc.
Certainly in my professional world there is no issue with younger (which at this point is nearly all of them) staff being poor or reluctant readers. I hire 20something/30something adults regularly; haven't yet had any difficulty finding applicants who on this point and others are more qualified than my peers and I were back in the day. My wife says the same.
As you note, the source of the alarum is overwhelmingly teachers. I notice also that all of the college-level jeremiads about this turn out to be from professors who are at or near retirement age. Being old enough to have had rotary dial telephones in my childhood home, many of my humanities professors when I was an undergraduate were certain that we read much less and/or less well than they had. Rinse and repeat on that part.
There've also always been some K-12 teachers who were loudly certain about a literacy crisis in the young. Had a couple of those as members of my extended family over the decades...but I've also known plenty of schoolteachers who did not think this.
So I dunno. I do wonder about a selection bias: the frustrated teachers/professors of this era now have personal megaphones (social media) which those of previous eras did not. Our credulous mediots regularly get their social/cultural trend-story ideas from, of course, social-media feeds.
Hence the educators feeling that the damn kids don't read anymore become way more heard outside of the breakrooms/faculty lounges/dinner tables. Which may, or may not, reflect an actual increase in the fraction of teachers having that complaint.
I am an adult who works in a corporate IT setting, and I have to say that this is not a change limited to children/adolescents. I believe that there has been a shift in the willingness of the median American (or other high-internet-exposure Westerner) to spend time reading content which is boring/not actively engaging. We might see the effect more in adolescents since they're the population taking most of the tests and being directly graded on their literacy, but I don't think it's limited to them uniquely.
I work with a number of career adults who seem newly incapable of reading anything longer than a paragraph. Internal email open rates are below 30% even on short summaries of important subjects like "how to access your tax/healthcare information" and "how to open the programs you use every day". I know this is a classic IT guy complaint but I send 1 or 2 emails per quarter if that. Employees who 10 years ago could reliably handle self-service for things like password resets are now completely incapable of such and cannot learn or retain the information from text guides. It's not just admin stuff but even longstanding documented processes for ordering/selling products where people are now winging it when they didn't 10 years ago.
My guess is that it is due to some combination of shorter attention spans, more engaging alternatives to reading boring prose, and possibly "reading exhaustion" from the median person having to spend more time reading/responding to emails and such year over year. Plus, time spent on phones/social media apps is still increasing each year and that time has to come from somewhere. In the case for both employees & students I'm guessing they are eating into their discretionary & "functional" reading time (along with time spent outside or on family/friends/hobbies).
Scrolling gets better every year and reading is more or less the same as it was 2000 years prior!
I know several college professors and they all say this is an accelerating problem for freshmen. You simply can't have people read something and expect most of them to have learned anything by doing so. Even giving them written instructions for how to do a lab or fill out an assignment will be spotty.
It's not just reading either. I saw an article saying that even *film students* are no longer watching the movies assigned in class, or not paying attention if they do.
I guess everyone has their own perspective. For me, in EEurope in higher education, we are hit by a demographics crisis (TRF way below 2.0 for 2-3 decades now) without immigration. And since we are not the top-top Uni in the country (at least in our field), we teach less and less capable kids (since there are less good students, and they all fit to the "top" unis, making those the "top" unis simply by this), even if there is no illiteracy crisis. There might be, tho.
(Our primary-secondary education is also quite selective, so my son got into the good schools, and his schoolmates are of similar background.)
A completely different perspective, but US faculty at the Professors subreddit mention the illiteracy crisis all the time.
Children definitely seem to be worse at the kind of intellectual abilities reading gives you, but they seem to be more aware of facts in general. For example a 13 year old in my family is aware of class warfare, which she almost certainly learned online. Even as someone who was online at that age, I don't think I was aware of such issues.
Sorry that this is a bit of a gotcha, since you were probably just talking about children's exposure to public discourse, but it's pretty controversial whether class warfare is indeed a "fact". And no offense to your 13 year old relative, but I don't expect a 13 year old's understanding of class warfare to be particularly meaningful, let alone any sort of compensation for a loss of literacy.
I think you just got triggered by a shibboleth and decided to launch into your own headcanon. Being aware of class and its relationship to the world is a signal of intellectual thought especially at that age.
Well, it is certainly a fact that there have been quite a few examples of class warfare in the history of 20th century, both in a fairly literal sense of the term (people belonging in camps divided by social class forming armies and fighting civil wars) and in a less direct sense (strikes, other industrial action).
It's clearly a fact, though you could reasonably quibble about the term "class". The more powerful frequently (ab)use their power against those less powerful. These days the blame is frequently publicly assigned to corporations, but the decisions were made by people who were powerful within those corporations.
Note that this is a common human behavior pattern, and is one of the things democracy was supposed to prevent. (It doesn't, it just alters who is powerful.) It's what anti-trust laws were about. It's what unions were created to solve. That the solutions didn't really solve the problem doesn't imply that the problem doesn't exist.
Warfare to me implies coordinated struggle between reasonably well-defined polities. Rich people and large corporations can have various negative effects on the rest of us, but, for example, someone seeking a monopoly over some industry sees himself as fighting his own class, i.e., competitors.
Also, we need to not get the term confused with the concept. A 10 year old reading Animal Farm two decades ago might be aware of the basic idea behind class warfare, but not have a term for it while the 13 year old today might have the term (and maybe not have read Animal Farm :-)).
German here. We've seen it happen with my niece & nephew. Their elementary school has dropped marks entirely and increasingly adopted an unstructured learning system that was sold has making reading pleasurable for kids as opposed to boring letter-learning. In practice, this passed on the part of actually learning anything to parents. My wife's sister had to do extensive homework with them, but to just get them capable of reading in third/fourth grad at a level that was formerly reached by the end of the first year she had to sit down with them on weekends, too.
Of course, I also know teachers and day care workers. From their PoV, the parents are a massive pain, and I can completely understand them. Troublemaker kids that in the past would have been simply scolded by their parents, are instead supported by them. Teaching basic competencies like dressing yourself or be on time are passed on to the day care workers who have 20 kids at once, while the mom with an only child complains that well he doesn't want to, what can she do? So the teachers and day care workers are way too busy just babysitting and actual education falls by the wayside.
Not to mention the troubles of immigration; Having a classroom where one-third can't even speak proper german to begin with obviously causes issues. Unfortunately many teachers still have trouble openly talking about this, too. Most of the ones I know well enough will admit that even the aforementioned problems are also at least disproportionally related to immigration, though not entirely.
I'm doing lectures at university myself, and here as well it's also a simple fact that immigrants are a disproportional amount of problem cases. We also get pressure from the university admin to drop standards, for various reasons: First, they get paid by student, so they want to give a good "learning experience". Second, the demographic stats always look bad with high standards, immigrants fail and germans get through (plus a few asians and western immigrants). Not to mention the constant complaining by the students themselves. So, to avoid this stupid headache, better have no standards. A professor I work with literally said to me "I'll let them pass in the oral exam eventually anyway, so why not mark the written exam easy to save some time?" I'm sad to say that, but in many university classes I'm aware of we do not make sure that the students actually learn anything meaningful. Who wants to learn will still do so, but you can easily get through without.
And also from other parents themselves I hear stories that make me think. They have a son who is rowdy, can't sit still, has trouble emphasizing with others, and has a tendency to get obsessed with random stuff like tractors. Literally what every single boy was like back when I was a boy, including myself. So they go from therapist to therapist, until finally someone gives them some diagnosis. Usually ADHS or ASD or both. Then the parents and the teachers have an excuse why he isn't learning much, and he has to go to therapy regularly to talk about his feelings. The end result of this process is a young adult who is exceptionally adept at claiming issues, getting sympathy and navigating benefits systems, but who is behind on most measurable skills.
I can go at this for hours, but you get the gist: As far as I can see, we've developed a culture of everyone passing on the responsibilities for kid's development to everyone else. They still end up learning things, somehow, but everything is much slower and more complicated than necessary. Credentials are increasingly unreliable at showing actual proficiency.
When I went to school in Germany unstructured learning schools were not the default. You could send your child to a Walldorfschule or something similar, but it would've been an active choice on the part of the parents.
It's not the default now, either. But school move into the direction of less structure and less testing, based on what I hear from both parents of school-aged children and teachers. Especially elementary schools. People vary more on whom/what to blame (lazy parents, progressive ideology, immigration, take your pick) than on disputing that characterisation itself.
My impression is that the situation would be similar without immigration as well -- unless you would be allowed to teach only the current "better half" of the classes, which is really unlikely. Your class would simply be filled up with "bio-German" students who now are not accepted to your Uni.
Yes and no. If my last paragraph doesn't make it clear, I don't think the situation is entirely caused by immigration. But it definitely sped some things up and made them harder to resist; Without immigration, enforcing standards makes you "just" the unpopular hardass. With the actual immigration as it exists in germany, you're now a racist to boot.
I have heard the same thing from basically every teacher I know (ranging from elementary school through college undergraduate level). I still don't take the claim "students are showing up to increasingly advanced grades without basic reading comprehension abilities or other severe limitations" for granted but based on so much agreement from these teachers I am inclined to think it is true or closer to true than not.
I have children, I'm married to an English teacher, and I work in a school (I'm in the UK).
Both my girls are completely avid readers, fiction and non-fiction, but perhaps they're unrepresentative.
As for students, I don't strongly get the feeling that there is a specific literacy crisis here. I think there is a feeling that as a cohort students are less independent, more needy, perhaps more apathetic and less resilient. There are a few measurable ways in which this (or something like it) is true: there is increasing demand for additional support for students due to their education needs (starting in early years), and we see more accommodations being requested for exams and tests (extra time, reading support, rest and movement breaks and so on).
These sorts of things *might* manifest as students claiming work is too hard for them, or the pressure of makes them anxious, or being somewhat indifferent to certain types of reading asked of them. I feel teachers complain more that students don't want to do the reading than that they genuinely can't.
But I don't think what I'm talking about is overwhelmingly obvious to the extent that you'd be sure there's an underlying change in students, rather than it being due to more demand from parents, changes is school protocols, increasing competitiveness at the top, diagnosis drift or other factors.
One relevant thing in the UK perhaps: here we have an externally validated examination system, so if it's true that in the US students can be passed for classes where they actually by any reasonable standard should be failing, and thus be genuinely unaware of their deficiencies, it would be very difficult for that to happen here. A university might accept a student with a grade D in a subject, but they would have a good idea what they were getting (and the student would understand their own level of attainment).
"As for students, I don't strongly get the feeling that there is a specific literacy crisis here."
I think the specific crisis here is supposed to be that the kids can't (and therefore won't...) read long works. Long works like entire novels rather than just excerpts.
Yes I'm familiar with the debate, OP was asking for personal/anecdotal evidence and I gave my experiences.
My wife complains that students skip her classes and sometimes they're rude to her and there are times when they don't have much to say. She doesn't complain that they're incapable of engaging with the big ideas at all or that they have massive gaps in their ability to comprehend text.
If OP had said: 'people with experience of kids, do you see evidence there's a mental health crisis?' I'd have said 'it's complicated, but maybe'. But my answer to 'do you see evidence there's a literacy crisis?' is 'it's complicated, but not particularly'.
The way this was explained to me internally when I worked for Google and Microsoft when they moved towards ideograms is that the main driver is the cost of internationalization and localization. If you label your UX with text, you needs to translate the strings to every language you support and test the UX to make sure it still looks okay and is usable with the translated text that might be different lengths, difference character sets, different reading direction, etc. But if you use abstract symbols, you can use exactly the same UX for different languages and locales.
We try to make the blog as general as possible but sometimes not everything applies to everybody, this seems like it unconsciously slipped in. We live in London and Amsterdam, where this is possible.
I love things like this, and have some 80/20 money tips. DM me! for instance: did you know it's possible to get most of your 401k and IRA contributions out before 59, with no penalty, for any reason, meaning you should basically always max out those contributions to avoid tax.
Ask Claude to explain roth laddering. There is a catch. Basically, you can choose to take your prior contributions to IRA or 401k and convert them to a new Roth IRA. Now you have to wait 5 years (sorry ) you can then extract that transfer amount penalty free. Learned this on the r FIRE subreddit.
1. Iran’s foreign minister, Abbas Araghchi, said that the Strait of Hormuz would be open during the truce, but the IRGC stepped in and said NO!
As tankers and container ships tried to exit, IRGC sent out armed speed boats to herd them back into the Persian Gulf. The IRGC targeted at least three commercial cargo vessels with projectiles, causing damage to the containers and triggering a fire on at least one ship. No crew were injured. And two Indian-flagged tankers were fired upon (the Sanmar Herald and the Jag Arnav). We have a recording of the Sanmar Herald incident: “Sepah Navy! Sepah Navy! This motor vessel Sanmar Herald. You gave me clearance to go! My name number two on your list! You gave me clearance to go!” The captain was pissed! I wonder if they’ll give his company a refund for the toll they paid. No injuries or major damage were reported, and the crews are safe.
2. During the confusion as to whether the Strait was open, five cruise ships slipped through the Strait of Hormuz, hugging the coast of Oman. The first was the Celestyal Discovery late Friday night. Then four other cruise ships (Mein Schiff 4, Mein Schiff 5, MSC Euribia, and the Celestyal Journey) transited the Strait of Hormuz on April 19, 2026, beginning around 00:30 GMT. They sailed at full speed through the declared minefield zone with no stops. The passengers had been unloaded weeks ago. There were only crew aboard these ships. I wonder if the captains and crew will get bonuses from their companies.
3. Despite some garbled reporting by the MSM, the US Navy is not operating in the Persian Gulf or in the Strait (though they sent two destroyers through the Strait last week). The US has posted its Abraham Lincoln carrier group about 125 miles outside the Strait of Hormuz — they're trying to stay out of range of Iranian drones, and they're not engaging directly with Iranian gunboats. They just forced about a half-dozen tankers with Iranian oil to turn back toward the Strait. But Iran still has control of the Strait. There was talk about our LCS assets clearing the minefield(s), but if they are, they're sort of far from immediate support by the carrier group. And the Iranian gunboats seem to be zipping around the Strait without being challenged by the US Navy. Maybe we're waiting for the Bush carrier group to arrive on the scene?
4. And this just in: The US just disabled an Iranian container ship, the Tousca, inbound for Iran by hitting its engine room. Reportedly, US Marines boarded it and took control of it. Iran's military command says it will retaliate for the US shelling and seizure of the Iranian-flagged cargo ship Touska earlier today, calling the blockade action "armed piracy by the US military.” I’ll be curious what the Navy will do with the Tousca. Are they going to tow it to port somewhere? Skuttle it?
5. Also, it’s not clear what the US will do if they encounter a Chinese tanker. And BTW, although the Chinese navy stays pretty close to home, the Chinese have sent their three aircraft carriers out on some blue-water exercises over the past few months. Normally these stay pretty close to their home bases. They’re not nuclear powered, but the next one the one they’re building will be. Google Gemini says their current carriers have the range to reach the Gulf of Oman and return to China without refueling, though. But they’d need all sorts of support ships. I doubt if China has developed the logistics to handle a long-range deployment. Even though our nuclear-powered carriers don’t need refueling, all the support ships in the carrier group do. And it’s a complex logistical operation to move a carrier group from one ocean to another.
6. Various Iranian parliamentarians called for the impeachment of Foreign Minister Araghchi. Control of the Strait is Iran’s biggest strategic bargaining chip in the current game of chicken, and his statement was perceived as a surrender. There were mass demonstrations in major cities against him. As of this writing, Araghchi is still Iran’s Foreign Minister. No word from Supreme Leader Mojtaba Khamenei. In fact, Khamenei has not made a public appearance since his election as Supreme Leader over a month ago. Lots of rumors flying about. One rumor is that he lost either a foot or a leg in the Israeli strike that killed his father. Another rumor is that he’s in a Moscow hospital being treated for the rumored wounds. The fact that the IRGC and the government aren’t coordinating their actions and statements makes me wonder whether he’s in a condition to give orders.
7. Saudi Arabia has maxed out the oil that it’s pumping across its east-west pipeline to its ports on the Red Sea to 7 million barrels per day. Unfortunately, their ports on the Red Sea can’t handle that volume of oil. I assume they’re storing it there? But some Saudi oil is getting out. The Houthis seem to be silent at the moment, so tankers can get through the Bab el-Mandeb Strait. In case you’re wondering, the Suez Canal cannot handle VLCC (Very Large Crude Carriers) and ULCC (Ultra Large Crude Carriers). They can only accommodate mid-size tankers. Iraq has restarted work on an oil pipeline to Jordan that was halted for some reason. The UAE is going to build a pipeline over to Oman. I can’t make the math work myself, but about 9 mb/d is coming out of the Middle East vs 20mb/d that’s not coming through the Strait of Hormuz, and experts suggest we’re facing a shortage of approximately 11 million barrels per day.
Thanks I always like reading these. This is unrelated, but I heat my house with oil, about 700 gallons a year. I called and had my 1000 gallon tank filled up, the price may go up or down (was ~$5.20/ gal), but I feel better without the uncertainty. Uncertainty is expensive.
Regarding bonuses for the crew members on the cruise ships, considering that cruise lines mostly use crew from poorer countries that aren’t paid much, a reasonable guess is “no.”
But the captains and officers are well-paid, indeed. Cabin crew was probably offloaded with the passengers because, well, the companies wouldn't have had to pay them even their minimal minimum wages.
Point six is the most significant IMO. It feels like there's an internal crack or power struggle within the Iranian government over how to react. I get the sense that the pragmatists are willing to make big concessions and the hardliners are refusing, which suggests the pragmatists might think their situation is desperate.
True. But the Saudis and Gulf states face the same problem if they have to shut down their oil wells. It's very hard to restart oil wells that have been shut down.
But let's face it. Even if we fuck Iran's economy to the point of capitulation, the world economy will likely be fucked for years to come.
Well, Russia provided about 12% the world's oil supply, and the Gulf States provided something like 30+ percent. The Gulf States had the ability to increase their production significantly whenever they needed to. My understanding was that they increased their production somewhat to meet the increased world demands with the decrease in Russian oil production (but i'm willing to be corrected on that point if I'm wrong). But now the blockades have reduced oil output from the region by about 60% (?). So, there's nowhere in the world that will make up for that production blockage.
As for Europe, it was willing to apply sanctions to Russia because of the threat it posed to Ukraine and the other countries that bordered Russia. AFAIK, European sanctions against Russian oil and gas are still in effect. But it doesn't matter anymore because hasn't Ukraine knocked all the Druzhba pipeline feeds to Europe off line?
This may be a stupid question -- but would it be safe to say that the US Navy is unable to counter Iranian drone tactics, and is thus unable to take control of the Hormuz Strait ? Or are there other reasons for them to stay out of it ?
It doesn't really have anything to do with drone tactics. Protecting civilian shipping would require navy ships to operate right off the coast of Iran moving in predictable patterns escorting slow, fragile targets. The US cannot guarantee that Iran would be unable to move enough AShMs into position to seriously threaten sinking a Navy ship.
Does the answer have to do with what political risks the Navy/administration are willing to take? Even if you're pretty sure your defenses can fend off a drone attack without major losses., if you think that losing a single support ship or ten sailors would be a political disaster, you may not be willing to risk it.
Right, it seems that if the US Navy was capable of effectively countering drone warfare, then losing even a single support ship would be a remote possibility.
Are you implying that the damage was actually due to an Iranian attack? Or are you implying that the loss of the oiler is inhibiting the US fleet from commencing active operations to clear the Strait of Hormuz? ... Or both?
My point is that it doesn't look like Iran has a cohesive leadership at this point. And like Donald Trump, their right hand doesn't know what their left hand is doing.
I had an epiphany the other day on immigration. I've been told by a couple people (I first posted it in the Discord) that some of this is easily known, but I consider myself pretty steeped in the culture war, once from the left wing perspective and more recently from the right, and never heard this laid out clearly at all by either side.
It is a one hundred percent fair characterization of people who oppose 'illegal immigration', oppose immigration reforms, and oppose H-1B/PERM fraud, that they are 'anti-immigrant', and not just 'anti-illegal-immigrant'.
This is because the existing immigration laws, as written, aren't actually designed to accept immigrants, the general class.
We accept relatives of existing Americans, we accept the cream of the crop (EB-1/2, O-1, etc), we have the diversity lottery, and we have various consolation prize statuses like political asylum, victims of abuse, etc. But the category that is explicitly for 'work in general' - anything below 'advanced degree' - is reserved for PERM tests, where the company must prove that they attempted to find an American for a position and failed.
An 'average Joe' does not have a pathway to America, common descriptions of his employment-based pathway implicitly describe fraud (typically pretending to have looked for an American by only posting a job ad in the local paper) and the system was explicitly designed to block off such a pathway by default. I've previously stated in many arguments that 'we're letting in a good number a year and the only problem is the length of the line' and I've now realized that this just isn't true at all.
I think most "anti-illegal-immigrant" people are in fact just xenophobic, and the actions of the past year have demonstrated that vividly, but even so, I don't think what you presented is a fair argument. And there are a non-zero number of people out there who honestly are anti-illegal immigration.
One of the more refreshing immigration comments I've seen was on a Hanania H1B post, where someone stated he opposed H1B because he's an entry level tech worker and H1Bs lower his earning potential.
IMO the immigration debate would be more fruitful if everyone could be as direct.
"I'm a single working class male in my 20s, and I oppose mass migration because it's disproportionately men under 40 and I don't need increased sexual competition."
"We're the only African family in our area, and we'd like more African families to migrate so we feel more comfortable practicing our cultural traditions, religion, and heritage."
Simple way to tell someone who is "honestly anti-illegal immigration" from a dog-whistling xenophobe: "Would you consider (not automatically agree with, just consider) doubling current legal immigration quotas?" Which would suddenly turn a lot of illegal immigrants into legal.
The latter would automatically reject this idea. The former will at least listen to arguments.
Agreed. Ideally I'd like a referendum on how many immigrants to let in each year, with the number set by the median voter. Half the electorate will view the number as too high and half will view it as too low.
And if the median preference is to double the number to let in - ok, vox populi vox dei.
( Separately, a _much_ more complicated question is how to unravel the mess of ad hoc patchwork exceptions that immigration law contains. Personally, I'd prefer a system that effectively ranked immigrants by their anticipated net benefit to the people already here (including, as a factor, the potential immigrants' intention to assimilate into American culture) and then let them in in rank order till the number from the referendum is reached. Sort-of kind-of like a lower bar version of the O-1s )
This is in the direction of correct, although there's still a confounder: we don't know how many immigrants can be naturally assimilated. It's not simply a matter of what the law permits. If the law authorizes a million immigrants into a community of 100,000, that community will not be the group doing the assimilating.
If a community of 300 million authorizes immigration of 3 million (the usual US legal rate IIRC), it's a lot easier, but still non-zero effort. Someone has to hail those 3 million immigrants and introduce them to US society and explain how it works and ensure that they mostly function like the actual US citizens they are expected to be. And a lot of those immigrants will want to move into enclaves filled with immigrants like themselves, and mostly keep living they way they lived in the old country, including how they drive, how hard they work, how hard they try to pay back loans, how they react to police, and a thousand other things. Even if they're conscientiously trying to integrate, and even if they're sterling citizens who were pulling all their weight and then some in their prior country, it's hard. And far from all of them are trying.
A lot of the rhetoric I'm seeing from people opposing illegal immigration reads as quietly downstream of this understanding. I'll notice in passing that they're perfectly fine with people who are obviously immigrants, because those people are actively trying to behave like the Americans born and raised here.
"It is a one hundred percent fair characterization of people who oppose 'illegal immigration', oppose immigration reforms, and oppose H-1B/PERM fraud, that they are 'anti-immigrant', and not just 'anti-illegal-immigrant'."
The "oppose immigration reforms" part is doing a lot of heavy lifting here. Without that, all you've got is an accusation that anyone who wants the laws to be enforced is anti-immigrant, because the laws are supposedly anti-immigrant, and no, that's not how it works. Lots of people who don't like certain laws and would prefer that the currently-illegal thing be more common and accepted than it is, still want the laws to be enforced because Rule Of Law is a thing. A generally good thing.
And, it's kind of nonsensical to say that people "oppose immigration reform", because nobody opposes immigration reform. Or any other kind of reform; it's implicit in the definition that "reform" points to things that one wants to happen. Unless a person believes that America's policies are absolutely perfect, they're going to want to see changes, and they're going to see those changes as reforms.
They may have a very different idea than you as to what constitutes a "reform". But that's where you need to develop your argument. Because right now, I'm mostly seeing just a denunciation of people as "anti-immigrant" solely on the basis that the reforms they'd like to see are not the sort of reforms that you'd like to see.
> Unless a person believes that America's policies are absolutely perfect
Which is exactly what I see in "anti-immigrant" people. The more stridently they denounce illegal immigrants, the less thought they give to ANY immigration reform -- in any direction. Often they also heap praises on "immigrants who follow the rules" -- the CURRENT rules, of course. The idea that these rules could be changed is studiously not mentioned.
In *any* direction? You really think the strident anti-immigrant people don't want to see tighter visa requirements, more deportations, and maybe a big, beautiful wall on the southern border?
I suspect you are defining "reform" to include only changes in the subset of possible directions that you favor. In which case, you're basically just arguing "these people disagree with me and they're *wrong*", but with an obfuscatory and inflammatory phrasing. And I don't think that's very helpful.
> You really think the strident anti-immigrant people don't want to see tighter visa requirements
I am sure they do, but they almost never openly SAY it.
> more deportations, and maybe a big, beautiful wall on the southern border?
They frame these things as "enforcing existing laws" and "cracking down on the illegals" -- emphasis on "illegals"
I rarely if ever see anyone in US openly call to decrease LEGAL immigration; as I said in an earlier post, the same people who yell loudest about "illegals" often go out of their way to praise immigrants who "follow the rules". Openly calling to lower legal immigration seems to be... uncouth.
> I suspect you are defining "reform" to include only changes in the subset of possible directions that you favor
And what if they don't want to? They're anti-immigrants-who-wouldn't-be-allowed-in-if-laws-were-being-followed-as-written. That doesn't mean they're anti-immigrant. It means they're against any immigration that would not explicitly benefit the country.
Except it is very easy to show that giving legal status to say, illegal immigrants who pick strawberries, WOULD benefit the country. "Explicitly benefit the country" is a bad-faith argument.
Given some of the angry comments I've read elsewhere, anti-immigrant sentiment isn't aimed at strawberry pickers. Rather, it's aimed at actual criminals, including everything from vandals and people who get in car accidents while driving uninsured to actual rapists and murderers, with a side helping of immigrants who benefit from government services they obviously cannot have paid for. This anger is then coupled with "what do we do about it?", which turns into "well, in this list of cases, they were obviously here illegally, so let's tighten up the border".
AFAIK, no one has proposed seeking out strawberry pickers and awarding them legal status (the closest I've heard is doing this for people who manage to join the military), and while no one's making much noise about people who maybe pick strawberries by day and then assault people by night, the main complaint is still whether and how they're getting benefits at taxpayer expense.
Nothing bad faith in that, and to assume it is is to ensure the problem only gets worse.
One can argue that for every illegal immigrant rapist, there are a hundred thousand who just pick strawberries, but hopefully you can see how that would be cold comfort to the one suburbanite who lost her daughter, to say nothing of her hundred or so friends and acquaintances, many of whom are struggling to make ends meet and pay taxes and watch their representatives make a show of rolling out the red carpet for illegal immigrants.
All of which is essentially red herring. According to FBI statistics, illegal immigrants commit both violent crime and benefit fraud, at LOWER rate than native-born Americans. (Lower still among legal immigrants.) Yes, this suburbanite is angry. Focusing on him, as opposed to on his several neighbors who lost their daughters to violent US citizens, is nothing but deflection.
People are not struggling to make ends meet because their representatives roll out the red carpet for illegal immigrants. They struggle to make ends meet because for last 45 years US has been implementing policies which divert economic benefits away from the working class, and toward the owner class. This image sums it up perfectly:
Even when you factor in social cohesion? Racial resentment and conflict do not suddenly appear out of nowhere. It is a product of circumstance. We cannot change humanity, but we can change the circumstances that guide their actions.
I am a US citizen. I am also white, Jewish, and originally from Russia. My parents and I immigrated legally -- yet today a Russian family exactly like mine (two parents with professional degrees, one teenager) would have almost no chance of legally immigrating into US. We followed the rules -- luckily it was before the rules were changed.
Having said that, I feel rather more "cohesion" with a Brazilian family I've known for about 8 years -- which is in US illegally, -- than with white Christian Nationalists. The former never did anything bad for me (actually, helped me a great deal and vice versa), which is not something I can say about the latter.
99% of US citizens only have a very cursory understanding of the US immigration system - simply because they don’t need to understand it. Nobody reasons in terms of “PERM” or “EB-2”, people just reason off vibes.
Yes, other than “very online” people and think tank employees nobody knows what H2A is, how much the salaries are, or what taxes are being docked vs evaded. Policy is decided based off vibes.
The federal minimum wage is 7.25/hour. If you're saying $18 is three times "the federal minimum wage minus taxes", that's nice, but the person making $18/hour will also pay taxes, so it's irrelevant.
Correct, an agricultural job sucks way more than working at McDonald's. This is why garbage-men also tend to make more than cashiers. The job sucks more and it requires a higher wage to find someone to do it.
Picking strawberries or raspberries by hand all day, outside, exposed to the elements and the bugs, for away from any entertainment which one could spend money on after work?
Yeah, it's worse.
Using the latest BLS nonfatal injury/illness rate table, the comparison looks like this:
Fast food worker proxy — Limited-service restaurants, NAICS 722513
Total recordable rate: 2.6 per 100 full-time workers
DART rate: 1.0
Days-away-from-work rate: 0.8
Crop production worker — Crop production, NAICS 111
Total recordable rate: 4.1
DART rate: 2.5
Days-away-from-work rate: 1.4
Animal production worker — Animal production & aquaculture, NAICS 112
I'm Australian and don't think the US visa system is well structured. I also don't see why a diversity lottery system for average Joes plus a merit system for exceptional Joes would be anti-immigration.
Why would you deliberately take normal people if there was a line of the world's best artists and professionals stretching out the door? And if you, for some reason, held a lottery to allow normal people in, why would you want to make the pool of winning tickets larger?
The merit system stops right under "world's best artists and professionals" (i.e. EB-1 or O-1). Anything lesser, such as a mere credentialed and experienced SRE who is not literally one of the best in the world, is not qualified by merit alone.
I have worked on O-1s and EB-1s, and referring to recipients thereof as "the world's best" implies that the bar is much higher than it is in practice. Though given the current Administration's antipathy to all immigrants, who knows what the current practice is.
It has been a while, and it would be unethical to give specific names, but I can recall getting either an O-1 or an EB-1 for someone who designed and sold women's shoes, a young filmmaker who had never made a feature film and was coming to work at a film school for high school students; a few other artists who had a couple of shows scheduled at galleries in New York, several animators who worked on big films at Dreamworks or the like (somehow, each of them played a critical role in the success of Shrek IV, or whatever), the inventor of some sort of pharmaceutical compounding doohickey, and a forensic pathologist who worked at a county medical examiner's office.
Not that these people were not quite talented, but in the grand scheme of things, they were indistinguishable from a zillion other quite talented people.
Another eg: I am very sure that Melania Trump qualified for her EB-1, despite knowing little about her. High pay (pretty easy bar to reach, given that the median model makes very little) plus was on the cover of European Vogue ("Evidence that your work has been displayed at artistic exhibitions or showcases") plus judged a couple of fashion shows ("Evidence that you have been asked to judge the work of others") could easily be enough because you only need to meet three criteria. https://www.uscis.gov/working-in-the-united-states/permanent-workers/employment-based-immigration-first-preference-eb-1
"...but I can recall getting either an O-1 or an EB-1 for someone who designed and sold women's shoes, a young filmmaker who had never made a feature film and was coming to work at a film school for high school students..."
Also, the O-1 visa is used for foreign minor league baseball players. So a foreign national playing high-A minor league baseball will be here on an O-1 visa.
In some sense these *are* the world's best. I'd guess that most of the guys playing minor league baseball are in the top 25,000 in the world at that. In another sense ... where are we drawing the cut-line for "world's best" again?
PhD folks also often come in on O-1 visas. I don't know how many of the world's PhDs would qualify. Probably a lot more than 25,000 ...
Right, this is where the US visa system is weird, because it takes (for example) a quota of 10,500 migrants in speciality occupations without some of the H-1B restrictions, but only if they're from Australia. Or you could get a normal H-1B. Either way you'd be on a "non-immigrant" basis, except actually you can just keep applying for permanent residency because an I140 is somehow not intent to immigrate.
So you have a situation where immigrants from certain countries are effectively allowed to work in the US (because of the H-1B, or the magical E-3) and stay there until they're either rich enough to buy a visa/green card (E-2/EB-5), attractive enough to marry (I130), or lucky enough to win the green card lottery, which can have reasonable odds over a few decades depending on the country of origin. Australia is sitting at 5.39% chance of winning, so if I played the green card lottery for a couple of decades I'd have a 2 in 3 chance of winning permanent residency
This is obviously a very silly system but I'd struggle to call it completely anti-immigrant. It's designed by madmen but it does provide pathways to some people sometimes, although God help you if you're Indian.
>because an I140 is somehow not intent to immigrate.
Of course an I-140 is an intent to immigrate; it is an application for permanent residence. But there is no problem forming an intent to immigrate after you arrive; it is only fraudulent if you have the intent to immigrate when you arrive on certain non-immigrant visas. And only certain ones; for example, an H-1B is a dual intent visa, as are many others. https://www.nolo.com/legal-encyclopedia/what-are-dual-intent-visas.html
Doesn't this just assume that the only way to not be anti-immigrant is to be willing to accept more-or-less unlimited numbers of immigrants who meet some very minimal standard like "not a known criminal or terrorist, not carrying a contagious disease, not likely to be immediately a dependent of the state?" I mean, yeah, this works in the sense that I can call everyone in poltiics a communist by defining communist as "anyone who wants any regulation at all of markets." But I can't imagine that this convinces anyone who didn't start out agreeing with you.
An example of a person who cannot immigrate to the US by default without fraud being committed on their behalf is a British accountant with a bachelor's degree and ten years of experience. I think many people who say "they should come the legal way" would be surprised by that fact.
"Unlimited" is not implied at all, a numeric cap would make plenty of sense and many are already deployed all over the place in the immigration system.
If your definition of anti-immigrant is supporting a policy that does not allow most people to immigrate to your country, you will indeed find that most voters in most desirable countries are anti-immigrant by your definition. But again, is there some benefit to this definition other than defining the other side as bad?
Like, if a billion perfectly fine human beings want to immigrate to the US this year, but I don't think it would be a good idea for us to take in a billion immigrants next year, by your definition I'm anti-immigrant. But also, so are everyone except the literal open-borders folks.
You’re pointing out that the system is broken. There hasn’t been a push for immigration reform that would allow for more consistent monitoring on the southern border while also allowing for easier immigration by the sort of average Joe you’re talking about here. Instead the options seem to be almost uncontrolled immigration under Biden, or completely shut down immigration under Trump.
Unless there’s a viable solution presented that allows for normal productive people like you’re describing to get in, then I think it’s a legitimate position to take to be opposed to massive illegal immigration, while still supporting more legal immigrants that are vetted.
Now you're arguing something different than OP, not just that everyone should have a chance but that the probability should be high too. (Of course even without the lottery everyone would have a chance by making the right career choices, or marrying right. And if you believe in "accident of birth"-thinking then everyone had a chance of being born in the US already, about 4%.)
Also, 1% per year adds up. Much higher and it would almost be guaranteed one would get in sooner or later. That certainly wouldn't work if applied to the whole world.
No, I am arguing exactly what OP argued, which is that there is no practical immigration avenue for a British accountant with a bachelor's degree and ten years of experience.
>Also, 1% per year adds up
Not very quickly. It takes 69 years to reach a 50 pct lifetime chance of winning. 1-0.99^69 = .5002.
I know a lot of things. Most of them are not relevant. I even know a lot of things about Afghanistan. Even those are not relevant. Your posts have a tendency to go off the rails if the topic. I previously complained that you on your own had 1 out 9 posts in previous threads. Could you limit yourself as a start to 1 out of 20? 5% of total posts?
I am not familiar with the law. I know a few people who went to USA for PhD, and stayed there. Some of them work in IT, some of them are university teachers. What do you assume was the mechanism they got there, and would it match your explanation? (Let's assume they *didn't* acquire citizenship by marriage.)
This would probably be EB-2, the employer-sponsored permanent residence visa for advanced degree holders. For the university professors, their participation in public research qualifies for a National Interest Waiver from the PERM test, so they just need the employer sponsor (the university). I don't know whether IT workers qualify for NIW, it probably depends on the company. If not, then PERM fraud was probably involved for those guys.
And, you are assuming that OP's friends have green cards; they can be here long term on nonimmigrant visas. O-1s can be renewed repeatedly, for example.
H-1Bs are subject to the same requirement (with no waiver opportunity) though note also that they are a nonimmigrant visa. They fill work shortages proven by looking for two months for an American at a market rate salary and failing; this requirement is generally cheated by posting an ad somewhere nobody will find it (before jobs.now started aggregating these).
At least to start. But the fact that it's very very difficult for an H1-B visa holder to get their H1-B transferred to another company, studies have shown that over the salaries of the visa-holders are suppressed over time. While US citizens can freely move from employer to employer for a better salary. My experience was that each job move would garner me an immediate 10% salary increase. While staying with an employer year after year, I garnered piddling 1-2% salary increases.
The people who are in to personality tests (I am not a person who is in to personality tests) seemed to think that tests for clinical psychology (e.g. MMPI) are an entirely different kind of thing from Meyers Briggs type personality psychology. Certainly, they purport to measure something different.
ENTP here as well even if I "believe" in it all very little, its more like a fun fantasy to connect with; like all these categorization frameworks (astrology, etc) they do a really good job with making the writing "feel" like its exactly what I think I am! lol
now my unnecessary and unceasing desire to debate.....
There seems to be something wrong with Google Search. When I search for "overcoming bias respectable rants", Google does not return the link to the well-known article: https://www.overcomingbias.com/p/respectable-rantshtml.
This seems to happen to only this specific article. Why?
A win for DuckDuckGo, which returns the article as the first hit. Google didn’t find it when I tried it. (I only checked the first four pages of results.) Google did return the top level page of overcomingbias as the third result. Google returned the article as the first result when I searched for "overcoming bias respectable rants site:overcomingbias.com".
When I tried the search in a Firefox private browsing window, it worked (returning the article as the first hit). But I couldn't get the search to work in a regular browser window even after deleting cookies and local data.
Normally, I use DuckDuckGo first, and only try Google when I don't find what I'm looking for using DuckDuckGo, so I only see DuckDuckGo's failures. I still suspect that Google is better than DuckDuckGo, but I now realize I don't have good evidence of that.
Googling the words "overcoming bias respectable rants", in that order without quotation marks, got the Hanson article as the top response in three browsers on two different devices, regular and incognito mode.
If that was a ploy to get me to google that term, well played, you understand how social media works. Other than that, I had zero issues with that search, it was the #1 result, and its comment section was #2.
Did you put "overcoming bias respectable rants" in quotes, like that? Because that got zero hits for me, probably because that is not the title. But a search for "overcoming bias" "respectable rants" brought it right up.
Not weird, because putting the entire thing in a set of quotes creates a search for that specific string of words. There is no particular reason to think that that set of words exists.
Just a couple of days ago I wanted to reread the Meditations On Moloch and had the same problem. The first three pages are a bunch of sites talking about it, readings on spotify and youtube, or even tweets. Then I gave up and just searched my browsing history (which goes back years and I should probably clean it up).
I just assumed that google's search algo has been bent so heavily in the name of pushing marketing to the top that now when it's looking for something that can't be monetized in any way it just comes off as completely lobotomized.
My first result is SlateStarCodexAbridged, then a link to the effective altruism forum, then a tweet from some random guy, the actual post is nowhere to be seen within the first five pages of results.
I guess fingerprinting plays a role. Are you logged into google? Do you get the same results in a private window?
Regardless, this is not the behaviour I'd expect from a supposedly good search engine.
I had a similar experience in the past, trying to find something, instead finding dozens of news articles talking about the thing... and of course none of them actually links to the thing they talk about, instead they just link to other articles talking about the thing.
Google search sucks, but also this SEO approach (never link X, if you can instead link another article about X on your own server) sucks.
Even if they did link to it, Google might not be indexing it anyway. For example, Google refuses to index my blog for no apparent reason, and this seems to be pretty common. In fact, Google is even failing to index *Github repositories* now.
So Claude 4.7 is better than 4.6 at almost everything. But notice that Mythos is about the same amount better than 4.7, and none of us are going to get to use Mythos now, and maybe not ever. So that got me ruminating about the emergence of a dividing line, between improvements in AI (such as, presumably, 4.7) that are shared with the public and improvements like Mythos that are not. It made me realize that I've gotten used to the improvements being passed on to the public -- some in the form of access to new models that's free or cheap for all, some sold to manufacturers, etc., to integrate into their products and processes.
Seems clear that no good can come of making Mythos available to all who want it. It's like passing out the keys to a bunch of doors. But what about future models? Seems like many of them, maybe most of them, will have capabilities that it's dangerous to make widely available. So then what? I'm not crazy about the idea of either the AI companies or the government being the sole entity that can use these models. Are we entering an era of a division between the AI haves and the have nots? How's that going to play out?
It's quite reasonable that Mythos-level bug finding 1) is genuinely new 2) no one else will invent it for at least a few months and 3) can be used defensively first.
1. Anthropic puts a lot of effort into making their standard Opus line unable to do cyberwarfare. They fail to train it in that, they potentially untrain it in that (complicated but theoretically possible), and they align it very heavily to refuse cyberwarfare requests.
2. Anthropic maintains the Mythos brand as AIs slightly better than the best public Opus. The white hat cyber companies always have the best Mythos, and the public always has the best Opus. I still haven't been able to figure out whether Mythos is simply bigger than Opus in the same way Opus is bigger than Sonnet, but it seems like this must be part of the story (since it apparently costs a lot per token to use). If Mythos is so big that it would be cost-prohibitive for the public to use it anyway, and Anthropic would never have bothered training a model that big for the public, then maybe not much is lost by keeping it out of public hands.
I agree that "exactly how is Opus different from Mythos?" is one of the more interesting questions in practical AI right now.
I think we could maybe add a third issue, which is that Anthropic appears to be pretty compute constrained and is having a hard time keeping a lot of customers happy already, so maybe it would have been impracticable to serve Mythos to the public at this time. At least, Ed Zitron quotes the Financial Times (paywall) to that effect here: https://www.wheresyoured.at/four-horsemen-of-the-aipocalypse/
> If Mythos is so big that it would be cost-prohibitive for the public to use it
I think this varies greatly for different parts of the public. The person on the street who just uses Claude as a search engine probably wouldn't pay a fortune to upgrade from Opus to Mythos. But companies that are already spending tens of thousands of dollars on Claude API every month might well be willing to pay for improved performance.
I can see how that would work. But there are a lot of other abilities future AI could have that are not cyberwarfarish the way Mythos’s coding abilities are, but could be still used to do great harm. For example:
-Ability to influence people, or ability to learn how to influence an individual user (currently AI can’t really acquire permanent info and skills from interactions with users, but maybe a future advance will make that possible.)
-Ability to make deep biological interventions: The AI can understand and change a cell the way a clockmaker can a clock. Or it can function as an extra organ, fully integrated with all the connections and feedback loops of the rest of the body. (Or if those sound implausible to you, imagine some lessor abilities that are still way beyond what we and AI can do now.)
In fact it seems quite possible to me that we’re entering an era where AI’s will have many abilities that can be used to do great harm. Will it be feasible for Anthropic to handle the dangers they pose using the model you suggest for Mythos?
And, of course, not all of these models with extraordinary abilities will be coming out of Anthropic.
I am refraining here from making crass and savage jokes about ways our president and various other fools and monsters with power would want to use the new powers they'd have access to.
Rumor suggests that up until "GPT-5", each successive GPT-number release used another order-of-magnitude compute during training. GPT-4.5 was the next step along that path, but was so large that it was cost-prohibitive to serve at scale, and so they only made it available as a research preview to paying subscribers. Progress then all moved to test-time compute / reasoning models, verifiable rewards, etc, rather than brute scaling, giving us the GPT-5 series.
My strong suspicion is that Mythos is another true order-of-magnitude step along the brute pre-training scaling line, plus all the advancements in post training. The name fits into their schema of "literary work sizes" and suggests it's larger than Opus in terms of raw parameter count, which then also suggests additional compute was used (via Chinchilla / compute-optimal training).
(I realize I'm sort of conflating OpenAI and Anthropic's model strategy here, but I'm confident the same dynamics affected both).
I'm a little skeptical that this is 'the beginning of the AI haves and have-nots'. Making a really expensive model and then not selling it to hardly anyone seems like it defeats the point for these for-profit companies.
(At least short of ASI scenarios where possibly everything goes out the window)
I do think we'll see more of this sort of Glasswing-style rollouts where we give the 'good guys' a head-start on the bad-actors, at least for awhile, but I do think it's "head start" and not "permanent advantage". I don't know how long of a head-start it'll be and I suspect Anthropic is also playing-it-by-ear, too. I'd guess less than six months and quite possibly significantly less.
And it's possible this is just the new normal... but actually I'd guess we see shorter and shorter delays, while in some sense 'the security holes you will always have with you', I do think there's going to be a diminishing returns aspect to these where even as models get smarter they'll have a harder and harder time finding new holes to exploit in existing code.
Not necessarily, but thus far I am not getting a clear picture on what those companies are (allegedly) using Mythos for, nor how effective it is in practice (as opposed to on benchmarks), nor what kind of relationship those companies have with Anthropic. By analogy, if there are a bunch of YouTube influencers proclaiming that BugCo-brand stain remover is the best stain remover they've ever seen, but none of them show real-world effects on BugCo on common stains, then this alone should not cause you to update greatly in favor of BugCo (though it still should make you update a little).
I also don't get why this level of skepticism about Mythos is in the water. It's one thing to say that Anthropic's subjective descriptions of Mythos's abilities and impact may be exaggerated...
... but the claims that "Mythos might not exist" and "these 3rd party benchmarks may be fake" seem *highly* unlikely to me - just making up a model or lying about benchmark results would be an incredibly short-sighted thing for Anthropic to do and I see no reason why they'd be doing it. Yes "something something, marketing" but "let's have a marketing minute for a few weeks then completely crater our credibility for no reason" does not seem realistic to me, even before you consider questions like "wouldn't one of the third parties involved in Glasswing come out and say 'it's fake' or 'it's highly exaggerated'".
I think marketing is exactly the point: brewing up some hype for their upcoming product by claiming that it's too powerful to ever be allowed into human hands. Oh no, anyone who gets his hands on this AI would instantly become king of the world ! Too bad we're not releasing it, not for any amount of money ... ... well, ok, you twisted our arm...
Mythos does most likely exist, in the sense that it's Claude Opus 4.8 or whatever; it might even exist in the sense of being some new revolutionary model and not a slight upgrade to the previous version (though this is less likely). And it most likely does perform reasonably well on benchmarks, just like all the other models (more or less). But in total absence of evidence for its performance in real-world scenarios (other than what's presented by the people who are selling it), I am going to reserve judgement. I don't think doing so constitutes an outrageous leap of logic.
Brewing up hype for a fake product you made up (or highly exaggerated) is not a good marketing strategy and I doubt that's what Anthropic is doing.
And I'm not sure why we're talking hypothetically about "probably performs reasonably well on benchmarks, like other models". Anthropic has published the specific benchmark numbers, they're linked in the comment you originally replied to.
If you think those numbers are just lies, again, I think that is an obviously bad marketing strategy and I don't think it's likely that Anthropic is doing it. "Marketing" only makes sense if you actually sell the thing eventually at which point people will run those same benchmarks and if they get significantly worse results, that's *huge* egg on Anthropic's face for no reason - if you're going to lie, why do it in a way that people can easily check for later?
And maybe a semantic point, but accusing the company of outright lying doesn't feel like "reserving judgement" to me. That's not a neutral default position, that's one I'd only take with evidence and I just haven't seen any evidence for it.
> Brewing up hype for a fake product you made up (or highly exaggerated) is not a good marketing strategy...
Isn't this the essence of marketing, or at least advertisement ? Think of any car/medicine/stain remover/software commercial...
> Anthropic has published the specific benchmark numbers, they're linked in the comment you originally replied to.
Was anyone who doesn't work at Anthropic able to verify them ?
> If you think those numbers are just lies...
No, probably not; instead, they're probably exaggerations. By analogy, that stain remover probably does remove some stains.
> if you're going to lie, why do it in a way that people can easily check for later?
Depends on how easily they can do it. Consider video card benchmarks back in their heyday (before NVidia ate everyone's lunch), or Volkswagen's diesel emissions numbers, or (for a more relevant scenario) all those spectacular "AI-generated artworks" that took human users months of work to create by using AI. There's lots of precedent.
> And maybe a semantic point, but accusing the company of outright lying doesn't feel like "reserving judgement" to me.
I guess maybe that's the main point of our disagreement: whenever any company claims to have created a revolutionary new product yet refuses to let people independently evaluate it, my default neutral position is that they're either lying outright or massively exaggerating. I've been burned too many times before. But it's possible that my judgement is too harsh, and/or that Anthropic is an outlier.
People who don't work at Anthropic have been able to verify its performance. That's Project Glasswing. The proof is in the pudding - before the public announcement, open source maintainers noticed a sudden step-change in the quality of vulnerability reports they'd been getting. Where previously they'd been receiving an annoying stream of bad, clearly AI-generated reports of fake vulnerabilities which just wasted everyone's time, sometime in late February/March, something changed, and suddenly they were getting lots of high quality reports of true vulnerabilities, often very subtle ones that had been latent in the codebase for years. "It's probably fake/overblown marketing" is bad mental model.
...As opposed to Anthropic having, what, discovered all those security vulnerabilities manually? Or there being some large-scale fraud and none of the bugs actually even existed?
