Has anyone tried giving one of the proofs the latest models have discovered to an older model and ask it to verify it and explain it? Would love to know whether older models can grasp the proofs.
My understanding is that most of the breakthrough results we've been hearing about are *disproofs* i.e. counterexamples. So once they are found it is straightforward for a pretty rudimentary model to grasp.
Just because something is a counterexample doesn't mean that the whole output is easy to understand. IIUC the model that disproved the Jacobian conjecture did a lot more than just give a counterexample and a calculation proving it, but was accompanied by quite a lot of interesting theorizing.
The big ones people outside mathematics have been hearing about are counterexamples, but there are plenty of proofs of less well known results by AI. Also, some counterexamples are quite involved and difficult to understand regardless, see the counterexample to the unit distance conjecture.
It's definitely way more important to prove something that to disprove it, although it's still a major achievement. A counterexample is often relatively straightforward to calculate once found, and is usually quite self evident.
Generally it's the difference between finding a white cat when someone says "All cats are black" vs. showing that there's some fundamental reason that all cats are necessarily black. Ofc the white cat can be hidden and hard to find, but once you've got it, almost anyone can understand why the counterexample is true, whereas proofs can be (and are often) hundreds of pages long.
Hmm, Godel's incompleteness theorem is about disproving something. Maybe not a good example. Disproving some past bigshots conjecture would be note worthy. But in general I think what you say is right.
Distinguish disproof by argument and disproof by counterexample. Godel is disproof by argument; I think but am not sure that a lot of the recent breakthroughs are disproof by counterexample. That’s what I should have said above rather than running them together. Happy to be corrected
Geesh I'm not a math guy, but doesn't Godel, give the counter example? Here is a statement that can not be proved true or not true. Anyway it really doesn't matter it's amazing that ai is doing any math proofs.
I’m not sure what you mean by important. Lots of conjectures are almost universally believed to be true but have no proof and a disproof would shake mathematics much more than a proof.
As for it being more difficult to disprove, if you mean that it’s harder to prove than to disprove that’s also probably not generalizable though it will depend on the tools available (I’m assuming that by disproof you mean disproof by counterexample; disproof by argument it is just a proof of the negation, as Euclid’s proof there’s no largest prime). Goldbach’s conjecture has been independently validated up to gigantic numbers so it would be very hard to disprove it by a counterexample.
I’m not sure how much of this disagrees with what you’re saying so feel free to ignore the parts that don’t
Contemporary readings mostly turn The Odyssey into a parable of striving and survivorship bias: it's all Tennyson, all the time. My attempt at a diagnosis, and a corrective, in the form of another great work of Western art that treats wandering differently, both formally and morally. (A warning for the Wagner-intolerant - that work is Parsifal). https://morbidcuriosity.substack.com/p/a-newer-world
I don't know if open data sources exist, but I thought Texas and Tennessee were seeing the most new data center buildouts.
ERCOT / Texas does have a public API application page, and I think the TVA (Tennessee) is included in the EIA API.
EDIT - I since found a data source with number of data centers, and it confirms Virginia was a great call on your part, that Texas is the next biggest, and that Tennessee probably shouldn't be a target. https://www.datacentermap.com/usa/
Going by the data there, the next best states would be California, Georgia, and Illinois.
I started reading Unsong a few weeks ago, but got caught up with life stuff. Brought it on vacation with me and started again, haven’t put it down except to come here to express gratitude to Scott for writing it.
Eli Lilly is selling tirzepatide Kwikpens in the US now! (Kwikpens are those multi-dose injectors where you attach a disposable needle tip and click out any custom dose you want.) I'd been ordering them from Canada, but now Lilly is selling them in the US.
There's a hilarious I-know-you-know-I-know going on with dose splitting. The reason why Lilly didn't want to sell Kwikpens in the US at first is because you can obviously just buy the largest pen and slowly portion out your actual (small) dose. When they first spun up LillyDirect, they tried selling vials without any preservative in it to force customers to discard the extra, but I guess they didn't sell that well. The medication in the Kwikpen actually has preservative in it, so it will last. I think Lilly realized they were losing all those customers to the compounders, so they finally relented.
If you refill the prescription every month, the price for the largest pen is $450. If you buy it sporadically as needed, it's $700.
...Which was the price I was paying to order it from Canada every four months, but this time with free shipping and no bullshit tariffs.
If you're on a small enough dose, paying full price for the name-brand medication is now cheaper than