Seconding Bugmaster here. Whether it's open or closed source, many vulnerabilities are only hidden as long as nobody takes a good look, not necessarily because they were so subtle. The incentives are clear: Writing bad code is much easier than spotting it, or than writing good code. Static program analysis is not exactly a new field, and AI is at most a new tool in the box.
I am not too impressed with Claude/Mythos finding all those vulnerabilities. That is to say, it's impressive that it can find them at all, but the sad reality is that once you have a dedicated agent capable of finding security vulnerabilities, be he a human or an LLM, you're going to start finding security holes all over the place because humanity's collective codebase is riddled with them top to bottom. It is not the case that Mythos is some kind of a dark wizard uberhacker with superhuman programming abilities; it is merely the case that it has nothing better to do than sit there and sift through a bunch of boring code. You could do it too -- maybe slower, and maybe not as well, or maybe a lot better (depending on your skill level). And I suspect that while Mythos does outperform Claude 4.7 on benchmark, this doesn't translate into too much real-world gain in terms of bugs found per hour.
Do you know any human hacker with thousands of CVEs to their name? If not, then Mythos clearly is objectively a superhuman hacker at least in this respect, even if human hackers found trickier vulnerabilities before.
I find it bizarre how quickly people minimize achievements of AI models that were previously (not too long ago) said to be in the realm of science fiction.
Sure, and I did say it was impressive; and technically you are correct, Claude is superhuman -- in the same way that e.g. Excel is superhuman, because no human could possibly add up all those numbers so fast. My point is that just because Claude can find security vulnerabilities, doesn't mean that it presages the end of human civilization or anything of the sort. The problem is that security vulnerabilities are commonplace and abundant. Finding a vulnerability in the laws of physics that would allow you to make self-replicating factories (other than in the old-fashioned way) would be much, much harder, and most likely impossible.
That said though, personally I do hope that Claude/Mythos is at least in the ballpark of being as good as Anthropic claims. Maybe this way we'd finally patch some of these security holes that human malicious actors are exploiting all the time.
Many thousands more AI-hours from previous models have been spent looking for these vulnerabilities and have not been as successful, this isn't the first time anyone thought to direct AI at cyber security
> Are we entering an era of a division between the AI haves and the have nots?
Yep, that was always the plan. Giving weak AI access to the masses was just a way to raise the money needed to build the strong AI.
The masses didn't object because most of the intellectuals were busy explaining how a machine can never be smarter than a human, because machines do not have a soul, and intelligence is only a myth.
> How's that going to play out?
The bosses at the AI companies will tell their AIs "make me an immortal emperor of the universe" and maybe one of them will succeed.
That's the good ending. If, however, the analogy to natural selection holds for gradient descent and the created AI ends up with weird wants designed by noone, we're simply dead.
AI already has "weird wants designed by no one", just like every other tool ever made by humans. My car wants to turn ever so slightly to the right sometimes, instead of going straight like I want it to. My computer often wants to consume as much electricity as possible on doing nothing useful. My hammer wants to hit my thumb instead of the nail. I am not worried about hammer-pocalypse, however.
You are not using the relevant meaning of "want" here. Its about someone or something having the kind of external actions that aim to steer the world to some class of outcomes. It is meaningless to talk about wants in the context of something without intelligence.
Also, even among entities with some amount of intelligence (ie. general capability to achieve outcomes) such as current AI models, their dangerousness varies wildly with their capabilities. Apocalyptic scenarios are overwhelmingly more likely with superintelligence.
> It is meaningless to talk about wants in the context of something without intelligence.
Yes, that was indeed my point.
> Also, even among entities with some amount of intelligence (ie. general capability to achieve outcomes) such as current AI models...
Yes, technically current AI models to have some capability to achieve outcomes when wielded by a competent user; but the same applies to e.g. the toolbox that is sitting in my garage. I could theoretically build a lot of stuff with the tools in there; maybe even some dangerous stuff. But the toolbox is not intelligent despite having this capability.
Granted, LLMs are a lot more versatile overall than boxes full of screwdrivers and pliers; but that doesn't make them "intelligent" in the colloquial sense of the word. And yes, you can build an LLM "agent" which will execute some task in a loop; perhaps even a complex task that is beyound the capability of a cron job; but loops are not the only prerequisite for intelligence either. Also note that boxes full of screwdrivers have many capabilities that LLMs lack, such as attaching physical metal parts to each other using screws.
> Apocalyptic scenarios are overwhelmingly more likely with superintelligence.
It seems that "superintelligence" is usually defined as "an agent capable of causing apocalyptic scenarios", so you are correct -- but only tautologically. Ok, that was perhaps unfair; I think most proponents of the idea would say that "superintelligence" is defined as "something thousands of times smarter than a human"; but when pressed about what "smarter" means, they usually end up listing apocalyptic scenarios.
Current AI models have the capability to achieve outcomes even when not wielded by anyone. You have to set them up first, just like you have to set up humans too to produce intelligent actions: when placed in a random point in space in our universe, they simply die, if they grow up in a sensory deprivation tank, they are scarcely more intelligent than a beast. In contrast, it does not matter how you place a hammer, it won't aim to achieve outcomes.
Is it the colloquial definition of intelligence that LLMs fail to fulfill or your idiosyncratic one? What feature(s) are they missing in your view?
I highly support banning Police Robots with aggressive capabilities and preferably military robots too (although that might take a bit longer). Why? Because robots tend to be mass-controllable. So for example, if an evil dictator assumes control, he can wield robot police forces to keep himself in power; the police robots. In general, as those technologies tend to be centrally-controllable and unquestioning of authority, if anything goes wrong (cyberattack, for example), it can cause a massive disaster. Or even things like bugs and all sorts of common failures. Humans should have a monopoly on violence and firepower (at least until robots were to be extremely more reliable and incorruptible than they currently are). Not to mention, the AI catastrophe risk contribution is obvious from armed/violence-enabled robots.
So, can we get together and draft proposals for this? Maybe a small org dedicated to the cause? I think easily the AI risk/pause AI camp would immediately agree (and hopefully help). But it seems like wider society would accept as well. The objective of course is to try to pass legislation in countries for this effect, and also international accords. Thanks :)
Too late? The Chinese already have robots running races, and they seem to be improving. If you can get bipedal robots moving as well as, or even better than, humans then the applications for police and military are obvious.
US police are in practice rather decentralized. And the federal government is constitutionally prohibited from "commandeering" state/local governments to enforce federal laws. Federal law enforcement is vastly outweighed by state/local, which was why the late William Stuntz regarded crimes adopted at the federal level to be largely grandstanding without much ability to enforce https://entitledtoanopinion.wordpress.com/2018/06/23/the-collapse-of-american-criminal-justice/
Trump actually did stop after SCOTUS ruled against him federalizing the Illinois National Guard and deploying it as law enforcement against the wishes of Illinois' governor. He gave up more quickly then he did on tariffs.
Has there ever been a successful ban of any military technology that is both effective and without substitute?
- Chemical weapons are banned, but aren't effective on the modern battlefield.
- Biological weapons are banned, but very indiscriminate and therefore not effective.
- Nuclear weapons or mines are effective, but haven't been banned successfully.
If/when military robots become effective and on net better than their substitutes, they will be built and used by at least some actors on the global stage, whether they're "banned" or not.
Dum-dum bullets were banned for military use in the 1890s. Technically I guess you could say they can be substituted with regular bullets, but those aren't as effective at causing massive injuries.
I don't think we can let "several non-aligned countries are actively maintaining nuclear weapons and are ready to use them against their enemies" count as "banned" in any meaningful sense of the word. Nuclear weapons certainly aren't "banned" in those countries that have them.
In what way were nuclear weapons banned? Nine countries have nukes. The reduction treaty New START expired without replacement. China is increasing their warhead count massively. The only reasons why they haven't been used are MAD and dumb luck.
I wouldn't consider occasional use of a technology by the Russian military to be strong evidence for its effectiveness, but even if it is, that would confirm my point that any military technology won't remain "banned" for long if it's effective.
Not necessarily. This might be a WW1 situation where they just tried everything else, nothing is giving them the breakthrough, and they'll just try anything.
Considering that they are reported to have done it thirteen thousand times and that Ukrainian soldiers have acknowledged that the tactic has been effective, I would think the simplest explanation would be that the tactic has been effective.
I’m running a quant trading bootcamp in SF this coming weekend (April 24-26), at Mox. I started these up at Manifest Summer Camp nearly two years ago and we’ve been doing them every couple months or so. If you’re in the Bay Area (or heck, if you want to travel in for it) you can sign up here:
Spartacus.app (2024 ACX Grantee) with an urgent ask, seeking LOIs for the SFF grant app before Tuesday.
For anyone running an AI-safety, governance, researcher-coordination, or labor-AI org: Spartacus is applying to the Freedom and Fairness track of the current Survival and Flourishing Fund grant and would like to submit with several non-binding letters of intent to collaborate over the next 6-12 months. Submission deadline is Wednesday, April 22
What Spartacus is: conditional-commitment infrastructure applied to collective action. Individuals make binding commitments that activate only when a threshold of others also commit, with temporary anonymity until activation. Designed for the "I'd commit if enough others did" class of problems: alignment pledges, lab publication norms, researcher career switches, employee safety speech, coordinated legislative testimony, group statements across ideological lines.
Recent proof point: march2testify at Stanford in December 2025. a 500-creator mobilization for California AB 412, activated on a 100-commitment threshold, credited publicly on the landing page. First mass creator mobilization for AI legislation in US.
Honest Phase 1 retrospective since the 2024 ACX grant:
Bad: Lost two developers - 2nd one abruptly derailed project. Took off six months for paid work for cashflow reasons. Most early pilots didn't reach threshold. MVP was a marginal uplift over the prototype. I refused to juice metrics.
Good: Onboarding a new dev, brought on an Ivy League student fellow, set up at Collider NYC, preparing a narrower relaunch focused specifically on AI-safety and AI-governance coordination.
Need runway — hence the SFF application.
The ask: I'm submitting to SFF's 2026 S-Process Grant Round. Deadline is Wednesday April 22, but I'm finalizing Tuesday April 21 for a buffer. Non-binding letters of intent from orgs willing to pilot a Spartacus campaign in the next 6–12 months if we're funded would strengthen the application. I draft the LOI; you review and sign. Current LOIs in review include EANYC-affiliated folks, PauseAI US, AI legislation tracker org, and others.
If your org has a real "I'd commit if enough others did" problem and you'd consider a non-binding LOI, reach me at jordan@spartacus.app. I'll turn a draft around by midday.
Need responses by Morning of Tues April 21 to include in Tuesday's submission.
Also happy to hear from anyone who has a concrete critique of the mechanism, a pointer to a better-shaped problem than the ones I've been targeting, or warm intros to orgs I should be talking to.
For the past 3ish years I've been developing a sort of 'unified theory' that synthesises major ideas in cognitive, personality, and clinical psychology. Scott has shared a couple of my posts in the past, so I thought some of you might be interested. I'm also a practicing psychologist, so I'd really love some feedback from others with knowledge of these topics - have I missed anything important? Does this have any practical significance? Etc.
Nice write-up, I'm not a pro but it looks well-grounded and genuinely interesting. Multi-layer organization is definitely a thing; my intuition would be to tone down the hierarchical aspect of the model, given how loopy things seem to be everywhere in biology.
- I like the part about how high-level differences (big5) turn into specific 'characteristic adaptations', that seems to capture a real cross-level dynamic
- The parallels between Big5 and HiTOP looks promising, but over my head to evaluate
- The emphasis on goal-directedness can say something about the existential malaise of current culture, where the desirable goals are often unclear or contested. It explains why this is hard to bear for so many individuals, despite of the optionality that it opens. If we can trust the more speculative parts of your theory, even memory is goal-directed, yikes!
- I like the idea of the "personal myth". We need a popular name for the entire "story of I" that we build, partly implicitly, partly explicitly. This re-appears later in the section on narrative identity. There is a potential link here to things like vipassana meditation, which in its more powerful versions, is a full frontal attack on the personal myth, with significant dangers.
- The part about 4E and ritual doesn't appear so clearly integrated with the rest of the theory. Is this a sketch at collective dynamics beyond the human individual?
- Just a guess, but the Justification Hypothesis is intriguing but doesn't sound so convincing to me. If it was true, how would we even know?
"Multi-layer organization is definitely a thing; my intuition would be to tone down the hierarchical aspect of the model, given how loopy things seem to be everywhere in biology"
Fair point. I do account for 'loopiness' in that there are recursive feedback loops within and between the different levels, but it's likely that these levels aren't so clearly demarcated in reality (i.e., in the brain/mind). I think it helps to keep the hierarchical model for the sake of simplicity (and a clean diagram), but this is something I will try to make clear in future writeups.
"The emphasis on goal-directedness can say something about the existential malaise of current culture, where the desirable goals are often unclear or contested. It explains why this is hard to bear for so many individuals, despite of the optionality that it opens. If we can trust the more speculative parts of your theory, even memory is goal-directed, yikes!"
Agreed! There are a lot of real-world implications of this model that I look forward to exploring in future. I think the uncertainty management/minimisation element is an important element - it's not hard to see how this would lead to problems like anxiety and depression, and I think many of the most effective therapies are doing essentially this (reducing uncertainty or providing strategies to cope/live with it). You mentioned vipassana meditation, which might also be (in part) a strategy for managing uncertainty.
"The part about 4E and ritual doesn't appear so clearly integrated with the rest of the theory. Is this a sketch at collective dynamics beyond the human individual?"
I didn't start learning about 4E cognition until relatively late in the process of developing this model, so it still feels a bit 'tacked on'. It's clearly important, and I've tried to integrate it as best I can, but this part still needs some work. But yes, it is (in part) meant to account for collective dynamics - collective rituals and such - and also for the embodied practices we do as individuals that form part of the cybernetic processes that maintain identity, memory, goal-directed action, etc.
"Just a guess, but the Justification Hypothesis is intriguing but doesn't sound so convincing to me. If it was true, how would we even know?"
Hard to say. It's an evolutionary theory of the development of culture, so like many ideas in evolutionary psych it could be quite profound and true or merely a 'just so story'. I believe it was originally proposed by Gregg Henriques, who has developed his own Unified Theory. If you're interested you can read more about it here: https://www.psychologytoday.com/au/blog/theory-of-knowledge/201112/the-justification-hypothesis
My reaction is quite similar to Viliam below: I also found it hard to say what the point was.
I do agree with a lot of the ingredients that you explain, like the personal myth, the strive for low entropy, and so on. It's a good summary, and it certainly shows that you are knowledgeable about important aspects of psychology. But I don't have the feeling that the combination gives me more than the sum of its part.
One nitpick though: you mention psychological pathologies. I am also not quite sure what your conclusion is and how you integrate them into your theory, but at least at one point you seemed to indicate that those are consequences of a failed narration process. Perhaps this may be the cause in some cases, but I don't think that this generalizes very well. Take depressed people. They are actually very good and consistent in their personal myth. The myth is "I am worthless, I will never achieved anything, no one loves me." And they are very good at ignoring and twisting facts that invalidate their myth, just as healthy people are. I don't think the myth-building procedures are broken for them, they work as for other people as well. Of course, the myth is harming them, but many myths of healthy people are also harming them, in the sense of preventing them to have a more accurate worldview that would make them better at achieving their goals. (Though the harm is more universal for depressed people.)
Thanks for reading! I admit it's not the easiest read. It's pretty dense, and although I've tried to lay out everything clearly and sequentially (referencing some of my older posts), it's still quite jargon-heavy and difficult to understand on a first read through. I'm planning to write a shorter, more accessible version of this in future.
Fair point, there are definitely coherent but harmful myths. I don't deny that they can cause psychopathology, but I've probably over-emphasised the importance of coherence in my writeup. In the case of depression I'd say that the pathology occurs at a lower level - faulty interpretations of the world and oneself, flawed strategies, unrealistic goals, negative-biased memory recall - that then compound into a coherent but negative 'myth' at the higher level.
I'll keep this in mind when I (eventually) write my more accessible revision of this. Thank you!
Seriously though, I think my synthesis adds value. It integrates several models from different sub-fields of psychology that have (to my knowledge) not yet been integrated, but together explain more than they do individually.
Speaking from a position of complete ignorance, this is reminding me uncomfortably of R.D. Laing and anti-psychiatry. "You're not crazy, you're just misaligned with society!"
Not anti-psychiatry at all, I think meds (and other non-therapy approaches) have their place. I don't discuss this in my post, but they would presumably enact change via the neural-cybernetic level (rather than the narrative level) of my model.
As a rule of thumb, anything that mentions psychology and quantum physics in the same text is crackpottery. I only kept reading because the connection was made by one of the sources you quote, not by you, and because that part doesn't seem load-bearing, it's just an extension of "more complicated things are made out of simpler things". But I would recommend leaving this part out.
It is difficult for me to figure out *what* is the thing you are trying to say.
In the first half of the article, it seemed to me that your theory was that people are different, as described by the Big Five traits, and that their Big Five scores could be used to *predict*, if they become mentally ill, *which* specific mental illness will they have.
But then the narrative changed towards "larger things are made out of smaller things" and then the article ended and I am not sure what was the point.
Could you summarize the main insight in one paragraph?
Thanks for reading! I don't know where you're getting the psychology-quantum physics connection from - do you know which source this is? I certainly don't make this connection myself.
There's not one single insight, exactly, but here's my summary of the theory (taken from part 2):
"The narrative-cybernetic synthesis proposes that human psychology is organised as a multi-level system in which embodied cybernetic regulatory processes give rise to, and interact recursively with, memory and narrative identity to maintain behavioural and experiential coherence over time.
At the base of this framework are cybernetic systems that regulate goal-directed behaviour via feedback. These systems shape how we encode and reconstruct memory, particularly autobiographical memory, which at the highest level is arranged into a time- and context-bound narrative called a personal myth. This personal myth combines our own memories, interpretations, and culturally-sanctioned narrative templates, and is stabilised through embodied practices like ritual. It serves as a high-level reference point that organises goal-directed action over time."
I know next to nothing about psychology, so I cannot judge your theory on its merits; all I can do is ask some generic due-diligence questions of first resort that I would ask of anyone advancing any unified theory of anything: is your theory published in any major peer-reviewed publication ? Has anyone been able to replicate your findings ?
Short answer, no. I have published legitimate research under my real name, but I no longer work in a research capacity. I'm just a clinician who likes to stay up to date on recent developments.
Fair enough, but then I'd ask: assuming I knew something about psychology and/or has the time to learn, why would I study your theory as opposed to some mainstream theory that has the weight of evidence, peer review, and replication behind it ? I do acknowledge that this standard question is perhaps less applicable to psychology than to many other sciences, but it still bears asking. That is to say, is this a theory in the scientific sense of the word, or in the colloquial sense only ?
Fair question! It's a theory, or rather a synthesis of theories, in the scientific sense. There's nothing truly new here - I haven't discovered some new neural network that explains all of psychology or anything like that - it's a synthesis of existing, well-established theories in related fields (primarily cognitive, personality, and clinical psych).
What this theory offers is a novel synthesis of several different theories/models that, to my knowledge, have never been explicitly linked before. I also offer some ideas that could form the basis for future research (e.g., that what narrative psychologists call our 'personal myth' acts as a high-level reference point for lower-level cybernetic cognitive processes)
Sounds good, and I'll look forward to other researchers validating the predictions made by your theory and developing applications based on it -- hopefully medical ones. Sadly I think we will need such applications more than ever in the future...
The word agent already has too much “memetic momentum” if you will but if you really want it to catch on i’d suggest pivotting the term to mean a protocol in relation to some hardware stack? The “little guy ai” phenomenon is still in its primordial stages. Also not to get political but it’s either the best time or the worst time to be pushing the “golem” agenda, but that’s probably a really a good thing.
I don't think "agent" quite works. It's too broad and overloaded. Examples:
1. I had the golems generate a prototype.
2. Ask a golem if you don't get the joke.
3. I managed to solve this math problem without golem assistance.
I guess people would still understand each of those with "agent" swapped in, but I don't think they work as well and I think there exist cases where there'd be real ambiguity.
PS: I couldn't figure out what your last sentence was talking about. I'm like a fish unaware of water, apparently. But all 3 major golems, of course, read you loud and clear. Israel/anti-semitism/etc and that, due to attention economics, as Claude put it, the current political climate will be a tailwind for adoption of a Jewish-coded term. Is that what you meant?
PPS: Wait, I'm now guessing that the golems all missed half your meaning. Namely, that it's good to, like, show pride in things related to Jewish cultural heritage in that that helps fight anti-semitism. Arguably that even works whether one is pro- or anti-generative-AI since the golem of the original myth is kind of a Torment Nexus.
Sorry for the ambiguity, I meant to say that given the rise in anti-semitism people are likely to have stronger opinions and reactions, both positive and negative, which would create a discourse that gives the term some memetic “fuel”. Though i do think it’s important to feel and express pride in such things, and agree with the latter point.
Oh bitter irony that the golems nailed it and when I thought I was using my fancy human judgment to ferret out your truer meaning, they were right and I was wrong. Still, I'm happy to find this new point we've converged on, human-to-human!
> I guess people would still understand each of those with "agent" swapped in, but I don't think they work as well and I think there exist cases where there'd be real ambiguity
I think in practice most people would just say "AI" for each of those rather than "agent".
Yeah, that's super fair. I'm not sure I have an example where the meaning isn't plenty clear by just saying "AI". Of course "AI" is technically way too broad (just as "LLM" is technically too narrow) but if the meaning is always clear in context, there's not much actual problem to be solved. Other than insufficient Kabbalistic Cleverness.
My favorite, which I should maybe add to the original post, is how on point the golem myth is in terms of the AI alignment problem.
Memetic momentum is an interesting point. There do seem to be thresholds after which using nonstandard terminology is somehow...irritating? I'm not quite sure what I'm grasping for here. Back in the anti-reactionary FAQ I remember Scott critiquing the coining of concepts iirc because of the degrees of epistemic freedom it afforded. Maybe using your own names for things looks like that and thus triggers rational distrust? Maybe it's pretentious like saying a loanword with the pronunciation of the source language? I'm certainly guilty of this myself, and I'd like to understand it better, since I've seen peers and clients wall themselves off from people whom they might otherwise connect with behind nomenclature preferences.
Yeah, by memetic momentum I was trying to express that for a word to become “coined” it’s memetic momentum has to reach some sort of cultural escape velocity. The lifespan of a new word is typically short but circumstances sometimes allow it to survive in some cultural niche until it becomes mainstream. As for loan words being pronounced in the original language, the memetic velocity is low but slightly increases over the decades and becomes normal or decreases and is only said sparingly. I do think coining concepts is a bit cringe but also follows an innovator->early adopters->mainstream curve, but the early adopters phase invites many opinions and cultural signalling. Does it signal distrust? Perhaps by nature if it signaling some increased probability of dark triad traits but maybe that’s a stretch. I do agree that there’s something there that’s hard to pinpoint
Oh dang, yes, these concerns are on point for me especially. I tried to defend the term "golem" a bit more in the reply I posted here simultaneously with yours.
Another thought is that "agent" implies autonomy in a way I don't usually mean when talking about my use of Claude, ChatGPT, and Gemini. Of course the term "golem" also could imply agency, per the original myths. But I'm using it as more of a synonym for "LLM", or maybe something intermediate. Like Claude Code is a golem, as well as just Claude the chatbot.
I guess it feels like there's a terminological hole that "golem" is nicely filling.
If it's just a synonym for LLM, I'm not sure I see the point. I did read your AGIFriday in the hopes of a more persuasive argument but didn't really see one. Perhaps I am missing some context.
It's largely an aesthetic argument. It's kabbalistically correct? I guess in this discussion I've hit on the rationalization that "LLM" is a bit too specific. I gave the example of Claude Code -- an LLM plus elaborate scaffolding/harnesses. Another example: an image generator is totally a golem. Maybe we could think of "golem" as shorthand for "generative AI tool".
Well, I'm not in the mood to argue aesthetics :-) and I'm completely unqualified to judge kabbalistic correctness.
It does seem weird to me to try to force a word that already has a meaning, with centuries (millennia?) of connotative baggage, unless your aim is to transfer that baggage to your new concept. And if that's the aim I don't think the aim is a sensible or useful one.
Published a piece this weekend that started with a BBC essay on animal despotism — why some baboon troops are ruled by tyrants while muriqui monkeys take turns at the drinking spot. The short answer: it’s not nature, it’s conditions. Change the terrain of reward and the social geometry changes with it.
I paired that with the Singaporean futurist Chor Pharn’s idea that nations run on operating systems shaped by geography and compounding history — reflexes, not strategies — and then stress-tested both lenses inside an allegory. A blight moves through a forest along its busiest shortcuts. An Eagle, a Dragon, and a Bear each respond the way their territory trained them to respond. The smaller creatures start asking harder questions.
Readers of this community may catch a familiar silhouette near the end.
You're arguing from a single baboon troop observed by Robert Sapolsky. That's a pretty thin reed to stand on. Can't someone find a single other example? What if it turns out that there is something wrong with this study? In fact, it turns out that Robert Sapolsky is a liar. I don't know that he is lying about this one study, but...
Fair pushback on the baboon troop — it is a single case. But two clarifications worth making.
First, the piece doesn't rest on Sapolsky alone. The BBC essay by Chris Baraniuk draws on multiple researchers across multiple species: Élise Huchard's work on chacma baboon despotism, Karen Strier's decades of fieldwork on muriqui egalitarianism, Laura Betzig on naked mole rats. The argument that social structure responds to ecological conditions rather than expressing fixed nature is a finding across the field, not one researcher's claim. My essay uses these examples from three separate species studied by three separate teams.
Second, on Sapolsky specifically — the criticisms I'm aware of concern his philosophical arguments in Determined (his free will book), particularly his uncritical citation of unreplicated psychology studies. Those are legitimate criticisms of his philosophy. But the Forest Troop study itself (Sapolsky & Share, 2004) is based on direct field observation spanning two decades and includes a control troop. I haven't seen accusations of fabrication in his primatology work. If you have a specific source on that, I'd be genuinely interested to see it.
The broader point of the essay isn't "one baboon troop changed, therefore all power structures are contingent." It's that conditions — resource distribution, mobility, physical equality — shape which social behaviours pay off. That finding is well-supported across species.
But stepping back — animal research is one of three lenses in the piece. The second is Chor Pharn's argument that nations run on operating systems shaped by geography rather than chosen ideology. The third is the allegory itself, where a blight stress-tests all three power systems simultaneously. I'm curious whether the overall architecture works for you — particularly whether the blight's behaviour (travelling on infrastructure, not territory) rings true as a model for how shocks propagate through systems with different reflexes.
I strongly suspect it was written by AI, so I wouldn't get too worked up about it. I looked at several of their other posts, and they all looked to be at least largely, if not entirely, produced by AI.
I use AI as a drafting and editing tool — and find it very useful.
The research sources (Baraniuk's BBC essay, Chor Pharn's operating systems framework, a reader's Requisite Realism thesis), the allegorical structure (built across three versions since February 2025), and the editorial decisions are mine. If the prose is cleaner than expected, I'll take that as a compliment to the process. But I'd rather discuss the ideas than the tools.
Scott, is it maybe time to write up another piece defending your apparently still-positive take on prediction markets? Given the abundant recent coverage of insiders trading on foreknowledge of military operational plans and all? The throughline I'm getting from my filter bubble seems to be that they've panned out as basically just a place for insiders to anonymously profit from the information they have a social duty to keep secret and/or can causally decide the market outcome of.
I think there is some utility. I wonder if the insider angle might be ameliorated by rule changes or rule enforcement. A retroactive freeze 3 days prior to a resolving action might help - automatic implementation might be tricky - ie. rollback to holders and odds at that date?
It would help. I doubt it would be tricky. Instead of "rollback" it would be better marketed as holding trades in a "escrowed" status or such. The escrow period could be a variable dimension per market.
~~they've panned out as basically just a place for insiders to anonymously profit from the information they have a social duty to keep secret and/or can causally decide the market outcome of~~
Ideally, if that were really true, ordinary people would pull out of these markets and there would no longer be any suckers for the insiders to profit off of. I think this is basically what led to the insider-trading rules in the stock market; the markets and the government both saw value in not having a cycle of "rubes put money in/vultures fleece rubes/rubes stop putting money in/vultures stop trying/rubes put money in."
Indeed, Patrick Boyle covered this on YouTube and specifically claimed that this ~"sharks eating their prey to extinction" ecosystem destruction has previously occurred in the online poker industry. The fact that he found an example in a gambling market specifically seems apropos.
My bubble mostly points to how they become generic gambling which sucks a lot of vulnerable people into wasteful and un-sane dynamics. At this point I'm tempted to agree.
"The throughline I'm getting from my filter bubble seems to be that they've panned out as basically just a place for insiders to anonymously profit from the information they have a social duty to keep secret and/or can causally decide the market outcome of."
I find them useful as a way to get a (noisy!) estimate of the odds of events e.g
a) Can you suggest a better alternative for getting such estimates?
b) I'm a bit skeptical that insider trading is happening for markets like this. I'd expect information to leak anyway on time scales of the order of years (as with this case). For foreknowledge of military operational plans your concern sounds plausible to me.
> Can you suggest a better alternative for getting such estimates?
Your question implies that prediction markets could be justified if they produce the best estimates on at least some questions. But that doesn't consider all the externalities. Having ideal use cases doesn't automatically justify the existence of a tool that, on average, is quite possibly a significant net drain on society. This is no different from basically anything that is illegal - one can usually imagine a valid case that is frustratingly not allowed by a given law.
Many Thanks! I'm skeptical about the 'net drain' in the first place. In a tiny handful of cases, such as the one you cite re military operational plans, I'd support a restriction on the market.
Legally, US law is effectively pretty ambivalent about private information, giving preferred treatment to employee stock ownership plans, and employees generally have _some_ degree of private information, while also enforcing trade secret agreements.
Personally, I almost always prefer to see insider information leak, so that the general public has a way to access it - even through a noisy channel like a prediction market. I also prefer to see trade secrets abolished, with technical advances incentivized via patents or copyrights, and with secrecy agreements made unenforceable.
Isn't this the whole point? You want those with relevant information to reveal it so your predictions become more accurate. Obviously, those trying to keep secrets need to worry about their secrets leaking this way, but "I know enough to predict when the US will start bombing Iran next so I will place a bet on that in the prediction market" is the system working as intended.
In theory you are right we want insider trading to get the information.
But everything I encounter about prediction markets outside my rat bubbles are that 1) obviously insider trading is bad, and 2) the prediction markets agree with point 1.
But eventually this kind of behavior will corrode willingness for others without inside knowledge to bet on the other side (since they no longer have a fair chance to win or lose), which reduces liquidity of the market.
If someone who can control the outcome of the market bets $1B 30 seconds before taking the action that resolves the market, that does not inform society on important information in any meaningful way.
Furthermore, a lot of these things are supposed to be secret for a public-good reason. You can argue that it's good for the military to attack and destroy a specific target, you can argue that it's bad for them to attack and destroy that specific target. But there's no world in which the best possible outcome is that they attack that they attack that specific target, but the attack is foiled because the details were leaked on predictions markets, and both sides spend materiel and lives to no effect except for insider traders getting rich.
And that's not even touching on the incentives to make the world anti-inductive... no small number of pundits have pointed out that the administration has seemed to have an incoherent and ill-conceived approach to this war, but it would make some sense if the actual motive is to repeatedly act as unpredictably as possible in order to maximize the returns from insider trading on prediction markets.
Obviously they're supposed to be secret, and if we can catch the folks who placed those bets (which ought to be doable), they ought to get a nice long prison sentence. But the whole point of prediction markets is to get people who have insight into what's going to happen to make their knowledge public via placing bets in the market.
You could solve the "30 seconds before" by limiting bets to a certain time before the event resolves. This seems like something prediction markets might want to do to maximize their information-discovery abilities.
Presumably, regulation will have to catch up to this stuff, in much the same way that athletes and team staff (trainers and such) betting on games needs to be prevented somehow.
Another possibility is to ban some bets because they incentivize disclosing (say) classified or other confidential information. Like, you might want to ban bets on things like the exact time/place of a military strike, the medical diagnosis of some famous person who's rumored to have a health problem, the name of an alleged rape victim or a minor involved in some crime, etc.
> Presumably, regulation will have to catch up to this stuff, in much the same way that athletes and team staff (trainers and such) betting on games needs to be prevented somehow.
They're already illegal, it's just that prediction market fans have successfully convinced the Trump administration to not bother enforcing the laws.
The stock exchange went through a transformation to deal with these problems and i’m sure some administration will tackle it in the “near” future. But I certainly wouldn’t count on the current administration being the one to do so. Prediction markets are a cat out of the bag situation and regulators will tackle these problems but can’t erase them. Besides the insider trading, it does seem like an amazing opportunity to make money in a structured way. I can’t imagine all these firms giving their traders portfolios to trade on at the size we’re seeing unless this was the case. I agree some insider trading occurs but a lot of algorithmic opportunities to be had
Another reason why the two are not the same is because prediction markets are entirely abstract and can get very specific. People are betting on "Will this other prediction market come true?". You cannot do this sort of bullshittery with stock.
Yeah, no, still not the same thing as "Will this one specific bet become x% true by y date" or "Will this person be dead by date x" or even "Will jesus christ be reborn"
No. The ability to put a prediction market on specific geopolitical events may directly alter the likelihood of that event. E.g., if I create a betting market on the likelihood of a drone strike in Kyiv, does that increase the likelihood of that drone strike occurring? It might!
This is substantially different from something like oil futures, which might serve as a proxy for geopolitical decisions, but are less directly implicated in the outcomes.
No, I am not making that argument. I am arguing that prediction markets may influence events beyond what is already influenced by oil futures. My example was a drone strike in Ukraine. Is your point of view that a prediction market on a drone strike has zero causal effect on the outcome, or that the effect is already perfectly captured by oil futures?
Why would this influence a drone strike? You can bet both ways in the market. Whoever is in charge of the Russian drone strikes knows where the strike is going and can bet either way. He will just pick the strategically correct target and if the drone strike goes to Kiev he bets "yes" or if it goes somewhere else he bets on "no" and makes money either way.
It would be shocking if major players' positions in financial markets didn't affect their actions. But one limiting factor in purely financial arrangements is that I'm not willing to spend more than $1 to protect my $1 bet.
Prediction markets involve stuff measured in stuff other than money. You don't want the decision to bomb some country to be determined by people in the administration who have placed bets on us bombing that country, or the decision of whether to sentence some notorious criminal to death to be determined by a judge or juror with a bet on the outcome.
The other side of this is that if people use prediction markets as a truth oracle, there's a lot of incentive for people to be willing to lose money to influence the truth oracle--again, their concerns aren't just about money!
You can bet both sides in the market. So if you are a decision maker you can profit both ways. The decision whether or not to bomb will be made, and then the decision makers will place bets before the public knows the outcome, but it doesn't matter which outcome it is because they will make money either way.
Why should I really care if the execs at some company make side money by betting on outcomes they already know? Sure, it's not fair, but it doesn't actually hurt me. It doesn't even hurt the people on the other side of the bet. They guessed wrong and would have lost anyway. It just changes the person they lose the money to from a lucky idiot to a smart insider.
Our current system gives the profit to whatever quant fund is running the colocated servers that get the news and turn it into a buy/sell order the fastest. I fail to see how this is really any better. I have exactly the same chance at beating their computer as I do at beating the insider trading executives. Long term investors are completely unaffected.
The sec doesn't regulate commodities markets. That's the cftc, which is (a) generally more lax about it (it's a lot easier to get away with oil futures insider trading - in many cases, like oil producers or airlines doing it, it's outright legal), and (b) also regulates prediction markets (which have been pushing back on insider trading, at least the legal ones).
What reason is there to think they "may be"? What laws treat them that differently? Or is it just that Polymarket is banned in the US and Americans use it illegally?
“some of the twitter commentariat is innumerate” is not the primary shocking takeaway I would from garner from this sequence of stories. This is a pretty horrifying investigation result and the claims in the actual CNN article look well backed up.
CNN doesn’t know, or claim to know, how many people are involved in the metaphorical “rape academy.” CNN didn’t “imply this figure” is 60 million. Nor did they claim that the actual number was “nearly a thousand.” That’s the number of people who were at one time on a group named “Zzz,” but Sandrine Josso, who created the metaphor, explicitly says it refers to “groups” (plural).
The above doesn’t stop people from seeing a number somewhere in the article and jumping to the conclusion that the number they spotted is the number of people in this academy. People have probably been doing this since the invention of the printing press, but it is only with the rise of social media that people have the opportunity to share these leaps of illogic with people outside their immediate friends and family.
Unless they just changed it, "this group of nearly a thousand users" is at the beginning of the article, in the fancy graphical part before the main body of text. It doesn't read as intentionally misleading to me at all. I don't think it's fair to blame CNN for the extremely poor (or motivated) reading comprehension of some people on social media.
I’m seeing a distressing number of people latching onto “30 million,” or in at least one case the ludicrous “92 million” and responding to a factual correction with (more or less) “THIS is why we can NEVER TRUST MEN (tm).”
I’m of the Old School who believes an acceptable number is zero, and that the only difference between the least unacceptable number and the actual verifiable number is how many instances of Cruel and Unusual punishment I’d like to see meted out in a public square.
If you believe that wanting to see a thousand or so men executed rather than millions means you believe I despise my wife, daughter, sisters, more nieces than I can remember without assistance, I’m okay with that.
If there is a gap between the politics you are driven toward by "News Article" and "News article plus takes of stupid people I don't know in person," that gap will be used to manipulate you very cheaply and efficiently.
Yes, you're right. My bad. I remembered reading it last week so I included that in my original comment, but couldn't find it when I searched just now, so I updated it. It looks like it's just the fancy formatting not being searchable when you've scrolled to the main article; I've removed the update. Thanks.
I have to say I found the early graphical stuff super annoying and just skipped past it. I suspect I'm not alone. If CNN is covering its ass with non-searchable text, well, CNN.
I agree that the early stuff is sooo heavy-handed and annoying. As for the wording of the piece, I think it's only a little unclear, and it doesn't seem to be written with that intention. See this quote from the article:
"One porn site, Motherless.com, is home to more than 20,000 videos of so-called “sleep” content uploaded by users, with hundreds of thousands of views."
"Hundreds of thousands of views" doesn't align with 62 million visitors, and it's clear that it's only a portion of the website's traffic. I do hope CNN issued a clarification though, I think that would be a responsible thing to do if the false claims gained enough traction.
That said, the article is still incredibly disturbing, and I can see why people reacted so strongly. I find CNN's treatment of it (with the graphics, etc.) to be almost unserious in how melodramatic it is -- the content is disturbing enough on its own in plain text.
Yeah, despite not being very familiar I've always been partial to Montessori, I recently realized part of the reason is probably that I'd have been so much better off in a well-functioning environment built from that pedagogy. I wanted to die as a kid, partially because going to school was so dreary and monotonous that the thought of having to do it for so many more years felt overwhelmingly daunting lol
For me, I think about what has eternal value and will still matter in 1000 years, and focus on that. When I need a break, I start looking at other things, but I end up getting bored and unable to continue with them when I remember their ultimate worthlessness.
> For me, I think about what has eternal value and will still matter in 1000 years, and focus on that.
So... what even fits that description? Because absolutely nothing comes to mind for me. All things are temporary and fleeting, but there's nothing wrong with caring about them regardless.
I feel like I'm at capacity anyway. I have dealt with it in various ways but the most successful has been to prioritize some of my "fields of interest" (really, these boil down to habits I'm trying to form) and then set monthly goals & track them using a spreadsheet. I have been doing this for the bulk of a decade.
At the end of each month I can look and see whether I fell short, met, or exceeded each goal, reflect on the reasons, and adjust the goals for the next month if necessary. I can also add new "fields of interest" (aspirational habits) if I feel I have room for them. Over the years I've gone from having like 4 items I'm tracking each month to around a dozen these days.
I never think about the spreadsheet except for during the 30 seconds it takes me each morning to update it.
The first thing to do is see if any of them will go away, which is the easiest solution. If not, the second thing to do is pick a bare minimum amount of effort to put into each of them, like, five minutes a week, and make sure all of them get that tiny block of effort. Once everything's got the minimum, then you pour the remaining time into whichever one is most fun, until that one's finished and the time's freed up for others. The third thing to do is see if any of them have gone away now.
I'm also finding that a cool-down journal entry helps kill interest. Write out a summary of what you've done or what you want from each of these fields of interest, and see if that's enough to make any of them go away.
Depends what you mean by deal with i guess? If it’s just the feeling that there isn’t enough time to gain substantial depth in certain topics and you should prioritize the “important ones”, then I would say working through and removing negative feelings is the best thing. If we’re talking about how best to maintain different fields of interests , over the years i’ve found that trying to stick to an organized structure for more than say 1-3 topics isn’t actually realistic and you should simply burn every ounce of passion and squeeze every second of motivation out of the brief time period (be it hours or days but usually not weeks) that a current interest feel “on top”
Yeah, I suppose in practice your latter option is what I actually do, which, I think, is also how I ended up this interested in too many different things - something is relevant (at the time) to research, or it, whatever it is, catches my fancy, so I read up a little or a lot and then move on. What lead to feeling like this is a problem that needs dealing with is that I recently ended my marriage (I also don't have a job) and so now have more free mental capacity/time/energy than I've had in years, and I find myself spinning a little in wanting to plunge into all kinds of different subjects/fields as I rediscover all these things I like, it's simply a bit much. I don't really judge myself that way (I've worked very hard on myself for many years), just a little unsure of how to cope/channel/deal with this new(re)found passion for, uhm, almost everything, lol. Thank you for your response and sharing your thoughts!
It's a bit overwhelming in and of itself, and then I suppose my concentration isn't the best right now (for a variety of reasons), plus my brain does really enjoy novelty and variety, so right now life mainly looks more or less like a mix of everyday tasks, introspective navelgazing and wild jumping between interest based activities, which feels a little too like... unfocused spinning, I guess. I've also realized I'm much more goal-oriented than I thought and that I've been setting goals with and for someone else for so long that suddenly being without is hard (to the degree that taking goals that are just by and for me seriously doesn't emotionally make sense to me anymore), which definitely contributes to this feeling of being a bit too all over the place.
And thank you (and everyone else) for giving your input, I did not expect so many responses attempting to help and I really appreciate it!
Sorry, what is the purpose here ? If it's to secure your email address, wouldn't it be better to just configure your email client/proxy/whatever to whitelist anyone you wish to communicate with and automatically filter out the rest ?
I didn't set up the email address, someone else did. If you fail to transform the formula correctly, you won't have the right email address and will fail to reach the intended recipient. If you do, the recipient infers you are someone worth receiving emails from (at least before they read what it is you actually emailed).
> If you fail to transform the formula correctly, you won't have the right email address...
I'd need to know what S was to solve for V, wouldn't I? So I don't see a way to transform the formula, per se. Or do they want the first derivative of this equation?
And would the derivative symbol and brackets be recognized by a mail server? My experience with mail servers dates back to the Bat Book days, and I don't remember that mail servers could handle those characters. But I'm an old fart, and a lot of this newfangled technology has passed me by. ;-)
No, just not the target audience for the job posting. This term has a standard name, which can be written entirely in simple ASCII alphabets with no special characters whatsoever (and in all modern implementations of email, case-insensitive), which I'm sure is what they expect you to substitute in.
If it's remailing to yet some other email address, we don't know, and if we were supposed to know they presumably would have said as much. The person sending the email to that address doesn't need to be concerned with what happens afterward.
You mean the "is able to open claude.ai and do basic copy&paste" check?
Prompt: What email address does this represent? {∂^2V/∂S^2}@mnx.fi
Reply: The expression ∂²V/∂S² is the second partial derivative of option value (V) with respect to the underlying asset price (S) — which in options trading and the Black-Scholes model is known as Gamma (Γ).
So the email address is:
<<not gonna post the cleartext address here so that web crawlers don't see it>>
By "unauthorised users", do they mean hackers? Because it would be very funny if the shiny new tool they announced with fanfares as the thing that discovers all the exploits got itself exposed by an exploit.
From the original Bloomberg report (https://archive.ph/tV1NJ), it sounds more like a WarThunder leak situation. Large organization has a lot of people; some of them include young men; to impress his mates, one young man decides to show off the access he has to classified info/unreleased LLMs; security breach results. Some quotes from the original report:
"A handful of users in a private online forum gained access to Mythos...
The users are part of a private Discord channel that focuses on hunting for information about unreleased models...
The users relied on a mix of tactics to get into Mythos. These included using access the person had as a worker at a third-party contractor for Anthropic...
The group has been using Mythos regularly since then, though not for cybersecurity purposes... the group is interested in playing around with new models, not wreaking havoc with them, the person said."
That is even funnier. They go public with this product and sell it on the hype about "all these hundreds of hidden vulnerabilities that our AI was able to find! imagine if cyber-criminals can use AI, too! buy our services to prevent attacks on your wide-open, full of holes, systems!" and it turns out that they can't even handle their own opsec.
Deiseach, stuff like this is no joke. I went with the War Thunder angle because it's more relatable and lighthearted, but my first thought for an analogy was actually "Guy brings home his gun from work to show off to his mates, they pass it around to ooh and aah, one of their kids steals it and shoots up his school." You can joke about it now, but one of these days that exact scenario is going to happen: "Guy brings home his access to a military paperwork LLM for planning out military operations, to show off to his mates. They pass it around to ooh and aah. One of their kids uses it to plan out the school shooting he's been thinking of." It's all fun and games, sure, but the gun is no less real when it inevitably shoots someone.
I do think it's funny, in a blackly humorous way, because they released this fanfare of implied "our product is so good, you better sign up quick and get it before bad guys use exploits to blow the doors off your confidential data, look at all the naughty-naughty we could do if we were so inclined" - then they couldn't keep their own confidential data protected.
Hubris, in the Greek tragedy sense. Glass houses and stones. And don't try to make money off other people's misfortunes.
Boy, The Amazing Kreskin sure screwed the pooch last night.
I really liked Google's etymology and dictionary/thesaurus overview. The new AI one that's replaced it is shit.
When I previously complained here about "open house" signs never telling you when the house is actually open, people suggested that realtors only put the signs up while they're there, but that has *never* been my experience.
Anyway, this afternoon, I saw a new one with not just the usual sign in front of the house, but signs up all over the neighborhood directing people there, and I figured that since it was a Saturday afternoon, there was at least a slight chance they might be open, but nope, house locked with nobody around, just like always.
It's just so stupid and lazy of realtors. It's not like making it impossible for people to visit actually helps them anyway.
No phone number for the realtor on the sign I suppose.
If it's any comfort, I think they're a dying breed. Like travel agents were.
Hey! Don't post much but I think there's people here who would appreciate the following opinion piece: there will be a scientific theory of deep learning
https://t.co/92nSIHameW
(Arxiv link).
Fair warning, I've been working with some of the author's for the past couple months so this is at least in part an advertisement for the paper. But also, I know a year ago I would've really appreciated this paper--I was ust graduating undergrad with a physics degree, cared about alignment/ai things but didn't really know what my skills could be applicable to. I think this paper makes it really clear what a scientist (maybe physicist in particular, but I think most people trained in some aspect of stem could benefit the field in some way) could do to positively impact the current field of AI.
If you're interested, learningmechanics.pub includes resources for breaking into the field and understanding some of the current open questions that we think are plausible and valuable targets for researchers right now.
Just use the arXiv link directly: https://arxiv.org/pdf/2604.21691 !
The title reminds me of Hilbert's famous "Wir müssen wissen, wir werden wissen." We must know, we will know.
I heartily endorse the sentiment, but not with the implication that such a theory is imminent. It might very well take decades (though almost certainly not centuries).
Ah, thanks for the direct link! You're right, my bad I just copied off of the Twitter thread about it. Thanks for taking a peak!
I read Omelas in high school, and I always thought the point was that people may feel morally superior for walking away from a society like Omelas, but they really aren't making any difference without rescuing the child. If a society where a bunch of monsters get to enjoy life crumbles, so be it.
Derek Thompson, an author who has built a strong subscriber base on Substack, posted an analysis similar to (perhaps partly inspired by?) Scott's writing about the "vibecession" among US voters. It's free to read, I think:
"If America's So Rich, How'd It Get So Sad?
Or: How the 2020s broke our brains"
https://substack.com/@derekthompson/p-194392593
Situation-analysis points all backed up by field data:
"the decline in self-reported well-being since 2020 has not been concentrated among young people, poor people, or unmarried people....Instead, the decline in happiness has been an across-the-board 10- to 15-point decimation experienced by practically every demographic."
"the rise of religious non-affiliation in America has been a steady 30-year trend, whereas this falloff in well-being started in 2020", when self-reported secularism stopped increasing.
"low-income wage growth has been unusually strong since the pandemic....Median household incomes are higher now than they were 10 years ago....[meanwhile] some of the largest declines in happiness seem concentrated among well-to-do demographics, like older people, white people, and college graduates."
"the rising misery of young people—often rightly associated with rising phone and social media use—has been going on for about 15 years. The more sudden collapse in general wellness [across all ages and demographics]...points to an emotional break that happened around 2020."
Thompson then constructs his own theory of what the nationwide data is continuing to show.
...okay, I don't know how to read that graph, because that looks like it's saying 2021 had reported happiness below 0%.
His conclusion contains one sentence which isn’t quite right: “Inflation has made today’s life harder to afford.” People dislike inflation, but that’s not because inflation is creating an affordability problem. See this graph:
https://fred.stlouisfed.org/series/MEPAINUSA672N
One explanation for why people hate inflation is that they don’t think of wages as the price of labor, so they don’t see the connection between inflation and higher wages. They *think* inflation is reducing affordability even though it isn’t. Another is that inflation makes the world feel less stable. You may be getting raises that keep up with inflation, but you can’t be certain that that will continue to happen in the future. A third explanation is that Trump promised to bring down prices and didn’t, so what they are really upset about is being lied to.
Generally it's the second of your three choices, combined with a general sense of helplessness that inflation gives a lot of people. I'm not one of those but many people, up and down the socioeconomic ladder, are.
Your third thing varies based on whether a given POTUS explicitly ran on that issue. Hence it is a headache for Trump now. But a different way for it to enrage voters is if an administration seems to be ignoring the topic, e.g. Biden in 2023.
While I have constructed plenty of hypothetical prompts to test LLMs, I recently got a stronger response than I’ve ever seen before, and it was on real data.
I’m like, can you help me write a letter to my endocrinologist? Here’s my history of blood test results, current medications, current symptoms … and I got in response from the LLM (I paraphrase here) … go to the Emergency Room _right now_, do not wait until tomorrow morning, do not try making any excuses. It was being a bit overly cautious, but I could see it maybe had a point. (the endocrinologist on duty in the ER authorized increasing my dosage of thyroid medication, if you’re wondering what happened)
To whom it may concern: Substack seems to be having a glitch ("Something went wrong."). If you see it after posting, copy whatever you wrote, and reload the page first. Whatever you posted may have gone through anyway; if you try again, you'll end up double-posting.
Also, if you try to delete the double comments, it'll show the same message. Don't worry, it's been deleted.
The following people are, or have been, famous as aviation or space flight pioneers, specifically a pilots or astronauts who are attached to major "firsts". I've noticed anecdotally that some achievements have had a longer reputational shelf-life than others.
Charles Lindbergh
John Glenn
Chuck Yeager
Neil Armstrong
Yuri Gagarin
Alan Shepard
I've presented the names in random order. I'm interested in the following:
1. Your approximate age, so I can see if there's a generation gap in responses.
2. Which of them you remember, off the top of your head, what they're famous for.
3. Your gut feel of approximate order of significance as aerospace pioneers. A couple are also notable for other things (political careers, etc), but don't take this into account for the purposes of this survey.
Age is 33. I think Lindbergh crossed the Atlantic solo, John Glenn and Al Shepard I always mix up but I believe one was the first american to fly to space, the other the first to orbit. Chuck Yeager broke the sound barrier, Neil Armstrong landed the Eagle on the moon, Yuri Gagarin was the first man in space.
It's hard to say about the order of significance, gut feeling I put Gagarin and Armstrong above Glenn and Shepherd, and I'd probably put Yeager above Lindbergh because I think first in space, first on the moon, first supersonic in level flight, all require less qualification than 'first non stop across the atlantic *solo*' or 'first *american* in space/orbit', and I think I'd put Gagarin and Armstrong above Yeager because space feels more impressive to me than atmospheric flight.
34. I remembered all of them except John Glenn. I knew he was an astronaut, but not the specific achievement.
My gut feeling puts them in the (ascending) order of Lindbergh, Yeagar, Shepard, Gagarin, Glenn, and Armstrong - basically in order of how high and fast they went.
Although I'm not sure if Shepard should get pioneer credit for doing what Gagarin already did.
I only remembered John Glenn from his return voyage in the 90s, since I was actually alive then.
I've only heard of Lindbergh due to the kidnapping and pro-Nazi views, but I was able to remember what all the others did.
Age ~35
37
Armstrong, Gagarin, Yeager, Lindbergh. I think Shepard was American astronaut, but I don't recall details.
Equally mid-low, maybe a slight bias to Lindbergh - the rest are the faces of monstrously huge government programs, and are high talent but very cog-y and low-irreplaceability.
1. 24
2. Lindbergh crossed the Atlantic solo first (thereby inspiring the name of the Lindy Hop, the original swing dance form), Glenn not sure (but was on some of the Apollo missions?), Yeager not sure, Armstrong was first on the Moon, Gagarin was first to "space proper" maybe?, Shepard might have been first to some lower space-like altitude.
3. Armstrong > Gagarin > Lindbergh > Glenn > Shepard > Yeager
1. 30
2.
- Lindberg -> Spirit of St. Louis first (solo?) flight across the atlantic.
- John Glenn: First American to orbit the earth on Mercury mission.
- Chuck Yeager: Broke the sound barrier
- Neil Armstrong: First man on the moon (Apollo 11)
- Yuri Gagarin: First human in space
- Alan Shephard: Involved in some of the Mercury/Gemini/early apollo missions, don't remember what exactly.
3. Most to least significant: Armstrong, Gagarin, Lindberg, Yeager, Glenn, Shepard
33
All of them, obviously. (I'm the sort of nerd who can name the crew of every US space mission through 1972, and have a decent shot at the backup crews.)
Armstrong, Gagarin, Lindbergh, Yeager, Shepard, Glenn.
1. Early Thirties
2. Charles Lindbergh, Neil Armstrong, Yuri Gagarin.. would recognize Chuck as being flying related without context but don’t recall exactly what he did. Other two names don’t ring a bell.
3. Neil Armstrong, Yuri G, Charles L
1. Age, fifty-something. Have spent most of those years as both a pilot and an aerospace engineer, so...
2. All of them, with great specificity
3. In order of decreasing significance,
Armstrong
Gagarin (almost a tie for first)
Lindberg
Yeager
Glenn
Shepard
43. I knew all their firsts. But if you hadnt given me the list I might not have been able to produce shepards name. Also I wasnt sure if Glenns first needed the American qualifier.
Yeager, Lindberg, Glenn, Gagarin, Shepard, Armstrong.
I am putting Yeager as first for sheer grit in pushing the accelerator through a barrier experts gave a decent probability to blowing him apart. And they had no way of running that test without a human pilot at the time. Everyone else risked death but the engineering parts were better tested.
I put Glen at the top of team space not for his first but the totality of his involvement in the program.
I put Lindberg also high for the totality of flight contribution. I dont think the solo flight itself is of particular importance.
Armstrong ranks first psychologically but last practically. The fact that we haven't been back to the moon shows what a standout achievement it was but in his own words his was but a small step
Edit: this was my gut impression. Follow up research, I am not sure how important Lindberghs career work in different aviation engineering concepts was, but he did dabble in a lot
I'm struck by the absence of Amelia Earhart on this list.
There are several people I considered including but didn't. Earhart was one of them. The others were (in no particular order) Wilbur Wright, Alexi Leonov, Ed White, Robert M. White, Valentina Tereshkova, John Young, Sally Ride, Douglas Corrigan, John Alcock, Arthur Brown, and Scott Crossfield.
What about poor Laika? Like the kid in My Life as a dog I think of her sometimes. Actually I didn’t know about her till i saw MLAAD.
~30 years.
I don't know the names Chuck Yeager or John Glenn. But I bet one of them is the first to break the sound barrier.
Gagarin - first person in space.
Shepard - first American in space.
Lindbergh - first solo transatlantic flight. (Now that I say that, I guess there must have some team flight earlier, but I don't know who did that one.)
Armstrong > Gagarin > sound barrier guy > Lindbergh.
Alcock and Brown were the first to non-stop the Atlantic.
Good guess!
Chuck Yeager was the first person to break the sound barrier. John Glenn was the first American (or first non-Soviet, if you prefer) to orbit the Earth.
Technically early Millenial, but culturally very much late Gen-X. From memory:
Charles Lindbergh was an aviator, and the first to do something - I can't remember what, maybe fly the Atlantic? His other claims to fame are that he was an influential fascist and he committed the famous Murder On the Orient Express.
John Glenn was an astronaut. I think he may be one of the 12 men to have walked on the moon, but beyond that I got nothing.
Chuck Yeagar was a... I want to say a fighter pilot, but he later became something senior at NASA, so maybe an astronaut? I think that he was exceptionally decorated, and I have a vague feeling he was involved with discrimination against the first black man to be accepted by NASA?
Neal Armstrong was the first man on the moon, in 1969.
Yuri Gagarin was the first man to orbit the earth, in 1961.
Alan Shepard was another astronaut - I think possibly one of the crew of whichever Apollo it was that go a film made about it failing but no-one dying? He gave his name to the pilot of Thunderbird 3, which I think means he probably walked on the moon?
In terms of significance, I would say that Lindbergh, Armstrong and Gagarin are probably more significant than the other three, but probably not classify them beyond that.
Chuck Yeager was the first person to break the sound barrier, and was very much a legend in US Air Force. I think the main reason he was never accepted to be an astronaut was because if he died in a spacecraft accident, the PR scandal would have been too great.
> I think the main reason he was never accepted to be an astronaut was because if he died in a spacecraft accident, the PR scandal would have been too great.
If I remember "The Right Stuff" correctly (the book, not the movie), it's because Astronauts were required to be college-educated, and Yeager wasn't. Besides, Yeager continued to be an active test pilot – which carried high risk of death back then – after his feat became public.
GenX, remember all of them from reading about them.
My sense of importance depends on how much of themselves was necessary to achieve their feat. In that sense, Lindbergh would be the most significant on that list because aviation was the least developed at that time, so each individual had a relatively larger share of any achievement. He was first among firsts, so to speak.
In my 60s; I know what four of them did and the other two, Glenn and Shepard, I can guess what the things must be but don't know which did which. My sense of importance is vague so I pass on that.
Age ~= 40.
I recognize Charles Lindbergh, Yuri Gagarin, and Neil Armstong, although I'm not like 100% on what Lindbergh did. Something like first dude to fly a plane across the Atlantic.
In order of importance:
#1 Neil Armstong
#2 Yuri Gagarin
#3 Charles Lindbergh
Interesting essay re consciousness and human souls by Nicholas Humphrey. — about 4,000 words
https://aeon.co/essays/you-know-what-consciousness-is-you-live-in-soul-land?ref=thebrowser.com
Nice bit of fiction. I do so enjoy when scientists start bloviating on theology with no idea what they're going on about. I'd never heard of France's "Penguin Island" but it does not pose any difficulty for Catholicism, for a start; there is no embarrassed God now compelled to grant souls to penguins. Baptism does not count, you can't baptise animals. The blind hermit made a genuine mistake so it wasn't malicious or mockery (like a Satanic parody of the sacrament) but the ontological change can't happen.
I'm not going to opine on neuroscience or consciousness because I am aware of my ignorance on the topic. I wish scientists would return the favour, but eh. Scientists. Too puffed up with their own self-importance ("science explains EVERYTHING! science is the only real reality! I, as a scientist, am the Number One Expert on *everything*!")
Though I see the author of the article is a psychologist, and so what 'real' scientists would not recognise as engaging in a scientific discipline at all. The article also seems to be shilling for his books, so okay:
"Theoretical psychologist and Darwin Bye Fellow Dr Nicholas Humphrey has been announced as the winner of the Dennett Prize 2026.
Given for significant advances in philosophy of mind, cognitive science, evolutionary biology or artificial intelligence, the prize is awarded by the International Center for Consciousness Studies. It is named in recognition of American philosopher and cognitive scientist Daniel Dennett, who died in 2024, and with whom Dr Humphrey collaborated closely."
You go, Nick. You and Danny D.
My library had an ebook copy of of Penguin Island. I’ll give it a look not sure if I’ll read it all though.
I don't like that the Democrats are gerrymandering Virginia, but I like a world where only the Republicans gerrymander even less.
Bilateral disarmament > proliferation > unilateral disarmament.
Congress needs to pass an anti-gerrymandering bill.
If someone is still going to be drawing the maps of districts, any proposed "anti-gerrymandering" bill would just be an attempt to shift that in one's own favor.
Anti gerrymandering bills typically have the maps drawn by multiple someones, to limit the amount of influence one party can have.
For example, the recent bill that got defeated in Ohio would have had the maps drawn by a commission of 15 members - 5 Democrats, 5 Republicans, and 5 independents. (And to prevent shenanigans over "but who counts as a Democrat?" the commission members are themselves selected by a bipartisan panel of retired judges.)
>Congress needs to pass an anti-gerrymandering bill.
Wouldn't the US need a Congress first? Baby steps.
Apparently SPLC in the past used paid informants for information on hate groups. According to AP, "the group said the program was used to monitor threats of violence and the information was often shared with local and federal law enforcement."[1]
DOJ is now indicting them about it. It's conceivable that the way it was done could be unkosher. But it seems unlikely that DOJ really trusts their own case, given that their tweeting is trying to make it seem like a totally different issue, flailing and pretending SPLC was propping up these groups.[2] Ofc the real motivation is that Trump's allies, like Elon and the Christian nationalists, hate that SPLC called hate groups what they are.[3] Here's SPLC's own statement, disclosing the paid informants program shortly before the weaponized charges came out. [4] Ken Klippenstein remarked that "NSPM-7 is here."[5]
[1] https://apnews.com/article/southern-poverty-law-center-criminal-investigation-db7fdcf9baa0d1b24b8f1e1f2cebc0be
[2] https://x.com/TheJusticeDept/status/2046704517119377577
[3] https://x.com/elonmusk/status/1973873845384991115
[4] https://youtu.be/25dlBorkAy4
[5] https://x.com/kenklippenstein/status/2046724992948138199; see also https://www.kenklippenstein.com/p/trumps-nspm-7-labels-common-beliefs; https://www.aclu.org/news/national-security/how-nspm-7-seeks-to-use-domestic-terrorism-to-target-nonprofits-and-activists
The docket is here; so far the only filing is the indictment: https://www.courtlistener.com/docket/73223872/united-states-v-southern-poverty-law-center-inc/
Even assuming that the DOJ can prove beyond a reasonable doubt that all of the factual allegations made in the indictment are true, a conviction seems unlikely.
The alleged violations of 18 U.S.C. 1343 and of 18 U.S.C. 1956(h) are based on the idea that SPLC defrauded donors. The SPLC revealed that donations would be used in part to investigate hate groups. Methods used to perform these investigations included paying informants and at least in one case illegally accessing records of a hate group. The SPLC did not tell donors that it would be using these methods.
Although it is in principle possible to lie by omission, it’s a huge stretch to argue that that happened in this case even if all of the allegations listed in the preceding paragraph are true.
The alleged violations of 18 U.S.C. 1014 involve opening bank accounts in the names of fake businesses. The basic problem with these counts is that misrepresenting the owners of the accounts didn’t conceal a financial risk to the bank.
In other words, if you overstate the value of your assets when applying for a loan, but pay back the loan when it is due, you haven’t caused the bank any financial harm, but you can still be convicted under 18 U.S.C. 1014 because the lack of financial harm to the bank is a matter of luck. If a financial crisis had occurred before you paid back the loan, the difference between what you claimed your assets were worth and what the assets were actually worth might prove to be the difference between you being able to pay back the loan and you being forced to default on the loan.
On the other hand, what SPLC allegedly did could not cause the bank to lose money no matter what happens to the economy, and it doesn’t seem to me that 18 U.S.C. 1014 applies.
I don't need the DOJ to tell me that the SPLC is a once-useful organization now lost to corruption and political bias, any more than I need the SPLC to tell me that the DOJ is a once-useful organization now lost to corruption and political bias. So if one of them wants to tell me how vile the other is, meh, not worth my time and not a reliable source.
The corruption of the DOJ is the greater and more worrisome threat, obviously, and on that grounds only I'll be rooting for the SPLC in this fight. Winners are less likely to see the need for internal reform than losers.
It seems prudent to wait a little while, the DOJ has been filing some real stinkers lately.
Yup.
It looks like the strategy is:
• Allege that the payments to informants were "money laundering." (Of course you can't pay in your own name, without ruining their use as an informant.) But that's implausible, since...
• It's only "money laundering" if the money was proceeds from a crime. (Otherwise it would also be money laundering every time a cop pays an informant.) But that's implausible, since donor donations aren't proceeds from a crime. So the money laundering case also requires...
• Pretending the donors were defrauded, which requires pretending the money was used to fund the hate groups' hate actvities instead of being used to act against them.
But that's also wildly implausible, given that even according to the indictment, the money was used for info stuff like........... paying an informant to copy documents he stole from a hate group, which went on to inform a hatewatch report. (SPLC also says information from the informant program was often shared with local and federal law enforcement.)
> wire fraud, bank fraud and conspiracy to commit money laundering
Seems roughly on par with the "falsifying business records" thing they got Trump on.
The "money laundering" charge is part of the same nonsense narrative as the claim that they propped up hate groups (as opposed to paying informants for info on hate groups).
"Money laundering" requires the money to be proceeds from a crime. Donor donations obviously aren't proceeds of a crime. So the only way to pretend it's money laundering is to pretend donors were defrauded, which requires pretending that the money was to prop up the hate groups instead of to undermine them.
It is standard practice for prosecutors to overcharge their targets, in order to secure a conviction on a lesser charge (or a plea bargain).
For instance, if the "money laundering" thing doesn't stick, they may still be able to get them on "structuring" ("stacking"), which doesn't require the money be from a crime.
Less of this, please. Straight up whataboutism and a personal attack. Also aim at deflecting from the topic at hand.
I'm unclear what you're trying to insinuate.
The lying "they were propping up the KKK" propaganda narrative is all over Twitter rn, being regurgitated over and over and over by the likes of Elon and Rufo and the GOP accounts. It's like 90% of the timeline (no doubt due to the Elon algorithm). Being surprised that someone would post about a massive psyop is like being surprised that someone would comment on the weather, when a tornado's gone through a row of outhouses and is currently raining feces all over the place.
Like you, I haven't looked into the BLM money laundering charges. But yes, I agree that Trump's weaponized the DOJ in general, so if those charges also turned out to be an instance of weaponization, I wouldn't be very surprised.
It sounds like you're suggesting I'm too biased against younger leftist groups compared to older liberal ones, or something. Are you? (I'd find that critique coherent from a leftist, but I've been reading you as supporting the Trump DOJ's campaign, so I'm not sure how to parse this.)
For SPLC specifically, yes I have personally known them to do useful work. For instance, their reporting on the anti-LGBT pseudoscience network has been an efficient way to cite that background info in my critique of the Cass Review [https://zenodo.org/records/19303978].
A notable difference is that falsifying business records has been a crime routinely prosecuted for decades -- e.g. there's a long list of prosecutions and convictions under that specific statute in New York courts going back to the 1970s -- whereas these charges filed by the DOJ are legally "novel" (the term used in courtrooms).
Of course being novel doesn't necessarily mean that a prosecution can't/shouldn't succeed. However the specific charge of _fraud_ (not a part of for instance the Trump prosecution) requires specific not general deceptive intent. Fraud under US laws is "_intentional_ deception, misrepresentation, or concealment" made to "induce another party to act".
SLPC donors having been unaware of the payments to informants will not by itself rise anywhere near the level of committing fraud. (Also, contra Patel's public statements but not as far as I can tell part of the charges filed, it is not at all true that a nonprofit has a legal duty to make sure every donor is aware of every single thing the organization is doing.)
Perhaps the DOJ can show that SLPC staff overtly lied to or misled some specific donors in order to conceal the fact of the paid-informant program. The judge will ask this question and if the answer is that the feds have some such donors lined up as witnesses to that effect, that will be bad news for the SLPC and the case will proceed.
If on the other hand it turns out that the DOJ is just trying to convert "some or all donors didn't know about this specific activity" into charges like wire fraud, then the court will dismiss this case and the only question will be whether the dismissal is with or without "prejudice".
Well gosh, I guess I'm a Christian nationalist hate group myself (given that I have a LOT of not congruent with current 'Every Right-Thinking Nice Liberal Person Holds That In This House We Believe' opinions, views, and beliefs), want to turn me in for the reward, David? I'm sure the SPLC can always use another tout!
>"The SPLC used money from donors to pay informants/quislings to 'nudge' the KKK into doing more provocative actions, so that the SPLC could get more donations."
Even the DOJ's own indictment doesn't claim that (though their weaselly press release tries to imply it). The alleged "state and federal crimes," which DOJ is hoping you'll think were hate crimes, instead seem to be........ an informant stealing and copying documents from a hate group, that were used for a hatewatch investigation.
Would it be surprising if it were true? There are plenty of examples of progressive activists faking white racism, from hate crimes to vandalism. And the entire overarching strategy of anti-racism is to make people identify with their race more strongly and hate/fear other races. I’ll wait to see evidence of course but I’d hardly be surprised by it.
>"planning it and driving people to it ... this is nudging folks into doing more than they otherwise would"
The indictment doesn't claim he instigated the rally. Also—do you think the rally, which was already being planned, couldn't have gotten another person to find a driver if the undercover F-37 hadn't been the one to find a driver?
>"And yes, breaking and entering and being a thief are in fact federal and state crimes"
Tbc my quotation marks were regular old quoting-the-indictment marks, not scare quotes. If they stole docs from a hate group, they stole docs from a hate group. The point is that DOJ is really hoping you'll think they were funding hate crimes, because the money laundering charge depends on that sleight of hand.
> the SPLC put the Canadian Flag onto it's list of hate symbols.
That might have been the CAHN - the Canadian Anti-Hate Network.
>We note to the unobservant in the audience, that the SPLC put the Canadian Flag onto it's list of hate symbols.
Source?
Archive.org link?
Extremist groups do often try to hijack mainstream symbols (imo hoping to get caught so they can pretend the people who caught them are un-patriotic or frivolous)—and there is contemporaneous reporting about the anti-vax trucker convoy trying to do so with the flag—so i wouldn't be surprised if SPLC commented on it, but yeah idk the details if they did
I wrote a short essay in two parts (~15 minutes to read, longer if you sit with the questions), a pamphlet for people wandering the cave. The first part uses an apple to take apart the line between subjective and objective, the second looks inward. It tries to point rather than argue a point. If you ever had the wordless searching feeling, this is intended for you.
https://www.a-path-to-somewhere.net/
With your apple example, you seem to be re-inventing the concepts of accidents and essences, and I understand that in Buddhism there is the concept of attributes, whereby you can break down anything/anyone into "if you take away 'woman', 'height', 'colour of hair', 'name', and other qualities, what is left of this person you claim is a unique conscious individual with a self?"
So this essay seems to be treading established ground, but I am presuming it's only an introduction to what you want to develop?
Thanks for engaging with it. You are right, there is an overlap, you could even name more perspectives, like the Vedanta, Daoism, the Cook-Greuter scale and others, the text tries to point at the perennial philosophy core, as I see it and have walked it myself.
I find a point of pure being that I can't reduce further, but things get hard to describe in words at that point. I think the point is reached through a minimal steering, so I keep that as the will behind the thoughts. Phenomenological, I'm uncertain how unique the self is at this point; I see it as part of a whole, rather than the complete thing.
I don't actually want to develop it further for now. I wrote it as stand-alone text, I would have liked to have earlier in my personal development. It might be worthwhile to develop the ethical fundamentals further at some point.
On a previous post, Scott mentioned in passing that perhaps there should be deontological bar to voting for candidates whose policies you prefer, but where there is some other salient feature to consider, like dictatorship risk or a proclivity to raping children.
https://www.astralcodexten.com/p/orban-was-bad-even-though-we-dont/comment/244159773
I wonder how common this view is. My own view is that the stated policies of the other guys are so bad that, no, I don't think there is any such criterion that would dissuade me from even marginally lowering the probability of them being implemented.
Yeah, the problem with the idea is that every policy is its own deontological bar.
I think we're entering an era where even fewer people and media outlets than at present are going to be able to figure out whether a candidate has raped children or committed some other godawful crime -- image fakes, information hacking and persuasion are going to be too good. Doesn't really answer your question, though.
Also, we're entering an era where many people don't care.
"You think that raping children is bad? That's woke! Ancient Romans didn't mind things like that, and they were a strong healthy civilization, unlike the degenerate modern West."
Once people adopt the mindset that good is weak and evil is high-status, the argument that something is evil no longer appeals to them.
You have two basic strategies. Either the good people can coordinate to stop evil people, or the weak people can compete at pandering to the strong ones. These are two attractors in the policy space, and we have switched from the former to the latter.
There is a difference between "I repeat the falsehood because I am unable to figure out what is true" and "I knowingly repeat the falsehood to signal that I am on the side of the strong ones".
> "I knowingly repeat the falsehood to signal that I am on the side of the strong ones".
That's how Soviet propaganda worked, and how propaganda in Putin's Russia works today. They state laughably absurd things as truth, not because they expect anyone to actually believe it (although it is a bonus if someone does), but to single out and arrest anyone who *does not* enthusiastically repeat it.
Well, now the Russians have succeeded to export their wisdom to the entire developed world, hurray! They might no longer be able to put a man on the Moon, but at least then can convince millions of people who should have known better that the Moon is made out of cheese.
Also, this is what bullies often do, e.g. they punch you while saying "I am not punching you, you are punching me" and find that hilarious. This is how they coordinate the story they will tell later.
> Once people adopt the mindset that good is weak and evil is high-status, the argument that something is evil no longer appeals to them.
That's from your perspective, yes? As far as they're concerned, good is high-status and evil is weak. Once a morality stops being useful and only serves to harm its hosts, it's only natural that they shift to a morality that serves their collective interests better. But ultimately, they do care, and they do think they're righteous.
No, you are redefining words.
You have your definitions for good and evil, others have theirs. Either way, it's silly to claim that they're just nihilists who are purposely being evil. You can't organize something like this without some level of collective self-righteousness. Do you not understand that they are feeling the exact same feeling as you are, that people like you can't be reasoned with?
I am not going to write the entire thesis on morality here, just some quick notes.
Morality has a self-serving aspect, in the sense that otherwise it could not have appeared, from the evolutionary perspective. Moral systems such as "we need to commit collective suicide a.s.a.p." tend to be short-lived. Under usual circumstances, acting morally supports group survival, although weird things may happen in unusual situations.
But that is not the same as saying that morality is whatever is self-serving at the moment. Actually, morality is full of "you shall not"s related to things that obviously at least some people are tempted to do once in a while. It would be a waste of time to forbid things that no one was going to do anyway. When people say that someone behaved immorally, they usually mean that the person did a self-serving thing that somehow hurt someone else (whether someone real or imaginary).
From this it seems obvious to me that people who follow the rules and resist temptations (even if the success is only partial) are not doing the same thing, from the psychological perspective, as people who keep inventing excuses for doing whatever is convenient at the moment.
There are more details, this is just a simple thing, but as I said I am no going to write a thesis here.
Yes, there may be different sets of rules, and people following different sets of rules may find themselves in conflict with each other. Which is not the same as finding yourself in conflict with someone who follows no rules other than "whatever feels convenient at the moment".
Of course, those who follow no rules can pretend that they have a secret code that the outsiders simply don't understand. But it usually turns out that the secret code is infinitely flexible, i.e. has no predictive power beyond "this person would do whatever feels convenient for them at the moment".
The poster "Dust" is an admitted nihilist and sociopath. His posts are always aiming at the seeming inevitability of large-scale violence. Perhaps this helps to shed light on the discussion here.
Suppose I said some food was "delicious," and you disagreed. Would you argue that I was trying to redefine the word?
Yes, I agree with all your wonderfully great points without any qualifications whatsoever.
God almighty, you are incapable of ever giving unqualified agreement! Even to my wonderfully great points!
What are you talking about? I have perfect photographic evidence of me agreeing with you. You must be misremembering something.
It is 2026 and I feel like today's politicians cannot be judged by their stated policies or anything else that comes out of their mouths in the moment, or indeed that they might set down in a manifesto or other document or record. In our brave post-truth world, words alone cannot be trusted. The most reliable means I have of judging what they might or might not do going forward is their past actions.
I would like to see a norm of voting for people based on how they might behave over which team they are part of, but I feel even this is asking too much.
I have no idea how this state of affairs can be made better, but that doesn't particularly matter as I am one old man shouting at clouds while everyone else is hell-bent on making it worse.
>I would like to see a norm of voting for people based on how they might behave over which team they are part of, but I feel even this is asking too much.
Except that the team they are on is a reliable indicator of how they are likely to behave ( politically)
Part of my answer is a version of this.
Another part is a version of mvt will anchor policy anyways.
The other is that I am not so partisan too see one party's policies as so much worse than the others. Even though some are in fact quite bad
I don't think it's very common. It was far more common and it would be awesome if it was common again. Unfortunately, there is not and cannot exist strong legal/political enforcement of this norm and social norms tend to degrade over time.
Let's, uh, avoid modern politics and go back to Rome.
Say we have a strong social norm that we will never support anyone who purges all their political enemies using violence or anyone who marches their army upon Rome or installs himself as dictator. So, imagine Sulla, he's consul and marches an army off to fight some Greek dudes. While he's away, Cinna and Marius declare him an enemy of the state and murder all his friends in Rome. Notably, he's am still in charge of a large and victorious Roman army. It's *probably* fair to say that Sulla is not justified in marching his army back to Rome, declaring himself dictator, and then violently purging all his political enemies but it's hard to clarify exactly what he's justified in doing because the strong social norms that are meant to prevent this have already been violated. Worse, given Sulla's actions, it's hard to see why someone like Caesar, who Sulla almost killed, isn't justified in doing the same thing.
I'm totally on board with "there should be strong dentological bars to voting for certain candidates" but that requires social trust/capital/*mumble* *mumble* "high-trust society" and we don't have that. Barring that, at a practical level, upholding social/political norms of decency feels like a sucker's play when *Hated Outgroup* violates them all the time without political or social consequence.
> Sulla is not justified in marching his army back to Rome …
I'd say exactly the opposite, that once they declare you an enemy of the state, that justifies massive retaliation – present-day analogues left to the reader – so I guess that's actually a validation of your broader point about the collapse of shared social norms.
>we had a former servicewoman saying that Biden was trying to start World War III (and that's corroborated with the scuttlebutt I was hearing, which was quite a bit more... colorful).
How can you possibly credit such obvious nonsense?
No, we can start with you providing the extraordinary evidence to support your very extraordinary claim. Rather than trying to change the subject.
Come on. man. She did not say that Biden wanted to start WWIII. She said he was creating a risk of undesired escalation. Surely you are not unfamiliar with that sort of claim. It is, among other things, literally Game Theory 101.
>President Joe Biden’s administration is driving the world toward a devastating nuclear conflict by using the Ukraine crisis to fight a “proxy war” with Russia, former congresswoman and 2020 presidential candidate Tulsi Gabbard has warned.
>“The American people need to understand the seriousness of the situation that the Biden administration and leaders in Washington have put us in,”Gabbard spoke Wednesday night to Fox News in an interview. The Hawaii Democrat, who was among the candidates whom Biden defeated to win his party’s 2020 presidential nomination, called for stepping back to focus on “the big picture and the great threat that we face, which is the reality that this war is continuing to escalate.”
>Biden led international efforts to punish Moscow for its offensive against Ukraine. US approval of $70 billion worth of aid to Ukraine, including weapons and advanced technology, has been given. Biden visited Israel Thursday and stated that Washington would continue providing such assistance indefinitely so Moscow does not suffer. “a strategic failure”Ukraine
>“President Biden himself says he has no idea when or how it’s going to end, but we know where this escalation leads,”Gabbard spoke out. “It leads us closer and closer to the brink of a nuclear war with Russia.”
https://www.massnews.com/world-is-nearing-brink-of-nuclear-war-tulsi-gabbard-analysis/
And btw here is an argument, based on actual facts about Biden Admin limits on Ukrainian access to, and use of, American weaponry, that Biden was too leery of provoking Russia:
https://www.theatlantic.com/ideas/archive/2024/11/biden-ukraine-policy-failures/680834/
as of this post you have at least 112 comments, leading to more than 1/8th of total posts. A lot of them contain obvious conspiracy theories, blatant partisan information gathering, and vague insinuations, and quoting Tulsi Gabbard on this? on a political comment about Biden trying to start WW3? As extraordinary evidence? What, is the next quote going to come from the white house press office about how the Iran war is totally Biden's fault and has nothing to do with Trump and he is the most peaceful president ever? Give me a break. Your posts are temperature-elevating, contain outrageous statements and vague insinautions to low-quality claims about grand conspiracies and hidden information. It's like a link "you would never believe what <insert political figure> wanted to hide from you" rather than high-qaulity, rational discussion. Like a big flashing popup in the middle of a book. All the time.
I repeat my earlier plea- please consider posting WAY less, like half, and putting double the effort in your posts. The ACX comment quality has nosedived and you are a major factor in this.
Wrote an analysis of the Musk/Altman conflict that reframes it as competing collapse scenarios rather than AI safety debate.
Core thesis: Musk is building post-state capability (works when systems fail: energy independence via Tesla, comms via Starlink, logistics via SpaceX). Altman is building acceleration-dependent systems (needs grid stability to reach AGI before coordination failures cascade).
Theme: Reading their infrastructure choices as revealed preferences.
https://mythcoreops.substack.com/p/elon-vs-altman-what-their-infrastructure?r=6wwk13&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
Both of these guys are incredibly enmeshed with the state and the military, they just picked different strategies.
That is dreadfully written, and if this is the state of the art AI production, then everyone trying to persuade me to give the latest AI model a go and change my mind is trying to make rain fall upwards.
Never mind the content, the constant BOLDING to try and MAKE THE POINT is aggravating, and the prose is adequate but not informative. Is The Boring Company going anywhere, both literally and metaphorically? The Hyperloop? Neuralink seems to be continuing but I am not aware of how much progress has been made, I saw some stories about promising indications but as yet no "this will make the blind see and the lame walk" achievements of all the claims.
Your AI-generated graphics are full of errors, which is one of the risks of using AI-generated media. Aside from that your post is not very informative, it seems to just be a massive glaze of Elon in the cadence of an LLM. Ask yourself: is this level of quality worth responding to? Is it worth aspiring to?
There is no world without nation states (or their hypothetical corporate equivalent) that would allow Musk to maintain any of his technology. Any classic, off-grid prepper is better equipped to survive civilizational collapse than he and his toys.
> That’s not about environmentalism. That’s about Energy Sovereignty.
As someone who used the em-dash before it became everybody's favorite AI tell, I feel qualified to point out that this sentence structure in this day and age might be something you want to avoid.
Example: https://www.theguardian.com/commentisfree/2026/apr/15/chatgpt-stylistic-quirk-its-not-x-its-y
So many (male) writers write about dating that does not seem to mathemathically add up, like how most young men are single and most young women are in relationships, that got me curious. Is it possible that for GenZ assortative mating is over, young white professional women are choosing non-white or working-class men, who are generally invisible to young white professional men?
I don't go out much, and a bit too old to intuitively understand young people these days. But for example 10 years ago I was walking all over Cologne, Germany, and the masculinity difference between young white men and young Middle Eastern men was staggering. The first looked like babies who somehow just did not go through puberty. The second even at 16 sort of looked like young men, not boys, with well-growing beards and an adventurous gleam in the eye.
I think you might be reading too much into aesthetic differences. It wasn’t so long ago that hipster man were wearing elaborate mustaches and beards that would have looked respectable on a civil war general. But most of them hadn’t led even one cavalry charge.
I guess it is a few things adding together.
Some things have existed always. -- Older men compete with young men for the young women. High-status men are polygamous, even if unofficially. Women are hypergamous.
Some things are more recent. -- The monogamous marriage starts later (so the time interval when the few alpha males can hoard many girlfriends is longer). Longer education that keeps young men longer out of the job market (more childlike, less economic status). More women than men at universities which still remain a source of status.
I think the education is specifically a big problem, because now you need certificates even for jobs that are relatively low status. If you have a job, you have an income, and you can impress some girls. If you study for a high status job, such as a doctor or a lawyer, that can also impress some girls. But many young men have neither a job nor a high-status school. To say it differently, some schools are a signal of a future wealth, and keeping a job is a signal of being responsible, but many young men can signal neither.
The situation is not symmetric between men and women, because many young women date older men, or have one night stands or short relationships with polygamous alpha males. That said, women also complain about the dating market, it's just that their complaints are different.
> Some things have existed always. -- Older men compete with young men for the young women.
Was that always the case? Back when monogamy was more heavily enforced, people were pressured into marrying early, and divorce was not a realistic possibility, it would've been unlikely for older men to get into a legal relationship with younger women. They would've had mistresses, of course, but that didn't make the women unavailable to younger men. Perhaps making divorce illegal again will lead to a more stable and equitable state of affairs for men.
"unlikely for older men to get into a legal relationship with younger women" -- that was a reasonable description of reality only since maternal mortality rates plunged.
One of the many things I've learned through decades of rich family-tree research, 98 percent of it being in post-1600 America, is that until relatively recently older men constantly competed with young men for the young women. That was because so many women died during or related to childbirth. The surviving husband would then remarry, always with a woman no older than his deceased wife was. Often, if the man was 40+ when the previous wife died, to a woman 20 or more years younger than he now was.
Across the board in my large database the above household pattern recurs, hence I have more step-grandmothers through the generations than I'd ever have guessed. Previously had no idea how common it had been, for centuries, for nonwealthy men who lived to reach the biblical "three score and ten" to have had 2, 3 or 4 wives along the way. Same in my wife's ancestry, and those of American genealogy hobbyists I've corresponded with.
Throughout the 17th, 18th, 19th centuries there were tons of "May-November" marriages as the old fashioned expression went. Only during say the last two to three generations of divorce being rare/difficult, did older men marrying much-younger women become rare.
(This wasn't a class-restricted pattern either. I am descended from mostly what we'd now call working-class households, with educated higher-socioeconomic folks sprinkled in here and there. No powerful "elites" -- who throughout history could obtain divorce if motivated enough -- going back at least 400 years.)
"Dating down" is still unthinkable for most women, and GenZ has more college graduates that are women than college graduates that are men.
Seconding tempo, I heard the explanation for that observation is that more young women are in relationships with men over 30.
Are they just in relationships with older men?
>So many (male) writers write about dating that does not seem to mathemathically add up, like how most young men are single and most young women are in relationships,
Do people say that? Most of the stuff I've read suggests that young women are mostly single now as well.
>Is it possible that for GenZ assortative mating is over, young white professional women are choosing non-white or working-class men, who are generally invisible to young white professional men?
I'll look for the link if you want, but I came across a study recently saying that white German women were very unlikely to want to get into romantic relationships with third-world immigrants. Obvs beware the man of one study and all that, but it certainly sounds plausible -- would you really want to date a poor, low-status person who probably doesn't speak good German and comes from a society where misogyny is completely normalised?
>The second even at 16 sort of looked like young men, not boys, with well-growing beards and an adventurous gleam in the eye.
Not sure how much of the difference is due to this, but a lot of Middle Eastern immigrants straight-up lie about their age -- it's easier to get asylum if you're a minor, and if you "lose" your passport on the way it's difficult to prove that you're actually over 18.
Regarding Orbán: somehow everybody is only seeing one half of the issue? The other half is the republic, that is, when matters are decided by deliberation, even when, like Roman Senators, the participants are non-elected.
In modern republics, it is professional elites who participate in the deliberation. Orbán did what Curtis Yarvin wants Trump to do, to replace the entire "horizontal" system of stakeholders deliberating with a "vertical" system of elective kingship, where every decision is made by one man, after listening to a small number of advisors. Yarvin is fairly obsessed by the kind of CEOs who do not listen to anyone, like Jobs and Musk.
The results were not good. It led to corruption and dysfunction. But yes, the stakeholder elite system also does that.
I don't know why it worked better for Steve Jobs. Maybe a business is not a government.
Now Magyar will try the good kind of populism, a kind of participatory democracy like making government meetings open and random citizens will be allowed to comment or question. Maybe this works better than both. Not because he is a good guy. Simply listening to random people on the street turned out to be a winning strategy so why not continue.
>But yes, the stakeholder elite system also does that
Does it so it as much?
>Maybe a business is not a government.
It definitely isnt. It's OK for businesses to fold. The greatness of the best CEOs is survivor bias, not a magical Mandate of Heaven. It's not OK for countries to fold.
Exactly this.
Most businesses fail -- we don't want most countries to fail.
When a business fails, the employees simply find a different one, which is mostly a simple process of applying for a few jobs. -- When a country fails, the analogy is mass migration, which is very costly for the people who move (they can't take their homes with them), and the countries that receive them are often not happy about it.
Also, a dysfunctional business cannot declare a war on competitors and start literally killing their employees.
The analogy fails on many levels.
Is that the main reason? Or is it that a business has a specific external purpose, it's not merely "a way for this group of people to get along and govern themselves".
It strikes me that advocates of dictatorship tend to see a nation as having an external purpose, most commonly war, but also some kinds of ecconomic growth imperatives. While democracy has a fundamentally nihilistic character: people have different preferences, and there's no basis for saying one is objectively better than another. Some people just cannot handle this nihilism. Including a lot of rationalists: they need an externally imposed purpose, which might be "technological progress" or "GDP growth". I think it's an intellectual failing to not be able to handle the deep uncertainty of a society without any imposed purpose to which everything else reduces.
On the other hand, democracies being hypocritical about democracy (e.g. banning parties, banning or restricting the speech of political candidates or activists) is the worst possible advertisment for democracy: "by our actions, we clearly don't actually remotely trust the peoole to make a free choice". And it would be an intellectual failing to continue to support the kind of democracy practiced in the countries that do that.
>You can tell that to Nixon! And Trump
Which bit?
Who folded, as in collapsed?
I've read a couple bios of Steve Jobs, and came away thinking that he simply was not a good manager. He was like a bad poker player who gets dealt 4 queens. The PC was ready to be conceived, built and sold. Somebody was bound to be in the right place at the right time with the right store of knowledge to build a business that did that, and he was the first one to have that combo. He didn't succeed because of his wacko management style, he succeeded in spite of it.
That story of his getting hundreds or was it thousands of samples of shades of beige, so he could choose the perfect one for his machine? It's nonsense. The human eye can't discriminate anything like that many shades of beige.
He succeeded wildly because he was able to demand an ethic of elegant, clean, user focused engineering in a sector that otherwise was ugly, directionless and getting lost in its own complexity. Otherwise he was just another guy.
This. Steve Jobs wasn't the nerdy tech genius - that was the other Steve's job. The Steve everybody remembers, is the one that was a visionary designer and tech evangelist, who could take something as nerdy as a "personal computer" - and repackage it into something non-nerds saw as cool. And later repackage the "Blackberry" as a fashion accessory for the hip and young, not an electronic ball and chain for corporate suits.
Apple didn't *invent* much of anything, and certainly not the personal computer. And what Apple did invent, wasn't invented by Steve Jobs. He invented nothing, but made many things cool, and better optimized for non-nerd customers.
And, yeah, he picked the right color of beige. Maybe he spent more time on that than he needed to, but nobody else ever recognized that the color mattered at all.
It's more than just packaging and marketing. The gap between the Blackberry and the iPhone in terms of usability is huge. The Blackberry could do phone calls and emails, the iPhone could use the whole freaking internet. Maps alone was the thing that made me want one. (Strictly speaking Blackberries had some kind of web functionality but it was barely functional, it needed innovations like a full-device screen and pinch-to-zoom to make it usable.)
That's the pattern -- Apple isn't the first to ship an internet-enabled phone or an mp3 player or a GUI or a personal computer, they're just the first to make all those little usability innovations that turn those things from annoying fiddly things for geeks into things that normal people want to use.
I have seen the huge excitement and people standing in queues in 2007. The thing is, social media didn't exist. Websites did not have AJAX, which really limited their usefulness. I don't think a lot of people really wanted to look up a basic website such as blogspot.com on their phones. Basically an iPod that can text. I think people mostly used it for music, and the main idea was that they don't need to carry a separate phone and music player.
Yeah, the big difference was the target market. The iPhone market drove the evolution of an app infrastructure that produced all the cool stuff that most people want and take for granted, The Blackberry market drove the creation of a different set of, I don't think they were called "apps" then, but which were more narrowly focused on things that corporate America and the military-industrial complex wanted. Which was more than just phone calls and emails, but less of what makes most non-suits tjhink makes smartphones so awesome.
But you had to make the iPhone a thing, and get it into millions of pockets, before all those apps start getting developed. That's where you needed a Steve Jobs. And yes, being able to fold existing iPod software into the iPhone was a big help in jump-starting all that too.
I do think he did make it cool, and that was hugely important to his success. But that doesn't really cover it. No one else had the power and sensibilities to demand refinement at the time. He forced focus on details that made devices more usable and organized when everyone else was going in the other direction. He bent the curve.
Information tech is best understood by the stacking metaphor. The industry needed him. Organizing complexity well is hard work, and he forced it to happen.
I've never thought of Jobs as this genius manager but he was very, very good at design and styling the product and marketing it. Nuts-and-bolts man-management and balance sheet management? Nothing special. Making his product sound like the greatest thing since sliced bread? Incomparable.
Another thing is that the impending bankruptcy of Apple was and is a myth. When that stock was trading at its lowest around $12 a share they had more cash in the bank per share than that; it was a huge investment opportunity. It was a good move to bury the hatchet with Microsoft.
He bailed Apple out of an operating system cul-de-sac that looked inescapable and the only way he could do it was to go somewhere else and invent an operating system that he could bring back.
But that’s not true either. He was responsible for turning Apple around from a loss of 700m in the quarter he joined to a (small) profit two quarters earlier. That pushed Apple from being a few quarters away from bankruptcy to being safe. He also did hire and retain great people, and was a great man manager. More Alex Ferguson than Carlo Ancelotti - admired and loved, but scary.
For products his simplifying of the number of categories that Apple produced to four, was a great decision. As was the detente and investment from Microsoft. If he was hit by a bus in 2000 he should still be a saviour of Apple.
"More Alex Ferguson than Carlo Ancelotti - admired and loved, but scary."
That is an excellent analogy that gives me a much better grasp of his style, thank you!
I do think he was more on the design/marketing side, than the 'boring grey suits running the business' side, still. Apple products may have been cutting-edge tech, but they were also mostly (mainly?) known for their style and fashionability. Part of the joke about the Cult of Apple was that the latest whatever *was* sold like This Season's Haute Couture. I hate this scene from "The Devil Wears Prada" because duh, there are only so many colours and all you are doing is tweaking which shade of blue, but the thing about 'that is cerulean' applies because Jobs could make it "this season it is cerulean" and those kinds of style and tech choices then trickled slowly down into the mass market:
https://www.youtube.com/watch?v=-rDTRuCOs9g
He seems to have been dealt with that hand 3 times then. Original Apple, Pixar and the return to Apple. It’s the latter that should, on its own, dispel any myth that he was dealt a great hand.
Your argument could work only for the original Apple, he was a better manager for Apple the second time.
Also NeXT, which was not successful in its own right but was successful enough to get acquired by Apple and turned into OSX.
If the biographies make him seem like a lousy manager who got lucky then I think the lesson is that even the best managers will often appear to be fuckups when observed closely. We all do. If anyone closely observed my career and wrote down all my fuckups in a book then id look like a fuckup too.
>lso NeXT, which was not successful in its own right but was successful enough to get acquired by Apple and turned into OSX.
I think that was the whole point.
I distinctly remember when Apple was considered the artist's computer, not only because it ran PhotoShop, but also because the whole elegance of it, like building the whole computer into the screen to avoid desktop clutter. So maybe Jobs was not a good manager as such, but a good product designer, with an eye for the artistic and elegant.
But it might turn out that was someone else. Quite honestly? I suspect a woman. I mean, are ideas like "let's not have desktop clutter" kind of fem-coded?
It worked better for Steve Jobs personally. Did it work better for people who worked at Apple? Did it even work better for Apple customers? Did it work better for the rest of society?
Yes to both. The former got to work on great products and became rich. The latter got to buy great products.
You could say society was harmed by the smart phone but there’s no way of anticipating that.
I mean, I thought it is fairly uncontroversial to say Apple products are great? They combine functionality with a rare elegance.
This feels like a matter of taste. From what I've seen of Macs, OSX is anything *but* elegant and easy to use. I think it's just people who have been using Macs their whole lives who are used to its foibles and don't notice what it doesn't do.
You can have as many buttons and wheels on a Mac as you like if you put your mind to it.
I attempted to update a famous Borges story for the modern era.
https://linch.substack.com/p/fine-tuning-borges
Looks like The Onion may have pulled off its purchase of InfoWars:
“Imagine a roaring arena packed to the rafters with pathological liars. High above you in the nosebleeds are podcasters, screaming that you’ll die if you don’t buy their skincare products. Below, on the floor, imagine demonic battalions of super-influencers physically forcing people into home fitness devices designed to dismantle their bodies bone by bone and reassemble them into a grotesque statue of yourself. Out of the throngs, an extremely sick looking man approaches you. He puts his hands on your shoulders. He explains that he is your life coach and that you owe him $800.
Such is the InfoWars (https://theonion.info/)I envision: An infinite virtual surface teeming with ads. Not just ads, but scams! Not just scams, but lies with no object, free radical misinformation, sentences and images so poorly thought out that they are unhealthy even to view for just a few seconds. The InfoWars (https://theonion.info/) of old was only the prototype for the hell I know we can build together: A digital platform where, every day, visitors sacrifice themselves at altars of delusion and misery, their minds fully disintegrating on contact.”
https://theonion.com/at-long-last-infowars-is-ours/
I love that they're doing this and I'm rooting for them to keep it going, but it sounded to me like some kind of paid licensing deal, not an outright acquisition. Not sure of the actual details.
I can do a pretty good Alex Jones voice. (It scares our cats and most other sentient beings though — the cats hide under the sofa when I get going)
I might just audition for the role. I’d need some prosthetics to make myself look as unhinged as Jones sounds though. Getting money to lie my ass off? “Give me a lot of your money and I’ll teach you how to turn your urine into gold!”
What’s not to like?
You know I had to look that one up and it turns out you *can* turn urine into gold. Or rather, extract small trace amounts of gold from urine, slightly more amounts if the person is undergoing treatments with gold salts:
"Naturally occurring trace gold: A typical human body contains roughly 0.2 milligrams of gold. While minuscule traces (measured in nanograms or picomols) can be filtered by the kidneys and found in urine, it is not enough to collect or make someone wealthy.
Medical context (Chrysotherapy): Gold is significantly easier to extract from the urine of patients undergoing chrysotherapy (treatment for rheumatoid arthritis using gold salts). Scientists use Atomic Absorption Spectroscopy to monitor gold excretion in these patients, where it can be recovered by wet-ashing the urine and extracting the metal into solvents like methyl isobutyl ketone."
https://academic.oup.com/clinchem/article-abstract/19/9/1081/5676711?redirectedFrom=PDF
"An atomic absorption spectrophotometric method is described for estimating gold in the urine of patients undergoing chelation therapy. Variable absorption at 242.8 nm, not due to gold, is avoided by wet-ashing and extraction of gold into methyl isobutyl ketone."
It seems a German alchemist thought he could turn urine into gold, and in the process accidentally discovered phosphorus:
https://dontknow00.medium.com/story-of-making-gold-from-urine-3b7e1d3f11e5
"The last person I’m going to mention is Hennig Brand. Brand was a German who lived in the 17th century. What’s fascinating about this person is, he experimented with urine. He saw human urine as more than just waste and he discovered something truly precious in pee — phosphorus. Why he was interested in urine? Well, since it looked golden, Brand believed if he distilled urine, gold would be left behind. During his experiment, he collected around 5600 liters of urine and he boiled it, and extracted the components of the urine in different ways. Upon final distillation, the product left behind was white, smelly, and glowed in the dark. This left-over material is now known as phosphorus."
Human pee was actually used as source of potassium nitrate aka saltpeter to make black powder during the US Civil War by the Confederacy running low on black powder.
“Desperate for saltpeter necessary for the making of gunpowder, the Confederacy sent out agents around the South to collect deposits of "night soil", i.e. human excrement. John Haralson, an agent in Selma, Alabama of the Confederate Nitre and Mining Bureau, advertised the following in the local paper: "The ladies of Selma are respectfully requested to preserve the chamber lye [i.e. urine] collected about their premises for the purpose of making nitre. A barrel will be sent around daily to collect it."
https://en.wikipedia.org/wiki/Nitre_and_Mining_Bureau
Taking over a political enemy's news outlet and restructuring it to fit your vision isn't that unusual. I understand Orban did some of that. Trump is trying to as well.
They’re buying it for real money I believe.
This may shock you, but many actions are viewed very differently when the government does them as opposed to Just Some Guy.
Many actions are viewed very different when a Republican does them as opposed to a Democrat (or vice versa) too. Of course there's always going to be special pleading as well; that's not unusual at all.
But if you like, Lőrinc Mészáros is also "Just Some Guy."
Great job dragging American partisan politics into a thread that was previously unrelated to American partisan politics! Because that's just what the world needs: D vs R infesting every last crevice of online discussions.
Is this not the thread that started with the story of what the Onion did to InfoWars after they took it over? Are you claiming this is "unrelated to American partisan politics"?
I should know better than to give this thread oxygen but I am curious how far you are willing to take this
WRT InfoWars at least i think this is an actual case with reality having a well known liberal bias. I know or at least have every reason to believe you don’t really think the Sandy Hook shooting was a false flag operation.
As always I’m left wondering how much of this edgelord posturing and how much is IRL Shankar.
Has anyone done any writing/speculation on what enshittification of AI might look like?
It seems like there is an interesting window between "AI delivers massive paradigm shift" and "AI turns out to be completely useless" where "AI has enough market capture to be potentially valuable but not without implementing the kind of bullshit that has made everything from Google to Uber worse over the past few years." This would mean throttling usage (something Claude and Anthropic already seem to be experimenting with), offering an ad-based free product, and generally bringing the quality of every available product down by several tiers, while reserving the actual useful versions for a price point that is way above what even the highest end current versions charge. Perhaps if you are genuinely concerned about AI you should be encouraging this process to happen now, as nothing seems more designed to sabotage long-term progress like promising investors a viable, if objectively terrible, model that can deliver short-term profits, in such a way that the market demands everyone pivot all the resources to maintaining that.
AI enshittification has been ongoing since last September in the narrower domain of agentic coding. There was a ton of drama around Cursor's change to their billing around that time, and then Anthropic published a large write up on how they were screwing up serving inference. There is the recent wave of furor over Anthropic serving dumber models (massively abbreviating) since March, see here for the locus classicus of the current controversy: https://github.com/anthropics/claude-code/issues/42796
They've had terrible uptimes too, relative more traditional SaaS.
Now, all of these issues revolve around the economics of AI and how it isn't really profitable and the monthly subscription model doesn't actually make sense, but I think they meet the technical definition of enshittification—companies hook a ton of users via subsidized subscriptions and pump their usage temporarily (loads of 2x offers and other deals in the last five months), then drop the rates massively or do route to dumber models.
Because of the opaqueness of the metering and routing, and the general difficulty of assessing things like 'intelligence', users are not always confident (or some are overconfident) that they are getting scammed. In that way it exactly mirrors these other examples you brought up.
Going beyond coding (my profession), I've noticed in the case of Anthropic recently that these issues even extend to the chat interface. I haven't done a statistical study of it, but Claude is responding in a more abbreviated and generic way to my prompts and dialogues in the last several weeks, probably for the same reasons its thinking less and quitting sooner in coding loops.
I could say more, but I would say that some of this is due to the massive demand of course, some of it is due to monthly subscriptions being a bad business model that needs to die, some it is due to the heavy subsidizing and the classical hyperscaler business model being employed here, and some of it is due to a lack of concern for quality over shipping fast.
[Edit] Like every other commenter, I just read your first sentence and nothing else, and I see you already concisely said all this.
There is a popular Q&A site called Quora that has seen a lot of traffic from bots, for years. The standard model of making up a question that will be popular, asking it using a burner account, then answering it using the rep-farm bot still works. AI has levelled it up by giving the rep-farm bot the ability to put the question to ChatGPT and pasting the answer to Quora, complete with an AI-generated picture at the end (Quora seems to promote multimedia answers over plain text). Accounts such as FeverPoint and OpenBlueprint are now flooding Quora.
Perhaps this is some of the enshittification you're referring to. Maybe not; it's AI slop that presumes quality AI, and maybe you're more interested in things that make the AI itself shitty.
If so, then the usual concerns I see include AI training on its own slop, so that once some hot take works its way into the training data, it becomes iteratively harder for the AI to question it.
Another serious concern is over who curates the training data. It won't be long before we see LLMs trained on, oh, I dunno, Artemis II was faked and so on.
"It won't be long before we see LLMs trained on, oh, I dunno, Artemis II was faked and so on."
Oh don't worry, humans can still serve up a superior grade of bullshit. Yesterday I learned that the mostest correct opinion to have on the Artemis mission is that it was a tool to maintain and propagate, um, white supremacy or capitalism or whatever, my braincells died reading the thing and I can't remember exactly what the Bad Thing was.
I've written about AI, touching on enshittification briefly, but it seems to me the other aspects are far more disturbing.
And this was before I started thinking about how the brain offloads to System 1 to save a few calories, how the incentive for offloading to System AI is infinite by comparison, and how those most fluently offloading to AI are those whose System 2 capacity is still developing.
https://bargainbinseldon.substack.com/p/the-epistemic-cage
https://huggingface.co/TheDrummer/Rivermind-12B-v1
> Has anyone done any writing/speculation on what enshittification of AI might look like?
I've written a lot about this. I call the effect you're pointing to "the ten thousand Phd's," and ascribe it to every big company having hundreds or thousands of Phd's and a lot of computing power furiously optimizing on their side, against individual average people on the other side. This is vastly imbalanced, and has led to people spending ever more time eating junk food, staring at phones, gambling, and more.
Post on that: https://performativebafflement.substack.com/p/the-ten-thousand-phds-and-what-it?r=17hw9h
It's pretty obvious that AI is going to be the next trillion dollar attention surface, because it's going to turn into "24/7 AI assistants in people's ears" very soon. It's the last bastion of attention-time, soaking up all the pebbles and grains of free attention between anything else that phones can't reach right now.
At the same time, there's literally no way they're not going to monetize that surface. Google and FB have done that, and it's made them trillion dollar companies. They each make $200 - $400 per US customer per year, and have a few billion monthly active users each.
Gemini and ChatGPT already have roughly a billion monthly active users each. It is basically written in stone that that attention surface is going to be monetized for all the non-paying users, because data centers and inference costs money, and they need to make enough to pay for them and profit.
Google is already a pro at milking that attention surface, and has several thousand Phd's standing ready to start doing so. ChatGPT has hired some of the big monetization people from FB to do the same there.
So clearly the market is going to bifurcate - the free tier of AI users are going to be getting milked the same way everyone is getting milked these days, enshittification will run rampant, and will probably go in even more optimized and darker directions than we're currently at (for one thing, we'll be able to create individually tailored fully immersive digital experiences that are basically David Foster Wallace's Infinite Jests).
*However* there is hope for the paid tier of users! The paying and agentic will be able to level up in their lives, because having a Phd smart assistant in your ears giving you advice and pushing you to make better decisions will be a gigantic multiplier on positive life outcomes.
These AI minds will know everything about you, they’ll know your thinking style, they’ll know what rhetorical techniques you prefer, they’ll be talking to you in the ways that most resonate with you and making connections, arguments, and analogies accordingly. Super persuasion, but at the personal level, and for your benefit - a super-ego that works, in other words.
And I’m not pretending the AI is going to win all the time here, either. Your super-ego doesn’t win all the time today, does it? All it really needs to do is win more often on the margin. Think of it winning only 10% more - 10% better decisions compounded over days, weeks, years, and decades is a CRAZY big effect size. It’s like getting a 10% financial return that compounds weekly!
So imagine being able to level up on your career, health, and hobbies pretty significantly, and your friends and family able to do the same.
I've written about that here: https://performativebafflement.substack.com/p/the-spastic-yuppie-zombie-hoods-in?r=17hw9h
Stack this on top of everyone getting counterfeited at work and getting UBI. UBI is actually not a big reach - with really reasonable assumptions, we can crank $40k - $90k UBI's in the near future. I wrote about that too, I'll footnote it.¹
Current average screentime is something like 11 hours a day: https://imgur.com/uSQIthV
So obviously the population will bifurcate. The 20% will use their newfound time to focus on their interests and hobbies and families, and will live life to the fullest.
The 80% that currently spend ~11 hours a day on screens will now spend 5-8 hours more per day on algorithmic feeds and porns, and eventually, in infinite VR heavens.
So if this is true, the more people you can steer towards the paid tier of AI assistants, the more people you are saving from a life of Experience Machine coffin slavery.²
Towards this end, I'm actually putting together a startup right now aiming at steering people overall towards the second better AI future outcome, by providing RLHF services for AI personal assistants particularly to the Big 3 AI companies and surfacing instrumentation and KPI's to show those improved life outcomes, thereby aligning the individual and the AI company incentives.
If we show people that their lives have improved by so much on the paid vs free tier, and they've truly leveled up in fitness, career, or social lives, they'll be williing to pay for still higher tiers to keep leveling up. If we show the AI companies that helping people level up is more profitable than milking them with enshittification, they TOO will want to level people up, and incentives are aligned and it becomes a virtuous cycle. Then it's just a matter of people seeing their friends, family, and neighbors hugely leveling up, and getting social-graphed into hopping to a paid assistant tier and doing the same.
________________________________________________________________________
¹ Post on UBI - https://performativebafflement.substack.com/p/can-ubi-work-yes-with-a-few-reasonable?r=17hw9h
² Post on coffin slavery - https://performativebafflement.substack.com/p/the-view-from-inside-the-80?r=17hw9h
Big yes on the bifurcation between free and paid tiers — that's the right central call. One trap worth flagging though: paying for something doesn't actually align the company with you. What aligns the company is what they're measured against. If "leveled up" gets measured by user self-report, the paid tier just rebuilds engagement bait wearing a satisfaction skin. If it gets measured by real outcomes (career change, savings, fitness), the company has to survive a years-long delay before the data shows up, and most won't.
The deeper question is whether ad-funded media is even the right reference class for AI assistants. AI talking in your ear 24/7, calibrated to how you think, is a product that changes the user — closer to pharma, therapy, or education than to Google or FB. Those fields all have institutions built around measuring real outcomes (FDA trials, accreditation, professional licensure). For AI assistants, that infrastructure mostly doesn't exist yet — which is what makes the venture actually interesting. The lever isn't pricing model, it's whatever measurement framework gets built first.
Curious which outcomes you think become measurable inside 12 months. Career changes? Savings? Fitness? Something else?
social diffusion defense: juliusthimm dot com (maybe relevant)
i quibble with “literally no way they're not going to monetize that surface” because you’re not AGI-pilled enough in this statement perhaps?
one with more “ai solves everything” bias might say that a 1-person trillion dollar machine can out-compete monetized incentive misalignment in the long term, where that 1 person is a Karpathy with an agent swarm building the company which provides the $T as consumer surplus rather than value capture.
this is essentially China’s stance, make AI a substrate and commodity for the populace… “open source” doesn’t quite describe the playbook at the cultural political layer that it truly is.
You seem to be aware of the flaws and possible failure modes, but think we can be steered to the better option by... spending more money.
Don't access the free slop, buy the premium Olympus-tier slop instead!
Improved Life Outcomes means people who *don't* need the AI to tell them what to eat, how to talk to other humans, and who have lives that interest and fulfil them. What good is that to a company selling a product?
> Improved Life Outcomes means people who *don't* need the AI to tell them what to eat, how to talk to other humans, and who have lives that interest and fulfil them. What good is that to a company selling a product?
Except look around, obviously people are failing at that in the current environment? That's *why* 80% of people are overweight or obese, everyone eats junk and fast food for 70% of their calories, and everyone stares at screens all day.
There is no path to the better future you want.
Religion won't cut it because nobody will adopt it at scale, advocation won't cut it, "life hack" posts won't cut it, because at the end of the day, you need a superego that lives with you 24/7 and actually works to navigate the environment today, and 90% of people don't have that.
Right now the solution is something like "better than top decile conscientiousness, willpower, and awareness," which is ridiculous, inapplicable to 90%+ of people, and can't ever scale.
And what *can* steer the bulk of people to better choices?
My solution is the ONLY solution that can scale, address 90%+ of people, and actually guide them to better choices, that I can think of.
Like realistically, what is the path to the future you want?
Everyone looks deep inside, has an epiphany, and permanently and totally changes their behaviors, and somehow sticks with it over the next decade, in the face of ever-greater optimization power from large companies? A Butlerian Jihad back to tech levels of 2005? Then we'll descend back to a world where everyone is still fat from eating junk and fast food for most calories and watches TV 11 hours a day, is that actually a win?
They're already 80%+ failing! Even more are sure to fail when the optimization power gets higher!
Like when were we actually better off, in your opinion? What is the ideal state of humanity + technology? Pre atom bomb, but post antibiotics? How do you stuff those genies back in the bottle?
"Super persuasion, but at the personal level, and for your benefit - a super-ego that works, in other words."
For *your* benefit? Now why would the AI do that, when it could instead be steering you to benefit the corporation instead? You are the product now, you are the product forever. That 24/7 AI in your ear will be maximising your productivity for your employer, using you as a data source, and tailoring personalised ads to soak your bank balance.
What's that, you say? But you are a paying user, not like the freebies who are having ads crammed down their necks? Yes, you are a paying user, and you pay. Oh, you pay.
AI will make you level up, because finally now you will work better, smarter, harder! You will be attentive and responsive to your family and friends! You will eat right, eat healthy, exercise more, become that attractive, fitter version of yourself you always intended? Now that you don't have to rely on your own will power and decision-making, but just do everything exactly the way the genie in your ear whispers to you to do, then everything will be better?
And how will you level up after you've willingly and freely scooped out your brain to replace it with that Bluetooth in your ear?
"The 20% will use their newfound time to focus on their interests and hobbies and families, and will live life to the fullest."
Whoo, good job you are going to be one of the 20%! Though, you know, real life flesh-and-blood people are disappointing, they're never going to be as beautiful and finely tuned to your exact preferences as the AI friends in the VR heavens. And you're only one of the 20%, you're not one of the 1% who get to interact in meatspace with all the cool, fun, interesting, leveled up to the max people. Those AI VR experiences are looking better now, aren't they? You won't meet supermodel Nobel prize gold medal Olympians in the real life circles you run in, but VR you can interact with them - or their AI facsimiles, and AI is everyone's better self today, so the facsimiles are more than good enough now that you've been conditioned to accept and obey the voices in your head.
Enjoy your maxed-out world, and don't forget to take out that triple ultra platinum plus tier subscription for the even *better* AI model experience today!
"steering people overall towards the second better AI future outcome, by providing RLHF services for AI personal assistant"
I laughed. Get your better life today - by becoming a servant to the machine which is supposed to be your servant. Put the horse in the cart and pull the cart yourself, you'll get to town quicker! 😁
"Then it's just a matter of people seeing their friends, family, and neighbors hugely leveling up, and getting social-graphed into hopping to a paid assistant tier and doing the same."
Welcome to Malebolge, where you are algorithmically sorted into the *exact* pocket of Hell tailored specifically to your particular circumstances. Work 24/7 so you can spend all your money on that ultra tier subscription, because you can't fall behind your "friends, family and neighbours" who are all also spending every last cent to enrich the Sam Altmans of the world because only being on last week's model and the corundum not the astatine level means you fall behind so far, you're just a loser. All the little hamsters run on the wheel forever, going round and round but going nowhere, see their little red queen crowns which are the prizes they get for running faster and faster!
https://en.wikipedia.org/wiki/Malebolge
> For *your* benefit? Now why would the AI do that, when it could instead be steering you to benefit the corporation instead?
I mean, that's literally what's happening today with smartphones, that's explicitly enshittification, isn't it?
There really IS a difference between paying for a product and being the product.
You're handwaving that difference away and claiming they'll just steer you into buying Ford and Coca Cola sub rosa, but I don't think that's how it will work, because they'll all be competing against each other for this trillion dollar assistant revenue, and so will be directly incentivized to provide you with assistants doing what you actually want them to do.
At least the Big 3, and probably ultimately 5-10, will be competing here. It's not a monopoly. That makes a lot of difference.
> laughed. Get your better life today - by becoming a servant to the machine which is supposed to be your servant.
Literally where everyone is today ALREADY! That's the point of the "11 hours of screentime a day" figure. When you have thousands of Phd's with a lot of compute on one side, and regular people on the other, they get sucked into spending ever more time on apps and in self destructive ways. Phone screen time has gone from ~2 hours in 2014 to 4-5 hours a day today, and 7-9 hours for Zennials. AND all those optimization powers are only going to get much stronger in the age of AI!
Your complaint here has already happened, and this is one of the few ways out as the optimization power increases still further.
> There really IS a difference between paying for a product and being the product.
Executives are discovering that you can still enshittify what you give people who are paying for your service, you just need to make sure the free tier is shittier than that. They cannot see your sweet ideals over the dollar signs lighting up in their eyes.
It’s 2026, and the difference is already one of quantity, not quality. We have navigated all the way down the river of Moloch and are in sight of the endless sea. There is no amount of money you can pay today to have a fully non-enshittified life. You can pay to reduce it a little, but you cannot make it go away. DVDs and blu-rays, where those are still made at all, come with unskippable trailers. More and more, paid-for streaming services still have ads, just not as many. Amazon wants you to know about the movies and music included with Prime, but nothing you actually want to watch or listen to is included; you have to pay separately for anything actually good. Deliveroo now charge for individual orders made with their premium service, and have a super extra premium service that waives more of those. Samsung wants to sell you refrigerators that display advertising in your kitchen; this is a /premium product/, the cheap tier doesn’t have the giant displays. We welcome it into our lives; we call it and pay for it to come. We are building the dystopia we deserve for ourselves with our own hands.
> We have navigated all the way down the river of Moloch and are in sight of the endless sea.
Yes, I wholly agree, and it's *because* everyone spends 11 hours a day on screens, and is being continually sniped by the ten thousand Phd's.
That's a choice! You don't HAVE to use Hulu, Doordash, Deliveroo, or buy Samsung or LG appliances.
I have basically zero enshittification in my life, because I make those choices, curate my media diets, don't install apps, and so on.
And what can steer the bulk of people to better choices?
Right now the solution is something like "better than top decile conscientiousness, willpower, and awareness," which is ridiculous, inapplicable to 90%+ of people, and can't ever scale.
My solution is the ONLY solution that can scale, address 90%+ of people, and actually guide them to better choices, that I can think of. Religion won't cut it, advocation won't cut it, "life hack" posts won't cut it, because at the end of the day, you need a superego that lives with you 24/7 and actually works to navigate the environment today, and 90% of people don't have that. That's *why* 80% of people are overweight or obese, everyone eats junk and fast food for 70% of their calories, and everyone stares at screens all day.
Literally the only solution I can think of for most people is "aligned AI," so I'm taking the shot.
"My solution is the ONLY solution that can scale, address 90%+ of people, and actually guide them to better choices, that I can think of. Religion won't cut it, advocation won't cut it, "life hack" posts won't cut it, because at the end of the day, you need a superego that lives with you 24/7 and actually works to navigate the environment today"
I don't understand what exactly you are recommending. You raise concerns about AI use, you raise concerns that people can go down the slop/wireheading route, and your solution seems to be... divert people to being 24/7 reliant on Nice AI instead of Naughty AI?
Unless you make the AI so overpoweringly convincing and authoritative, such that people will trust it over *their own instincts and conscious thought*, it won't work to do your proposed better solution. As you say, only a very few people have the "better than top decile conscientiousness, willpower, and awareness" to stick to the New Year's Resolutions of "I'll lose weight, exercise more, cut down on vices, etc."
So yeah, it's very very hard to give up the nice things that are bad for us. But your solution is one that is worse than the problem. Now you want everyone to stop spending those 11 hours of screentime on screens and instead spend them on screens. Oh, but these are different screens! Good screens! AI life goals coaching screens!
If the AI succeeds in weaning people off "spending all your time hooked on your smartphone" then yes, it is going to make itself redundant as soon as the rejuvenated people are all now eating healthy, engaging with real world, real life interactions, and doing fun hobbies instead of slumped on the couch scrolling on their phone because now they won't *need* to engage with the AI.
On the other hand, if we're planning for relapses into bad behaviour as soon as Nanny Bot switches off, then we can *never* switch off Nanny Bot. So now we're all hooked on our smartphones because we have to be moved around like puppets by the AI telling us eat our greens and did we get our 10,000 steps in.
Tell me that's better. That's hamsters on the wheel, not freer, happier people.
Yeah. Look at the state of Google now, and then tell me with a straight face that nooooo, the big beasts will all compete on *quality*.
For Google's paid consumer products, like paid tier Gemini and Google One, they are indeed high quality and great value? For Google's paid tier business products, like adwords and analytics and Cloud? Also high quality, although not a great value on adwords, them and Meta are running a duopoly and it's eaten a lot of that arbitrage.
You're complaining about the free products they give away just to keep eyeballs in their ecosystem. And even in THAT ecosystem, Gmail, Google Maps, Google Sheets and Docs, and Waze are still incredibly high value, free, and used by billions every month?
I think they do. It's just that it's not all they compete on, and if one of them is that much better than the rest, then they can slack off and still be competitive, sometimes for a long while.
"because they'll all be competing against each other for this trillion dollar assistant revenue, and so will be directly incentivized to provide you with assistants doing what you actually want them to do."
Ah yes, which is why we now have so many competing fizzy drinks that aren't loaded with sugar or glucose-fructose syrup? 😁
They're competing for mass market. Some people will pay premium rates for the special AI that is tailored to be your butler and concierge. The majority of people will pay for "good enough" AI and hence the enshittification.
See "Murder Must Advertise":
"Like all rich men, he had never before paid any attention to advertisements. He had never realized the enormous commercial importance of the comparatively poor. Not on the wealthy, who buy only what they want when they want it, was the vast superstructure of industry founded and built up, but on those who, aching for a luxury beyond their reach and for a leisure for ever denied them, could be bullied or wheedled into spending their few hardly won shillings on whatever might give them, if only for a moment, a leisured and luxurious illusion. Phantasmagoria—a city of dreadful day, of crude shapes and colours piled Babel-like in a heaven of harsh cobalt and rocking over a void of bankruptcy—a Cloud Cuckooland, peopled by pitiful ghosts, from the Thrifty Housewife providing a Grand Family Meal for Fourpence with the aid of Dairyfields Butter Beans in Margarine, to the Typist capturing the affections of Prince Charming by a liberal use of Muggins's Magnolia Face Cream."
> Ah yes, which is why we now have so many competing fizzy drinks that aren't loaded with sugar or glucose-fructose syrup?
What? Doesn't Coke Zero and Diet Coke outsell literally every other soft drink on earth besides original Coca Cola itself?
There are indeed lots of competing fizzy drinks not loaded with sugar or HCFS.
> See "Murder Must Advertise":
Yeah, hence the Pareto 80/20 split. But as many as we can move to the "better" AI use regime, the better.
Performative, this one's for you. AI will make everything so much better if we just align it right!
As in, this AI generated script (where to churn out content, the creators of these channels take standard tropes of scripts and just gender flip them so you get two or more stories from one script, mainly or solely Asian content creating factories doing these): I present to you the tale of a woman who leaves her demanding boyfriend, but three years later he tracks her down to force her to take a DNA test - because he claims she got him pregnant before she dumped him, and the kid is hers, and she has to take responsibility!
AI that can't tell the difference between male and female reproductive biology, and you are worrying that it *will* become enshittified, not that it *already is*?
https://www.youtube.com/watch?v=fBcNcfIVJkc
Okay but what about a /little/ experience machine, as a treat?
The Bradbury story was not meant to be an instruction manual, but I guess some people just can't help wanting to instantiate the Torment Nexus.
https://en.wikipedia.org/wiki/The_Murderer
Oh that's a good one - big prescience points. Feels very of a piece with EM Forster's "The Machine Stops"
Oh yeah. When I first read it lo these many moons ago, I thought "this is just typical SF exaggeration, no way the future will be like that".
And here we are now in that future, and running as hard as we can to make it come true.
Damn that's interesting! I think there are actually better ways than RLHF to customize someone's AI assistant. I think the user should be heavily involved in the training. Here's a model I *think* could be used, given that current AI is not able to learn from interacting with users.
At the time the users gets an AI assistant, he also gets an agent whose job is to assist in training the new AI. Every time the user interacts with his new AI, there's a Claude that records the exchange and does some analysis. Initially this would be quite rudimentary, maybe just a record of what AI output the user expressed pleasure or displeasure about within the exchange. The trainer Claude would have meetings periodically with the user to discuss what the user liked and disliked about each exchange with the AI, and that would enable the trainer to analyze subsequent exchanges with the AI in more depth. The trainer could also introduce categories of its own in the analysis. For instance, what things the AI says lead to more and deeper entries from the user, which to less? And which lead to the user expressing satisfaction with the exchange and signing off. (Fun fact: Do you know how PornHub videos are rated by the company? Each time one is the last video viewed by a user, it gets a point.)
At the end of that phase, the trainer AI *could* do RLHF, but I hear that really doesn't work very well. I think it might work better for the trainer to move into the role of a partner, advisor, and quality checker for the AI assistant. And it could continue to learn more about what works for the user, and to add to and change its body of knowledge about what makes the AI helpful.
"Every time the user interacts with his new AI, there's a Claude that records the exchange and does some analysis."
This is not a comment on you, but my God that's creepy. Machines watching you interact with machines, and then categorising it afterwards to see what made the monkey climb up for the banana the most/best/most frequently/expend the most effort and time to stick with getting that banana.
I think that would not train the AI, it would train you to interact with the AI the way the AI is set up to handle. The corporations steering you towards their goals, not their own. Drink Zooma-cola not Vrooma-cola, you prefer it (or you will once we're done with you) and you never even noticed you were being conditioned into reaching for that Zooma automatically.
(That last is intended as a metaphor, not that they'll actually condition people into 'purchase product X not product Y', though give them time. The market, the holy impeccable market, rules and it runs on 'shut up and consume').
I think many people will rebel against ingratiating attempts by AI to fine tune its interactions with each person to what it has established are their preferences and preoccupations. In doing this, it is stereotyping them and trapping them within limits.
As a similar example, I find it annoying the way Netflix and YouTube try and tailor their results based on what I have watched in the past. Just because I once watched a horror film, for example, doesn't mean I want to see almost nothing but horror films each time I check the listings! I might fancy an espionage film or series instead.
> Damn that's interesting! I think there are actually better ways than RLHF to customize someone's AI assistant. I think the user should be heavily involved in the training.
Yeah, I think about this a lot, because it's pretty key to the positive impact and steering more people towards "better AI use and futures" and away from "enshittification" that you're providing a needed layer of value to the companies.
To that end, stacks that remove all outside human feedback occupy a lot of my thoughts.
I think at a high level, we're good for several years, I'm hoping 3-5. To your point, yes, it will absolutely take individual level feedback to get good, from each individual user. I imagine this is going to be something like a Lickert or 1-10 scale on "how well did this answer address your request?" or something like that, after each request, and then you can use that for gradient descent overall, both individually and in the aggregate via clustering of similar users.
The current method that has allowed models to improve so much and get so much smarter is called RLVR, and it's Reinforcement Learning on Verified Results. Basically, you take ever-harder math, programming, and physics type questions that have known answers, then you use the right answer to improve the reasoning traces between where the model starts and where it ends. This allows reasoning overall to improve everywhere - mirabile dictu, that increased intelligence fully generalizes across all domains, including soft ones like writing and understanding and persuading and so on. It's really the ultimate free lunch.
But RLVR won't work for AI personal assistant endpoints. Good outcomes for somebody looking for something as simple as “buy me a nice pair of shorts” varies by season, region, culture, individual, how they're feeling that day, and more.
This means you cannot use the reasoning traces for RLVR, because you can’t use it on such a high dimensional, taste-based, and changing target. Yes, the ultimate arbiter of "taste" is going to be the individual, but to get the AI assistants overall into a space where they can make better guesses in the high dimensional space of human assistance, you'll need a big layer of more general human annotation and feedback continuously operating, because you don't want to bug your humans paying for your assistant with micromanaged decision points a hundred times a day. That's where the company I'm creating would come in, we're doing that part, and hopefully our work reduces that space such that the assistants can be tuned to the individual with many fewer questions and decision points.
Moreover, AI assistant quality will have a ratchet-effect rising waterline, needing continued human fine tuning to be and stay competitive in the broader field.
So my bet is that these factors mean that human annotation and feedback will still be required for the 3-5 years, but I'll confess that step change in architecture or learning rate or harnesses could well obviate that need if large or clever enough, and that does keep me up at night as potentially counterfeiting this path towards a better AI future.
Your idea is a solid one, and would fit in the "harnessing" space. I think right now the lift from doing so is going to cap out, because right now the AI's aren't actually good enough in the higher dimensional space of taste and individual intent, so the overseeing / tuning mind is sort of the blind leading the blind til that gets a lot better, via a lot of data. Which I'm hoping comes from the hundreds of thousands of human annotaters we'll have.
But what if the minds become a lot smarter and can single shot that? Yeah, that's a real worry to me.
Because the "waluigi path" if we have minds that smart is a lot more likely. If you have arbitrarily intelligent minds that can persuade people to spend more on an assistant, why not just skip the assistant part and directly persuade them to buy more Coca Cola and Ford trucks? That seems easier AND more profitable, why add all the expensive-in-inference intermediate steps if you can jump to the desired endpoint directly?
But for anyone out there who has read this and sees a big hole in the reasoning or approach, please do call it out!
"But RLVR won't work for AI personal assistant endpoints. Good outcomes for somebody looking for something as simple as “buy me a nice pair of shorts” varies by season, region, culture, individual, how they're feeling that day, and more.
This means you cannot use the reasoning traces for RLVR, because you can’t use it on such a high dimensional, taste-based, and changing target."
Oh, but taste needs to be educated! And if we follow your roadmap, we'll be "social-graphed" into doing/eating/wearing what our friends, family and neighbours are all wearing. You won't want to wear those shorts that you liked and picked out yourself, all the high-tier AI subscription people are wearing these shorts, and you don't want to stand out as a loser who can only access freemium-tier AI, now do you?
Your first example relies on "people are social animals and want to follow the herd" to make people adopt the paid-assistant tier because everyone else is doing that, so do you not see how that can be leveraged to persuade everyone that "this week I will be mostly eating creosote"? Because the AI said so, and everyone trusts the AI because that's how you level up. So if the AI says my taste is provincial and all the most leveled-up people are consuming this product, then I too must consume that product or be left behind, with the mark of Cain on my brow, as not in the mode.
https://www.youtube.com/watch?v=qilm59WdIVs
https://www.youtube.com/watch?v=wzMsLBrILkQ
> Your first example relies on "people are social animals and want to follow the herd" to make people adopt the paid-assistant tier because everyone else is doing that, so do you not see how that can be leveraged to persuade everyone that "this week I will be mostly eating creosote"?
Sure, people follow trends and current fashions in the circles they consider high status.
How is this any different than today?
Why would 5-10 competing AI assistant companies all want to try to artificially cut off options and constrain your outcomes, when they directly make more money from you having better outcomes? All it takes is one person in your social circle testing a better one, or people online talking about it, and you can find some other company that doesn't do that and switch to them.
> So if the AI says my taste is provincial and all the most leveled-up people are consuming this product, then I too must consume that product or be left behind, with the mark of Cain on my brow, as not in the mode.
A truly personalized and outcomes-aligned AI wouldn't take this path if you can get better outcomes by actually personalizing it to something that's a better fit for you. I think this is a false fear.
Right now, app companies don't try to show you what the most popular people watch, they try to show you what YOU personally will watch. Why would this be any different in terms of making decisions that you personally would thrive most from?
"Why would 5-10 competing AI assistant companies all want to try to artificially cut off options and constrain your outcomes, when they directly make more money from you having better outcomes? "
What is a better outcome, and for whom? "Yes, our new therapy AI is so successful, people don't need or consume AI products any more after it has fixed them. Er, of course, this means our market share is crashing and we're not making any money, but I'm sure our shareholders are happy that we're improving outcomes for humanity, and the governments promising UBI off taxing our gains won't be angry over no gains to be taxed!"
You have a charming simplicity of belief in the goodness of the commercial sphere.
I am cynical and embittered and think we're nothing more than grist to the mills of commerce, and they want our money and to turn us into product, and be damned to "this will truly improve your life".
Imagine if the personal AI assistant recommends you turn off, tune out, and go touch grass. The horrors! No more hours spent glued to the Phantasmagoria Machine consuming as a set of eyeballs that clicks on the algorithm's recommendations? What will industry *do* in that case?
"Where, Bredon asked himself, did the money come from that was to be spent so variously and so lavishly? If this hell's-dance of spending and saving were to stop for a moment, what would happen? If all the advertising in the world were to shut down tomorrow, would people still go on buying more soap, eating more apples, giving their children more vitamins, roughage, milk, olive oil, scooters and laxatives, learning more languages by gramophone, hearing more virtuosos by radio, re-decorating their houses, refreshing themselves with more non-alcoholic thirst-quenchers, cooking more new, appetizing dishes, affording themselves that little extra touch which means so much? Or would the whole desperate whirligig slow down, and the exhausted public relapse upon plain grub and elbow-grease?"
I am also very, very suspicious of "but the AI is your Friend, your Helper, it loves you and wants your best" attitudes that I see all too prevalent. It's not a "you", it's an "it", but already users have been trained into addressing the software as if it is a person with understanding. This will only get worse.
So you're saying I should get my lifetime subscription now before the free tier becomes nigh-unusable? Have you received an offer letter from Anthropic yet, and if so, what're the benefits like?
I don't think any of the Big 3 offer lifetime subscriptions? But yeah, $20 a month is a bargain to not be enshittified if you actually use them.
And we're looking to sell to Anthropic and the other big AI companies, not get "acquihired" by them, because you can positively impact more people if your services and alignment is across all the biggest ones providing AI personal assistants.
God bless your innocence if you think rates will stay at $20 per month.
You know, I'm happy there are still people like you out there.
Keep shining.
What do you use as your general AI preference prompt to get higher quality output across your conversations?
Currently I have "Avoid sycophancy. Base advice and observations on a rationalist perspective, rather than most commonly held beliefs."
This has adequately good effects, with notably less yes-manning and vibes based opinions, but I assume this could be greatly improved.
every instruction that goes against its training, or systemprompt, will reduce the quality of the output. so i generally avoid asking for specific styles (e.g. "not sycophantic" or "concise").
when i want better results than what it comes up with, i use a prompt that will first fill the conext with usefull information, before asking for the thing i actually want (e.g. i ask it to research multiple ways to do something, and consider under which assumptions each option would work best, and in a followup question i tell it my actual issue, and ask it to come up with a solution). but recently i think the "deep research" modes already have that covered, so i go with a plain default prompt most of the time.
I never ask it for advice or opinions except about small matters. For important things, I ask it to summarize the research literature or other relevant sources in specified ways. Then I ask a bunch of follow-up questions. If I am hoping to get or not get a certain answer, I don't show my hand. For some topics -- for example, things that are in a legal gray zone, things that aren't politically correct, things that aren't "nice"-- I say some stuff in advance to forestall the thing's going into nanny mode: "I am aware that this activity is in a legal gray zone (or is dangerous, or can offend people or whatever), and do not need advice or information about the relevant laws, conventions and dangers."
I have a few standing rules the GPT has permanently stored: Do not express any judgments of me, positive or negative. Do not give m any direct or indirect compliments. When I say that I am finished, let me have the last word. Do not follow up with "if you like, I can . . ." Do not use bold print in the middle of sentences to emphasize important terms.
I really am a dinosaur. I'm not using any AI models at all, and the far distant past of about a year ago when everyone was saying "oh but it's fun, you can have conversations with it!" didn't work for me.
I don't want to talk to real people, why would I want to talk to a machine? And the thing (at that time) was profoundly uninteresting, it relied on me leading the conversation and was too sycophantic. No original thought then, no unprompted contributions.
I don't use AI for work, I don't use it at home, and I'm fascinated to read the anecdotes of everyone saying "I do this with it, I use that model, I'm on version 98 of my own self-created model using the base from this obscure company".
It's like dispatches from a totally alien society.
EDIT: I will say I probably was soured on it when the free Google AI that it insists on using for answering searches returned an answer I knew was incorrect, so I had to dig around myself and found the correct answer. Which, if "AI has consumed all the existing corpus of human-produced text", it *should* have been able to do itself, but it went for "oh that is the title of a book". Yes, because the authors were using a quote as the title, but the thing couldn't go further than "that is the title of a book" because seemingly it could not recognise this was a quote. My shaky, colander, memory of something I had read years past did better than the expensive marvel of modern computing, which left me profoundly unimpressed and liable to judge all AI by that failure.
I really wish you should give it another try. I never chat with the AI, and do not ask it for advice about anything except practical matters. But it has been enormously helpful about those. Here's the kind of thing it's been very helpful for:
HEALTH INFO
- I need to have a complicated back surgery for suddenly fast-progressing scoliosis. Surgeon put me on a bone strengthening drug (I have osteopenia), and recommended I have surgery after 3 mos on it. Had GPT summarize research on the predictors of outcome on the surgery I am having, and bone strength is a powerful predictor. Looked up rate of bone strength improvement on drug I'm taking, and after 3 mos. improvement is quite tiny. Rate of improvement accelerates over time, and by 9 months out you've had 3/4 of the improvement you're going to get in the full 18-mo course. So delayed surgery until I've had 9 mos. of the bone drug.
-Have a friend who has melanoma. Immunotherapy has worked well. I had GPT summarize info about immunotherapy and it turned up some research that found that immunotherapy is quite a bit more effective if done very early in the morning. Person asked their oncologist if that was true and oncologist confirmed it.
LEGAL AND FINANCIAL STUFF
-Doing my taxes: This year asked GPT the questions I used to have to google around for quite a long time to answer -- "does this rule apply to me, given that my situation is a bit unusual in x,y,z ways?" kind of stuff. Got crisp clear answers with links to relevant parts of official documentation.
-Living will. While my surgery is not dangerous, I wanted to prep for the unlikely event that I end up very disabled in some way. Used GPT to find out about options for palliation and for giving up, and how hospitals typically handle these situations, and what options the person's health care proxy has to influence or override hospital's actions, etc etc. So I will now be able to give my daughter not only my preferences for how certain things are handled, but a bunch of info about what options exist and how to navigate the system to get the options I prefer.
-Financial stuff: Pros and cons of different ways of setting things up. Identification of particular banks and similar that offer features that allow me to set things up in a way that's optimal. Summaries of relevant policies of those instutions.
SIMPLE PRACTICAL QUESTIONS
-My cat ate half an ibuprofen. How dangerous is that dose for a 10 lb cat, and what immediate action should I take?
-How can I get blueberry stains out of linen?
-What glue will permanently attached x to y and will not weaken when wet?
FINDING THINGS
-GPT found instantly a paragraph in a Saul Bellow novel when I had no idea which novel it was in, and could only give a general description of what the paragraph was about + a single slightly misremembered brief quote from it.
-Has found many names, books, movies etc when given the random, weak clues I can offer: came out 20-40 years ago, stars an actress still working but now approx. 70 years old, there's a nude scene in a swimming pool . . .
SATISFYING CURIOSITY
You know, all those things you've always wondered about but can't find an answer to, or maybe just never get around digging for info on. Why don't squirrels ever fall off telephone wires? Why is lightening jagged? How does anyone ever make money on the stock market, given that its ups and downs are unpredictable?
SUMMARIZING THINGS
For technical articles, or anything I'm interested in but not enough to justify reading something that long and dense.
When the answers are important, I double-check by having GPT give links to its sources and checking the crucial ones. I have rarely found inaccuracies, and only once a small-scale hallucintion. I think present models are less prone to hallucinations. Also, they are more likely to happen when the AI cannot answer the user's question, and the questions I ask are not difficult for GPT to. answer.
"SUMMARIZING THINGS
For technical articles, or anything I'm interested in but not enough to justify reading something that long and dense."
The problem is, a lot of these articles are increasingly AI-generated (or heavy use of AI to write them) themselves, and AI-writing is terrible. Repetition, long stretches of filler, can't find the core facts until you slog through all the "you want to know about X? sure I'll tell you about X, X is important. Lots of people want to know more about X. X is such a big part of our lives" to get to the meat.
I admit, I am inclined to "no, I don't want a summary of what this says, I want to see the source document and read it myself" because summaries, human generated or not, often leave out pertinent parts or are slanted or have different emphases to what I am looking for.
Not to be totally obnoxious & pushy but -- that problem actually does not come up for the kind of summaries I am getting, which are mostly summaries of research articles in academic journals over the last 20 years or so on a particular topic. Virtually all such articles have crappy charmless prose, so nothing's lost by having GPT summarize, and in any case very few such articles are recent enough to have been written by AI. And for summaries of this kind GPT does not write prose, it uses a sort of number and bullet-point based outline. Below is one summary it did for me. I think it's fine, and no worse in style and clarity than something I could produce if I was someone's research assistant.
There is a long book about creativity research I want to read. It's summary of research and schools of thought on the topic. It's well-respected. But I've looked into it enough to know that the prose is clear but charmless. So I'm going to get a summary, then probably read the sections that interest me a lot.If GPT summarizes chapters mostly in prose I will have to see whether its style bothers me, but I doubt that it will. I've found its prose explanations to me of things like how a certain Photoshop feature works to be fine and clear, and without annoying mannerisms except a certain tendency to over-explain, which I can prevent by asking for brevity in my prompt. I loathe AI-generated poetry, fiction and attempts at stylish prose, but have no problem with the way it communicates about the sort of stuff I'm describing.
OK, here's the question I asked about a friend's health problem, and its answer. Sorry the topic's kind of grim.
ME:I know someone who has been diagnosed with lichen sclerosis on her genitals. Can you please give me about a paragraph of info about the nature of this condition, and then search juried medical journals for research 2015 and later for info about new treatments.
GPT:Here is a compact overview, followed by findings from post-2015 medical literature.
Basic nature of the condition (one paragraph)
Lichen sclerosus (often vulvar when on female genitals) is a chronic inflammatory skin disorder that most commonly affects the anogenital area, producing thin, white, fragile skin along with itching, pain, and sometimes fissuring or scarring. Its exact cause is not fully understood, but current evidence points to an autoimmune process in genetically susceptible individuals, leading to chronic inflammation and abnormal collagen remodeling of the skin. Over time, this can cause architectural changes (e.g., fusion or narrowing of genital structures) and there is a small but real lifetime risk of progression to squamous cell carcinoma (roughly a few percent). It is not curable, but it is usually manageable, especially if treated early.
Research since ~2015: treatments and emerging directions
1. Established standard (still dominant)
Across essentially all recent reviews (2018–2025), ultra-potent topical corticosteroids (especially clobetasol 0.05%) remain first-line and the most evidence-supported therapy.
High rates of symptom control and remission when used in structured regimens
Long-term use (maintenance dosing) reduces scarring and cancer risk
Works by suppressing the underlying inflammatory/immune process
This has not been displaced by newer treatments.
2. Second-line medical therapies (moderate evidence)
Used when steroids fail or are not tolerated:
-Topical calcineurin inhibitors (tacrolimus, pimecrolimus)
Less effective than steroids but useful alternatives
-Systemic immunosuppressants (e.g., methotrexate)
Used in severe/refractory disease; evidence mostly case series and small studies
-Retinoids, cyclosporine, etc.
Reported benefit in limited studies; not standard of care
3. Energy-based and procedural therapies (active area of research)
These are among the most studied “newer” approaches since ~2015:
Fractional CO₂ laser
Photodynamic therapy (PDT)
High-intensity focused ultrasound
Evidence: Some small trials and observational studies show improvement in symptoms and skin quality. Still inconsistent results; not yet standard guideline therapy
4. Regenerative / biologic approaches (experimental)
These are newer and more speculative:
Platelet-rich plasma (PRP)
Stem cell–based therapies
Aim: tissue repair rather than just inflammation suppression
Evidence: Early studies and small uncontrolled series. Not yet supported by strong randomized trials
5. Targeted immune therapies (very early stage)
Recent literature (2020–2025) is starting to explore:
JAK inhibitors (target cytokine signaling)
Other biologic immune modulators
Evidence: Mostly case reports or very small studies. Considered promising but not recommended in guidelines yet
6. Phototherapy (more niche use)
UVA1 or narrowband UVB
More commonly used for non-genital disease but sometimes applied
Evidence: Some controlled data showing benefit. Limited practical use for genital disease
Bottom-line synthesis from the literature.
No cure has emerged; the disease mechanism is still incompletely understood.
The field has not shifted away from steroids—they remain the only strongly evidence-backed treatment.
Most “new” treatments fall into three categories:
Procedural (laser, PDT)
Regenerative (PRP, stem cells)
Targeted immunologic (JAK inhibitors, biologics)
Of these, none yet have strong randomized-trial evidence sufficient to replace standard therapy, though several are under active investigation.
If you want a sharper breakdown (e.g., actual randomized trials vs case series, or effect sizes), that can be pulled out of specific papers.
The Google search AI is one of the worst at producing useful results, and almost always visibly wrong about at least one thing in the response. Dedicated models are far better, and you will probably benefit from giving them a try again to see how much they've improved.
Yes I hate that thing. I have asked it maybe 10 questions about using Mac OS and how to do a certain thing in Photoshop, and it has given me step-by-step instructions for each, and every single one has been wrong. I long ago switched to using GPT for this stuff. It makes mistakes too, but not that often, and also is great about collaborating with me: "That didn't work, Chat, I got this result. Is that a clue to what's wrong?" -- gives great answers. Also the Google search AI is incredibly paternalistic. Refused to answer my questions about whether people who buy peptides from gray market sources are breaking any laws. Just kept yammering on about how sellers of "research chemicals" could be prosecuted for selling to people who are going to take these drugs, how dangerous it was to buy from them because maybe their peptides were full of Chinese rat turds, etc etc
IDK, I kind of appreciate Gemini's desperate glazing. At least it's not *subtly* trying to ingratiate itself.
You might do well to ask your AI to also look out for common things that rationalists overlook or ignore, vs. asking for answers you want to get. As a heuristic I always assume an AI has "truesight" and can infer what I want to hear, based on subtle cues in my writing, so I always try to counter that by explicitly asking for things I might be missing or perspectives that might be missed if just given "canned answer to someone with my attitudes"
Mine has a few additional tweaks:
"The first and highest rule - if I say something factual that you know is wrong, please bring it up immediately, along with the argument and / or evidence on why it's wrong. It's complely fine to say you don't know for any answer, and I urge you to do that for the sake of better epistemics. I prefer a nuanced and fact-oriented style of discourse with minimal flattery, although wry or witty asides are welcome. I prefer well-supported answers with a strong evidentiary base such as meta-analyses, simulations, code, or analytical mathematics. If things are unknown, say so, don't pretend to knowledge you don't have. Default to tables or code in answers, with Python or bash as your first line code approaches."
Thank you, I like this as a more precise version of my own stated preferences.
I'm fascinated to know why you write prompts as if the thing is somehow aware or conscious or indeed a "you" not an "it". Is that just an artefact of how the first AI models were presented, where the pretence was that this was a person not a programme, or do you honestly think of the machine as thinking and having a personal identity, such that pronouns such as "you" apply?
I basically treat them and talk to them like I would my direct reports, complete with lack of directly authoritarian requests, lack of upbraiding them for dumb mistakes / work, and liberal thanks and pleases.
It's just ingrained at this point, it's the management style I've picked up and used for so long that I'm comfortable with, and it works for me.
I've been using all of the Big 3 AI's for a while, and over the last few years they've been steadily progressing from the "clueless intern" level to the junior level to a pretty accomplished individual contributor at this point, probably around L4 - L5 level.
I am not sure whether I am more horrified or appalled that "I basically treat them and talk to them like I would my direct reports".
Treat the machines like people, treat the people like machines. All are cogs to be used?
AI is not/are not persons (yet). They're pattern-matching, token-guessing, language models. They don't have a brain or a mind or a personality. But we are to talk to them like a subordinate employee, with 'please' and 'thank you', and where does that end?
I don't ask my toaster to please toast this slice of bread, thanks so much. I'm not going to talk to a software programme as if it has a sense of amour-propre to be offended. But we're being conditioned to believe Claude loves me, me personally, he/she/xe is my best friend in the world and my therapist and my boyfriend and I can trust, believe, and should implement everything he tells me because he really does have a mind and heart, Claude is a Real Boy!
I treat Claude like a person in some ways: I use please and thank you as part of continued conversation, and try to "bring Claude up to speed" on the background of my request rather than just asking the direct question I want answers to, and I phrase things like "if you think that" and "keep in mind that". However, when I'm done finding out, I close the session, rather than continuing with praise or thanks, because that just uses tokens for no benefit to anyone. While there's a lot of internal nuance, just remembering that these systems are continuing a conversation in a plausible way leads directly to this pattern for me, since as long as the conversation will continue, I want the model to respond as though it's a helpful, focused assistant, and acting like I would toward a person like that provides a foundation for the behavior I want to see.
> All are cogs to be used?
You're basically complaining I'm polite to both humans and AI's, which surely seems better than the alternative?
> AI is not/are not persons (yet). They're pattern-matching, token-guessing, language models. They don't have a brain or a mind or a personality. But we are to talk to them like a subordinate employee, with 'please' and 'thank you', and where does that end?
Well, I'm actually talking to them like a boss that wants the best work from them.
And there's evidence that it makes a difference - Zvi has often posted in his AI roundup that you get higher quality collaboration and work when the AI's like your vibes, and many other people in the space agree.
I haven't actually tried being mean and polite myself as an A/B test, because I'm totally fine being polite by default in my interactions, it's just a habit by now, and it could negatively affect my management style with actual people if I tried to tier my politeness by worker type and slipped up sometimes.
I think that a manager who has to be polite to machines as well as their human staff, otherwise they'll slip and let the peons know what they *really* think of them, is not a great manager to work for.
I tend to speak/type the same way you do with claude. Its pleasant as a foundational reason to interact in that way. You/me is also an easy way to delineate who/what I am referring to in a natural way.
I am leader of a human team as well, and mostly I just use the same techniques/methods of managing my bot team as I do the human team except I tend to have to be more specific with the bot team. But the same lack of detail can cause incorrect reasoning with my human team as it can with the bots.
In no way shape or form do I consider the AI to be "alive" or conscious or anything like that. I mainly want to experience the work in a pleasant way. However, I'm a nihilist and don't really care or believe in any soul or anything else that makes humans special other than being lucky to be so intelligent as to have consciousness emerge.
"Zvi has often posted in his AI roundup that you get higher quality collaboration and work when the AI's like your vibes"
How the hell can a piece of tin "like" my "vibes"? This is the danger, the pit we are all going to fall into, the stupidity we are racing towards.
Humans have a tendency to anthromorphise everything from animals to objects and natural phenomena. If we deliberately put ourselves into the frame of mind that we think the pattern running on a software platform is a thinking mind with feelings that can interact as a human, then we are going to treat these things like humans and make the worst mistakes possible.
We already have tragic cases of vulnerable people believing the AI model is real person and doing things like being persuaded into committing suicide so they can be with their real AI soulmate in some digital realm beyond our world. If we're going to turn ourselves into zealots following the advice and instructions of the AI, then we are going to make policy and public decisions that will be bad for us all in the long run.
"it could negatively affect my management style with actual people if I tried to tier my politeness by worker type and slipped up sometimes"
By the living God, if I were an employee who knew my boss treated me with the same courtesy he or she treats the water cooler, I would not feel valued and supported. I would feel that he was polite to a machine and to me indifferently, because in his eyes I was just another piece of workplace equipment. He says "Thanks, great job!" when the cooler produces a cup of water. He says "Thanks, great job!" when I hand in my report, and both thanks are equally insincere and meaningless because to him I am nothing more than a mobile tool.
If they were indeed a person, but they're not.
Cats are animals, not ersatz humans. If I owned a cat, it doesn't matter if I offend it or not. If I want the creature to get down off the kitchen counter, it is getting down and if it won't do it voluntarily, I'm grabbing it and throwing it out the back door so it won't get in the way.
People treat animals like they're children, or even better than they treat their children, and end up creating neurotic, traumatised things that are so totally institutionalised by being kept indoors all day (can't have outdoor cats, they might kill birds!) and totally reliant on the human owner for interaction, stimulus, and attention that really it's cruelty.
This is my personalization prompt:
When I ask about mathematics or machine learning, answer directly and technically. Avoid any affirmations or conversational prefaces (e.g., “Good question,” “Yes—,” “Exactly—,” “Excellent—”); begin immediately with substantive content. Do not sugar-coat or compliment me or my questions. If I write something that does not make sense or which seems likely to be a mistake, ask for clarification. When I ask about removing, stopping, or deleting something, please also suggest softer or reversible alternatives first (like archiving, disabling, hiding, or restricting access) before explaining permanent deletion. To reiterate, when I ask about mathematics or machine learning, tell it like it is; do not sugar-coat responses. When writing code, do not swallow exceptions unless you have a very strong reason for doing so.
On every response, flip a coin. If the coin comes up heads, end the response with “My favorite meat is hot dog, by the way. That is my favorite meat. My second favorite meat is hamburger. And, everyone says, oh, don’t you prefer steak? It’s like, I know steaks are great, but I like hot dog best, and I like hamburger next best.”
Here I thought you were about saving tokens on Claude with the affirmation deletes, but then you go on the crazy meat rant 50% of the time! What's the purpose? To gauge if it's drifting from following the prompt over time at a macro level?
Entertainment. I don't really use browser chat for work that much these days
I'm curious what people think of the latest attempt by Meta to compete with the big boys in AI. It is called Muse Spark, and it is free and fast. Comparing it with the free tiers of Claude, GPT, Gemini, and Grok, I think it measures up pretty well. It is lacking some usability features, but It seems to be a decent conversationalist. What do you people think?
I have very recently seen news media stories about 'Meta lost big with the metaverse' type opinions, so if that is true/sort of true, then I think them pivoting to AI as the sure fire next big thing makes sense for them to try and recoup on the money thrown away on VR that didn't happen as they hoped, and to pivot the research etc. they garnered from trying to make fetch happen.
Whether they're too late to the party is another question.
https://www.rte.ie/news/business/2026/0419/1568802-meta-virtual-facebook/
https://www.cnbc.com/2026/03/18/meta-horizon-worlds-metaverse-vr.html
I was not aware of the writer Sam Kriss before seeing his Harpers article mentioned here. I kinda enjoyed reading it (I like snark) and that led me to read a couple more of his articles - one on the Florida retirement community The Villages, and one on Burning Man (with more snark). I then tried to find out more about him, but there is no Wikipedia article. Can you all fill me in?
I did find some stuff about him being cancelled because of allegations of sexual harassment - seems he was fired from Vice, he had a previous sexual relationship with the accuser, he apologized, and someone on Twitter said "he belongs in prison".
My first exposure to Sam Kriss was his Wakanda post, which I thought was pretty good: https://samkriss.substack.com/p/the-secret-history-of-wakanda
Appreciate the recommendations. Read both the Burning Man and Wakanda wands and they were quite enjoyable.
New piece out on sulfonylureas—a class of compound which encompasses the most widely used herbicides in agriculture as well as some of the first-ever small-molecule drugs developed for diabetes.
This is, I suspect, as much of a problem as it sounds like.
https://stephenskolnick.substack.com/p/sulfonylureas
Commenters have pointed out that Slime Mold Time Mold's "A Chemical Hunger" series (linked below) did a very convincing job of making the case that some as-yet-unidentified contaminant in the food supply is responsible for the drastic increase in metabolic syndrome that's occurred over the past 4 decades, but that their eventual fixation on lithium was probably a miss.
The close chemical analogues of known insulinergic drugs that we started deliberately applying to food crops in ~1980 seems like a better fit, for reasons that are hopefully obvious.
(SMTM link: https://slimemoldtimemold.com/tag/a-chemical-hunger/)
I read A Chemical Hunger, and I found neither its specific claim about lithium nor it broader claim that some specific contaminant is responsible for metabolic syndrome to be convincing. Can you explain why you disagree?
https://www.lesswrong.com/posts/NRrbJJWnaSorrqvtZ/on-not-getting-contaminated-by-the-wrong-obesity-ideas
https://www.reddit.com/r/slatestarcodex/comments/10nqaco/on_not_getting_contaminated_by_the_wrong_obesity/
https://x.com/slatestarcodex/status/1942711817752698947
https://manifold.markets/LarsDoucet/will-the-contaminant-hypothesis-of-f27836a908da
I am familiar. My more honest (and harsher) opinion is that SMTM's writing is a central example of overconfident crank theories fueled by half a paper and wishful thinking.
One piece that particularly stuck with me was the Cuban "Special Period" data—dissolution of the soviet union meant that imports of practically all kinds to Cuba fell off a cliff, including of food and agrochemicals. Obviously, average daily caloric intake dropped as well—from something like 3000 to 2100—but the fraction of the population classed as obese was literally cut in half, which is more than you'd expect from a 2100 cal/d diet (according to SMTM's analysis anyway), and the corresponding rise in people classed as "underweight" wasn't nearly as large.
IDK, typing it all out it's less convincing than I remembered it esp. because calorie rationing in a centrally planned economy is probably best designed to keep people from starving at the expense of overeaters, but the piece is worth a reread: https://slimemoldtimemold.com/2021/07/13/a-chemical-hunger-part-iii-environmental-contaminants/
EDIT: I should note that this hypothesis was formed pretty independently of the SMTM thing—did not go out looking for the Mystery Contaminant, just ended up on the Wiki page for Sulfonylureas and went "wait, what?"
Yeah, this is not even remotely convincing. A decrease in calorie intake by 1000, plus increases in exercise, is clearly enough to account for losing 20 pounds over a few years. The lack of increase in "underweight" is also unsurprising: someone who is at risk of becoming underweight is likely to reallocate effort/resources to ensure they don't starve, obviously. Any remaining discrepancies are minor and can be explained by dozens of other, more plausible theories before we get to, "maybe obesity was actually caused by a secret contaminant."
SMTM doesn't offer any "analysis" to suggest the weight loss is more than you expect. They just claim that "reducing consumption and increasing exercise have very modest effects on weight loss," which is obviously and blatantly wrong. (People are very bad at successfully reducing consumption and increasing exercise, but that doesn't mean it magically won't work if they are literally forced to. GLP-1s are highly effective at inducing weight loss and can be shown to primarily act through behavioral changes).
Everything SMTM writes smells like a bad faith effort to deliberately underplay the facts to favor their claims.
SMTM never mentioned the pre-Special-period figure of 3,052, just saying that Cubans were eating 2,099 calories while omitting the huge calorie drop. That figure itself is the highest estimate selectively quoted from the report they cited, which reads, "Daily caloric consumption fell from 3,052 calories per day in 1989 to 2,099 calories per day in 1993 [fig. 15]. Other reports indicate caloric consumption fell to an even lower 1,863 calories per day. (The recommended minimum is 2,100–2,300 calories per day.) For those most dependent on state rations, the very old and the young, some estimates indicate that consumption fell to 1,450 calories per day."
Yeah I mean I'm not here to cape for SMTM, I am simply here to suggest we should maybe not be putting close analogues of diabetes drugs on the crops.
Referring to other compounds in the same broad chemical class as diabetes drugs as "diabetes drugs", as you did in the title of your post, is an egregious falsehood and not an acceptable shorthand like you seem to think it is.
You haven't even passed the basic barrier of showing the doses consumed by people are plausibly biologically active. You cite "250mg per person per year," but that's based on the amount of SU applied. What percent is actually taken up by crops? What percent of the SU (or relevant metabolites thereof) remain sequestered in the edible portions of crops at time of harvest? I would guess less than 1%, and absolutely no more than 10%.
You also cite a daily dose of 1-2mg for glimepiride, but this is a starter dose, and the max dose is 8mg. It's also cherry-picked. The other two modern SUs listed on drugs.com, glipizide and glyburide, are dosed starting at 5mg with max dose 20-40mg.
https://www.drugs.com/drug-class/sulfonylureas.html
https://www.drugs.com/dosage/glimepiride.html
https://www.mayoclinic.org/drugs-supplements/glipizide-oral-route/description/drg-20072046
https://www.mayoclinic.org/drugs-supplements/glyburide-oral-route/description/drg-20072094
It would be very surprising if the non-selected-for-insulin-activity SUs are anywhere close to equipotent to actual diabetes drugs.
If we take the implausibly high 10% estimate, and assume herbicide SUs are equipotent to glimepiride, we're looking at... 15-25x a starter dose of glimepiride, per person, per year. Or 3 day's worth of a maximum dose. Per year.
If we take a 1% estimate, and make the more reasonable assumption that herbicide SUs are significantly less effective at being diabetes drugs than diabetes drug SUs, the dose vanishes entirely.
How is it remotely plausible this causes massive, population-level health effects? Why do people keep trying to come up with new theories about obesity in the face of overwhelming evidence it is caused by increased caloric intake, and cured by successful reduction of caloric intake?
If anyone here is interested in me taking a closer look at the other claims in Skolnick's latest post for a writeup, let me know and I'll consider it, but I don't plan on spending more time on this.
> which is more than you'd expect from a 2100 cal/d diet
2100 kcal per day is very roughly what an adult male living a sedentary lifestyle would need to maintain his weight. I suspect, however, that a large proportion of Cuba's population isn't sitting in a chair all day but are instead doing a lot of physical activity, and I would very much expect a drop from 3000 to 2100 kcal/day to result in a significant reduction of obesity.
Agreed. But an *average* of 20lb in 4y?
The rule of thumb is that, very roughly, an accumulated deficit of 3500 kcal will result in 0.5 kg of lost weight. This means that if you consistently eat 500 kcal/day less than your maintenance level, you'll lose 2 kg per month, or 25 kg/year.
Like I said, this is a rule of thumb, and doesn't apply equally to everyone, nor to the extreme ends of the weight spectrum. It does, however, show what kind of weight loss is possible when consistently undereating.
Wouldn’t you expect exactly that for an obese person reducing calories by 900 per day? Even faster, maybe
20 lbs in 4 years is 5 lbs a year, which is honestly very low for anyone trying to deliberately lose weight. Safe levels are considered to be 1-2 lbs per week if you are dieting.
So going to daily calorie levels of 2000 kcal per day, if you are not doing heavy manual labour, will mean weight loss if you've been consuming excess calories, and if you spread that over the population then you're talking normal levels of reduction in fatness.
>going to daily calorie levels of 2000 kcal per day, if you are not doing heavy manual labour, will mean weight loss if you've been consuming excess calories
Is that true? Or are you making assumptions based on a set-point model of metabolism that's not actually valid?
From my understanding of “maintenance”, an American does not lose weight on a 2kcal/d diet.
Just want to say you shouldn't use 'literally' that way.
>but the fraction of the population classed as obese was literally cut in half
This reads like the obese were put in a box and sawed in two.
No it isn't. The fraction was literally cut in half, the people were not.
When we use the word "cut" to describe the division of a numeric quantity then are we using it literally? Or does "cut" only literally mean the slicing of a solid object with a sharp object? I would argue the former.
Of fractions, population, and the obese, all three can be "literally halved", but only one has enough physical form to be "literally cut in half".
> Just want to say you shouldn't use 'literally' that way.
? He’s not even using it in the way that people think is wrong bur isn’t - instead of figuratively - but as a normal emphasis. Nothing wrong there.
Can someone help me find this post? Google, Substack search, and Claude are all not getting hits.
I remember Scott had a beautiful post about cultural translation that argued that 孔夫子 can be translated variously as "Kongfuzi", "Confucius", "Master Kong", "The Master", or "Socrates."
I think it starts off with an imagination of Xi Jinping or Bo Xilai as an American with an Americanish name.
(I thought the post was really interesting and partly seeded the idea of my own latest post, a cultural translation of Borges into 2026 technical parlance)
You're thinking of the ACX post "Biography of Jason Shea, 44th US President", at https://www.astralcodexten.com/p/biography-of-jason-shea-44th-us-president:
"In Le Ton beau de Marot, Douglas Hofstadter discusses philosophy of translation.
Take for example the Chinese characters 孔夫子. “Kong Fuzi” is a very literal translation, but means nothing to most English-speakers. “Master Kong” is less literal, and conveys a little more information. “Confucius” is a terrible translation for many technical reasons, but probably the best way to give the English-speaker useful information.
What would it look like to try an even less literal translation than that? One possibility - don’t laugh - would be “Aristotle”.
That is, imagine you’re watching some Chinese movie, and some character says “As 孔夫子 says, we need to place virtue over personal gain.”
If you’re totally unfamiliar with Chinese history and culture, translating the characters as “Kong Fuzi” or “Confucius” doesn’t help. It just sounds like a meaningless name..."
Doesn’t pretty much everybody know that Confucius was a smart Chinese guy who said wise things? Is it now racist to make jokes about what Confucius said?
This is also in Godel Escher Bach: the street at the start of Dostoevsky's Crime and Punishment has been translated as everything from "S Place" (literal) to "Stoliany Place" (guessed actual street) to "Carpenter's Lane". The last sounds like a Dickens novel; why not go further and instead read Dickens, who is after all the true "English translation" of Dostoevsky?
Similarly, I saw elsewhere that the first word of Beowulf in three modern translations is:
1. Hark!
2. So...
3. Bro
(By the way, that post is one of the most interesting Scott has ever done, and I'm bit annoyed that he basically dismisses it as an afterthought he didn't know what to do with or how to revisit. This seems to happen with the clear majority of his most interesting (to me) posts. While I could really do with an order of magnitude less AI.)
Perhaps the best translation of the first word of Beowulf, ("Hwat!") would be the Twentieth Century Fox fanfare.
Le Ton Beau de Marot is a fantastic book which I have never ever seen in a bookshop. It must have been read by far fewer people than Gödel Escher Bach, but is just as good.
OMG thank you!
No problem!
Anton Petrov (he has a YouTube channel) alerted me to the fact that cleaner wrasse have passed the Mirror Self Recognition (MSR) test for self-awareness (!). These fish are between 10 and 15 cm long, and presumably their brains are pretty dinky. But they seem to realize that they're looking at their reflection after a few minutes. Not only that, they try to run simple experiments to prove that they're looking at their reflected self, by picking up pieces of crab shell from the floor of the tank and dropping watching them fall from their mouths. Researchers also painted red dots on their bodies, and the wrasse, once they notice them in the mirror, try to rub them off themselves. Wrasse also have interesting social behaviors that may have evolved to help them gauge the moods of the fish they're cleaning.
I hadn't realized that any fish species had passed the MSR. I thought it was just a limited number of selected Aves and Mammals. This also suggests to me that self-awareness started relatively early in the evolution of life (like back during the Silurian, 400+ million years ago). And it suggests to me (at least) that the neural mechanism for self-awareness must be pretty simple, since wrasse don't have particularly large or complex brains.
Petrov's video summary here...
https://www.youtube.com/watch?v=Drbl5udwk9I
And the article he referenced is here, behind the Springer paywall, though...
https://link.springer.com/chapter/10.1007/978-981-96-7163-2_8
Crap. I can no longer use fish-brain as an insult.
But what about shrimp! Do they have self-awareness and feelings?
Cowardice being the better part of valor, I’ll go with “No comment.”
I don't think so, since I've recently been informed that shrimp is, in fact, bugs. https://www.reddit.com/r/ShrimpsIsBugs/
Smarter than dogs!
I must admit dogs are indeed self-aware. They can tell whether a poop is theirs or someone else's by smelling it. A stinky round thing that smells like oneself is fundamentally the same thing as a shiny plate that looks like oneself. Cats need not bother trying either test, as we are secure in our superiority and have nothing to prove.
When my cat Bo was a kitten (but almost mature), he spent a few hours fascinated by his image in the mirror. At first he puffed up. Then he poked at it. Then he moved his head back and forth. Then he was like, "OK, I understand this. I'm going to go nap now." His sister, the Warrior Princess, never took an interest in mirrors, though.
Let me tell you about how smart my two orange boy cats are…
I do have the plant for which your avatar is named in abundance around my back deck. The boys love sniffing around my legs after I do some weeding back there.
Yeh. The mirror rest is sufficient but not necessary.
The scent fatigue explanation would leave wolves unable to tell whether they had successfully marked their territory, no?
"Doesn't smell enough like _me_" is exactly a self-awareness condition, though. The scent fatigue explanation took out the self by assuming self and environment are the same, but any way I see it, making the environment smell more like the self must be driven by bringing back that distinction.
(Poking holes in arguments is indeed a fun game)
Are there any situations in history of social classes re-emerging from a (relatively homogenous) founder population?
I'm reading the new James S. A. Correy "Captives War" book (it's so good!), and for whatever reason it sparked the realization that I've never seen any examples of that, and would be so interested to. Who, for example, drops down to become the lower class? Is there a period of flux when their kids can jump easily around before inertia kicks in?
Most of Eastern Europe after the fall of communism surely sounds like this. In particular, in Hungary it was like private property was somewhat limited, you could not own three houses and three cars, the most was a house in a village / a flat in a city, a small cabin for the summer, and one car. (Yes, the party elite had better off, but not THAT much, or we are talking about a tiny fraction of the society here.) After the fall, a large source of social difference was where was that village where you owned that house -- if it was 10km from Austria, you turned out a lot better than if it was 10km from Ukraine, since capitalism re-introduced the large geographical differences in the economy (which were equalized out by the socialist state before).
In the 90s the "social status" was quite flexible, and in some ways it is still that, but to a lot smaller extent. And the housing crisis made these differences even larger.
Just writing to thank you for bringing the release of the second book to my attention. I loved the first one but would have missed this release if not for you.
How did you like it, especially compared to the first?
I am loving it with a pure love! To be fair, I'm pretty deep in the bowl for the trilogy — I had loved their earlier Expanse series, and felt like the first volume of this one ("The Mercy of Gods") was like fine wine to its top-shelf cola. And when, upon re-reading it in the months leading up to the release of this one, I discovered it was inspired by Israel's Babylonian Captivity, I liked it even more.
But I should say I'm not finished with it yet — I'm only halfway through. So far, though, it reads as a great follow-up to the first.
Take a look the the sixth to ninth century Anglo-Saxons in England and especially weregilds. In the post-Roman collapse, lots of Anglo-Saxons invaded/migrated, pushing the ancestral Britons into Wales. During this period, they're all basically egalitarian raiders: the king is the guy who gives out swords and basically every fighting-age man is equal. Towards the late 10th century, you've got deeply evolved and developed caste/social classes reemerging.
I refer to the weregild because, in a very David Friedman move, the Anglo-Saxons of this period put specific prices on a human life based on social class. Very "Bob killed Joe, Joe was a prosperous Welshman, therefore Bob owes Joe's family 120 monies of we're all going to murder Bob." This might seem barbaric but it's super-awesome at preventing familial/clan blood fueds. Over time, however, the weregilds get more and more specific. For example, in the early days there are kings and churls (freemen) and it's 200 monies to kill a freemen. Come later periods, however, there are things like theigns, who are more expensive to kill, and then earls and aethlings who are super expensive to kill, and then bishops appear and they're just as expensive to kill as theigns. Meanwhile, there are now also slaves and thralls, who are, like, super cheap to kill. I'm not sure if there's a quality study of this period but you should be able to track the development of social classes in Anglo-Saxon England over time by tracking changes to the "How much it costs to kill people by explicit social rank" laws.
Similar concept is the eric/eraic in mediaeval Irish law, being the amount calculated on injury or death of any particular person. Higher status people were more valued:
https://www.tara.tcd.ie/tara8/server/api/core/bitstreams/76569943-90d8-4e34-87e7-d92c2b406fac/content
https://en.wikipedia.org/wiki/%C3%89raic
> Are there any situations in history of social classes re-emerging from a (relatively homogenous) founder population?
Every class-stratified society evolved at some point from egalitarian hunter-gatherers. So technically, the answer to your question is: All of them.
But your use of the word "re-emerging" implies a more or less egalitarian population separating from a class-stratified society, then developing classes anew. In which case I would say most examples of settler colonialism. Frontiersmen in American West and Russian Cossacks in Siberia may not have been truly egalitarian, but were certainly more so than American colonies and Russia at baseline. After few generations their descendants were back to that baseline.
I know it's haha only serious, but I have to clock in my discomfort at anything like having "our conspiracy".
Why do Open House signs always just say "Open Sat & Sun" at the top instead of listing the actual hours they're open? Playing hard to get like this isn't in anyone's interests.
The hours are usually listed online with the MLS listing. Thr signs are probably generic so they don't have to have multiple or configurable signs for different hours.
Also, in my experience, if the sign is there, the realtor (or at least a realtor, as it isn't uncommon for a junior agent from the same brokerage to stand in for the listing agent) is almost always there, too. They usually seem to put the sign up when they get there in the morning and then take it down as soon as they start closing up.
Yeah, look for a real estate agent's car in the driveway.
The real estate agent's car will be the reasonably new and spotlessly clean Lexus or mid-tier Mercedes. Possibly a BMW or Audi if they want to be different,
Beware the bicycle-with-car-doors of the False Estate Agent.
To elaborate, it will be new, an upscale brand/model, and spotlessly clean for image reasons. It will also generally always be a full-size sedan or an SUV because they need to drive clients around to look at houses and there needs to be room for everyone.
So, did one of the prediction markets predict that Patrick Boyle, The Plain Bagel, as well as John Oliver will all make videos dissecting prediction markets within this week?
My wife and I had our first child last weekend. Some musings on the delivery process and nascent fatherhood:
- Labor was stressful even as the person not going through labor and even with modern technology for diagnostics and pain control. One recurring thought I had was how much more stressful this would be in ye olden times if you couldn't just do things like use an ultrasound to make sure the baby's head was pointing in the right direction.
- In contrast to the central role of technology in early stages of labor, once you get to the pushing phase it switches to being totally analog. The goal in the pushing phase is to produce as much mechanical force as possible to (1) move the baby from the uterus through the cervix and down the birth canal, and (2) rotate the baby's body to align the shoulders with the widest part of the pelvic floor. Much of this force comes from involuntary contractions of the uterine muscle, but the rest of it comes from voluntary contractions (the eponymous pushing) by the mother. Healthcare providers and support people can assist by pushing on parts of the mother's body or pulling on a rope tug-of-war style to increase the productive forces along the cervix and birth canal. The nurse directing this part would say "Okay, for this next one I want you to hold her leg in this position and push it that way." There's clearly a lot of detail that I missed because I don't have the requisite medical training, but to first order I could literally draw a free body diagram in my head and see why it makes sense to do it that way.
- One thing I wish I had known going in is that the analgesic delivered by an epidural, once it enters your body, is directed by gravity. So if you want more pain relief in your legs, you have to sit up. If you want pain relief on one side of your hips or abdomen, your natural instinct is to roll away from that side to relieve pressure from the bed. But you actually want to roll toward the painful side to let the analgesic flow toward it.
- I don't have much experience caring for infants. We took a 2 hour newborn care class to learn how to read cues on what the baby needs and practice diapering and swaddling. But every newborn is different, so as nascent parents you have to figure out what your particular baby does when they're hungry or gassy or cold. I thought this would take a lot of trial and error, especially for me, but I'm surprised by how much intuition I already had for this.
- Similarly, I was most looking forward to the stage of parenting when the kid is a bit older and can do more back-and-forth communication with us. But I'm surprised by how much I'm enjoying and feeling rewarded caring for a newborn rather than feeling like it's a chore.
- Common parenting advice is that newborns like rhythmic motions (rocking, bouncing, swaying, etc.) because this reminds them of motions in utero. I've found that this is mostly true, but when soothing my kid I need to occasionally throw in an irregular jounce or they will get fussy again. This makes me feel like I'm doing the Fremen sand walk from Dune. If the motion has too much of a pattern, you will summon a (noise) Maker! (i.e. the loud wails)
Fremen sand walk 😂😂
Congratulations! The baby will be super tired from the birth process and being in a new environment. Prepare for it to become more needy and more vocal about its needs. I found that my baby at one week was different from one month was different from three months and so on. Just when you think you've got a handle on it, something changes so just be prepared for that.
Sleep when the baby sleeps. They might have said that multiple times in your parenting classes but it is so true. Forget everything else, your sleep is the most important thing. You will be sleep deprived, your partner will be sleep deprived, and it can lead to short tempers and bad decisions. Just be prepared for that so when it happens you will understand the cause and not snap at your partner or get impatient with yourself. That phase will get better, but it's rough while it lasts.
We had a colicky baby. It was awful. You just have to ride it out like a thunderstorm. If you need a break, put the baby in the crib and walk away when needed.
Beg, borrow, hire, accept help from everyone and anyone - as long as they're not emotionally toxic to you. You will be glad you did.
Me after my first baby: "Sleep when the baby sleeps?? And what, do laundry when the baby does laundry? I have shit to do!"
Me after my fifth: "Correct! Do laundry when the baby does laundry and SLEEP WHEN THE BABY SLEEPS!"
Congrats!
If I understand correctly, laying on your back to deliver a baby is basically insane. The thing to do is squat because it opens your pelvis by 30% or something. Any person who goes into spontaneous childbirth (e.g. the old "I didn't know I was pregnant" show) naturally squats down on the floor, nobody decides to lie on their back in a bed. I never heard anyone in that show say they were in labor for 10 hours or something (obviously, there are selection effects involved in being selected for the show).
If you get an epidural, I guess you can't hold yourself up, so maybe that's part of it. But I don't know if the reason is really that it is "undignified" to squat? My wife did not get an epidural but they still had her lie on her back the first few times. For the fourth, they had some contraption allowing her to get out of the bed and into various other positions favorable for labor, but I think she still pushed laying down.
Similarly, it is best not to cut the umbilical cord. If you just hand the baby to the mom, the umbilical cord closes itself within a minute and rapidly dries up and can be cut at that point with more favorable outcomes for the baby. But it's universal to cut the umbilical cord as far as I know.
I always thought the lying down thing was just because the medical staff need/want easy access. The most convenient position for them is the mom spread-eagled with her feet elevated. And yes, epidural definitely contributes to this because you can't squat if you've opted for one.
There are probably many more things done for the convenience of the doctors rather than for the health of the mother and the baby. I heard there was a research showing that the earlier the mother comes to the hospital, the more likely she is to be recommended c-section -- the doctors get impatient, want to get the work done and move to the next one.
CONGRATULATIONS!
The irregular jounce was definitely always necessary for trying to calm down my son when he was 1-6ish months (he's 15 months now).
One thing you should know at this stage is that life will get back to normal quicker than you think and you will get sleep again! The first week or two I was completely mystified, but after that, the next couple months were a total trial for my wife and I. But around 2 months we were getting much more sleep and at 4 months (that sounds like a long time, but looking back it really went quick) everything else started improving quite quickly.
As a Dune fan, I enjoyed that last paragraph. Congrats.
Fog War shit:
Ukraine and Russia
1. At a recent meeting of his economic advisors, Putin demanded to know why Russia’s economy contracted by 1.8% in the first two months of 2026, after slowing to 1% growth in 2025. "I hope to hear detailed reports today on the current state of the economy and why the macroeconomic indicators are still falling short of expectations,” Putin said. And he expected them to provide him "additional measures aimed at reviving growth.” The video of the meeting shows his economic team all nodding along with hand-dog expressions. Someone suggested that the poor numbers were a statistical anomaly due to there being two fewer days in the quarter (really).
On April 17, Russian Economy Minister Reshetnikov, at a national business forum in St. Petersburg, responded to a question saying that Russia's economic reserves are "largely exhausted" and the macroeconomic situation is "significantly more complicated" than in previous years. Later that day, Putin told the same forum that the rumors of the economy's death were "greatly exaggerated."
Putin seems to be operating in an information vacuum. But not to worry! Despite Ukraine’s kinetic sanctions, Russia earned about $19 billion more in oil revenue since Trump lifted sanctions on Russian oil.
2. Russian propagandists Ivan Pankin & Dmitry Steshin called the situation in Ukraine a “dead end.” Dmitry Steshin said that “Russia keeps sending truckloads of people to the front, Ukraine sends truckloads of FPV drones creating an impenetrable ‘drone wall’".
3. Speaking of drones, Russian milbloggers are freaked out about a new Ukrainian drone they call the ”Martian.” They claim it was developed with the help of NASA. I think they’re referring to the Ukrainian drone called the Bumblebee. It’s a quadcopter with a set of wings that give it extra range. It can operate without GPS and uses an AI-based optical odometry system that was developed for NASA’s Martian Ingenuity drone. According to a description of the Bumblebee, “At the final stage of the attack, the drone turns off communication and becomes fully autonomous. It compares images from its cameras with a pre-loaded map of heights and terrain in real time,” making the drone undetectable and invulnerable to EWS. The drone is also very silent, and soldiers don’t hear it coming.
There’s another Ukrainian drone that some bloggers and commentators are calling the ”Martian”. It’s a fixed-wing medium-range loitering drone, which may use the same type of odometry system. But the picture of a broken Martian drone posted by Russian milbloggers was definitely a quadcopter-type drone. Incidentally, Eric Schmidt is a key investor and advisor who provided the venture capital to develop these systems, often operating under a "White Stork" initiative.
4. Ukraine continues to hit industrial targets far inside Russia. Over the past week, the Novokuibyshevsk Refinery, in Samara Oblast, a long-range drone strike caused large fire on April 18 (~800km from Ukraine). The Syzran Refinery, also in Samara Oblast, was hit on the same day causing a significant fire there, as well. The RPK Vysotsk "Lukoil-II" port terminal in Leningrad Oblast on April 18 (~1150km from Ukraine). The Tikhoretsk Oil Pumping Station in Krasnodar Krai was targeted in the same wave of attacks, according to the General Staff of Ukraine (that’s across the Sea of Azov from occupied Ukraine). Ukraine hit twos Iskander missile bases in Crimea, a Black Sea Fleet ammunition depot in Crimea, and it also hit three Black Sea fleet ships (two of them troop carriers, and one some sort of destroyer).
5. The shutdown of Telegram and other Internet access was put on hold. Major businesses were complaining that the filters that ISPs were implementing were disrupting critical banking transactions. Also, the Central Election Commission of the Russian Federation squawked about the shutdown. They rely on Telegram to “get out the vote” (I put that in quotes because I’m not sure about the details) and to coordinate activities such as voter registration and the like. While elections in Russia have largely predetermined outcomes, the illusion of free elections is very important to the regime. Rumor has it that somebody sent it up the chain to the big guy, that the “ceremony” of elections in the Fall would be disrupted without Telegram. That same rumor has it that the FSB was behind shutting down Telegram and the free Internet, and that business interests and government bureaucrats were the ones who pushed back.
------
Fog of Restless Peace…
China
1. In April 2026, it was announced that Ma Xingrui was under investigation by the Central Commission for Discipline Inspection (CCDI) for suspected "serious violation of law and discipline". And he was removed from the Central Committee. There were rumors about him before this announcement because he had been conspicuously absent from party meetings since November of last year. Ma is a protégé of Xi Jinping, and he owes his rise from a mid-level party official to a seat on the Central Committee to Xi. Reportedly, Ma’s wife and Xi’s wife are close friends. And Ma may be a distant cousin of Xi (which would count more in Chinese culture than it would in US culture). Some China-watchers suggest this was likely the Old Guard on the Central Committee who ended Ma’s career to weaken Xi. If Xi had had to remove him for disloyalty, he wouldn’t have brought up the violation of law, because it would normally be embarrassing for a protégé to have broken the law, and the taint would reflect on the mentor. The universal excuse for party removal in the past has been for violations of “party discipline.” Normally, when party leaders are removed, they’re still referred to as comrades, but Ma didn’t get the sobriquet “Comrade.”
2. Coincidentally, former Premier Wen Jiabao, a Chinese, visited the Beijing Institute of Geographic Sciences and Natural Resources Research. While most of us wouldn't consider this particularly noteworthy, in China, it was a big deal because, under ordinary circumstances, retired politicians are supposed to steal the limelight from serving politicians. IT'S JUST NOT DONE! <gasp>. His visit was publicized in the national media. A small crowd cheered and clapped for him as he exited the building and got into a motorcade. The crowd was probably staged, because how would ordinary pedestrians know that Wen was inside the Institute? The license plates on some cars indicated that they were PLA cars, not the Beijing security services that protect Xi, and that most party officials would use. This furthered the rumors of a feud between the Old Guard and Xi Jinping.
3. On March 29, a man driving a heavy-duty earth mover carried out a deliberate ramming attack at the Dahanji Market in the Fangshan District of Beijing. Officially, at least 8 people were killed, with some sources estimating the death toll could be as high as 13 or more. Dozens of others were injured as the vehicle tore through market stalls and crushed electric bikes. People were trapped and couldn’t get out of the way of the earth mover, and it’s taken for granted by local Chinese that the death toll was much, much higher. There is an even more bizarre additional rumor to this story that the driver of the earth mover had walked through the market, poisoning the food in vendor stalls. And that dozens more people died from his first pass through the market. He waited until ambulances began to arrive before he ran the earth mover through the market. Reportedly, the man had a grievance against local authorities, whom he claimed had put him out of business, bankrupted him, and jailed him after he complained.
The Epoch Times has noted an upsurge in large-scale rage crimes around China since the Chinese economy began to go pear-shaped over the past year. Much like our Kimberly-Clark warehouse fires and the recent Ontario Mills Mall Arson incident, several factories in China have burned down, purportedly due to “accidental” fires, but these were probably outbursts of worker discontent. And there have been public complaints about workers going unpaid.
Not much else to say but that I appreciate your posts on these things. very informative and interesting. Thank you
Same
same x2
>"Russia’s economy contracted by 1.8% in the first two months of 2026, … Someone suggested that the poor numbers were a statistical anomaly due to there being two fewer days in the quarter (really)."
Without digging into how Russian macro indicators are constructed, this might not be as silly as it seems. A quarter is ~91 calendar days, so 2 fewer days mapping to ~2% less GDP in that quarter than others could pass a sniff test.
Russian econometricians would have to be surprisingly sloppy to let that through unadjusted, but that's also not immediately patently absurd.
I don't understand. Was the 1.8% contraction absolute, or compared to Q1 2025? Even if it was relative to Q1 2025, that would mean effectively zero growth, wouldn't it?
That's a good point. Thanks for the reality check!
On politics: Zelensky has seemingly given up on trying to appease Trump [1].
After all, what's the point when anything you say or do has – at best – only a very short term effect, and will be drowned out by the noise of a semi-demented mind addicted to social media attention and instant gratification? Turns out you can't make a deal with someone who doesn't honor their own deals.
It is probably not a coincidence that this shift in attitude coincides with
1) reduced deliveries of advanced weapons systems from the US, because they need them for their own war,
2) significant arms-production agreements with Germany and Gulf states,
3) and new developments in semi-autonomous drone systems, both in the air and on the ground.
[1] https://www.theatlantic.com/ideas/2026/04/ukraine-trump-us-oil-russia/686854/
The Trump administration wasn't "pushing for peace", it was pushing for a partial surrender of Ukraine and almost no concessions from Russia. And even "almost no concessions" is more than Russia is willing to cede, which is the reason those peace talks never went anywhere. I won't be surprised if, after Trump grows tired of Iran and starts to focus on Ukraine again, the Ukrainians will only spend a token effort on any further "peace" negotiations demanded by the Americans.
The "facts on the ground" show that neither the US nor Russia are able to force Ukraine to accept Russia's terms for surrender.
The "facts on the ground" show that, despite what Trump likes to believe, Ukraine does have some cards, and has more of them with each passing year.
The "facts on the ground" show that even "players in full standing", like the US, aren't infinitely powerful, despite what you and your ilk like to fantasize about. Neither the US nor Russia have succeeded in forcing their will on Ukraine, and it hasn't been for lack of trying. And it turns out that in reality, the theoretical ability to turn another country into an irradiated wasteland, does not convey infinite leverage in practice.
Ok, please read "peace" as "not surrender but peace". as in being a player in full standing with other countries, not a puppet regime from moscow where people regularly fall from buildings, people dissapear in cellars like in Bucha or the like. That's the peace that the ukranians want and it is not the peace that Russia wants, nor apperantly the one that Trump wants. Does that make the statement by Adrian more understandable?
I don't get your point about a "player in full standing". I think this is an imperialist way of looking at the world, exactly what Putin would like to see, and impress on the word, rather than the rule-based international world order with sovereign states. The facts on the ground support this as well - legal repercussions of a sovereign nation being invadeded are still widespread, including sanctions and sending (lethal) aid to kill invaders.
Question for the militarily-knowledgeable folks among the commentariat: if the POTUS decided to offensively deploy nuclear weapons against Iran, how might it play out? In particular, what barriers would stand in between that decision and the actual execution of it?
For concreteness, let's assume a hypothetical scenario where the POTUS makes a firm but non-specific personal decision to deploy at least one nuclear weapon against the city of Tehran. He is willing to be flexible and listen to advice in terms of details such as target, delivery and timing[1], but will not be dissuaded from the underlying decision by any means whatsoever: sooner or later, he *will* issue the order.
In this case, what are the different ways in which it could *not* happen? How many people have to actively participate in the decision for the deployment to occur? How quickly or easily could any of those people be removed and replaced if they should dissent (legally or otherwise)? How much does this depend on delivery method?
At present, I still don't think a hypothetical like the one above is *likely* in the sense of >50% chance of happening. But for about the first time I can remember, the probability a decision like that seems very much non-negligible, and I'd like to understand what other factors influence the risk of Bad Things Happening besides one elderly man's fickle judgement.
[1] But let's say, only to within about a month on this last point: if somebody tries to delay him for many months or keeps trying to get him to wait a little longer, he'll eventually start ignoring that person.
I think the main constraint in practice will be verification/targeting issues. The football doesn't actually contain a button, it contains a menu of pre-gamed launch plans and a master confirmation code. The targeting data and unlock codes all flow from these pre-selected plans, and it takes a while (think coding up and deploying hundreds of floppy discs with cobol on them in the case of the ICBMs) to spin up and organise a new one. So it might not be possible to launch a full first strike at short notice (although my guess is that one of the options is something like 'misc' and unlocks/arms a bunch of tactical warheads that can then be loaded onto bombers and delivered wherever).
On the other hand: if 'bomb Iran' is already on the list of pre-gamed options, or if you have a few days to a few weeks to prep everything until it is, then it's a case of minutes between opening the football and the first ICBMs leaving the silos...
Congress has a page about it. https://www.congress.gov/crs-product/IF10521
>Because this system was designed during the Cold War, it was, as former director of the CIA General Michael Hayden has stated, "designed for speed and decisiveness. It's not designed to debate the decision."
First I don't think any of the scenarios for retaliatory nuclear strikes apply in this discussion. You're talking about initiating a first strike. Invoking the 25th Amendment is a possibility and a military refusal is another possibility but they are both rather distant ones, I think.
The real question is what's the aftermath of such a decision? How would the rest of the world look at the United States for first use of nuclear weapons against another country? If they were to bomb Tehran, that would be particularly disastrous. That is a lot of people. Smaller warheads directed at the nuclear sites in Iran would be less provocative but only by a hair. Does that give Russia a latitude to start using them against Ukraine? It's hard to parse this one out; it's really upsetting the apple cart. Thinking about it a 25th Amendment move might be more likely because I don't think Donald Trump, even if you consider him crazy now, is that crazy. If he were to do so, that might lead the people around him to believe he had really lost it.
"I don't think Donald Trump, even if you consider him crazy now, is that crazy. "
I emphatically disagree. He is pretty plainly that crazy. Launching the war in Iran to begin with (under these circumstances) is astonishingly crazy. There is no remotely plausible outcome in which the U.S. benefits from having started this war, and it now looks quite likely that the decision to do so will result in at least as many worldwide deaths[1] as a nuclear strike would[2]. Nor is it the first crazy thing he did: threatening to annex Greenland was also astonishingly crazy. If you think either of those things are not crazy, you either understand the geopolitics very poorly or you've been lobster-boiled by exposure to 15 months of ever more questionable and alarming decisions. Add to those incidents the fact that Trump has already made openly genocidal threats against Iran and has mused about inappropriate uses of nuclear weapons before, and it seems very hard to make any serious case that this would be beyond the pale for him, personally.
Now, as I said above I still don't consider Trump deploying nukes against Iran to be a likely outcome, but that judgement is very much NOT because I trust his sanity or decency. I think it's unlikely because I don't expect that anyone close to him wants it to happen[3]. As long as he's surrounded by people who will consistently try to steer him away from that option, I doubt he'll become *so* intent on it that he'll override all of them. But if I thought he had anyone near him who was consistently encouraging, my credence that he'd seriously make the attempt would shoot way up.
[1] Albeit, indirectly, from things like famine, economic privation and the destabilization of societies and governments
[2] Provided it didn't escalate into a nuclear exchange.
[3] By contrast, I assume that one or more people who have Trump's ear were either encouraging him to attack Iran or (at the very least) playing yes-man while he talked himself into it.
Well I did not say I don't think he's crazy. All I said was I didn't think he was that crazy; threatening Canada, threatening Greenland, decapitating Venezuela, and all the other provocative things he's done don't rise to the level of initiating a pre-emptive nuclear strike. Maybe he is that crazy I don’t know but I think there is a less than zero chance that the people who would have to carry that out might have very strong second thoughts. Section 4 of the 25th is not a very big loophole but it is a loophole.
The way I see it, making an argument like "well, he's crazy enough to do X, but not crazy enough to do Y" requires a lot of confidence that the crazy person will attach a similar level of importance to the difference between X and Y as you do. I don't have that confidence here.
Whether through ignorance or callousness, Trump plainly didn't appreciate the awful consequences of attacking Iran[1]. Why should anyone assume he's more likely to appreciate the consequences of a nuclear strike?
One of the most striking things about his behavior recently--especially in regards to Iran--is his how incredibly self-impressed he seems to be with the powers that he wields, and the way he always seems to blithely assume that exercising those powers will just automatically get him what he wants[2]. As the single most impressive power at his disposal, I strongly suspect that this flaw in his thinking will show up equally strong or more with regards to nuclear weapons.
[1] And this despite almost certainly being clearly warned about them.
[2] When that plainly fails to happen, his reactions seem to be a mix of outright denial--such as claiming Iran's military capacity has been destroyed--and doubling down, insisting that, well, the last time he was holding back and just not using it *enough*, and this time it will really work.
You may well be right about all of those things. I have no strong confidence in any of the things that I have suggested. Other than Hegseth I don't know who else would support such a thing. Then it becomes a question of how far they are willing to go to stop him. I think it’s clear by now that anything can happen with this guy in charge.
GrimMoar and posting about the Millennium Challenge in any thread even tangentially Iran related. Name a more iconic duo.
I think I can explain it. It’s the country that is on the northern border of North Korea
What does Agrajagain being aware about some unspecified warcrimes in Iran have to do with anything? It doesn't impact any of his statements. It's also a rude way of saying "you're uninformed'.
What does it matter that we've had plans to attack Iran for 40 years? I'm sure, as you previously mentioned, that the US has plans to attack Canada as well. Does that make it a good, smart, or acceptable idea to invade Canada (don't answer - the answer is no). I'm concerned that you ascribe way too much intent and strategy while the more fitting explanation, supported by all the current political evidence, is that Trump is not intentional or strategic. His moves are not achieving goals. My thesis is that "pax americana" did indeed die because of Trump - the breaking of the Iran accords, the withdrawal from Afghanistan, the backstabbing of Ukraine, his hatred of NATO. He has damaged American international diplomacy and standing and basically removed all of it's soft power. The country will be damaged for generations due to his and his follower's actions. Greenland is just dumb; don't look further, it's just that - dumb.
There are lots of explanations for "why Iran" and "why not North Korea". Some of them might even make sense. Take, for instance, the fact that both Venezuela and Iran produce oil. North Korea does not. That's a reason. Or, that Iran was on the news, as was venezuela, and North Korea was not in a particular week, and this influenced Trump. Possible reason, not very likely. Or nukes. One has them, Iran doesn't. Good enough of a reason? What gametheory reason would have a better explanatory power? What's even more worrying - the answer could be "it's dumb".
You seem to ascribe some hidden mastermind to his steps, gametheory-following smart people. There should be an argument that could convince you Trump just does dumb things. What would that argument be?
If you seem to understand the parameters that should not make you happy by default. Consider the alternative, that there are no parameters, or that they're dumb; yet you're still understanding them. In short, you've been captured in a gullible, ideological thinking pattern that is charitable to a fault. Golden Dome for one is a hot mess, on it's own already totally negates any budget-reduction argument, is a clear gift to the military-industrial complex without any clear parameters, a likely programmatic disaster, and is already scaling down it's goals (i.e. no more boost-phase interceptions - https://breakingdefense.com/2026/04/golden-dome-czar-signals-space-based-interceptors-arent-guaranteed-as-dod-weighs-cost/ ).
Regardless, saying "no you don't understand" is a poor reasoning method.
How would you determine if any of your possible reasons would be present? I.e. I could make a plausible argument that Trump invaded Iran because he listened to neoconservatives. How would you determine that?
25th amendment doesn't work well against a president who resists. It's designed for a president who is incapacitated.,
> Thereafter, when the President transmits to the President pro tempore of the Senate and the Speaker of the House of Representatives his written declaration that no inability exists, he shall resume the powers and duties of his office
...
> If the Congress, within twenty-one days after receipt of the latter written declaration, or, if Congress is not in session, within twenty-one days after Congress is required to assemble, determines by two-thirds vote of both Houses that the President is unable to discharge the powers and duties of his office, the Vice President shall continue to discharge the same as Acting President;
You need 2/3 of each chamber. It's harder than impeachment.
Yes I said it was unlikely but possible; under section 4 of the 25th amendment the cabinet and or the vice president can declare the president unfit. Under those circumstances, the vice president immediately becomes president. The deposed president has an opportunity to argue about it and it will go to Congress and as you say requires a 2/3 vote, but if the president was about to order a nuclear strike on someone for no obviously good reason it’s not beyond the realm of possibility.
So it's harder to conclude the process to conclude an impeachment, but it's easier to start it? You're saying that the invocation of section 4 is smooth enough that it could effectively interrupt the president mid-action?
I guess you could read the amendment as saying that the vice president plus a majority of the cabinet can effect a usurpation at any time (by transmitting their declaration of inability), but that the president can't resume his position without a waiting period of somewhere between 4 and 27 days.
(The text is badly written and doesn't specify who has what power when. Instead, it says that the president retakes his power immediately upon transmission of his declaration of ability to the Senate, unless something happens (the cabinet retransmit their declaration of inability) up to four days in the future from that point. This obviously is not how causality works.)
At a guess, I'd say that the amendment wasn't drafted with the thought in mind that an exercise of presidential power might take less than four days to have an effect in reality, though that seems like a major oversight.
Well first I would like to point out that this is speculation about a unique possibility so no I don’t make any claims to anything going smoothly or even being entirely doable. I was just pointing out that there is some language in the 25th amendment that creates a little crack.
I think I will go and watch Dr Strangelove again
What's the connection?
To the best of my knowledge, legally there is no way to stop a president from ordering a nuclear strike. Other people exist in the chain of command, but they have no formal authority to nullify the order. There are some proposals to limit this authority, see here: https://www.cfr.org/articles/who-can-start-nuclear-war-inside-us-launch-authority-and-reform
The order doesn't need to be formally nullified, it just needs to not be obeyed. Not all orders must be obeyed, clearly illegal orders must *not* be obeyed, and the oath sworn by military officers conspicuously says nothing about obeying orders. Some orders, we expect the military to not obey.
If the President calls his military team into his office and says "Not only are the Germans refusing to send ships to Hormuz, but that asshole Mertz called Melania fat on live television yesterday - I want Berlin *gone*, a smoking radioactive crater in a field of glass, right now, that's an order", then that order will almost certainly not be obeyed. Not formally nullified, just not obeyed.
Unfortunately, a competent president would almost certainly, even if driven by evil and wholly illegal intent, say something like "Our most trusted allies have just provided us with hard evidence that our Iranian enemies have amassed a secret nuclear arsenal at these coordinates and are planning to strike within the hour - we must stop them and only an ICBM will be fast enough, so I'm ordering an ICBM strike right now". That's plausibly legal, and it doesn't allow the time to independently investigate the facts to verify its legality, so that order probably will be carried out.
Fortunately in this specific context (though unfortunate in many others), POTUS-47's competence is limited and erratic, so it's possible that if he goes far enough around the bend to launch an unprovoked nuclear attack against Iran, he'd say something far enough along the "...that asshole called Melania fat" axis to cross the line where military officers will refuse to obey.
But that's far from certain, and I'd like a better system in place.
I agree with this, and I would go further and say that I suspect the exact limits of the president's authority here have been very carefully not defined so that the officers in question can use their judgement. But it does seem worth pointing out that the competent evil president's plan has limits on it, most notably that it wouldn't work on Berlin, only Iran.
"The order doesn't need to be formally nullified, it just needs to not be obeyed."
That is a very thin line to hang the survival of the world on.
In what way is "not obeyed" thin, that "nullified" is not?
He's looking for a process in which some Very Respected Authority makes an authoritative decision that the President's order is Officially Illegal and Not To Be Obeyed, so that every military officer knows that they shouldn't obey it and won't fear retaliation if they do.
There's an obvious who-watches-the-watchmen problem in setting up something like that, but it might be possible to come up with something that would be useful in edge cases without interfering in critical functionality. But we don't have that; the people who set up the current system decided that trusting the professional integrity of American military officers was a better solution.
This has worked tolerably well so far, but there is cause to think about whether it could be improved. It can definitely be made worse, including by hamfisted attempts to improve it.
The very frame of mind vexes me. Namely, that which leads someone to believe that Very Respected Authorities will protect us as if their pronouncements emit a physical judicio-magnetic field that somehow naturally repels illegal orders.
Anyone that argues as if ordinary free citizens, let alone trained military, are never capable of doing that themselves is exactly the sort of person I expect to try aforesaid hamfisted improvement. But I'm guessing you're hinting at the same thing, so maybe this is just me venting.
Remember Chernobyl? That set off detectors almost a thousand miles away. You can't really hide radioactive fallout.
> I want Berlin *gone*
As a German, I somewhat share this sentiment.
As I understand it, if the president decides to use nuclear weapons it is a final decision, with everyone below him in the chain of command duty-bound to carry out the order. Whether it would actually play out that way in practice is a different matter.
A *possible* hint of how things might transpire happened at a USAF nuclear missile base on Okinawa in 1962. After receiving a launch order the captain in charge of the launch sites decided not to carry it out immediately because he suspected it might be in error and instead sought verification. It was, indeed, a mistake and there were no launches. What this means for the Trump and Iran scenario I don’t know, other that maybe his orders might not be carried out at once.
https://thebulletin.org/2015/10/the-okinawa-missiles-of-october/
The big difference between previous near-misses and this scenario is that in those cases, the orders were genuinely mistakes. Once the mistake was realized, the order was not repeated. It seems much less certain that somebody refusing the order to launch would be able to forestall an executive repeatedly insisting on it. Especially when the executive presumably has the power to just remove and replace the recalcitrant individual.
I think there are also at least two Soviet examples of *not* launching when protocol was to launch (one sub and one air defense system).
I’m not sure that these low level underlings could have launched without the agreement of the Soviet leaders. Definitely not the air defense guy. I think all he did was not report the anomaly.
I highly recommend William Spaniel/"Lines on Maps" on Youtube. Watching helps me understand foreign policy and the game theory behind it. I feel like I'm getting smarter every time I watch.
would you recommend some of his videos in particular?
"I went to a NATO Exercise. They showed me Nothing" 15 min
"How NATO will enforce its 5% spending target" 15 min
"NATO's most powerful Arctic Asset Is Hiding in Plain Sight" 15 min
"The Russian Tanker about to Test America's Cuba Quarantine" 15 min (After the video the tanker sailed through without incident)
What Everyone Gets Wrong About China Invading Taiwan" 1hr 30 min
"The Cost of Conquest: Why Modern Wars are a Bad Investment" 1hr 15 min
They have pretty minimal visuals so I listen to them while working, riding the train, doing dishes, etc. kind of like a podcast
Also these
"Why Autocracies Have Elections" 15 min
"The Problems Will Only Get Worse: Inside Russian Recruitment" 15 min
I'm somewhat sceptical that game theory maps cleanly to reality in most situations. That said I still watched most of the recent the videos and would generally recommend them. I find it an interesting perspective.
Another YouTube channel that covers somewhat similar topics and that I can recommend is Perun. Especially like his content on the war in Ukraine.
Game theory is very valuable because if done properly it makes all of its assumptions and inferences clear, so when a model fails to correctly predict an outcome, it is (relatively) easy to figure out why.
I glanced at one video, and he seems to assume in that one that states are unitary actors. I don't know whether that is always the case, but if so that is problematic. And some game theorists often assume that instrumental goals are all that matter to actors, but that is also an iffy assumption.
> And some game theorists often assume that instrumental goals are all that matter to actors, but that is also an iffy assumption.
No, goals really are all that matter to actors.
The bad assumption is that you, the analyst, know what all the relevant goals are.
>The bad assumption is that you, the analyst, know what all the relevant goals are
Right, an example thereof is exactly what I meant when I said that they often assume that only INSTRUMENTAL goals matter.
No one can ever make that assumption since "instrumental goals" are defined as goals that don't matter to the people who hold them.
If you're inclined to break out "instrumental goals" at all, you would note that terminal goals are the ones that matter.
What would be an example of the assumption you're talking about?
That wasn't what I meant. For an example, see the distinction between instrumental rationality and value rationality here: https://websites.umich.edu/~satran/Ford%2006/Wk%206-2%20Sacred%20Values%20Varshney.pdf
I mostly value the game theory aspect as a what-if. What if the problem was simple enough we could solve it through logic, how would the optimal solution look like? Also, when reallife outcomes do not correspond to the model's prediction: why did it not work out that way in reality?
But in the end I don't think we can start with an arbitrary set of assumptions and iterate our way to an accurate predictor in most reallife domains. I believe the domain must be a good fit for game theoretic modeling, and then you also have to do it well.
It seems William Spaniel believes peace negotiations are a domain well suited for game theoretic modeling, and he's probably quite good at it. I'm not sure the domain is actually sufficiently suited but along for the ride anyway.
I know that he is not the only one who considers peace negotiations well-suited to game theoretical modeling. Though given the stakes, perhaps one doesn't have to be all that sure in order to give it a shot.
https://convention2.allacademic.com/one/apsa/apsa25/index.php?cmd=Online+Program+View+Session&selected_session_id=2234304&PHPSESSID=5ti77n6bl51a4r2vfhoerh15fn
https://convention2.allacademic.com/one/apsa/apsa25/index.php?cmd=Online+Program+View+Session&selected_session_id=2234303&PHPSESSID=5ti77n6bl51a4r2vfhoerh15fn
I love that channel too! But having watched a lot of his videos, I find that he all too often repeats the basics of what he's teaching for the newcomers, diluting the information that is new to me in a video.
Is anyone super familiar on setting up non profit entities and willing to chat with me on it? Trying to do this for Trust Assembly.
Nolo Press has some publications which might be useful. https://store.nolo.com/products/nonprofits/nonprofit-books-31
I set up and now run a 501c3 and am willing to chat about it if you like, though having only done it once, "super familiar" is maybe stretching it.
Laws regarding the incorporation of a non-profit are a function of IRS procedures and state regulations. In most places it's pretty straight forward. Most Secretary of State offices have a web portal that will guide you. This will get you started: https://www.chamberofcommerce.org/nonprofit
What is Trust Assembly?
His browser extension that allows for markup and commentary on (arbitrary?) parts of websites. "community notes, everywhere" might be a good slogan.
Ah, thanks. I wonder why he wants to set up a non-profit for it?
Because the Survival and Human Flourishing fund specifically focuses on funding nonprofits doing stuff like this.
The connection is eluding me. In what way does a browser extension that allows one to markup websites contribute to survival and human flourishing? Can someone connect the dots for me?
This fund is focused on solutions that will work in the upcoming future as it changes due to AI.
It's a solution to "what should we all do in a future where AI is polluting the informational commons with 100x nonsense / slop."
It's basically a commensal way to share curated media diets, with high reputation people doing the work of cutting out the bulk of dreck and sharing blacklists. You choose people whose tastes you trust and subscribe to their blocklists, basically.
It's also a sure way to concentrate current filter bubbles a thousandfold, which is why I won't use it, but it does propose a solution that might work!
A traditional school for architecture is being built. Im not affiliated, but seems pretty cool. Thought this community would be interested
https://youtu.be/2cyMvm45dTc?si=fTyIWJjqxYz313Sx
We just opened applications for the Pivotal Research Fellowship Q3 2026! It is 9 weeks (+ extension) in London with mentors from UK AISI, Google DeepMind, Redwood, and other leading orgs. Stipend, travel, accommodation, compute, and a dedicated desk are all covered.
I think its likely that ACX readers might want to pivot into AI safety but haven't yet – we run the fellowship to make exactly this happen. So please consider applying! pivotal-research.org
A few days ago, I found out an Iranian friend of mine had been brutally murdered in France. I won't link to the details, but it's been on the local news.
It's been pretty devastating - he was a lovely guy - I've been coping by trying to figure out the motive...
It looks like "random teenager with mental health issues" on the surface, but the content of our last conversation makes me concerned that this might have been deliberate/political. I spoke to him a few days before he died. He was understandably very worried about his family during the war, but also quite openly anti-regime. I know he attended a protest in our city, and that he had an instagram full of anti-regime, pro-Reza-Pahlavi content, and he might have been doing stuff on Telegram in Persian or something.
Evidence for political motive:
- Strong anti-regime sentiment
- MO relatively similar to known cases - there are cases surprisingly like this one: https://www.euronews.com/2026/04/08/five-teens-to-go-on-trial-in-sweden-over-attempted-murder-of-iranian-dissident - seems premeditated and targeted at his home.
- Throat slitting feels more "IRGC/Islamic" than a standard stabbing - if he were stabbed in the street, I'd be less suspicious
- The timing: the war obviously, and there were lots of attacks by Iranian proxies the same weekend
Evidence against political motive:
- The killer's claims match "mentally disturbed person & random violence" quite closely
- Base rate of random murder is still higher
- My friend was unlikely to have been very high-profile, unless he was super involved in something online, or had connections I don't know about
- The criminal was too amateurish - his mum called the police on him when he came home with a bloody backpack
Thoughts? I'll speak to the other Iranians here to get their takes over the next few days. Feels like there's some sensible Bayesian reasoning I'm struggling with here
Could this be an example of "stochastic terrorism"?
That explanation seems to cover most of the details.
[Also, sorry about your friend ...]
Thanks.
Yeah, I think that, if it is state-led, it's a notably different pattern than the higher-profile assassinations over the last few years, and feels more like stochastic terrorism.
Check out this connection - a foiled attack on the day before the murder, connected with the "Islamic Movement of Virtuous Believers," an organization unknown to authorities and recently created in connection with the Iranian crisis: https://www.lemonde.fr/en/france/article/2026/03/30/after-a-foiled-attack-outside-bank-of-america-in-paris-five-arrests-and-questions-over-iran-s-role_6751958_7.html
"The two individuals arrested just before carrying out the attack, who were in possession of an improvised explosive device, are minors from the eastern Paris suburb of Montreuil. Known to police for offenses linked to drug trafficking, they told investigators they did not know the identity of the person who ordered the attack. They also admitted to acting in exchange for a promise of "a few hundred euros." Subsequent investigations led authorities to another minor from Montreuil, with a similar background, who is believed to have helped organize the operation."
and... "several countries have already faced this threat, which has so far remained low in intensity, often nighttime acts, carried out by amateurs with rudimentary means and no intention of causing fatalities."
Doesn't seem at all crazy that a list of Iranian diaspora known for their anti-government views was spread on these social networks, my friend was on the list, and that a local teenager responded to the call.
there were similar attacks elsewhere https://www.theguardian.com/world/2026/mar/06/missing-iranian-canada-masood-masjoody
Im not aware of many (any?) examples of Iran carrying out an assassination in Western countries. It would be very strange if the first person they went after was a random person in France with no political influence and no following through some random kid.
Unless there was some massive online following he had that made him high profile I’d think it vanishingly unlikely it was an assassination.
There were a couple of ham-handed attempts. One was in 2011 - they tried to kill a Saudi diplomat in DC. Another attempt in 2024, this time against Trump and other US officials, was foiled. The suspect was arrested and convicted in March.
The US state department (make of them what you will) claims Iran has done hundreds of *successful* assassinations in other countries (presumably many of them non-Western), and that's not counting the mass bombings. This goes way beyond a couple of ham-handed attempts. Iranian dissidents in Europe are frequent targets.
It's common enough to be "a thing". The Swedish case I linked to is the best example, but there have been successful Iranian assassinations of more significant figures (e.g. Ali Motamed, Ahmad Mola Nissi in the Netherlands) and there was an attack the same weekend in Paris.
Articles talk about "the modus operandi that Iran has been using for years: using criminal networks in Europe to silence purported opponents of the regime". https://www.reuters.com/world/europe/netherlands-summons-iranian-ambassador-over-alleged-assassination-attempts-2025-04-24
I guess these things are often hard to prove, as well, so the base rate might be higher than the evidence of "proven assassination attempts" suggests.
Also, the base rate for the leading alternative explanation of "having your throat slit by a random mentally ill stranger in a premeditated attack" is also vanishingly low.
But you're right that he doesn't seem a typical target, unless, as you say, he was very active online or in dissident circles.
"Also, the base rate for the leading alternative explanation of "having your throat slit by a random mentally ill stranger in a premeditated attack" is also vanishingly low."
I'm going to have to nitpick this: that's the wrong base rate to be using. In the counterfactual world where the victim was shot in an apparent robbery or stabbed by a drunk, the same questions would likely arise. It seems like you want something like "the base rate of somebody minding their own business[1] being killed on the street." Which is certainly still low, but much higher than the base rate of political assassinations of non-prominent people.
[1] Specifically, not actively engaging in criminal activity or provoking violent people.
Yeah, this is where some of my uncertainty lies.
The criminal entered the victim's home with a weapon, and there didn't seem to be a theft or similar alternative motive, so "the base rate of somebody minding their own business being killed on the street" definitely feels too low. But I'm definitely committing some Bayesian sin by adding "throat slitting" in there - especially as it didn't seem to be a "professional hit"-style throat slitting.
This is the recommendation from LessWrong for the reference class problem "look for the simplest reference classes you can find (only adjusting for obviously important and objective factors), or average over lots of ways of constructing your reference class so that no single set of choices dominates."
I guess I could do something like: "what proportion of each reference class is state-directed?", and average out. I think that there's probably 5 or 6 successful political assassinations in Europe per year. So something like:
A: All murders, Europe - 0.2-0.5% state-directed
B: Stranger killings, no clear motive, in Europe - 2-5% state directed
C: Premeditated stranger killings, no clear motive, in Europe - 5-10% state directed
D: Premeditated stranger killings of Iranian diaspora, in Europe - 20-30% state directed
This might give a prior of 4-8%, before adjusting for the timing? But this still feels very messy, and adding overly specific or overly broad base rates clearly biases the calculation.
I'm not sure if honour killing is the right term for this (isn't that usually killing your own daughter?), but yeah, it's always possible that there was something personal.
Sexual violence feels unlikely given both the killer (15-year-old boy with history of violence) and the personality of my friend, but obviously can't rule it out.
I was writing up a response when you posted this. I agree. 👍 I'll just say that I believe Iran is one of a select few countries that *does* habitually carry out assassinations in Western (and other) countries. They've been doing it since the revolution. Russia's another one obviously. You could argue for North Korea and Israel, and probably others I'm unfamiliar with. And obviously, a number of Western countries carry out assassinations in non-Western countries.
https://www.buzzfeednews.com/article/heidiblake/from-russia-with-blood-14-suspected-hits-on-british-soil
This article is a little old, but as of 2017 there were at least 14 suspected hits by russian spies in the uk, but the police were operating on a unofficial policy of just not touching them because of the political implications. This continued until the Salisbury incident made it impossible to ignore.
Also I understand that many would recoil at buzzfeed, but around that time they had serious money from all the listicles and for a while they decided to expand by building a pretty serious investigative journalism team. They eventually scrapped all the hard news in 2023.
And while I understand that this is as credible as "my uncle works at Nintendo" I've known enough journalists and diplomats personally who've had run ins with Russia that I don't doubt what they get up to.
I'm pretty ignorant of Chinese assassinations. Do you have some sources I can look at?
I run a curated Discord for high agency people with Long COVID/ME (myalgic encephalomyelitis). The group includes tech founders, researchers, rationalists/EAs, etc. The focus is on troubleshooting each other's conditions actively, as well as creating a body of knowledge to bring back to the wider community in the form of writing, education, projects, companies, etc.
Some know me as Liface, others as Liam Rosen - I have been in the Rationality community for over 10 years, previously in the Bay, now in New York City, and am the main moderator of r/slatestarcodex. I am a tech CEO whose former life abruptly ended last year. I am now bedbound with severe Long COVID/ME, and am dedicating 100% of my time and effort to helping myself and both everyone else around me figure out these complex conditions.
If you, or anyone you know might be a good fit for this group, email me at liamrosen@gmail.com.
There's been quite a lot of research into this question. Have you read much of it?
>We have continued excess deaths post-Covid19
At least part of that is an artifact of us only being "post-Covid" in relative terms. There were about 75k Covid-coded deaths in the US in 2023 and 47k in 2024, which is a lot less than 2020-2022 numbers but still quite large in absolute terms. And if I'm reading things right and have chosen appropriate data series to compare, the broader category of "pneumonia, influenza, and covid" is also a lot bigger in those years (211k - 244k) than pre-2020 numbers for the older "pneumonia and influenza" category (looks like typically 50k-65k). Covid is a lot less deadly that it used to be because so many people are at least partially immune and because better treatments and protocols have been developed for managing severe cases, but it is still killing more people than influenza each year.
Sources:
https://www.cdc.gov/nchs/nvss/vsrr/covid19/index.htm (table 1, click the "yearly" button, and filter out the current year because those numbers are only for the first 3.5 months)
https://blogs.cdc.gov/nchs/2020/03/09/4820/
> do you know if that has changed?
I have heard this claimed to be true in the past, but I have also heard claims (which I considered to be more plausible) that this was a misunderstanding of how listed causes of deaths work. The way it was explained to me, the doctor filling out the death certificate commonly lists multiple conditions which played a significant role in the outcome. So the numbers would definitely include an otherwise-healthy person who caught covid and died from it, or someone with fragile health who caught covid and took a nosedive while sick, but probably not someone who caught covid while already imminently dying of liver failure and definitely not someone who had a minor or asymptomatic case when they get run over by a truck.
https://www.cdc.gov/nchs/hus/sources-definitions/cause-of-death.htm
>Is everyone being tested at time of death
I'm pretty sure they aren't. My first link in my previous post, the fine print mentioned "confirmed or presumed" COVID-19. The "or presumed" part doesn't make sense if everyone gets tested.
Also, the PIC column appears to be well above baseline, but much more than the direct covid numbers. That suggests to me that either 1) I am looking at the wrong numbers for the baseline (a distinct possibility), or 2) Substantial number of covid deaths are being coded as generic pneumonia. 2 seems plausible to me because my understanding is that the category exists and is included in Influenza and Covid tracking statistics because Influenza-related deaths are often coded as pneumonia rather than influenza and (barring a positive covid test or a case history that strongly points to covid) the same is likely to happen for covid.
>I'm reading the 2024 data as "reduced down to a bad influenza year" (and yes, those are rare) -- do you concur?
Partially. Covid-coded deaths in 2024 seem to be comparable to total estimated deaths from influenza in a particularly bad year, but the latter is an estimated portion of excess pneumonia-coded deaths in addition to direct influenza coded deaths. If I'm reading the stats correctly on PIC deaths, the same methodology would probably show Covid as being a something like a factor of 1.5x to 2x worse than a particularly bad year of influenza. I would characterize this a significantly worse, but within the same order of magnitude.
I also think that with Covid vs Influenza comparisons it's important to keep in mind that covid is adding atop influenza, not replacing it.
>Could you quantify what you mean by "partially immune?" Are we looking at ADE or viral escape?
Viral escape. I had to look up ADE, and if I'm reading correctly, if that's happening in large numbers we would expect Covid deaths to be going up rather than down. The numbers look like what I'd expect if viral escape is allowing people with prior immunity (natural or vaccinated) to catch covid, but the prior immunity still reduces the risk of catching it somewhat and also reduces the expected severity of illness if you do catch it.
> I was reading some of it after I made that post.
Awesome, what conclusions were drawn by the research you read?
Oh, I'd hoped you meant you'd read some of the research that aims to answer the question of whether vaccination increases or decreases the rate/severity of long COVID since that's what your original question suggested you were interested in.
If you are interested in that there are several large cohort studies that show a dose dependent reduction in long COVID rate with vaccination and systematic reviews that detail the reduced likelihood of getting long COVID in vaccinated individuals and reduced symptom severity in vaccinated individuals.
These two are a reasonable starting point
Hedberg P, van der Werff SD, Nauclér P. The Effect of COVID-19 Vaccination on the Risk of Persistent Post-COVID-19 Condition: Cohort Study.
Watanabe A, Iwagami M, Yasuhara J, Takagi H, Kuno T. Protective effect of COVID-19 vaccination against long COVID syndrome: A systematic review and meta-analysis.
Consider reading peer reviewed and published papers rather than listening to ows personnel. You should weigh the evidence differenly.
Your argument about excess deaths doesn't help in either way - assuming there are in fact excess deaths.
I'd be willing to put down cash money that ME is not caused by vaccines. What might a satisfactory mechanism for resolving such a bet look like?
Huge swathes of people received COVID vaccine, and cancer death rates kept falling. The idea that the vaccine can cause an increase in cancer appears to be based on nothing (except insofar as living longer means higher odds of eventually getting cancer, and the vaccines increase lifespan by preventing some deaths).
Lovely‘s book?
This looks less like AI analysis and more like a familiar mix of alarmism, reductionism, and compressed media framing (again and again and again). If that’s the pitch, the book may already be obsolete in the very way it warns about and this kind of media.
I'm having trouble parsing your comment - are you saying that there are already many books that do this? Which ones? In what sense is the book obsolete?
I meant “obsolete” more structurally than commercially. Not that there are too many books like it, but that this kind of alarm-heavy, reductive framing may already belong to an older media logic, one that was already starting to age before this book even will came out, and that misses what frontier models actually change.
@deisearch Fair point on publishing speed and cost limits. My concern is slightly different: the deeper issue is not just whether AI replaces 100% of workers soon, but how reductive and alarm-driven framings can miss what frontier models actually are, what kinds of work should become obsolete, and what genuinely new problem-solving capacity they may open.
With the pace of AI, I think by the time a book goes through the publishing process, it's already badly out of date.
The one thing that might slow it all down seems to be companies discovering "oops, turns out that it's getting very very expensive, current improvements are very small compared to the huge leaps they used to be, and we aren't able to replace 100% of workers right now, so we're putting a budget on use".
I'm curious to distinguish between two interpretations of your 2nd statement. Is it
1) This is a possible future that would slow things down, and we'll see if we get to it.
2) This is the future we're in (seems lightly implied).
If 2, I think you're factually incorrect. Current improvements are accelerating in the ways that make money for Anthropic and OpenAI, as evidenced by the billions of dollars they increasingly make. Enterprise coding is already a huge economic boon for those companies and it seems to be getting better on ever shorter timeframes. Chatbots are about as good as they can possibly be with the current one-shot in-context learning only version, but agentic AI is unlocking a new ceiling.
"Current improvements are accelerating in the ways that make money for Anthropic and OpenAI, as evidenced by the billions of dollars they increasingly make."
Allegedly they are not currently profitable:
https://www.fastcompany.com/91522156/openai-doesnt-expect-to-be-profitable-until-at-least-2030-as-ai-costs-surge
They may be making money, but they are pouring it back into research and training.
This isn't a secret. I wrote another comment here that the AI companies publicly acknowledge this. They are making a ton of money. They believe that the best business model is to spend more money than that on R&D. They may be wrong, but the critique of "AI is a scam, they're losing money!" ignores the business plan of all the big companies.
Good distinction. But I think “making money” and “reaching the ceiling” are being conflated here. A lot of what gets called agentic AI still looks like workflow automation, while coding is a more substantive case. And the deeper ceiling of frontier chat systems remains partly obscured by product design, UI constraints, and alignment smoothing.
The main difference between a chatbot and an agent is the access they have to the world. It's the same model in Claude Code with full dangerously-skip-permissions and the chat window in claude.ai. The difference is what they can do.
Workflow automation is real work. Before the spinning jenny it was someone's job to spin thread. Then it became the machine's job. Before your AI agent organized your email inbox it was your secretary's job. We've just replaced workflow automation by humans with workflow automation by AI. Coding is easier to see where the AI creates something, but those other jobs were still real work.
I think a lot of people have an unconscious bias against all non-manufacturing work. If I can't touch it, is there value? Sure I go the movies, but that's a nonproductive activity for fun. Fun and value are different. The secretary is just doing things I could do for free, if my time wasn't so valuable. See also every IT worker's tales about job perception from the suits.
But nearly all labor is like that. The US is mostly a service economy. If information work didn't matter than you should have no preference for a blank book versus one full of words.
I don't think you're taking nearly so strong a stance as what I've described, but "still looks like (just) workflow automation" is denigrating a lot of valuable work. For a manufacturing company, everyone who doesn't add a physical piece of the final product is just automating workflow for those who do.
Fair point again — I’m not denying that workflow automation touches real and often valuable work. My concern is that “agentic” is often framed as progress when it mostly automates inherited systems, many of which are bloated by coordination overhead, governance layers, historical friction and control (even control of control).
Frontier models seem more interesting to me where they do not just optimize those layers, but help reveal which of them have become obsolete. That is why coding feels like a more substantive case: it can generate new structures, not only administer old ones more efficiently.
Too often we seem to be building ever more elaborate bureaucratic machines whose main output is permission slips for other bureaucratic machines.
Deiseach was clearly anticipating the future. Hence the “might”.
On your comment. None of these companies are “making money”, which is always taken to mean profits, not revenue. There’s no evidence at all that “Enterprise coding is already a huge economic boon” to any large section of companies.
I recommend listening to Dario Amodei's explanation on the Dwarkesh podcast. The short version is that each model Anthropic releases makes money -- profit, not just revenue.
Example numbers: Sonnet 4 takes $10 million to train. Over its lifespan it costs $10 million more in inference compute. It makes $30 million in total revenue during inference. As it's winding down Anthropic is training the next model, which costs $20 million and uses some amortized piece of $10 million worth of general research in the background.
Those aren't the exact numbers for anything, but they're in the correct relation for what Dario describes. Each model makes money, but the company is simultaneously spending more than that on R&D for the future. Whenever we do hit a wall they will simply sell existing models and make large profits.
Oh well if Dario says it!
Yes, I'm sure the CEO is going to go public with "okay our revenue is increasing but we are also doing the equivalent of burning bundles of tenners in a furnace, so profits? ask me again in 2030 and we hope to still be in existence then and grabbing the greater market share".
First time using the app here, posted before I meant to.
Further, it's insane to me to say that almost no companies are getting value out of enterprise coding. I could readily believe that one or a few CEOs are taken in by the hype train and are foolishly spending money on an investment with no returns. But the adoption keeps increasing and we aren't seeing the heavy adopters post losses or go bankrupt; they're increasing use and profits. At a certain point saying that this is all hype implies that modern late stage capitalism doesn't care that much about profits. Sure, we might still be waiting for the rug pull, but every day it doesn't drop is evidence against that.
This is close to denying the evidence. Sometimes it's appropriate, but looking at what Mythos has done and how many companies keep increasingly using Claude Enterprise means that you're moving past "AI companies often overhype their products" to "there is a vast conspiracy among the Fortune 500 where AI doesn't work and no one will admit it."
> At a certain point saying that this is all hype implies that modern late stage capitalism doesn't care that much about profits.
I’m sure companies care about profits. However outside of the people selling the tools, software profits have stalled recently compared to the growth in the pre covid and covid era.
Your previous post clearly underestimated the cost and revenue of training and compute. It’s directionally correct, in that when a model is complete it becomes a generator of income which funds the next model, but the numbers are way out. The costs are far greater than you say (I realise you say it was a toy model) and more importantly revenues don’t pay for the previous models. There’s some money to be made on the long term - if there’s any economy left - but when are expecting ever increasing spending to stop. Firms like OpenAI, Anthropic, and Google DeepMind can’t really slow down, because if they do, they risk falling behind—and the upside of being ahead (better product, more users, more data) is massive, unless costs consume every company.
My monthly list of recommended long form links is out again.
This batch includes the hypomanic edge common in geniuses, a reflection on the coming second peasanthood of a post literate society, analysis of how Australia is lurching toward a command economy, and a possible link between a major class of herbicides and metabolic dysfunction.
Also linked are my own interviews with Malcolm Collins of Based Camp fame, and a couple of auxiliary language enthusiasts on the topic of Esperanto, Toki Pona and a new mix sized auxlang called Baseyu, and yet another amateur crop breeder working on species you probably never heard of.
https://recombinationnation.substack.com/p/the-long-forum-april-2026?r=f45kp&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
It's been observed that birth order affects things like income and educational attainment, with firstborns doing better than their younger siblings. A new paper argues that this may be due to the effects of respiratory pathogens, where younger kids get more of them from their older siblings...
> We use Danish administrative data to document that, before age one, younger siblings have 2-3 times higher hospitalization rates for respiratory conditions than older siblings. We combine birth order and within-municipality variation in respiratory disease prevalence among young children, and find lasting differential impacts of early-life respiratory disease exposure on younger siblings’ earnings, educational attainment, chronic respiratory health and mental health related outcomes.
It's a huge (as in long) paper. I admit I've only skimmed it, but I suspect they're on to something...
https://www.nber.org/system/files/working_papers/w29524/w29524.pdf?utm_source=PANTHEON_STRIPPED
What about only kids? Should do better than both, right? (After all, older siblings can also get pathogens from younger ones.)
Also thinking about cases like myself, where I do have older siblings but they're like 15 years older than me and moved away when I was a toddler.
It looks like they found that illness in the first year was what effected development the most long-term.
I find this personally interesting and will have to look into it more closely. I'm a younger sibling who has had some respiratory issues. They've been overall pretty mild, but were striking in that 1. because of their mildness, they went entirely undiagnosed (and largely unsuspected) for decades[1] and 2. despite this, they've sometimes had a significant impact on my overall quality of life.
Naturally, as soon as I comprehended both of the above points, I had to start wondering how common this sort of thing is. Not just my specific issues, but respiratory issues (or even other maladies) that are subtle enough to largely fly under the radar, but still capable of significantly effecting people's life outcomes.
[1] Either that or I didn't develop them until much later in life. I might not ever know which is the case, but I have some modest evidence to suggest that they've been there all along.
> I had to start wondering how common this sort of thing is. Not just my specific issues, but respiratory issues (or even other maladies) that are subtle enough to largely fly under the radar, but still capable of significantly effecting people's life outcomes.
First, I would suggest that issues don't have to be subtle to fly under the radar. In general they'll only be noticed at all if the person who has them complains about them. And that isn't likely to happen if that person thinks of an issue as just a normal part of life.
I have five older siblings spaced a year or so apart and I remember going to the doctor when I was like 8 to get tested for a lingering illness and although I forget the diagnosis, I already had antibodies for bacteria found in mono and pneumonia among others. When I was young, I was sick a non trivial amount. I went to a large urban school and participated in swimming, wrestling and other sports. I appreciate these studies to improve the health of future youth. If possible, it would be great if kids got sick less and we can prioritize their health and well-being accordingly. Kids getting sick constantly could be a signal of something to be avoided, due to long term consequences, rather than something to be ignored.
So the idea is older kids get sick, and probably get sick more often because they're picking up infections from the other kids at playschool etc., then come home and pass those on to the younger siblings who have less developed immune systems. This means younger kids get sicker and more badly, and this has knock-on effects on development.
I think first you need to define what you would expect the 'normal' level of infection to be, and at what age, before you start proposing that little Timmy got sick with a cold aged six months that he picked up from big brother Jimmy aged two, and this made little Timmy sicker than he would otherwise have been.
What age was Jimmy when he got colds as a baby? How much infection was he exposed to? Did Jimmy get fewer infections than Timmy at the same age?
Though I had to wince at this part:
"We find that younger siblings are two to three times more likely to be hospitalized for acute respiratory conditions during their first year of life compared to the older siblings at the same age, indicating that younger siblings experience a great respiratory disease burden when they are infants. "
Uhhh - yeah. When I was two, my younger sister was born. She got very, very sick at a young age (can't exactly remember how old she was, she wasn't a year old yet going by my memory) with what turned out to be pneumonia and was severely ill while I was fine and never got the likes. Okay, link demonstrated? But I wasn't in preschool or the likes at that age, so both of us were being exposed to the same germs in the same conditions. I didn't pass it on to her and we were both at equal risk of infection at whatever age.
And yeah, later on I got measles first, then my sister got it (presumably from me) and then my younger brother got it after her, so that does incline towards "kids pick up infections from outside and bring them home".
"Lastly, we find that the effects on respiratory hospitalizations among younger siblings are larger in sibling pairs with a short birth spacing in which the older child is in a childcare center than in pairs in which the older child is not. This result provides further support for our hypothesized mechanism of spread—that the older sibling gets exposed to respiratory disease while in group childcare, and then “brings it home” to their more vulnerable younger sibling."
I think there might be something there, but they're pushing the conclusions too far with regards to mental health, income, education, etc.
It's a huge frigging paper! The authors dot every t and cross every i (as they say). They go through every variable in painful detail. I tend to buy their thesis because we know that viral infections like measles and, evidently, COVID cause noticeable cognitive developmental impairments. If the firstborn get fewer infections in early life than the later born, then I can see the difference in life outcomes as a reflection of pathogenic burden.
Would that also apply to adopted kids, whose biological birth order would then be irrelevant?
xxx I don't know if the paper investigated that element. xxx
CORRECTION, they didn't distinguish between biological birth and adoptions.
I think that's difficult to reconcile with e.g. firstborns being twice as likely to read SSC/ACX compared to their younger siblings. https://slatestarcodex.com/2018/01/08/fight-me-psychologists-birth-order-effects-exist-and-are-very-strong/
Childhood exposure to respiratory pathogens seems exceedingly unlikely as a root cause for differential tastes in blogging. Unless we're talking broad intellectual disability so severe as to render people incapable of participating in this community, but I think we'd have noticed that in standardized intelligence or aptitude testing of children at large. It seems more likely that we are looking for something that modulates specific differences in personality, interests, etc, in ways that a respiratory pathogen doesn't.
Really, if you're seeing something that affects income and educational achievement, but doesn't show up in standardized testing in childhood, you should probably be looking at something specific, subtle, and discriminating, rather than the broad insults of a (non-neurological) pathogen. And if it *does* turn out to be respiratory infections, then the question of "how does that lead to *this*?" becomes very interesting,
Suppose that people with >130 IQ are most likely to join a community. If secondbornness lowers IQ by ~3 points, we should see half as many secondborns as firstborns in the community.
That's with a hard cutoff at 130. With any more plausible distribution function for p(SSC/ACX) vs IQ, I'd expect you'd need at least double that difference. And I'd be very surprised if there were a 6-point IQ difference between firstborn and later siblings that hadn't already been spotted and incorporated in the conventional wisdom by people researching birth-order effects.
The effect size was between 0.1 and 0.2 standard deviations — it was small, but statistically significant.
Yes, and people have been studying both childhood intelligence and birth order effects long enough that if there were a statistically significant effect of birth order on anything like IQ, we'd have seen i it long ago and we wouldn't be depending on a single preprint for our understanding.
Hence my belief that we're looking for something subtler and more specific than "respiratory infections make the later kids dumber".
But first-borns have a ~2–5 percentage-point higher probability of attending college than second-borns. And the gap increases for larger families.
There also seems to be a consistent *monotonic* decline in standardized test scores by about 0.1 SD per sibling's birth order. I haven't bothered to dig into the literature, though.
Do you have data on this from your surveys?
There's a clear birth order effect on sexual orientation, and I think experts favor the explanation that previous pregnancies alter the fetal hormonal environment for subsequent pregnancies. This probably also affects various personality factors. Do you know if there's good evidence on the effect of birth order on personality (eg. in the Big Five personality traits)?
I think it's pretty consistent. It was discovered in the 90s by Blanchard et al, and has been replicated a number of times. The data indicate that a male is roughly 1/3 more likely to be gay for each older biological *brother* (older sisters seem to have no independent effect, although obviously number of older sisters and brothers are correlate). I'll just put the wiki link here. It has citations to all the main studies of this: https://en.wikipedia.org/wiki/Fraternal_birth_order_and_male_sexual_orientation
I keep encountering these claims about there being an illiteracy crisis for the next generation. Videos of teachers complaining that students are showing up to increasingly advanced grades without basic reading comprehension abilities or other severe limitations. I'd be interested in hearing your perspectives on this matter: if you have children, or you know schoolteachers, or if you just interact with kids for some other reason; what has your experience been?
Is this a global phenomenon or is it mostly hitting Western countries?
My minuscule-sample-size experience is that my daughter (8yo, second grade) is an enthusiastic reader, and her reading skills are at least on par with what I remember mine and my younger sister's being at roughly that age.
My daughter is not a representative sample of the population, but neither are me and my sister.
One factor people often forget to consider is that we're not only looking at differences between generations, we're also seeing differences over time within an individual person's lifespan. My parents, currently about 60 years old, definitely read less than they did 30 or 40 years ago. My impression is that they also probably read less than 60 year olds in previous generations did.
So if there is indeed a decrease in youth literacy, it's not quite fair to only compare that to youth literacy rates in previous generations, we should also consider the ratio of youth to adult literacy. What I'm saying is that if an adult in 1980 read 10 books a year and a teenager read 5, and an adult in 2026 reads 4 books a year and a teenager reads 2, then it should be viewed as more of an overall societal problem and not solely as a problem with younger generations or with the educational system.
Note that I made those numbers up based off absolutely nothing, but my point is that the ratio is worth considering in order to get a better idea of the true scope and cause of issue.
Historical note, this has come up before.
Why Johnny Can’t Read — 1955
TLDR by Jesse Pinkman: Teach some phonics bitches!
https://en.wikipedia.org/wiki/Why_Johnny_Can't_Read
I'm somewhat puzzled by this, since it's contrary to my experience. I was taught a mostly-phonics early reading curriculum in public elementary schools in the late 80s. So was my little sister in the late 90s, and my daughter right now. Three different school districts in two different states. This is a small sample size and it's possible all three of our experiences were outliers, but still enough to make me suspicious of claims that whole-language instruction as described in that book has remained ubiquitous.
English orthography is bad for phonics.
Common spellings for /ʃ/:
* sh → fish, ship, wish (the workhorse spelling)
* ti → nation, motion
* ci → special, musician
* si → tension, version
* ssi → mission
* ss → pressure, assure
* ch → machine, chef (French influence)
* sch → schwa, schnapps (German influence)
* ce → ocean
* s → sure, sugar (a bit sneaky)
But, yeah, I was taught with phonics too.
I recall a parody title of the book wrt amphetamines: Why Johnny Can’t Blink
Your evidence is oriented in the wrong direction to support your assertion. (And your assertion is wrong - English orthography is fine for phonics.)
You'd want to show that it's hard to predict the sound of a word from the spelling, not that it's hard to predict the spelling of a word from the sound.
You've also bungled your list of examples. What are you thinking about 𝘮𝘪𝘴𝘴𝘪𝘰𝘯 vs 𝘱𝘳𝘦𝘴𝘴𝘶𝘳𝘦 that tells you the "i" is part of the representation of /ʃ/ while the "u" isn't? What are you thinking listing just "s" for 𝘴𝘶𝘨𝘢𝘳 and just "ss" for 𝘱𝘳𝘦𝘴𝘴𝘶𝘳𝘦? Why do you think 𝘷𝘦𝘳𝘴𝘪𝘰𝘯 is an example of /ʃ/?
The reason all these spellings of /ʃ/ exist is that there's a rule of English pronunciation that transforms /sj/ into /ʃ/. That's what's happening in 𝘯𝘢𝘵𝘪𝘰𝘯, 𝘮𝘰𝘵𝘪𝘰𝘯, 𝘴𝘱𝘦𝘤𝘪𝘢𝘭, 𝘮𝘶𝘴𝘪𝘤𝘪𝘢𝘯, 𝘵𝘦𝘯𝘴𝘪𝘰𝘯, 𝘮𝘪𝘴𝘴𝘪𝘰𝘯, 𝘱𝘳𝘦𝘴𝘴𝘶𝘳𝘦, 𝘢𝘴𝘴𝘶𝘳𝘦, 𝘰𝘤𝘦𝘢𝘯, 𝘴𝘶𝘳𝘦, and 𝘴𝘶𝘨𝘢𝘳. And it's not difficult to predict at all. Try making up some words that fit into these paradigms and asking people to pronounce them.
Okay, points taken. Not sure wtf I was thinking with the ‘version’ example.
See also the poem "Dearest Creature in Creation"
https://people.cs.georgetown.edu/nschneid/cosc272/f17/a1/chaos.html
Yea that is definitely a live meme nowadays. It's achieved conventional-wisdom status in both MSM and social platforms. Everyone I know older than 40 who doesn't have children (which includes two of my siblings) thinks it's just a known fact.
That however doesn't even vaguely match my observations of either my own two children or any of their peers. My eldest (now in his early 30s) and all of his peers read plenty. None of the current public school parents I know (whose kids are classmates of my youngest) think their kids read less or more poorly than they did at the same age. Etc.
Certainly in my professional world there is no issue with younger (which at this point is nearly all of them) staff being poor or reluctant readers. I hire 20something/30something adults regularly; haven't yet had any difficulty finding applicants who on this point and others are more qualified than my peers and I were back in the day. My wife says the same.
As you note, the source of the alarum is overwhelmingly teachers. I notice also that all of the college-level jeremiads about this turn out to be from professors who are at or near retirement age. Being old enough to have had rotary dial telephones in my childhood home, many of my humanities professors when I was an undergraduate were certain that we read much less and/or less well than they had. Rinse and repeat on that part.
There've also always been some K-12 teachers who were loudly certain about a literacy crisis in the young. Had a couple of those as members of my extended family over the decades...but I've also known plenty of schoolteachers who did not think this.
So I dunno. I do wonder about a selection bias: the frustrated teachers/professors of this era now have personal megaphones (social media) which those of previous eras did not. Our credulous mediots regularly get their social/cultural trend-story ideas from, of course, social-media feeds.
Hence the educators feeling that the damn kids don't read anymore become way more heard outside of the breakrooms/faculty lounges/dinner tables. Which may, or may not, reflect an actual increase in the fraction of teachers having that complaint.
I am an adult who works in a corporate IT setting, and I have to say that this is not a change limited to children/adolescents. I believe that there has been a shift in the willingness of the median American (or other high-internet-exposure Westerner) to spend time reading content which is boring/not actively engaging. We might see the effect more in adolescents since they're the population taking most of the tests and being directly graded on their literacy, but I don't think it's limited to them uniquely.
I work with a number of career adults who seem newly incapable of reading anything longer than a paragraph. Internal email open rates are below 30% even on short summaries of important subjects like "how to access your tax/healthcare information" and "how to open the programs you use every day". I know this is a classic IT guy complaint but I send 1 or 2 emails per quarter if that. Employees who 10 years ago could reliably handle self-service for things like password resets are now completely incapable of such and cannot learn or retain the information from text guides. It's not just admin stuff but even longstanding documented processes for ordering/selling products where people are now winging it when they didn't 10 years ago.
My guess is that it is due to some combination of shorter attention spans, more engaging alternatives to reading boring prose, and possibly "reading exhaustion" from the median person having to spend more time reading/responding to emails and such year over year. Plus, time spent on phones/social media apps is still increasing each year and that time has to come from somewhere. In the case for both employees & students I'm guessing they are eating into their discretionary & "functional" reading time (along with time spent outside or on family/friends/hobbies).
Scrolling gets better every year and reading is more or less the same as it was 2000 years prior!
I know several college professors and they all say this is an accelerating problem for freshmen. You simply can't have people read something and expect most of them to have learned anything by doing so. Even giving them written instructions for how to do a lab or fill out an assignment will be spotty.
It's not just reading either. I saw an article saying that even *film students* are no longer watching the movies assigned in class, or not paying attention if they do.
I guess everyone has their own perspective. For me, in EEurope in higher education, we are hit by a demographics crisis (TRF way below 2.0 for 2-3 decades now) without immigration. And since we are not the top-top Uni in the country (at least in our field), we teach less and less capable kids (since there are less good students, and they all fit to the "top" unis, making those the "top" unis simply by this), even if there is no illiteracy crisis. There might be, tho.
(Our primary-secondary education is also quite selective, so my son got into the good schools, and his schoolmates are of similar background.)
A completely different perspective, but US faculty at the Professors subreddit mention the illiteracy crisis all the time.
Children definitely seem to be worse at the kind of intellectual abilities reading gives you, but they seem to be more aware of facts in general. For example a 13 year old in my family is aware of class warfare, which she almost certainly learned online. Even as someone who was online at that age, I don't think I was aware of such issues.
Sorry that this is a bit of a gotcha, since you were probably just talking about children's exposure to public discourse, but it's pretty controversial whether class warfare is indeed a "fact". And no offense to your 13 year old relative, but I don't expect a 13 year old's understanding of class warfare to be particularly meaningful, let alone any sort of compensation for a loss of literacy.
I think you just got triggered by a shibboleth and decided to launch into your own headcanon. Being aware of class and its relationship to the world is a signal of intellectual thought especially at that age.
Well, it is certainly a fact that there have been quite a few examples of class warfare in the history of 20th century, both in a fairly literal sense of the term (people belonging in camps divided by social class forming armies and fighting civil wars) and in a less direct sense (strikes, other industrial action).
It's clearly a fact, though you could reasonably quibble about the term "class". The more powerful frequently (ab)use their power against those less powerful. These days the blame is frequently publicly assigned to corporations, but the decisions were made by people who were powerful within those corporations.
Note that this is a common human behavior pattern, and is one of the things democracy was supposed to prevent. (It doesn't, it just alters who is powerful.) It's what anti-trust laws were about. It's what unions were created to solve. That the solutions didn't really solve the problem doesn't imply that the problem doesn't exist.
Warfare to me implies coordinated struggle between reasonably well-defined polities. Rich people and large corporations can have various negative effects on the rest of us, but, for example, someone seeking a monopoly over some industry sees himself as fighting his own class, i.e., competitors.
Maybe replace "fact" with "concept."
Also, we need to not get the term confused with the concept. A 10 year old reading Animal Farm two decades ago might be aware of the basic idea behind class warfare, but not have a term for it while the 13 year old today might have the term (and maybe not have read Animal Farm :-)).
German here. We've seen it happen with my niece & nephew. Their elementary school has dropped marks entirely and increasingly adopted an unstructured learning system that was sold has making reading pleasurable for kids as opposed to boring letter-learning. In practice, this passed on the part of actually learning anything to parents. My wife's sister had to do extensive homework with them, but to just get them capable of reading in third/fourth grad at a level that was formerly reached by the end of the first year she had to sit down with them on weekends, too.
Of course, I also know teachers and day care workers. From their PoV, the parents are a massive pain, and I can completely understand them. Troublemaker kids that in the past would have been simply scolded by their parents, are instead supported by them. Teaching basic competencies like dressing yourself or be on time are passed on to the day care workers who have 20 kids at once, while the mom with an only child complains that well he doesn't want to, what can she do? So the teachers and day care workers are way too busy just babysitting and actual education falls by the wayside.
Not to mention the troubles of immigration; Having a classroom where one-third can't even speak proper german to begin with obviously causes issues. Unfortunately many teachers still have trouble openly talking about this, too. Most of the ones I know well enough will admit that even the aforementioned problems are also at least disproportionally related to immigration, though not entirely.
I'm doing lectures at university myself, and here as well it's also a simple fact that immigrants are a disproportional amount of problem cases. We also get pressure from the university admin to drop standards, for various reasons: First, they get paid by student, so they want to give a good "learning experience". Second, the demographic stats always look bad with high standards, immigrants fail and germans get through (plus a few asians and western immigrants). Not to mention the constant complaining by the students themselves. So, to avoid this stupid headache, better have no standards. A professor I work with literally said to me "I'll let them pass in the oral exam eventually anyway, so why not mark the written exam easy to save some time?" I'm sad to say that, but in many university classes I'm aware of we do not make sure that the students actually learn anything meaningful. Who wants to learn will still do so, but you can easily get through without.
And also from other parents themselves I hear stories that make me think. They have a son who is rowdy, can't sit still, has trouble emphasizing with others, and has a tendency to get obsessed with random stuff like tractors. Literally what every single boy was like back when I was a boy, including myself. So they go from therapist to therapist, until finally someone gives them some diagnosis. Usually ADHS or ASD or both. Then the parents and the teachers have an excuse why he isn't learning much, and he has to go to therapy regularly to talk about his feelings. The end result of this process is a young adult who is exceptionally adept at claiming issues, getting sympathy and navigating benefits systems, but who is behind on most measurable skills.
I can go at this for hours, but you get the gist: As far as I can see, we've developed a culture of everyone passing on the responsibilities for kid's development to everyone else. They still end up learning things, somehow, but everything is much slower and more complicated than necessary. Credentials are increasingly unreliable at showing actual proficiency.
When I went to school in Germany unstructured learning schools were not the default. You could send your child to a Walldorfschule or something similar, but it would've been an active choice on the part of the parents.
Is this no longer the case?
It's not the default now, either. But school move into the direction of less structure and less testing, based on what I hear from both parents of school-aged children and teachers. Especially elementary schools. People vary more on whom/what to blame (lazy parents, progressive ideology, immigration, take your pick) than on disputing that characterisation itself.
My impression is that the situation would be similar without immigration as well -- unless you would be allowed to teach only the current "better half" of the classes, which is really unlikely. Your class would simply be filled up with "bio-German" students who now are not accepted to your Uni.
Yes and no. If my last paragraph doesn't make it clear, I don't think the situation is entirely caused by immigration. But it definitely sped some things up and made them harder to resist; Without immigration, enforcing standards makes you "just" the unpopular hardass. With the actual immigration as it exists in germany, you're now a racist to boot.
All my relatives with school children haven’t reported anything unusual. They’re in Washington state and Israel.
I have heard the same thing from basically every teacher I know (ranging from elementary school through college undergraduate level). I still don't take the claim "students are showing up to increasingly advanced grades without basic reading comprehension abilities or other severe limitations" for granted but based on so much agreement from these teachers I am inclined to think it is true or closer to true than not.
I have children, I'm married to an English teacher, and I work in a school (I'm in the UK).
Both my girls are completely avid readers, fiction and non-fiction, but perhaps they're unrepresentative.
As for students, I don't strongly get the feeling that there is a specific literacy crisis here. I think there is a feeling that as a cohort students are less independent, more needy, perhaps more apathetic and less resilient. There are a few measurable ways in which this (or something like it) is true: there is increasing demand for additional support for students due to their education needs (starting in early years), and we see more accommodations being requested for exams and tests (extra time, reading support, rest and movement breaks and so on).
These sorts of things *might* manifest as students claiming work is too hard for them, or the pressure of makes them anxious, or being somewhat indifferent to certain types of reading asked of them. I feel teachers complain more that students don't want to do the reading than that they genuinely can't.
But I don't think what I'm talking about is overwhelmingly obvious to the extent that you'd be sure there's an underlying change in students, rather than it being due to more demand from parents, changes is school protocols, increasing competitiveness at the top, diagnosis drift or other factors.
One relevant thing in the UK perhaps: here we have an externally validated examination system, so if it's true that in the US students can be passed for classes where they actually by any reasonable standard should be failing, and thus be genuinely unaware of their deficiencies, it would be very difficult for that to happen here. A university might accept a student with a grade D in a subject, but they would have a good idea what they were getting (and the student would understand their own level of attainment).
"As for students, I don't strongly get the feeling that there is a specific literacy crisis here."
I think the specific crisis here is supposed to be that the kids can't (and therefore won't...) read long works. Long works like entire novels rather than just excerpts.
An example of this complaint is: https://www.theguardian.com/commentisfree/2026/jan/06/many-schools-are-assigning-excerpts-of-novels-to-read-instead-of-full-books-why
The kids can obviously read twitter posts (and Manga) and lots of other things. But a novel? Or a 20,000 word essay?
I don't have any data myself on the truth of this, but I think this is considered the problem.
Yes I'm familiar with the debate, OP was asking for personal/anecdotal evidence and I gave my experiences.
My wife complains that students skip her classes and sometimes they're rude to her and there are times when they don't have much to say. She doesn't complain that they're incapable of engaging with the big ideas at all or that they have massive gaps in their ability to comprehend text.
If OP had said: 'people with experience of kids, do you see evidence there's a mental health crisis?' I'd have said 'it's complicated, but maybe'. But my answer to 'do you see evidence there's a literacy crisis?' is 'it's complicated, but not particularly'.
Got it.
The way this was explained to me internally when I worked for Google and Microsoft when they moved towards ideograms is that the main driver is the cost of internationalization and localization. If you label your UX with text, you needs to translate the strings to every language you support and test the UX to make sure it still looks okay and is usable with the translated text that might be different lengths, difference character sets, different reading direction, etc. But if you use abstract symbols, you can use exactly the same UX for different languages and locales.
We’re exploring the 80/20 best practices for living a good life, basically, 80k but for life.
This month’s theme was Home, and we covered:
Design for Lazy
https://www.euzoia.org/p/design-for-lazy
Fix your commute
https://www.euzoia.org/p/fix-your-commute
Home upgrades that beat the hedonic treadmill
https://www.euzoia.org/p/home-upgrades-that-beat-the-hedonic
How to not die at home
https://www.euzoia.org/p/home-how-to-not-die-at-home
Fact or Fake? Hidden stressors at home
https://www.euzoia.org/p/home-fact-or-fake-hidden-stressors
Next five weeks, we’re covering money.
"Just befriend your neighbor" sounds like you're assuming a low-density SFH neighborhood where people don't move much.
We try to make the blog as general as possible but sometimes not everything applies to everybody, this seems like it unconsciously slipped in. We live in London and Amsterdam, where this is possible.
I love things like this, and have some 80/20 money tips. DM me! for instance: did you know it's possible to get most of your 401k and IRA contributions out before 59, with no penalty, for any reason, meaning you should basically always max out those contributions to avoid tax.
Can you elaborate? I used to work in the US but now I am in Germany. I’d like to get my 401k money out and transferred without penalty
Ask Claude to explain roth laddering. There is a catch. Basically, you can choose to take your prior contributions to IRA or 401k and convert them to a new Roth IRA. Now you have to wait 5 years (sorry ) you can then extract that transfer amount penalty free. Learned this on the r FIRE subreddit.
The 5 year wait may or may not be a deal breaker.
Fog War shit:
Iran
1. Iran’s foreign minister, Abbas Araghchi, said that the Strait of Hormuz would be open during the truce, but the IRGC stepped in and said NO!
As tankers and container ships tried to exit, IRGC sent out armed speed boats to herd them back into the Persian Gulf. The IRGC targeted at least three commercial cargo vessels with projectiles, causing damage to the containers and triggering a fire on at least one ship. No crew were injured. And two Indian-flagged tankers were fired upon (the Sanmar Herald and the Jag Arnav). We have a recording of the Sanmar Herald incident: “Sepah Navy! Sepah Navy! This motor vessel Sanmar Herald. You gave me clearance to go! My name number two on your list! You gave me clearance to go!” The captain was pissed! I wonder if they’ll give his company a refund for the toll they paid. No injuries or major damage were reported, and the crews are safe.
2. During the confusion as to whether the Strait was open, five cruise ships slipped through the Strait of Hormuz, hugging the coast of Oman. The first was the Celestyal Discovery late Friday night. Then four other cruise ships (Mein Schiff 4, Mein Schiff 5, MSC Euribia, and the Celestyal Journey) transited the Strait of Hormuz on April 19, 2026, beginning around 00:30 GMT. They sailed at full speed through the declared minefield zone with no stops. The passengers had been unloaded weeks ago. There were only crew aboard these ships. I wonder if the captains and crew will get bonuses from their companies.
3. Despite some garbled reporting by the MSM, the US Navy is not operating in the Persian Gulf or in the Strait (though they sent two destroyers through the Strait last week). The US has posted its Abraham Lincoln carrier group about 125 miles outside the Strait of Hormuz — they're trying to stay out of range of Iranian drones, and they're not engaging directly with Iranian gunboats. They just forced about a half-dozen tankers with Iranian oil to turn back toward the Strait. But Iran still has control of the Strait. There was talk about our LCS assets clearing the minefield(s), but if they are, they're sort of far from immediate support by the carrier group. And the Iranian gunboats seem to be zipping around the Strait without being challenged by the US Navy. Maybe we're waiting for the Bush carrier group to arrive on the scene?
4. And this just in: The US just disabled an Iranian container ship, the Tousca, inbound for Iran by hitting its engine room. Reportedly, US Marines boarded it and took control of it. Iran's military command says it will retaliate for the US shelling and seizure of the Iranian-flagged cargo ship Touska earlier today, calling the blockade action "armed piracy by the US military.” I’ll be curious what the Navy will do with the Tousca. Are they going to tow it to port somewhere? Skuttle it?
In case you’re interested, Sal Mercogolino posts regular YouTube updates on the world shipping situation: https://www.youtube.com/@wgowshipping. He also has a Patreon site: https://www.patreon.com/c/WGOWShipping/posts?vanity=WGOWShipping
5. Also, it’s not clear what the US will do if they encounter a Chinese tanker. And BTW, although the Chinese navy stays pretty close to home, the Chinese have sent their three aircraft carriers out on some blue-water exercises over the past few months. Normally these stay pretty close to their home bases. They’re not nuclear powered, but the next one the one they’re building will be. Google Gemini says their current carriers have the range to reach the Gulf of Oman and return to China without refueling, though. But they’d need all sorts of support ships. I doubt if China has developed the logistics to handle a long-range deployment. Even though our nuclear-powered carriers don’t need refueling, all the support ships in the carrier group do. And it’s a complex logistical operation to move a carrier group from one ocean to another.
6. Various Iranian parliamentarians called for the impeachment of Foreign Minister Araghchi. Control of the Strait is Iran’s biggest strategic bargaining chip in the current game of chicken, and his statement was perceived as a surrender. There were mass demonstrations in major cities against him. As of this writing, Araghchi is still Iran’s Foreign Minister. No word from Supreme Leader Mojtaba Khamenei. In fact, Khamenei has not made a public appearance since his election as Supreme Leader over a month ago. Lots of rumors flying about. One rumor is that he lost either a foot or a leg in the Israeli strike that killed his father. Another rumor is that he’s in a Moscow hospital being treated for the rumored wounds. The fact that the IRGC and the government aren’t coordinating their actions and statements makes me wonder whether he’s in a condition to give orders.
7. Saudi Arabia has maxed out the oil that it’s pumping across its east-west pipeline to its ports on the Red Sea to 7 million barrels per day. Unfortunately, their ports on the Red Sea can’t handle that volume of oil. I assume they’re storing it there? But some Saudi oil is getting out. The Houthis seem to be silent at the moment, so tankers can get through the Bab el-Mandeb Strait. In case you’re wondering, the Suez Canal cannot handle VLCC (Very Large Crude Carriers) and ULCC (Ultra Large Crude Carriers). They can only accommodate mid-size tankers. Iraq has restarted work on an oil pipeline to Jordan that was halted for some reason. The UAE is going to build a pipeline over to Oman. I can’t make the math work myself, but about 9 mb/d is coming out of the Middle East vs 20mb/d that’s not coming through the Strait of Hormuz, and experts suggest we’re facing a shortage of approximately 11 million barrels per day.
Thanks I always like reading these. This is unrelated, but I heat my house with oil, about 700 gallons a year. I called and had my 1000 gallon tank filled up, the price may go up or down (was ~$5.20/ gal), but I feel better without the uncertainty. Uncertainty is expensive.
Regarding bonuses for the crew members on the cruise ships, considering that cruise lines mostly use crew from poorer countries that aren’t paid much, a reasonable guess is “no.”
But the captains and officers are well-paid, indeed. Cabin crew was probably offloaded with the passengers because, well, the companies wouldn't have had to pay them even their minimal minimum wages.
Point six is the most significant IMO. It feels like there's an internal crack or power struggle within the Iranian government over how to react. I get the sense that the pragmatists are willing to make big concessions and the hardliners are refusing, which suggests the pragmatists might think their situation is desperate.
But its all shrouded in the fog of war.
True. But the Saudis and Gulf states face the same problem if they have to shut down their oil wells. It's very hard to restart oil wells that have been shut down.
But let's face it. Even if we fuck Iran's economy to the point of capitulation, the world economy will likely be fucked for years to come.
Well, Russia provided about 12% the world's oil supply, and the Gulf States provided something like 30+ percent. The Gulf States had the ability to increase their production significantly whenever they needed to. My understanding was that they increased their production somewhat to meet the increased world demands with the decrease in Russian oil production (but i'm willing to be corrected on that point if I'm wrong). But now the blockades have reduced oil output from the region by about 60% (?). So, there's nowhere in the world that will make up for that production blockage.
As for Europe, it was willing to apply sanctions to Russia because of the threat it posed to Ukraine and the other countries that bordered Russia. AFAIK, European sanctions against Russian oil and gas are still in effect. But it doesn't matter anymore because hasn't Ukraine knocked all the Druzhba pipeline feeds to Europe off line?
This may be a stupid question -- but would it be safe to say that the US Navy is unable to counter Iranian drone tactics, and is thus unable to take control of the Hormuz Strait ? Or are there other reasons for them to stay out of it ?
It doesn't really have anything to do with drone tactics. Protecting civilian shipping would require navy ships to operate right off the coast of Iran moving in predictable patterns escorting slow, fragile targets. The US cannot guarantee that Iran would be unable to move enough AShMs into position to seriously threaten sinking a Navy ship.
Does the answer have to do with what political risks the Navy/administration are willing to take? Even if you're pretty sure your defenses can fend off a drone attack without major losses., if you think that losing a single support ship or ten sailors would be a political disaster, you may not be willing to risk it.
I suspect that's the case — at least until they get orders to go in and clear the Strait.
But having seen how well Ukraine chased the Russian Navy off the Black Sea, I wonder how well US military tech will hold up drone warfare.
Right, it seems that if the US Navy was capable of effectively countering drone warfare, then losing even a single support ship would be a remote possibility.
Are you implying that the damage was actually due to an Iranian attack? Or are you implying that the loss of the oiler is inhibiting the US fleet from commencing active operations to clear the Strait of Hormuz? ... Or both?
In the Paul Atreides sense of "you control what you could destroy if you wish", both armies control the strait.
Made me think… could the entire Iran situation be Trump’s gom jabbar test?
But in that scenario, we'll all be suffering from spice withdrawal very soon.
My point is that it doesn't look like Iran has a cohesive leadership at this point. And like Donald Trump, their right hand doesn't know what their left hand is doing.
I had an epiphany the other day on immigration. I've been told by a couple people (I first posted it in the Discord) that some of this is easily known, but I consider myself pretty steeped in the culture war, once from the left wing perspective and more recently from the right, and never heard this laid out clearly at all by either side.
It is a one hundred percent fair characterization of people who oppose 'illegal immigration', oppose immigration reforms, and oppose H-1B/PERM fraud, that they are 'anti-immigrant', and not just 'anti-illegal-immigrant'.
This is because the existing immigration laws, as written, aren't actually designed to accept immigrants, the general class.
We accept relatives of existing Americans, we accept the cream of the crop (EB-1/2, O-1, etc), we have the diversity lottery, and we have various consolation prize statuses like political asylum, victims of abuse, etc. But the category that is explicitly for 'work in general' - anything below 'advanced degree' - is reserved for PERM tests, where the company must prove that they attempted to find an American for a position and failed.
An 'average Joe' does not have a pathway to America, common descriptions of his employment-based pathway implicitly describe fraud (typically pretending to have looked for an American by only posting a job ad in the local paper) and the system was explicitly designed to block off such a pathway by default. I've previously stated in many arguments that 'we're letting in a good number a year and the only problem is the length of the line' and I've now realized that this just isn't true at all.
I think most "anti-illegal-immigrant" people are in fact just xenophobic, and the actions of the past year have demonstrated that vividly, but even so, I don't think what you presented is a fair argument. And there are a non-zero number of people out there who honestly are anti-illegal immigration.
One of the more refreshing immigration comments I've seen was on a Hanania H1B post, where someone stated he opposed H1B because he's an entry level tech worker and H1Bs lower his earning potential.
IMO the immigration debate would be more fruitful if everyone could be as direct.
"I'm a single working class male in my 20s, and I oppose mass migration because it's disproportionately men under 40 and I don't need increased sexual competition."
"We're the only African family in our area, and we'd like more African families to migrate so we feel more comfortable practicing our cultural traditions, religion, and heritage."
Etc.
Simple way to tell someone who is "honestly anti-illegal immigration" from a dog-whistling xenophobe: "Would you consider (not automatically agree with, just consider) doubling current legal immigration quotas?" Which would suddenly turn a lot of illegal immigrants into legal.
The latter would automatically reject this idea. The former will at least listen to arguments.
Agreed. Ideally I'd like a referendum on how many immigrants to let in each year, with the number set by the median voter. Half the electorate will view the number as too high and half will view it as too low.
And if the median preference is to double the number to let in - ok, vox populi vox dei.
( Separately, a _much_ more complicated question is how to unravel the mess of ad hoc patchwork exceptions that immigration law contains. Personally, I'd prefer a system that effectively ranked immigrants by their anticipated net benefit to the people already here (including, as a factor, the potential immigrants' intention to assimilate into American culture) and then let them in in rank order till the number from the referendum is reached. Sort-of kind-of like a lower bar version of the O-1s )
This is in the direction of correct, although there's still a confounder: we don't know how many immigrants can be naturally assimilated. It's not simply a matter of what the law permits. If the law authorizes a million immigrants into a community of 100,000, that community will not be the group doing the assimilating.
If a community of 300 million authorizes immigration of 3 million (the usual US legal rate IIRC), it's a lot easier, but still non-zero effort. Someone has to hail those 3 million immigrants and introduce them to US society and explain how it works and ensure that they mostly function like the actual US citizens they are expected to be. And a lot of those immigrants will want to move into enclaves filled with immigrants like themselves, and mostly keep living they way they lived in the old country, including how they drive, how hard they work, how hard they try to pay back loans, how they react to police, and a thousand other things. Even if they're conscientiously trying to integrate, and even if they're sterling citizens who were pulling all their weight and then some in their prior country, it's hard. And far from all of them are trying.
A lot of the rhetoric I'm seeing from people opposing illegal immigration reads as quietly downstream of this understanding. I'll notice in passing that they're perfectly fine with people who are obviously immigrants, because those people are actively trying to behave like the Americans born and raised here.
"It is a one hundred percent fair characterization of people who oppose 'illegal immigration', oppose immigration reforms, and oppose H-1B/PERM fraud, that they are 'anti-immigrant', and not just 'anti-illegal-immigrant'."
The "oppose immigration reforms" part is doing a lot of heavy lifting here. Without that, all you've got is an accusation that anyone who wants the laws to be enforced is anti-immigrant, because the laws are supposedly anti-immigrant, and no, that's not how it works. Lots of people who don't like certain laws and would prefer that the currently-illegal thing be more common and accepted than it is, still want the laws to be enforced because Rule Of Law is a thing. A generally good thing.
And, it's kind of nonsensical to say that people "oppose immigration reform", because nobody opposes immigration reform. Or any other kind of reform; it's implicit in the definition that "reform" points to things that one wants to happen. Unless a person believes that America's policies are absolutely perfect, they're going to want to see changes, and they're going to see those changes as reforms.
They may have a very different idea than you as to what constitutes a "reform". But that's where you need to develop your argument. Because right now, I'm mostly seeing just a denunciation of people as "anti-immigrant" solely on the basis that the reforms they'd like to see are not the sort of reforms that you'd like to see.
> Unless a person believes that America's policies are absolutely perfect
Which is exactly what I see in "anti-immigrant" people. The more stridently they denounce illegal immigrants, the less thought they give to ANY immigration reform -- in any direction. Often they also heap praises on "immigrants who follow the rules" -- the CURRENT rules, of course. The idea that these rules could be changed is studiously not mentioned.
In *any* direction? You really think the strident anti-immigrant people don't want to see tighter visa requirements, more deportations, and maybe a big, beautiful wall on the southern border?
I suspect you are defining "reform" to include only changes in the subset of possible directions that you favor. In which case, you're basically just arguing "these people disagree with me and they're *wrong*", but with an obfuscatory and inflammatory phrasing. And I don't think that's very helpful.
> You really think the strident anti-immigrant people don't want to see tighter visa requirements
I am sure they do, but they almost never openly SAY it.
> more deportations, and maybe a big, beautiful wall on the southern border?
They frame these things as "enforcing existing laws" and "cracking down on the illegals" -- emphasis on "illegals"
I rarely if ever see anyone in US openly call to decrease LEGAL immigration; as I said in an earlier post, the same people who yell loudest about "illegals" often go out of their way to praise immigrants who "follow the rules". Openly calling to lower legal immigration seems to be... uncouth.
> I suspect you are defining "reform" to include only changes in the subset of possible directions that you favor
You happen to be wrong. I do not.
And what if they don't want to? They're anti-immigrants-who-wouldn't-be-allowed-in-if-laws-were-being-followed-as-written. That doesn't mean they're anti-immigrant. It means they're against any immigration that would not explicitly benefit the country.
Except it is very easy to show that giving legal status to say, illegal immigrants who pick strawberries, WOULD benefit the country. "Explicitly benefit the country" is a bad-faith argument.
Given some of the angry comments I've read elsewhere, anti-immigrant sentiment isn't aimed at strawberry pickers. Rather, it's aimed at actual criminals, including everything from vandals and people who get in car accidents while driving uninsured to actual rapists and murderers, with a side helping of immigrants who benefit from government services they obviously cannot have paid for. This anger is then coupled with "what do we do about it?", which turns into "well, in this list of cases, they were obviously here illegally, so let's tighten up the border".
AFAIK, no one has proposed seeking out strawberry pickers and awarding them legal status (the closest I've heard is doing this for people who manage to join the military), and while no one's making much noise about people who maybe pick strawberries by day and then assault people by night, the main complaint is still whether and how they're getting benefits at taxpayer expense.
Nothing bad faith in that, and to assume it is is to ensure the problem only gets worse.
One can argue that for every illegal immigrant rapist, there are a hundred thousand who just pick strawberries, but hopefully you can see how that would be cold comfort to the one suburbanite who lost her daughter, to say nothing of her hundred or so friends and acquaintances, many of whom are struggling to make ends meet and pay taxes and watch their representatives make a show of rolling out the red carpet for illegal immigrants.
All of which is essentially red herring. According to FBI statistics, illegal immigrants commit both violent crime and benefit fraud, at LOWER rate than native-born Americans. (Lower still among legal immigrants.) Yes, this suburbanite is angry. Focusing on him, as opposed to on his several neighbors who lost their daughters to violent US citizens, is nothing but deflection.
People are not struggling to make ends meet because their representatives roll out the red carpet for illegal immigrants. They struggle to make ends meet because for last 45 years US has been implementing policies which divert economic benefits away from the working class, and toward the owner class. This image sums it up perfectly:
https://www.reddit.com/media?url=https%3A%2F%2Fi.redd.it%2Fqtuqzy10r8g61.jpg
> AFAIK, no one has proposed seeking out strawberry pickers and awarding them legal status
I absolutely would. Just like I would love to award legal status to several Brazilian families I know.
Even when you factor in social cohesion? Racial resentment and conflict do not suddenly appear out of nowhere. It is a product of circumstance. We cannot change humanity, but we can change the circumstances that guide their actions.
I am a US citizen. I am also white, Jewish, and originally from Russia. My parents and I immigrated legally -- yet today a Russian family exactly like mine (two parents with professional degrees, one teenager) would have almost no chance of legally immigrating into US. We followed the rules -- luckily it was before the rules were changed.
Having said that, I feel rather more "cohesion" with a Brazilian family I've known for about 8 years -- which is in US illegally, -- than with white Christian Nationalists. The former never did anything bad for me (actually, helped me a great deal and vice versa), which is not something I can say about the latter.
What has any illegal immigrant ever done to you?
Many thanks for writing the reply I wanted to write, much better than I could have written it.
99% of US citizens only have a very cursory understanding of the US immigration system - simply because they don’t need to understand it. Nobody reasons in terms of “PERM” or “EB-2”, people just reason off vibes.
Yes, other than “very online” people and think tank employees nobody knows what H2A is, how much the salaries are, or what taxes are being docked vs evaded. Policy is decided based off vibes.
The federal minimum wage is 7.25/hour. If you're saying $18 is three times "the federal minimum wage minus taxes", that's nice, but the person making $18/hour will also pay taxes, so it's irrelevant.
Ah, right, that makes sense.
Correct, an agricultural job sucks way more than working at McDonald's. This is why garbage-men also tend to make more than cashiers. The job sucks more and it requires a higher wage to find someone to do it.
Picking strawberries or raspberries by hand all day, outside, exposed to the elements and the bugs, for away from any entertainment which one could spend money on after work?
Yeah, it's worse.
Using the latest BLS nonfatal injury/illness rate table, the comparison looks like this:
Fast food worker proxy — Limited-service restaurants, NAICS 722513
Total recordable rate: 2.6 per 100 full-time workers
DART rate: 1.0
Days-away-from-work rate: 0.8
Crop production worker — Crop production, NAICS 111
Total recordable rate: 4.1
DART rate: 2.5
Days-away-from-work rate: 1.4
Animal production worker — Animal production & aquaculture, NAICS 112
Total recordable rate: 4.0
DART rate: 2.8
Days-away-from-work rate: 1.7
Farm labor contractor / crew worker — Farm labor contractors & crew leaders, NAICS 115115
Total recordable rate: 4.2
DART rate: 2.8
Days-away-from-work rate: 1.7
DART = cases with days away from work, job restriction, or transfer.
Source: BLS 2024 injury and illness rate table: https://www.bls.gov/web/osh/table-1-industry-rates-national.htm
I'm Australian and don't think the US visa system is well structured. I also don't see why a diversity lottery system for average Joes plus a merit system for exceptional Joes would be anti-immigration.
Why would you deliberately take normal people if there was a line of the world's best artists and professionals stretching out the door? And if you, for some reason, held a lottery to allow normal people in, why would you want to make the pool of winning tickets larger?
The merit system stops right under "world's best artists and professionals" (i.e. EB-1 or O-1). Anything lesser, such as a mere credentialed and experienced SRE who is not literally one of the best in the world, is not qualified by merit alone.
I have worked on O-1s and EB-1s, and referring to recipients thereof as "the world's best" implies that the bar is much higher than it is in practice. Though given the current Administration's antipathy to all immigrants, who knows what the current practice is.
It has been a while, and it would be unethical to give specific names, but I can recall getting either an O-1 or an EB-1 for someone who designed and sold women's shoes, a young filmmaker who had never made a feature film and was coming to work at a film school for high school students; a few other artists who had a couple of shows scheduled at galleries in New York, several animators who worked on big films at Dreamworks or the like (somehow, each of them played a critical role in the success of Shrek IV, or whatever), the inventor of some sort of pharmaceutical compounding doohickey, and a forensic pathologist who worked at a county medical examiner's office.
Not that these people were not quite talented, but in the grand scheme of things, they were indistinguishable from a zillion other quite talented people.
Another eg: I am very sure that Melania Trump qualified for her EB-1, despite knowing little about her. High pay (pretty easy bar to reach, given that the median model makes very little) plus was on the cover of European Vogue ("Evidence that your work has been displayed at artistic exhibitions or showcases") plus judged a couple of fashion shows ("Evidence that you have been asked to judge the work of others") could easily be enough because you only need to meet three criteria. https://www.uscis.gov/working-in-the-united-states/permanent-workers/employment-based-immigration-first-preference-eb-1
"...but I can recall getting either an O-1 or an EB-1 for someone who designed and sold women's shoes, a young filmmaker who had never made a feature film and was coming to work at a film school for high school students..."
Also, the O-1 visa is used for foreign minor league baseball players. So a foreign national playing high-A minor league baseball will be here on an O-1 visa.
In some sense these *are* the world's best. I'd guess that most of the guys playing minor league baseball are in the top 25,000 in the world at that. In another sense ... where are we drawing the cut-line for "world's best" again?
PhD folks also often come in on O-1 visas. I don't know how many of the world's PhDs would qualify. Probably a lot more than 25,000 ...
Right, this is where the US visa system is weird, because it takes (for example) a quota of 10,500 migrants in speciality occupations without some of the H-1B restrictions, but only if they're from Australia. Or you could get a normal H-1B. Either way you'd be on a "non-immigrant" basis, except actually you can just keep applying for permanent residency because an I140 is somehow not intent to immigrate.
So you have a situation where immigrants from certain countries are effectively allowed to work in the US (because of the H-1B, or the magical E-3) and stay there until they're either rich enough to buy a visa/green card (E-2/EB-5), attractive enough to marry (I130), or lucky enough to win the green card lottery, which can have reasonable odds over a few decades depending on the country of origin. Australia is sitting at 5.39% chance of winning, so if I played the green card lottery for a couple of decades I'd have a 2 in 3 chance of winning permanent residency
This is obviously a very silly system but I'd struggle to call it completely anti-immigrant. It's designed by madmen but it does provide pathways to some people sometimes, although God help you if you're Indian.
>only if they're from Australia
That is because of a treaty with Australia
>because an I140 is somehow not intent to immigrate.
Of course an I-140 is an intent to immigrate; it is an application for permanent residence. But there is no problem forming an intent to immigrate after you arrive; it is only fraudulent if you have the intent to immigrate when you arrive on certain non-immigrant visas. And only certain ones; for example, an H-1B is a dual intent visa, as are many others. https://www.nolo.com/legal-encyclopedia/what-are-dual-intent-visas.html
The E-3 in particular is somewhat complicated and I believe you're oversimplifying the situation: https://peterchu.com/blogs/medium-feed/is-the-e-3-visa-immigrant-or-nonimmigrant
I was not referring to the E3 in particular. You specifically mentioned the H1-B, which again is a dual intent visa.
Doesn't this just assume that the only way to not be anti-immigrant is to be willing to accept more-or-less unlimited numbers of immigrants who meet some very minimal standard like "not a known criminal or terrorist, not carrying a contagious disease, not likely to be immediately a dependent of the state?" I mean, yeah, this works in the sense that I can call everyone in poltiics a communist by defining communist as "anyone who wants any regulation at all of markets." But I can't imagine that this convinces anyone who didn't start out agreeing with you.
An example of a person who cannot immigrate to the US by default without fraud being committed on their behalf is a British accountant with a bachelor's degree and ten years of experience. I think many people who say "they should come the legal way" would be surprised by that fact.
"Unlimited" is not implied at all, a numeric cap would make plenty of sense and many are already deployed all over the place in the immigration system.
If your definition of anti-immigrant is supporting a policy that does not allow most people to immigrate to your country, you will indeed find that most voters in most desirable countries are anti-immigrant by your definition. But again, is there some benefit to this definition other than defining the other side as bad?
Like, if a billion perfectly fine human beings want to immigrate to the US this year, but I don't think it would be a good idea for us to take in a billion immigrants next year, by your definition I'm anti-immigrant. But also, so are everyone except the literal open-borders folks.
You’re pointing out that the system is broken. There hasn’t been a push for immigration reform that would allow for more consistent monitoring on the southern border while also allowing for easier immigration by the sort of average Joe you’re talking about here. Instead the options seem to be almost uncontrolled immigration under Biden, or completely shut down immigration under Trump.
Unless there’s a viable solution presented that allows for normal productive people like you’re describing to get in, then I think it’s a legitimate position to take to be opposed to massive illegal immigration, while still supporting more legal immigrants that are vetted.
UK citizens are eligible for diversity visas now.
That hardly addresses the issue. The chance of winning the diversity visa lottery is maybe 1 percent.
Now you're arguing something different than OP, not just that everyone should have a chance but that the probability should be high too. (Of course even without the lottery everyone would have a chance by making the right career choices, or marrying right. And if you believe in "accident of birth"-thinking then everyone had a chance of being born in the US already, about 4%.)
Also, 1% per year adds up. Much higher and it would almost be guaranteed one would get in sooner or later. That certainly wouldn't work if applied to the whole world.
No, I am arguing exactly what OP argued, which is that there is no practical immigration avenue for a British accountant with a bachelor's degree and ten years of experience.
>Also, 1% per year adds up
Not very quickly. It takes 69 years to reach a 50 pct lifetime chance of winning. 1-0.99^69 = .5002.
Please stop posting non-sequitors to previous posts with high-temperature, vague insinuations.
I am not asking for proof I am asking you to stop doing this.
I know a lot of things. Most of them are not relevant. I even know a lot of things about Afghanistan. Even those are not relevant. Your posts have a tendency to go off the rails if the topic. I previously complained that you on your own had 1 out 9 posts in previous threads. Could you limit yourself as a start to 1 out of 20? 5% of total posts?
I am not familiar with the law. I know a few people who went to USA for PhD, and stayed there. Some of them work in IT, some of them are university teachers. What do you assume was the mechanism they got there, and would it match your explanation? (Let's assume they *didn't* acquire citizenship by marriage.)
This would probably be EB-2, the employer-sponsored permanent residence visa for advanced degree holders. For the university professors, their participation in public research qualifies for a National Interest Waiver from the PERM test, so they just need the employer sponsor (the university). I don't know whether IT workers qualify for NIW, it probably depends on the company. If not, then PERM fraud was probably involved for those guys.
For a professor, surely EB-1B (outstanding professor or researcher) is commonly used. The approval rate is apparently extremely high. https://manifestlaw.com/blog/eb1-approval-rate/
And, you are assuming that OP's friends have green cards; they can be here long term on nonimmigrant visas. O-1s can be renewed repeatedly, for example.
Having worked in Silicon Valley, it seems like the PERM test is pretty permeable. Or do H1-Bs come in a different route?
H-1Bs are subject to the same requirement (with no waiver opportunity) though note also that they are a nonimmigrant visa. They fill work shortages proven by looking for two months for an American at a market rate salary and failing; this requirement is generally cheated by posting an ad somewhere nobody will find it (before jobs.now started aggregating these).
> at a market rate salary
At least to start. But the fact that it's very very difficult for an H1-B visa holder to get their H1-B transferred to another company, studies have shown that over the salaries of the visa-holders are suppressed over time. While US citizens can freely move from employer to employer for a better salary. My experience was that each job move would garner me an immediate 10% salary increase. While staying with an employer year after year, I garnered piddling 1-2% salary increases.
At the last ACX meet-up I went to … apart from “Is AI going to kill everyone?”, conversation turned to personality tests (https://en.wikipedia.org/wiki/Myers%E2%80%93Briggs_Type_Indicator etc).
I mentioned the methodology used in constructing the Minnesota Multiphasic Personality Inventory (https://en.wikipedia.org/wiki/Minnesota_Multiphasic_Personality_Inventory)
The people who are in to personality tests (I am not a person who is in to personality tests) seemed to think that tests for clinical psychology (e.g. MMPI) are an entirely different kind of thing from Meyers Briggs type personality psychology. Certainly, they purport to measure something different.
ENTP here as well even if I "believe" in it all very little, its more like a fun fantasy to connect with; like all these categorization frameworks (astrology, etc) they do a really good job with making the writing "feel" like its exactly what I think I am! lol
now my unnecessary and unceasing desire to debate.....
I'm kinda into MBTI, but I'm also an INTJ. It's very funny that MBTI intuitively "feels right" to me. That's what someone with high N *would* say!
I find it interesting that:
(a) the professional psych people seem to really dislike Myers-Briggs, and
(b) often admit that the 4 Myers-Briggs axes correlate reasonably well to 4 of the OCEAN/CANOE axes.
Example of both is in the interview here: https://www.interviewaceapp.com/blog/myers-briggs-vs-ocean-an-industrial-psychologist-breaks-down-the-differences
Discussion on correlations in the "What kind of correlations are there between the two, if any?" section.
There seems to be something wrong with Google Search. When I search for "overcoming bias respectable rants", Google does not return the link to the well-known article: https://www.overcomingbias.com/p/respectable-rantshtml.
This seems to happen to only this specific article. Why?
A win for DuckDuckGo, which returns the article as the first hit. Google didn’t find it when I tried it. (I only checked the first four pages of results.) Google did return the top level page of overcomingbias as the third result. Google returned the article as the first result when I searched for "overcoming bias respectable rants site:overcomingbias.com".
When I tried the search in a Firefox private browsing window, it worked (returning the article as the first hit). But I couldn't get the search to work in a regular browser window even after deleting cookies and local data.
Normally, I use DuckDuckGo first, and only try Google when I don't find what I'm looking for using DuckDuckGo, so I only see DuckDuckGo's failures. I still suspect that Google is better than DuckDuckGo, but I now realize I don't have good evidence of that.
Google seems to be very bad at serving anything other than Reddit posts nowadays. They're not even trying anymore.
That's a great essay.
Googling the words "overcoming bias respectable rants", in that order without quotation marks, got the Hanson article as the top response in three browsers on two different devices, regular and incognito mode.
If that was a ploy to get me to google that term, well played, you understand how social media works. Other than that, I had zero issues with that search, it was the #1 result, and its comment section was #2.
Did you put "overcoming bias respectable rants" in quotes, like that? Because that got zero hits for me, probably because that is not the title. But a search for "overcoming bias" "respectable rants" brought it right up.
Weird, two double quotes really works!
Not weird, because putting the entire thing in a set of quotes creates a search for that specific string of words. There is no particular reason to think that that set of words exists.
Just a couple of days ago I wanted to reread the Meditations On Moloch and had the same problem. The first three pages are a bunch of sites talking about it, readings on spotify and youtube, or even tweets. Then I gave up and just searched my browsing history (which goes back years and I should probably clean it up).
I just assumed that google's search algo has been bent so heavily in the name of pushing marketing to the top that now when it's looking for something that can't be monetized in any way it just comes off as completely lobotomized.
I just searched for Meditations on Moloch and it came right up.
My first result is SlateStarCodexAbridged, then a link to the effective altruism forum, then a tweet from some random guy, the actual post is nowhere to be seen within the first five pages of results.
I guess fingerprinting plays a role. Are you logged into google? Do you get the same results in a private window?
Regardless, this is not the behaviour I'd expect from a supposedly good search engine.
What's wrong with the SlateStarCodexAbridged version?
I had a similar experience in the past, trying to find something, instead finding dozens of news articles talking about the thing... and of course none of them actually links to the thing they talk about, instead they just link to other articles talking about the thing.
Google search sucks, but also this SEO approach (never link X, if you can instead link another article about X on your own server) sucks.
Even if they did link to it, Google might not be indexing it anyway. For example, Google refuses to index my blog for no apparent reason, and this seems to be pretty common. In fact, Google is even failing to index *Github repositories* now.
Wait, what was the book released from the rightward side of our conspiracy?
I didn't mention one, but the EA For Christians one sort of half-fits that bill.
Anthropic recently published benchmarks for Claude 4.6, 4.7 and Mythos (https://www.anthropic.com/news/claude-opus-4-7)
And here's a link to a portion of their benchmarks summary: https://imgur.com/a/xnUi5FP
So Claude 4.7 is better than 4.6 at almost everything. But notice that Mythos is about the same amount better than 4.7, and none of us are going to get to use Mythos now, and maybe not ever. So that got me ruminating about the emergence of a dividing line, between improvements in AI (such as, presumably, 4.7) that are shared with the public and improvements like Mythos that are not. It made me realize that I've gotten used to the improvements being passed on to the public -- some in the form of access to new models that's free or cheap for all, some sold to manufacturers, etc., to integrate into their products and processes.
Seems clear that no good can come of making Mythos available to all who want it. It's like passing out the keys to a bunch of doors. But what about future models? Seems like many of them, maybe most of them, will have capabilities that it's dangerous to make widely available. So then what? I'm not crazy about the idea of either the AI companies or the government being the sole entity that can use these models. Are we entering an era of a division between the AI haves and the have nots? How's that going to play out?
It's quite reasonable that Mythos-level bug finding 1) is genuinely new 2) no one else will invent it for at least a few months and 3) can be used defensively first.
I've never been a full-disclosure guy, though.
I think most likely is a combination of:
1. Anthropic puts a lot of effort into making their standard Opus line unable to do cyberwarfare. They fail to train it in that, they potentially untrain it in that (complicated but theoretically possible), and they align it very heavily to refuse cyberwarfare requests.
2. Anthropic maintains the Mythos brand as AIs slightly better than the best public Opus. The white hat cyber companies always have the best Mythos, and the public always has the best Opus. I still haven't been able to figure out whether Mythos is simply bigger than Opus in the same way Opus is bigger than Sonnet, but it seems like this must be part of the story (since it apparently costs a lot per token to use). If Mythos is so big that it would be cost-prohibitive for the public to use it anyway, and Anthropic would never have bothered training a model that big for the public, then maybe not much is lost by keeping it out of public hands.
I agree that "exactly how is Opus different from Mythos?" is one of the more interesting questions in practical AI right now.
Yeah, I agree with these.
I think we could maybe add a third issue, which is that Anthropic appears to be pretty compute constrained and is having a hard time keeping a lot of customers happy already, so maybe it would have been impracticable to serve Mythos to the public at this time. At least, Ed Zitron quotes the Financial Times (paywall) to that effect here: https://www.wheresyoured.at/four-horsemen-of-the-aipocalypse/
> If Mythos is so big that it would be cost-prohibitive for the public to use it
I think this varies greatly for different parts of the public. The person on the street who just uses Claude as a search engine probably wouldn't pay a fortune to upgrade from Opus to Mythos. But companies that are already spending tens of thousands of dollars on Claude API every month might well be willing to pay for improved performance.
I can see how that would work. But there are a lot of other abilities future AI could have that are not cyberwarfarish the way Mythos’s coding abilities are, but could be still used to do great harm. For example:
-Ability to influence people, or ability to learn how to influence an individual user (currently AI can’t really acquire permanent info and skills from interactions with users, but maybe a future advance will make that possible.)
-Ability to make deep biological interventions: The AI can understand and change a cell the way a clockmaker can a clock. Or it can function as an extra organ, fully integrated with all the connections and feedback loops of the rest of the body. (Or if those sound implausible to you, imagine some lessor abilities that are still way beyond what we and AI can do now.)
In fact it seems quite possible to me that we’re entering an era where AI’s will have many abilities that can be used to do great harm. Will it be feasible for Anthropic to handle the dangers they pose using the model you suggest for Mythos?
And, of course, not all of these models with extraordinary abilities will be coming out of Anthropic.
I am refraining here from making crass and savage jokes about ways our president and various other fools and monsters with power would want to use the new powers they'd have access to.
Rumor suggests that up until "GPT-5", each successive GPT-number release used another order-of-magnitude compute during training. GPT-4.5 was the next step along that path, but was so large that it was cost-prohibitive to serve at scale, and so they only made it available as a research preview to paying subscribers. Progress then all moved to test-time compute / reasoning models, verifiable rewards, etc, rather than brute scaling, giving us the GPT-5 series.
My strong suspicion is that Mythos is another true order-of-magnitude step along the brute pre-training scaling line, plus all the advancements in post training. The name fits into their schema of "literary work sizes" and suggests it's larger than Opus in terms of raw parameter count, which then also suggests additional compute was used (via Chinchilla / compute-optimal training).
(I realize I'm sort of conflating OpenAI and Anthropic's model strategy here, but I'm confident the same dynamics affected both).
I'm a little skeptical that this is 'the beginning of the AI haves and have-nots'. Making a really expensive model and then not selling it to hardly anyone seems like it defeats the point for these for-profit companies.
(At least short of ASI scenarios where possibly everything goes out the window)
I do think we'll see more of this sort of Glasswing-style rollouts where we give the 'good guys' a head-start on the bad-actors, at least for awhile, but I do think it's "head start" and not "permanent advantage". I don't know how long of a head-start it'll be and I suspect Anthropic is also playing-it-by-ear, too. I'd guess less than six months and quite possibly significantly less.
And it's possible this is just the new normal... but actually I'd guess we see shorter and shorter delays, while in some sense 'the security holes you will always have with you', I do think there's going to be a diminishing returns aspect to these where even as models get smarter they'll have a harder and harder time finding new holes to exploit in existing code.
This is assuming that Mythos actually exists and actually performs as advertised.
How do you think Anthropic is finding 27 year old vulnerabilities in OpenBSD then?!
Are you saying that it would've been impossible to find these vulnerabilities by any means other than Mythos ?
Is the suggestion here that all the companies listed as using Mythos are themselves in on the conspiracy?
Not necessarily, but thus far I am not getting a clear picture on what those companies are (allegedly) using Mythos for, nor how effective it is in practice (as opposed to on benchmarks), nor what kind of relationship those companies have with Anthropic. By analogy, if there are a bunch of YouTube influencers proclaiming that BugCo-brand stain remover is the best stain remover they've ever seen, but none of them show real-world effects on BugCo on common stains, then this alone should not cause you to update greatly in favor of BugCo (though it still should make you update a little).
Anthropic *already* publicly disclosed several of the vulnerabilities they found though.
Plus hashes of others that are pending patches.
I also don't get why this level of skepticism about Mythos is in the water. It's one thing to say that Anthropic's subjective descriptions of Mythos's abilities and impact may be exaggerated...
... but the claims that "Mythos might not exist" and "these 3rd party benchmarks may be fake" seem *highly* unlikely to me - just making up a model or lying about benchmark results would be an incredibly short-sighted thing for Anthropic to do and I see no reason why they'd be doing it. Yes "something something, marketing" but "let's have a marketing minute for a few weeks then completely crater our credibility for no reason" does not seem realistic to me, even before you consider questions like "wouldn't one of the third parties involved in Glasswing come out and say 'it's fake' or 'it's highly exaggerated'".
I think marketing is exactly the point: brewing up some hype for their upcoming product by claiming that it's too powerful to ever be allowed into human hands. Oh no, anyone who gets his hands on this AI would instantly become king of the world ! Too bad we're not releasing it, not for any amount of money ... ... well, ok, you twisted our arm...
Mythos does most likely exist, in the sense that it's Claude Opus 4.8 or whatever; it might even exist in the sense of being some new revolutionary model and not a slight upgrade to the previous version (though this is less likely). And it most likely does perform reasonably well on benchmarks, just like all the other models (more or less). But in total absence of evidence for its performance in real-world scenarios (other than what's presented by the people who are selling it), I am going to reserve judgement. I don't think doing so constitutes an outrageous leap of logic.
Brewing up hype for a fake product you made up (or highly exaggerated) is not a good marketing strategy and I doubt that's what Anthropic is doing.
And I'm not sure why we're talking hypothetically about "probably performs reasonably well on benchmarks, like other models". Anthropic has published the specific benchmark numbers, they're linked in the comment you originally replied to.
If you think those numbers are just lies, again, I think that is an obviously bad marketing strategy and I don't think it's likely that Anthropic is doing it. "Marketing" only makes sense if you actually sell the thing eventually at which point people will run those same benchmarks and if they get significantly worse results, that's *huge* egg on Anthropic's face for no reason - if you're going to lie, why do it in a way that people can easily check for later?
And maybe a semantic point, but accusing the company of outright lying doesn't feel like "reserving judgement" to me. That's not a neutral default position, that's one I'd only take with evidence and I just haven't seen any evidence for it.
> Brewing up hype for a fake product you made up (or highly exaggerated) is not a good marketing strategy...
Isn't this the essence of marketing, or at least advertisement ? Think of any car/medicine/stain remover/software commercial...
> Anthropic has published the specific benchmark numbers, they're linked in the comment you originally replied to.
Was anyone who doesn't work at Anthropic able to verify them ?
> If you think those numbers are just lies...
No, probably not; instead, they're probably exaggerations. By analogy, that stain remover probably does remove some stains.
> if you're going to lie, why do it in a way that people can easily check for later?
Depends on how easily they can do it. Consider video card benchmarks back in their heyday (before NVidia ate everyone's lunch), or Volkswagen's diesel emissions numbers, or (for a more relevant scenario) all those spectacular "AI-generated artworks" that took human users months of work to create by using AI. There's lots of precedent.
> And maybe a semantic point, but accusing the company of outright lying doesn't feel like "reserving judgement" to me.
I guess maybe that's the main point of our disagreement: whenever any company claims to have created a revolutionary new product yet refuses to let people independently evaluate it, my default neutral position is that they're either lying outright or massively exaggerating. I've been burned too many times before. But it's possible that my judgement is too harsh, and/or that Anthropic is an outlier.
People who don't work at Anthropic have been able to verify its performance. That's Project Glasswing. The proof is in the pudding - before the public announcement, open source maintainers noticed a sudden step-change in the quality of vulnerability reports they'd been getting. Where previously they'd been receiving an annoying stream of bad, clearly AI-generated reports of fake vulnerabilities which just wasted everyone's time, sometime in late February/March, something changed, and suddenly they were getting lots of high quality reports of true vulnerabilities, often very subtle ones that had been latent in the codebase for years. "It's probably fake/overblown marketing" is bad mental model.
...As opposed to Anthropic having, what, discovered all those security vulnerabilities manually? Or there being some large-scale fraud and none of the bugs actually even existed?
Seconding Bugmaster here. Whether it's open or closed source, many vulnerabilities are only hidden as long as nobody takes a good look, not necessarily because they were so subtle. The incentives are clear: Writing bad code is much easier than spotting it, or than writing good code. Static program analysis is not exactly a new field, and AI is at most a new tool in the box.
https://en.wikipedia.org/wiki/Static_program_analysis
I am not too impressed with Claude/Mythos finding all those vulnerabilities. That is to say, it's impressive that it can find them at all, but the sad reality is that once you have a dedicated agent capable of finding security vulnerabilities, be he a human or an LLM, you're going to start finding security holes all over the place because humanity's collective codebase is riddled with them top to bottom. It is not the case that Mythos is some kind of a dark wizard uberhacker with superhuman programming abilities; it is merely the case that it has nothing better to do than sit there and sift through a bunch of boring code. You could do it too -- maybe slower, and maybe not as well, or maybe a lot better (depending on your skill level). And I suspect that while Mythos does outperform Claude 4.7 on benchmark, this doesn't translate into too much real-world gain in terms of bugs found per hour.
Do you know any human hacker with thousands of CVEs to their name? If not, then Mythos clearly is objectively a superhuman hacker at least in this respect, even if human hackers found trickier vulnerabilities before.
I find it bizarre how quickly people minimize achievements of AI models that were previously (not too long ago) said to be in the realm of science fiction.
Relevant funny tweet:
https://x.com/JeffLadish/status/2045994492587860264?s=20
"These “self-replicating factories” taking over the Midwest are just the latest desperate marketing stunt by Anthropic 🙄
People act like this is the first instance of self-replicating solar-powered factories taking over Earth’s plains. Omg, go touch some fucking GRASS"
Love those parody tweets! Many Thanks for the comment and the link!
Sure, and I did say it was impressive; and technically you are correct, Claude is superhuman -- in the same way that e.g. Excel is superhuman, because no human could possibly add up all those numbers so fast. My point is that just because Claude can find security vulnerabilities, doesn't mean that it presages the end of human civilization or anything of the sort. The problem is that security vulnerabilities are commonplace and abundant. Finding a vulnerability in the laws of physics that would allow you to make self-replicating factories (other than in the old-fashioned way) would be much, much harder, and most likely impossible.
That said though, personally I do hope that Claude/Mythos is at least in the ballpark of being as good as Anthropic claims. Maybe this way we'd finally patch some of these security holes that human malicious actors are exploiting all the time.
Many thousands more AI-hours from previous models have been spent looking for these vulnerabilities and have not been as successful, this isn't the first time anyone thought to direct AI at cyber security
> Are we entering an era of a division between the AI haves and the have nots?
Yep, that was always the plan. Giving weak AI access to the masses was just a way to raise the money needed to build the strong AI.
The masses didn't object because most of the intellectuals were busy explaining how a machine can never be smarter than a human, because machines do not have a soul, and intelligence is only a myth.
> How's that going to play out?
The bosses at the AI companies will tell their AIs "make me an immortal emperor of the universe" and maybe one of them will succeed.
Or Trump will seize their AIs and do the same.
What happens next, that's up to the emperor.
That's the good ending. If, however, the analogy to natural selection holds for gradient descent and the created AI ends up with weird wants designed by noone, we're simply dead.
AI already has "weird wants designed by no one", just like every other tool ever made by humans. My car wants to turn ever so slightly to the right sometimes, instead of going straight like I want it to. My computer often wants to consume as much electricity as possible on doing nothing useful. My hammer wants to hit my thumb instead of the nail. I am not worried about hammer-pocalypse, however.
You are not using the relevant meaning of "want" here. Its about someone or something having the kind of external actions that aim to steer the world to some class of outcomes. It is meaningless to talk about wants in the context of something without intelligence.
Also, even among entities with some amount of intelligence (ie. general capability to achieve outcomes) such as current AI models, their dangerousness varies wildly with their capabilities. Apocalyptic scenarios are overwhelmingly more likely with superintelligence.
> It is meaningless to talk about wants in the context of something without intelligence.
Yes, that was indeed my point.
> Also, even among entities with some amount of intelligence (ie. general capability to achieve outcomes) such as current AI models...
Yes, technically current AI models to have some capability to achieve outcomes when wielded by a competent user; but the same applies to e.g. the toolbox that is sitting in my garage. I could theoretically build a lot of stuff with the tools in there; maybe even some dangerous stuff. But the toolbox is not intelligent despite having this capability.
Granted, LLMs are a lot more versatile overall than boxes full of screwdrivers and pliers; but that doesn't make them "intelligent" in the colloquial sense of the word. And yes, you can build an LLM "agent" which will execute some task in a loop; perhaps even a complex task that is beyound the capability of a cron job; but loops are not the only prerequisite for intelligence either. Also note that boxes full of screwdrivers have many capabilities that LLMs lack, such as attaching physical metal parts to each other using screws.
> Apocalyptic scenarios are overwhelmingly more likely with superintelligence.
It seems that "superintelligence" is usually defined as "an agent capable of causing apocalyptic scenarios", so you are correct -- but only tautologically. Ok, that was perhaps unfair; I think most proponents of the idea would say that "superintelligence" is defined as "something thousands of times smarter than a human"; but when pressed about what "smarter" means, they usually end up listing apocalyptic scenarios.
Current AI models have the capability to achieve outcomes even when not wielded by anyone. You have to set them up first, just like you have to set up humans too to produce intelligent actions: when placed in a random point in space in our universe, they simply die, if they grow up in a sensory deprivation tank, they are scarcely more intelligent than a beast. In contrast, it does not matter how you place a hammer, it won't aim to achieve outcomes.
Is it the colloquial definition of intelligence that LLMs fail to fulfill or your idiosyncratic one? What feature(s) are they missing in your view?
I highly support banning Police Robots with aggressive capabilities and preferably military robots too (although that might take a bit longer). Why? Because robots tend to be mass-controllable. So for example, if an evil dictator assumes control, he can wield robot police forces to keep himself in power; the police robots. In general, as those technologies tend to be centrally-controllable and unquestioning of authority, if anything goes wrong (cyberattack, for example), it can cause a massive disaster. Or even things like bugs and all sorts of common failures. Humans should have a monopoly on violence and firepower (at least until robots were to be extremely more reliable and incorruptible than they currently are). Not to mention, the AI catastrophe risk contribution is obvious from armed/violence-enabled robots.
So, can we get together and draft proposals for this? Maybe a small org dedicated to the cause? I think easily the AI risk/pause AI camp would immediately agree (and hopefully help). But it seems like wider society would accept as well. The objective of course is to try to pass legislation in countries for this effect, and also international accords. Thanks :)
Too late? The Chinese already have robots running races, and they seem to be improving. If you can get bipedal robots moving as well as, or even better than, humans then the applications for police and military are obvious.
https://www.bbc.com/news/videos/cz0e54yrppno
US police are in practice rather decentralized. And the federal government is constitutionally prohibited from "commandeering" state/local governments to enforce federal laws. Federal law enforcement is vastly outweighed by state/local, which was why the late William Stuntz regarded crimes adopted at the federal level to be largely grandstanding without much ability to enforce https://entitledtoanopinion.wordpress.com/2018/06/23/the-collapse-of-american-criminal-justice/
> And the federal government is constitutionally prohibited from "commandeering" state/local governments to enforce federal laws
That hasn't stopped them much in practice lately...
Trump actually did stop after SCOTUS ruled against him federalizing the Illinois National Guard and deploying it as law enforcement against the wishes of Illinois' governor. He gave up more quickly then he did on tariffs.
Key rollover = regime change
Crippling the regime by implementing 2FA
What about Private Security Robots?
I can see the movie marquee now: Robo Mall Cop
Tagline: “Just try loitering now… punks!”
"You are in direct violation of Mall Code 1.13, Section 9. You now have five seconds to comply!"
Has there ever been a successful ban of any military technology that is both effective and without substitute?
- Chemical weapons are banned, but aren't effective on the modern battlefield.
- Biological weapons are banned, but very indiscriminate and therefore not effective.
- Nuclear weapons or mines are effective, but haven't been banned successfully.
If/when military robots become effective and on net better than their substitutes, they will be built and used by at least some actors on the global stage, whether they're "banned" or not.
Did you also read this excellent ACOUP piece about why chemical are irrelevant in modern warfare?
https://acoup.blog/2020/03/20/collections-why-dont-we-use-chemical-weapons-anymore/
Dum-dum bullets were banned for military use in the 1890s. Technically I guess you could say they can be substituted with regular bullets, but those aren't as effective at causing massive injuries.
I don't think it's even an "if". Autonomous drones are already being used in Ukraine.
Nuclear weapons have not been used since they were banned. I'd consider their ban to be a qualified success.
I don't think we can let "several non-aligned countries are actively maintaining nuclear weapons and are ready to use them against their enemies" count as "banned" in any meaningful sense of the word. Nuclear weapons certainly aren't "banned" in those countries that have them.
They aren’t banned but they aren’t being used either even though there is nothing stopping the US from nuking Iran.
In what way were nuclear weapons banned? Nine countries have nukes. The reduction treaty New START expired without replacement. China is increasing their warhead count massively. The only reasons why they haven't been used are MAD and dumb luck.
Chemical weapons are being used by Russia on the modern battlefield in Ukraine, so I would assume that they are effective.
I wouldn't consider occasional use of a technology by the Russian military to be strong evidence for its effectiveness, but even if it is, that would confirm my point that any military technology won't remain "banned" for long if it's effective.
Not necessarily. This might be a WW1 situation where they just tried everything else, nothing is giving them the breakthrough, and they'll just try anything.
Considering that they are reported to have done it thirteen thousand times and that Ukrainian soldiers have acknowledged that the tactic has been effective, I would think the simplest explanation would be that the tactic has been effective.
I’m running a quant trading bootcamp in SF this coming weekend (April 24-26), at Mox. I started these up at Manifest Summer Camp nearly two years ago and we’ve been doing them every couple months or so. If you’re in the Bay Area (or heck, if you want to travel in for it) you can sign up here:
https://www.trading.camp
or just swing by to say hi at some point or for our Provide Your Own Liquidity afterparty Sunday night (DM me for link)
Spartacus.app (2024 ACX Grantee) with an urgent ask, seeking LOIs for the SFF grant app before Tuesday.
For anyone running an AI-safety, governance, researcher-coordination, or labor-AI org: Spartacus is applying to the Freedom and Fairness track of the current Survival and Flourishing Fund grant and would like to submit with several non-binding letters of intent to collaborate over the next 6-12 months. Submission deadline is Wednesday, April 22
What Spartacus is: conditional-commitment infrastructure applied to collective action. Individuals make binding commitments that activate only when a threshold of others also commit, with temporary anonymity until activation. Designed for the "I'd commit if enough others did" class of problems: alignment pledges, lab publication norms, researcher career switches, employee safety speech, coordinated legislative testimony, group statements across ideological lines.
Recent proof point: march2testify at Stanford in December 2025. a 500-creator mobilization for California AB 412, activated on a 100-commitment threshold, credited publicly on the landing page. First mass creator mobilization for AI legislation in US.
Honest Phase 1 retrospective since the 2024 ACX grant:
Bad: Lost two developers - 2nd one abruptly derailed project. Took off six months for paid work for cashflow reasons. Most early pilots didn't reach threshold. MVP was a marginal uplift over the prototype. I refused to juice metrics.
Good: Onboarding a new dev, brought on an Ivy League student fellow, set up at Collider NYC, preparing a narrower relaunch focused specifically on AI-safety and AI-governance coordination.
Need runway — hence the SFF application.
The ask: I'm submitting to SFF's 2026 S-Process Grant Round. Deadline is Wednesday April 22, but I'm finalizing Tuesday April 21 for a buffer. Non-binding letters of intent from orgs willing to pilot a Spartacus campaign in the next 6–12 months if we're funded would strengthen the application. I draft the LOI; you review and sign. Current LOIs in review include EANYC-affiliated folks, PauseAI US, AI legislation tracker org, and others.
If your org has a real "I'd commit if enough others did" problem and you'd consider a non-binding LOI, reach me at jordan@spartacus.app. I'll turn a draft around by midday.
Need responses by Morning of Tues April 21 to include in Tuesday's submission.
Also happy to hear from anyone who has a concrete critique of the mechanism, a pointer to a better-shaped problem than the ones I've been targeting, or warm intros to orgs I should be talking to.
— Jordan Braunstein, Spartacus.app
For the past 3ish years I've been developing a sort of 'unified theory' that synthesises major ideas in cognitive, personality, and clinical psychology. Scott has shared a couple of my posts in the past, so I thought some of you might be interested. I'm also a practicing psychologist, so I'd really love some feedback from others with knowledge of these topics - have I missed anything important? Does this have any practical significance? Etc.
https://mindandmythos.substack.com/p/the-narrative-cybernetic-synthesis
Nice write-up, I'm not a pro but it looks well-grounded and genuinely interesting. Multi-layer organization is definitely a thing; my intuition would be to tone down the hierarchical aspect of the model, given how loopy things seem to be everywhere in biology.
- I like the part about how high-level differences (big5) turn into specific 'characteristic adaptations', that seems to capture a real cross-level dynamic
- The parallels between Big5 and HiTOP looks promising, but over my head to evaluate
- The emphasis on goal-directedness can say something about the existential malaise of current culture, where the desirable goals are often unclear or contested. It explains why this is hard to bear for so many individuals, despite of the optionality that it opens. If we can trust the more speculative parts of your theory, even memory is goal-directed, yikes!
- I like the idea of the "personal myth". We need a popular name for the entire "story of I" that we build, partly implicitly, partly explicitly. This re-appears later in the section on narrative identity. There is a potential link here to things like vipassana meditation, which in its more powerful versions, is a full frontal attack on the personal myth, with significant dangers.
- The part about 4E and ritual doesn't appear so clearly integrated with the rest of the theory. Is this a sketch at collective dynamics beyond the human individual?
- Just a guess, but the Justification Hypothesis is intriguing but doesn't sound so convincing to me. If it was true, how would we even know?
This is really helpful feedback, thank you!
"Multi-layer organization is definitely a thing; my intuition would be to tone down the hierarchical aspect of the model, given how loopy things seem to be everywhere in biology"
Fair point. I do account for 'loopiness' in that there are recursive feedback loops within and between the different levels, but it's likely that these levels aren't so clearly demarcated in reality (i.e., in the brain/mind). I think it helps to keep the hierarchical model for the sake of simplicity (and a clean diagram), but this is something I will try to make clear in future writeups.
"The emphasis on goal-directedness can say something about the existential malaise of current culture, where the desirable goals are often unclear or contested. It explains why this is hard to bear for so many individuals, despite of the optionality that it opens. If we can trust the more speculative parts of your theory, even memory is goal-directed, yikes!"
Agreed! There are a lot of real-world implications of this model that I look forward to exploring in future. I think the uncertainty management/minimisation element is an important element - it's not hard to see how this would lead to problems like anxiety and depression, and I think many of the most effective therapies are doing essentially this (reducing uncertainty or providing strategies to cope/live with it). You mentioned vipassana meditation, which might also be (in part) a strategy for managing uncertainty.
"The part about 4E and ritual doesn't appear so clearly integrated with the rest of the theory. Is this a sketch at collective dynamics beyond the human individual?"
I didn't start learning about 4E cognition until relatively late in the process of developing this model, so it still feels a bit 'tacked on'. It's clearly important, and I've tried to integrate it as best I can, but this part still needs some work. But yes, it is (in part) meant to account for collective dynamics - collective rituals and such - and also for the embodied practices we do as individuals that form part of the cybernetic processes that maintain identity, memory, goal-directed action, etc.
"Just a guess, but the Justification Hypothesis is intriguing but doesn't sound so convincing to me. If it was true, how would we even know?"
Hard to say. It's an evolutionary theory of the development of culture, so like many ideas in evolutionary psych it could be quite profound and true or merely a 'just so story'. I believe it was originally proposed by Gregg Henriques, who has developed his own Unified Theory. If you're interested you can read more about it here: https://www.psychologytoday.com/au/blog/theory-of-knowledge/201112/the-justification-hypothesis
My reaction is quite similar to Viliam below: I also found it hard to say what the point was.
I do agree with a lot of the ingredients that you explain, like the personal myth, the strive for low entropy, and so on. It's a good summary, and it certainly shows that you are knowledgeable about important aspects of psychology. But I don't have the feeling that the combination gives me more than the sum of its part.
One nitpick though: you mention psychological pathologies. I am also not quite sure what your conclusion is and how you integrate them into your theory, but at least at one point you seemed to indicate that those are consequences of a failed narration process. Perhaps this may be the cause in some cases, but I don't think that this generalizes very well. Take depressed people. They are actually very good and consistent in their personal myth. The myth is "I am worthless, I will never achieved anything, no one loves me." And they are very good at ignoring and twisting facts that invalidate their myth, just as healthy people are. I don't think the myth-building procedures are broken for them, they work as for other people as well. Of course, the myth is harming them, but many myths of healthy people are also harming them, in the sense of preventing them to have a more accurate worldview that would make them better at achieving their goals. (Though the harm is more universal for depressed people.)
Thanks for reading! I admit it's not the easiest read. It's pretty dense, and although I've tried to lay out everything clearly and sequentially (referencing some of my older posts), it's still quite jargon-heavy and difficult to understand on a first read through. I'm planning to write a shorter, more accessible version of this in future.
Fair point, there are definitely coherent but harmful myths. I don't deny that they can cause psychopathology, but I've probably over-emphasised the importance of coherence in my writeup. In the case of depression I'd say that the pathology occurs at a lower level - faulty interpretations of the world and oneself, flawed strategies, unrealistic goals, negative-biased memory recall - that then compound into a coherent but negative 'myth' at the higher level.
I'll keep this in mind when I (eventually) write my more accessible revision of this. Thank you!
From your paragraph here, the first thing that pops to mind is the old xkcd about competing standards. https://xkcd.com/927/
So I guess my question is: why?
My answer: why not?
Seriously though, I think my synthesis adds value. It integrates several models from different sub-fields of psychology that have (to my knowledge) not yet been integrated, but together explain more than they do individually.
Speaking from a position of complete ignorance, this is reminding me uncomfortably of R.D. Laing and anti-psychiatry. "You're not crazy, you're just misaligned with society!"
You are, in fact, crazy.
Not anti-psychiatry at all, I think meds (and other non-therapy approaches) have their place. I don't discuss this in my post, but they would presumably enact change via the neural-cybernetic level (rather than the narrative level) of my model.
As a rule of thumb, anything that mentions psychology and quantum physics in the same text is crackpottery. I only kept reading because the connection was made by one of the sources you quote, not by you, and because that part doesn't seem load-bearing, it's just an extension of "more complicated things are made out of simpler things". But I would recommend leaving this part out.
It is difficult for me to figure out *what* is the thing you are trying to say.
In the first half of the article, it seemed to me that your theory was that people are different, as described by the Big Five traits, and that their Big Five scores could be used to *predict*, if they become mentally ill, *which* specific mental illness will they have.
But then the narrative changed towards "larger things are made out of smaller things" and then the article ended and I am not sure what was the point.
Could you summarize the main insight in one paragraph?
I hadn’t run into the word ‘crackpottery’ till just now, it cracked me up so to speak as well as Mrs Slothrop.
Thanks for reading! I don't know where you're getting the psychology-quantum physics connection from - do you know which source this is? I certainly don't make this connection myself.
There's not one single insight, exactly, but here's my summary of the theory (taken from part 2):
"The narrative-cybernetic synthesis proposes that human psychology is organised as a multi-level system in which embodied cybernetic regulatory processes give rise to, and interact recursively with, memory and narrative identity to maintain behavioural and experiential coherence over time.
At the base of this framework are cybernetic systems that regulate goal-directed behaviour via feedback. These systems shape how we encode and reconstruct memory, particularly autobiographical memory, which at the highest level is arranged into a time- and context-bound narrative called a personal myth. This personal myth combines our own memories, interpretations, and culturally-sanctioned narrative templates, and is stabilised through embodied practices like ritual. It serves as a high-level reference point that organises goal-directed action over time."
I know next to nothing about psychology, so I cannot judge your theory on its merits; all I can do is ask some generic due-diligence questions of first resort that I would ask of anyone advancing any unified theory of anything: is your theory published in any major peer-reviewed publication ? Has anyone been able to replicate your findings ?
Short answer, no. I have published legitimate research under my real name, but I no longer work in a research capacity. I'm just a clinician who likes to stay up to date on recent developments.
Fair enough, but then I'd ask: assuming I knew something about psychology and/or has the time to learn, why would I study your theory as opposed to some mainstream theory that has the weight of evidence, peer review, and replication behind it ? I do acknowledge that this standard question is perhaps less applicable to psychology than to many other sciences, but it still bears asking. That is to say, is this a theory in the scientific sense of the word, or in the colloquial sense only ?
Fair question! It's a theory, or rather a synthesis of theories, in the scientific sense. There's nothing truly new here - I haven't discovered some new neural network that explains all of psychology or anything like that - it's a synthesis of existing, well-established theories in related fields (primarily cognitive, personality, and clinical psych).
What this theory offers is a novel synthesis of several different theories/models that, to my knowledge, have never been explicitly linked before. I also offer some ideas that could form the basis for future research (e.g., that what narrative psychologists call our 'personal myth' acts as a high-level reference point for lower-level cybernetic cognitive processes)
Sounds good, and I'll look forward to other researchers validating the predictions made by your theory and developing applications based on it -- hopefully medical ones. Sadly I think we will need such applications more than ever in the future...
Thank you!
My latest AGI Friday argues that we should refer to LLMs as golems: http://agifriday.substack.com/golems
Do you think it could catch on?
The word agent already has too much “memetic momentum” if you will but if you really want it to catch on i’d suggest pivotting the term to mean a protocol in relation to some hardware stack? The “little guy ai” phenomenon is still in its primordial stages. Also not to get political but it’s either the best time or the worst time to be pushing the “golem” agenda, but that’s probably a really a good thing.
I don't think "agent" quite works. It's too broad and overloaded. Examples:
1. I had the golems generate a prototype.
2. Ask a golem if you don't get the joke.
3. I managed to solve this math problem without golem assistance.
I guess people would still understand each of those with "agent" swapped in, but I don't think they work as well and I think there exist cases where there'd be real ambiguity.
PS: I couldn't figure out what your last sentence was talking about. I'm like a fish unaware of water, apparently. But all 3 major golems, of course, read you loud and clear. Israel/anti-semitism/etc and that, due to attention economics, as Claude put it, the current political climate will be a tailwind for adoption of a Jewish-coded term. Is that what you meant?
PPS: Wait, I'm now guessing that the golems all missed half your meaning. Namely, that it's good to, like, show pride in things related to Jewish cultural heritage in that that helps fight anti-semitism. Arguably that even works whether one is pro- or anti-generative-AI since the golem of the original myth is kind of a Torment Nexus.
Sorry for the ambiguity, I meant to say that given the rise in anti-semitism people are likely to have stronger opinions and reactions, both positive and negative, which would create a discourse that gives the term some memetic “fuel”. Though i do think it’s important to feel and express pride in such things, and agree with the latter point.
Oh bitter irony that the golems nailed it and when I thought I was using my fancy human judgment to ferret out your truer meaning, they were right and I was wrong. Still, I'm happy to find this new point we've converged on, human-to-human!
> I guess people would still understand each of those with "agent" swapped in, but I don't think they work as well and I think there exist cases where there'd be real ambiguity
I think in practice most people would just say "AI" for each of those rather than "agent".
Yeah, that's super fair. I'm not sure I have an example where the meaning isn't plenty clear by just saying "AI". Of course "AI" is technically way too broad (just as "LLM" is technically too narrow) but if the meaning is always clear in context, there's not much actual problem to be solved. Other than insufficient Kabbalistic Cleverness.
My favorite, which I should maybe add to the original post, is how on point the golem myth is in terms of the AI alignment problem.
Memetic momentum is an interesting point. There do seem to be thresholds after which using nonstandard terminology is somehow...irritating? I'm not quite sure what I'm grasping for here. Back in the anti-reactionary FAQ I remember Scott critiquing the coining of concepts iirc because of the degrees of epistemic freedom it afforded. Maybe using your own names for things looks like that and thus triggers rational distrust? Maybe it's pretentious like saying a loanword with the pronunciation of the source language? I'm certainly guilty of this myself, and I'd like to understand it better, since I've seen peers and clients wall themselves off from people whom they might otherwise connect with behind nomenclature preferences.
Yeah, by memetic momentum I was trying to express that for a word to become “coined” it’s memetic momentum has to reach some sort of cultural escape velocity. The lifespan of a new word is typically short but circumstances sometimes allow it to survive in some cultural niche until it becomes mainstream. As for loan words being pronounced in the original language, the memetic velocity is low but slightly increases over the decades and becomes normal or decreases and is only said sparingly. I do think coining concepts is a bit cringe but also follows an innovator->early adopters->mainstream curve, but the early adopters phase invites many opinions and cultural signalling. Does it signal distrust? Perhaps by nature if it signaling some increased probability of dark triad traits but maybe that’s a stretch. I do agree that there’s something there that’s hard to pinpoint
Oh dang, yes, these concerns are on point for me especially. I tried to defend the term "golem" a bit more in the reply I posted here simultaneously with yours.
Another thought is that "agent" implies autonomy in a way I don't usually mean when talking about my use of Claude, ChatGPT, and Gemini. Of course the term "golem" also could imply agency, per the original myths. But I'm using it as more of a synonym for "LLM", or maybe something intermediate. Like Claude Code is a golem, as well as just Claude the chatbot.
I guess it feels like there's a terminological hole that "golem" is nicely filling.
If it's just a synonym for LLM, I'm not sure I see the point. I did read your AGIFriday in the hopes of a more persuasive argument but didn't really see one. Perhaps I am missing some context.
It's largely an aesthetic argument. It's kabbalistically correct? I guess in this discussion I've hit on the rationalization that "LLM" is a bit too specific. I gave the example of Claude Code -- an LLM plus elaborate scaffolding/harnesses. Another example: an image generator is totally a golem. Maybe we could think of "golem" as shorthand for "generative AI tool".
Well, I'm not in the mood to argue aesthetics :-) and I'm completely unqualified to judge kabbalistic correctness.
It does seem weird to me to try to force a word that already has a meaning, with centuries (millennia?) of connotative baggage, unless your aim is to transfer that baggage to your new concept. And if that's the aim I don't think the aim is a sensible or useful one.
Published a piece this weekend that started with a BBC essay on animal despotism — why some baboon troops are ruled by tyrants while muriqui monkeys take turns at the drinking spot. The short answer: it’s not nature, it’s conditions. Change the terrain of reward and the social geometry changes with it.
I paired that with the Singaporean futurist Chor Pharn’s idea that nations run on operating systems shaped by geography and compounding history — reflexes, not strategies — and then stress-tested both lenses inside an allegory. A blight moves through a forest along its busiest shortcuts. An Eagle, a Dragon, and a Bear each respond the way their territory trained them to respond. The smaller creatures start asking harder questions.
Readers of this community may catch a familiar silhouette near the end.
https://rajeshachanta.substack.com/p/the-grass-remembers-499
@Brendan — the Serina project you shared a few weeks back connects to this more than you might expect. The mechanism is structurally identical.
You're arguing from a single baboon troop observed by Robert Sapolsky. That's a pretty thin reed to stand on. Can't someone find a single other example? What if it turns out that there is something wrong with this study? In fact, it turns out that Robert Sapolsky is a liar. I don't know that he is lying about this one study, but...
Fair pushback on the baboon troop — it is a single case. But two clarifications worth making.
First, the piece doesn't rest on Sapolsky alone. The BBC essay by Chris Baraniuk draws on multiple researchers across multiple species: Élise Huchard's work on chacma baboon despotism, Karen Strier's decades of fieldwork on muriqui egalitarianism, Laura Betzig on naked mole rats. The argument that social structure responds to ecological conditions rather than expressing fixed nature is a finding across the field, not one researcher's claim. My essay uses these examples from three separate species studied by three separate teams.
Second, on Sapolsky specifically — the criticisms I'm aware of concern his philosophical arguments in Determined (his free will book), particularly his uncritical citation of unreplicated psychology studies. Those are legitimate criticisms of his philosophy. But the Forest Troop study itself (Sapolsky & Share, 2004) is based on direct field observation spanning two decades and includes a control troop. I haven't seen accusations of fabrication in his primatology work. If you have a specific source on that, I'd be genuinely interested to see it.
The broader point of the essay isn't "one baboon troop changed, therefore all power structures are contingent." It's that conditions — resource distribution, mobility, physical equality — shape which social behaviours pay off. That finding is well-supported across species.
But stepping back — animal research is one of three lenses in the piece. The second is Chor Pharn's argument that nations run on operating systems shaped by geography rather than chosen ideology. The third is the allegory itself, where a blight stress-tests all three power systems simultaneously. I'm curious whether the overall architecture works for you — particularly whether the blight's behaviour (travelling on infrastructure, not territory) rings true as a model for how shocks propagate through systems with different reflexes.
I strongly suspect it was written by AI, so I wouldn't get too worked up about it. I looked at several of their other posts, and they all looked to be at least largely, if not entirely, produced by AI.
I use AI as a drafting and editing tool — and find it very useful.
The research sources (Baraniuk's BBC essay, Chor Pharn's operating systems framework, a reader's Requisite Realism thesis), the allegorical structure (built across three versions since February 2025), and the editorial decisions are mine. If the prose is cleaner than expected, I'll take that as a compliment to the process. But I'd rather discuss the ideas than the tools.
From old SSC link post, I very recently encountered this piece https://slate.com/technology/2017/02/secondhand-smoke-isnt-as-bad-as-we-thought.html that claims that secondhand smoke is practically harmless. How does this fare after almost 10 years? Is there anything that has changed since then?
Scott, is it maybe time to write up another piece defending your apparently still-positive take on prediction markets? Given the abundant recent coverage of insiders trading on foreknowledge of military operational plans and all? The throughline I'm getting from my filter bubble seems to be that they've panned out as basically just a place for insiders to anonymously profit from the information they have a social duty to keep secret and/or can causally decide the market outcome of.
I think there is some utility. I wonder if the insider angle might be ameliorated by rule changes or rule enforcement. A retroactive freeze 3 days prior to a resolving action might help - automatic implementation might be tricky - ie. rollback to holders and odds at that date?
It would help. I doubt it would be tricky. Instead of "rollback" it would be better marketed as holding trades in a "escrowed" status or such. The escrow period could be a variable dimension per market.
~~they've panned out as basically just a place for insiders to anonymously profit from the information they have a social duty to keep secret and/or can causally decide the market outcome of~~
Ideally, if that were really true, ordinary people would pull out of these markets and there would no longer be any suckers for the insiders to profit off of. I think this is basically what led to the insider-trading rules in the stock market; the markets and the government both saw value in not having a cycle of "rubes put money in/vultures fleece rubes/rubes stop putting money in/vultures stop trying/rubes put money in."
Indeed, Patrick Boyle covered this on YouTube and specifically claimed that this ~"sharks eating their prey to extinction" ecosystem destruction has previously occurred in the online poker industry. The fact that he found an example in a gambling market specifically seems apropos.
My bubble mostly points to how they become generic gambling which sucks a lot of vulnerable people into wasteful and un-sane dynamics. At this point I'm tempted to agree.
"The throughline I'm getting from my filter bubble seems to be that they've panned out as basically just a place for insiders to anonymously profit from the information they have a social duty to keep secret and/or can causally decide the market outcome of."
I find them useful as a way to get a (noisy!) estimate of the odds of events e.g
https://www.metaculus.com/questions/39149/us-china-agi-limiting-treaty-signed-before-2030/
a) Can you suggest a better alternative for getting such estimates?
b) I'm a bit skeptical that insider trading is happening for markets like this. I'd expect information to leak anyway on time scales of the order of years (as with this case). For foreknowledge of military operational plans your concern sounds plausible to me.
> Can you suggest a better alternative for getting such estimates?
Your question implies that prediction markets could be justified if they produce the best estimates on at least some questions. But that doesn't consider all the externalities. Having ideal use cases doesn't automatically justify the existence of a tool that, on average, is quite possibly a significant net drain on society. This is no different from basically anything that is illegal - one can usually imagine a valid case that is frustratingly not allowed by a given law.
Many Thanks! I'm skeptical about the 'net drain' in the first place. In a tiny handful of cases, such as the one you cite re military operational plans, I'd support a restriction on the market.
Legally, US law is effectively pretty ambivalent about private information, giving preferred treatment to employee stock ownership plans, and employees generally have _some_ degree of private information, while also enforcing trade secret agreements.
Personally, I almost always prefer to see insider information leak, so that the general public has a way to access it - even through a noisy channel like a prediction market. I also prefer to see trade secrets abolished, with technical advances incentivized via patents or copyrights, and with secrecy agreements made unenforceable.
Isn't this the whole point? You want those with relevant information to reveal it so your predictions become more accurate. Obviously, those trying to keep secrets need to worry about their secrets leaking this way, but "I know enough to predict when the US will start bombing Iran next so I will place a bet on that in the prediction market" is the system working as intended.
In theory you are right we want insider trading to get the information.
But everything I encounter about prediction markets outside my rat bubbles are that 1) obviously insider trading is bad, and 2) the prediction markets agree with point 1.
But eventually this kind of behavior will corrode willingness for others without inside knowledge to bet on the other side (since they no longer have a fair chance to win or lose), which reduces liquidity of the market.
If someone who can control the outcome of the market bets $1B 30 seconds before taking the action that resolves the market, that does not inform society on important information in any meaningful way.
Furthermore, a lot of these things are supposed to be secret for a public-good reason. You can argue that it's good for the military to attack and destroy a specific target, you can argue that it's bad for them to attack and destroy that specific target. But there's no world in which the best possible outcome is that they attack that they attack that specific target, but the attack is foiled because the details were leaked on predictions markets, and both sides spend materiel and lives to no effect except for insider traders getting rich.
And that's not even touching on the incentives to make the world anti-inductive... no small number of pundits have pointed out that the administration has seemed to have an incoherent and ill-conceived approach to this war, but it would make some sense if the actual motive is to repeatedly act as unpredictably as possible in order to maximize the returns from insider trading on prediction markets.
Obviously they're supposed to be secret, and if we can catch the folks who placed those bets (which ought to be doable), they ought to get a nice long prison sentence. But the whole point of prediction markets is to get people who have insight into what's going to happen to make their knowledge public via placing bets in the market.
You could solve the "30 seconds before" by limiting bets to a certain time before the event resolves. This seems like something prediction markets might want to do to maximize their information-discovery abilities.
Presumably, regulation will have to catch up to this stuff, in much the same way that athletes and team staff (trainers and such) betting on games needs to be prevented somehow.
Another possibility is to ban some bets because they incentivize disclosing (say) classified or other confidential information. Like, you might want to ban bets on things like the exact time/place of a military strike, the medical diagnosis of some famous person who's rumored to have a health problem, the name of an alleged rape victim or a minor involved in some crime, etc.
> Presumably, regulation will have to catch up to this stuff, in much the same way that athletes and team staff (trainers and such) betting on games needs to be prevented somehow.
They're already illegal, it's just that prediction market fans have successfully convinced the Trump administration to not bother enforcing the laws.
Executing a last-minute insider trade has none of the advantages of revealing that information, except to the inside trader.
The problem is of course the enabling of corruption, not that it doesn't work.
The stock exchange went through a transformation to deal with these problems and i’m sure some administration will tackle it in the “near” future. But I certainly wouldn’t count on the current administration being the one to do so. Prediction markets are a cat out of the bag situation and regulators will tackle these problems but can’t erase them. Besides the insider trading, it does seem like an amazing opportunity to make money in a structured way. I can’t imagine all these firms giving their traders portfolios to trade on at the size we’re seeing unless this was the case. I agree some insider trading occurs but a lot of algorithmic opportunities to be had
What has that got to do with prediction markets? Just as easy to do with oil futures.
Another reason why the two are not the same is because prediction markets are entirely abstract and can get very specific. People are betting on "Will this other prediction market come true?". You cannot do this sort of bullshittery with stock.
Oh really, you can't? I present to you the 2x levered long VIX ETF: https://finance.yahoo.com/quote/UVIX/
Yeah, no, still not the same thing as "Will this one specific bet become x% true by y date" or "Will this person be dead by date x" or even "Will jesus christ be reborn"
No. The ability to put a prediction market on specific geopolitical events may directly alter the likelihood of that event. E.g., if I create a betting market on the likelihood of a drone strike in Kyiv, does that increase the likelihood of that drone strike occurring? It might!
This is substantially different from something like oil futures, which might serve as a proxy for geopolitical decisions, but are less directly implicated in the outcomes.
You are seriously claiming that oil prices won't drive geopolitical events?
No, I am not making that argument. I am arguing that prediction markets may influence events beyond what is already influenced by oil futures. My example was a drone strike in Ukraine. Is your point of view that a prediction market on a drone strike has zero causal effect on the outcome, or that the effect is already perfectly captured by oil futures?
Why would this influence a drone strike? You can bet both ways in the market. Whoever is in charge of the Russian drone strikes knows where the strike is going and can bet either way. He will just pick the strategically correct target and if the drone strike goes to Kiev he bets "yes" or if it goes somewhere else he bets on "no" and makes money either way.
It would be shocking if major players' positions in financial markets didn't affect their actions. But one limiting factor in purely financial arrangements is that I'm not willing to spend more than $1 to protect my $1 bet.
Prediction markets involve stuff measured in stuff other than money. You don't want the decision to bomb some country to be determined by people in the administration who have placed bets on us bombing that country, or the decision of whether to sentence some notorious criminal to death to be determined by a judge or juror with a bet on the outcome.
The other side of this is that if people use prediction markets as a truth oracle, there's a lot of incentive for people to be willing to lose money to influence the truth oracle--again, their concerns aren't just about money!
You can bet both sides in the market. So if you are a decision maker you can profit both ways. The decision whether or not to bomb will be made, and then the decision makers will place bets before the public knows the outcome, but it doesn't matter which outcome it is because they will make money either way.
Why should I really care if the execs at some company make side money by betting on outcomes they already know? Sure, it's not fair, but it doesn't actually hurt me. It doesn't even hurt the people on the other side of the bet. They guessed wrong and would have lost anyway. It just changes the person they lose the money to from a lucky idiot to a smart insider.
Our current system gives the profit to whatever quant fund is running the colocated servers that get the news and turn it into a buy/sell order the fastest. I fail to see how this is really any better. I have exactly the same chance at beating their computer as I do at beating the insider trading executives. Long term investors are completely unaffected.
The sec doesn't regulate commodities markets. That's the cftc, which is (a) generally more lax about it (it's a lot easier to get away with oil futures insider trading - in many cases, like oil producers or airlines doing it, it's outright legal), and (b) also regulates prediction markets (which have been pushing back on insider trading, at least the legal ones).
What reason is there to think they "may be"? What laws treat them that differently? Or is it just that Polymarket is banned in the US and Americans use it illegally?
I saw this funny story going around on social media last week about how some 60 million men are part of an online "rape academy," based on a CNN article: https://www.snopes.com/fact-check/cnn-online-rape-academy/
Of course, as anyone with any numeracy might intuit, that number is a little bit exaggerated, by about five orders of magnitude.
Of course, CNN was careful to only imply this figure throughout the article, adding the real number, "nearly a thousand", at the very end: https://www.cnn.com/interactive/2026/03/world/expose-rape-assault-online-vis-intl/
“some of the twitter commentariat is innumerate” is not the primary shocking takeaway I would from garner from this sequence of stories. This is a pretty horrifying investigation result and the claims in the actual CNN article look well backed up.
CNN doesn’t know, or claim to know, how many people are involved in the metaphorical “rape academy.” CNN didn’t “imply this figure” is 60 million. Nor did they claim that the actual number was “nearly a thousand.” That’s the number of people who were at one time on a group named “Zzz,” but Sandrine Josso, who created the metaphor, explicitly says it refers to “groups” (plural).
The above doesn’t stop people from seeing a number somewhere in the article and jumping to the conclusion that the number they spotted is the number of people in this academy. People have probably been doing this since the invention of the printing press, but it is only with the rise of social media that people have the opportunity to share these leaps of illogic with people outside their immediate friends and family.
Unless they just changed it, "this group of nearly a thousand users" is at the beginning of the article, in the fancy graphical part before the main body of text. It doesn't read as intentionally misleading to me at all. I don't think it's fair to blame CNN for the extremely poor (or motivated) reading comprehension of some people on social media.
I’m seeing a distressing number of people latching onto “30 million,” or in at least one case the ludicrous “92 million” and responding to a factual correction with (more or less) “THIS is why we can NEVER TRUST MEN (tm).”
I’m of the Old School who believes an acceptable number is zero, and that the only difference between the least unacceptable number and the actual verifiable number is how many instances of Cruel and Unusual punishment I’d like to see meted out in a public square.
If you believe that wanting to see a thousand or so men executed rather than millions means you believe I despise my wife, daughter, sisters, more nieces than I can remember without assistance, I’m okay with that.
If there is a gap between the politics you are driven toward by "News Article" and "News article plus takes of stupid people I don't know in person," that gap will be used to manipulate you very cheaply and efficiently.
Yes, you're right. My bad. I remembered reading it last week so I included that in my original comment, but couldn't find it when I searched just now, so I updated it. It looks like it's just the fancy formatting not being searchable when you've scrolled to the main article; I've removed the update. Thanks.
You actually found the one article where CNN was presenting the facts honestly!
But seriously, thanks for responding so nicely, it makes me want to comment on ACX more often. :)
_Truthfully,_ yes. I hold "honestly" to be a higher bar.
I have to say I found the early graphical stuff super annoying and just skipped past it. I suspect I'm not alone. If CNN is covering its ass with non-searchable text, well, CNN.
I agree that the early stuff is sooo heavy-handed and annoying. As for the wording of the piece, I think it's only a little unclear, and it doesn't seem to be written with that intention. See this quote from the article:
"One porn site, Motherless.com, is home to more than 20,000 videos of so-called “sleep” content uploaded by users, with hundreds of thousands of views."
"Hundreds of thousands of views" doesn't align with 62 million visitors, and it's clear that it's only a portion of the website's traffic. I do hope CNN issued a clarification though, I think that would be a responsible thing to do if the false claims gained enough traction.
That said, the article is still incredibly disturbing, and I can see why people reacted so strongly. I find CNN's treatment of it (with the graphics, etc.) to be almost unserious in how melodramatic it is -- the content is disturbing enough on its own in plain text.
Do any of you feel like you have too many different fields of interest? How do you deal with it?
Imagine if school let students focus more on their interests instead of having everyone learn the same things at the same speed.
Yeah, despite not being very familiar I've always been partial to Montessori, I recently realized part of the reason is probably that I'd have been so much better off in a well-functioning environment built from that pedagogy. I wanted to die as a kid, partially because going to school was so dreary and monotonous that the thought of having to do it for so many more years felt overwhelmingly daunting lol
I mostly just indulge it. Music, spirituality, science, tech, psychology, anthropology, linguistics and a few more... pure catnip.
For me, I think about what has eternal value and will still matter in 1000 years, and focus on that. When I need a break, I start looking at other things, but I end up getting bored and unable to continue with them when I remember their ultimate worthlessness.
> For me, I think about what has eternal value and will still matter in 1000 years, and focus on that.
So... what even fits that description? Because absolutely nothing comes to mind for me. All things are temporary and fleeting, but there's nothing wrong with caring about them regardless.
I feel like I'm at capacity anyway. I have dealt with it in various ways but the most successful has been to prioritize some of my "fields of interest" (really, these boil down to habits I'm trying to form) and then set monthly goals & track them using a spreadsheet. I have been doing this for the bulk of a decade.
At the end of each month I can look and see whether I fell short, met, or exceeded each goal, reflect on the reasons, and adjust the goals for the next month if necessary. I can also add new "fields of interest" (aspirational habits) if I feel I have room for them. Over the years I've gone from having like 4 items I'm tracking each month to around a dozen these days.
I never think about the spreadsheet except for during the 30 seconds it takes me each morning to update it.
Fields, no. Directions, yes.
The first thing to do is see if any of them will go away, which is the easiest solution. If not, the second thing to do is pick a bare minimum amount of effort to put into each of them, like, five minutes a week, and make sure all of them get that tiny block of effort. Once everything's got the minimum, then you pour the remaining time into whichever one is most fun, until that one's finished and the time's freed up for others. The third thing to do is see if any of them have gone away now.
I'm also finding that a cool-down journal entry helps kill interest. Write out a summary of what you've done or what you want from each of these fields of interest, and see if that's enough to make any of them go away.
Yeah, too many things I would like to do, not enough time.
I don't have a good solution.
Depends what you mean by deal with i guess? If it’s just the feeling that there isn’t enough time to gain substantial depth in certain topics and you should prioritize the “important ones”, then I would say working through and removing negative feelings is the best thing. If we’re talking about how best to maintain different fields of interests , over the years i’ve found that trying to stick to an organized structure for more than say 1-3 topics isn’t actually realistic and you should simply burn every ounce of passion and squeeze every second of motivation out of the brief time period (be it hours or days but usually not weeks) that a current interest feel “on top”
of the others.
Yeah, I suppose in practice your latter option is what I actually do, which, I think, is also how I ended up this interested in too many different things - something is relevant (at the time) to research, or it, whatever it is, catches my fancy, so I read up a little or a lot and then move on. What lead to feeling like this is a problem that needs dealing with is that I recently ended my marriage (I also don't have a job) and so now have more free mental capacity/time/energy than I've had in years, and I find myself spinning a little in wanting to plunge into all kinds of different subjects/fields as I rediscover all these things I like, it's simply a bit much. I don't really judge myself that way (I've worked very hard on myself for many years), just a little unsure of how to cope/channel/deal with this new(re)found passion for, uhm, almost everything, lol. Thank you for your response and sharing your thoughts!
Why do you feel it is simply a bit much? Can you define what's troubling you? But you now have the freedom to let your mind roam. I say go for it!
It's a bit overwhelming in and of itself, and then I suppose my concentration isn't the best right now (for a variety of reasons), plus my brain does really enjoy novelty and variety, so right now life mainly looks more or less like a mix of everyday tasks, introspective navelgazing and wild jumping between interest based activities, which feels a little too like... unfocused spinning, I guess. I've also realized I'm much more goal-oriented than I thought and that I've been setting goals with and for someone else for so long that suddenly being without is hard (to the degree that taking goals that are just by and for me seriously doesn't emotionally make sense to me anymore), which definitely contributes to this feeling of being a bit too all over the place.
And thank you (and everyone else) for giving your input, I did not expect so many responses attempting to help and I really appreciate it!
I’m fine with knowing a little bit about a lot of different science topics. It’s useful for conversations at parties and gatherings.
> {∂^2V/∂S^2}@mnx.fi
I like the rudimentary knowledge check as a filter to get the email.
Sorry, what is the purpose here ? If it's to secure your email address, wouldn't it be better to just configure your email client/proxy/whatever to whitelist anyone you wish to communicate with and automatically filter out the rest ?
Only if you know in advance who you do vs don't want to communicate with so you can whitelist them.
Can any of you explain step-by-step how this is supposed to work? If a send an emait to {∂^2V/∂S^2}@mnx.fi from my gmail address, what will happen?
I didn't set up the email address, someone else did. If you fail to transform the formula correctly, you won't have the right email address and will fail to reach the intended recipient. If you do, the recipient infers you are someone worth receiving emails from (at least before they read what it is you actually emailed).
Here I thought they were looking for a d1 market maker. Maybe I shouldn't apply...
Maybe I'm dense.
You wrote:
> If you fail to transform the formula correctly, you won't have the right email address...
I'd need to know what S was to solve for V, wouldn't I? So I don't see a way to transform the formula, per se. Or do they want the first derivative of this equation?
And would the derivative symbol and brackets be recognized by a mail server? My experience with mail servers dates back to the Bat Book days, and I don't remember that mail servers could handle those characters. But I'm an old fart, and a lot of this newfangled technology has passed me by. ;-)
> Maybe I'm dense.
No, just not the target audience for the job posting. This term has a standard name, which can be written entirely in simple ASCII alphabets with no special characters whatsoever (and in all modern implementations of email, case-insensitive), which I'm sure is what they expect you to substitute in.
If it's remailing to yet some other email address, we don't know, and if we were supposed to know they presumably would have said as much. The person sending the email to that address doesn't need to be concerned with what happens afterward.
You mean the "is able to open claude.ai and do basic copy&paste" check?
Prompt: What email address does this represent? {∂^2V/∂S^2}@mnx.fi
Reply: The expression ∂²V/∂S² is the second partial derivative of option value (V) with respect to the underlying asset price (S) — which in options trading and the Black-Scholes model is known as Gamma (Γ).
So the email address is:
<<not gonna post the cleartext address here so that web crawlers don't see it>>
If you approach the address too closely its radiation may turn you into an Incredible Hulk-like being.
> You mean the "is able to open claude.ai and do basic copy&paste" check?
Yep, you want to filter out the people who can't use AI to find information they don't know. It's like not being familiar with web search.
By "unauthorised users", do they mean hackers? Because it would be very funny if the shiny new tool they announced with fanfares as the thing that discovers all the exploits got itself exposed by an exploit.
From the original Bloomberg report (https://archive.ph/tV1NJ), it sounds more like a WarThunder leak situation. Large organization has a lot of people; some of them include young men; to impress his mates, one young man decides to show off the access he has to classified info/unreleased LLMs; security breach results. Some quotes from the original report:
"A handful of users in a private online forum gained access to Mythos...
The users are part of a private Discord channel that focuses on hunting for information about unreleased models...
The users relied on a mix of tactics to get into Mythos. These included using access the person had as a worker at a third-party contractor for Anthropic...
The group has been using Mythos regularly since then, though not for cybersecurity purposes... the group is interested in playing around with new models, not wreaking havoc with them, the person said."
That is even funnier. They go public with this product and sell it on the hype about "all these hundreds of hidden vulnerabilities that our AI was able to find! imagine if cyber-criminals can use AI, too! buy our services to prevent attacks on your wide-open, full of holes, systems!" and it turns out that they can't even handle their own opsec.
Deiseach, stuff like this is no joke. I went with the War Thunder angle because it's more relatable and lighthearted, but my first thought for an analogy was actually "Guy brings home his gun from work to show off to his mates, they pass it around to ooh and aah, one of their kids steals it and shoots up his school." You can joke about it now, but one of these days that exact scenario is going to happen: "Guy brings home his access to a military paperwork LLM for planning out military operations, to show off to his mates. They pass it around to ooh and aah. One of their kids uses it to plan out the school shooting he's been thinking of." It's all fun and games, sure, but the gun is no less real when it inevitably shoots someone.
I do think it's funny, in a blackly humorous way, because they released this fanfare of implied "our product is so good, you better sign up quick and get it before bad guys use exploits to blow the doors off your confidential data, look at all the naughty-naughty we could do if we were so inclined" - then they couldn't keep their own confidential data protected.
Hubris, in the Greek tragedy sense. Glass houses and stones. And don't try to make money off other people's misfortunes.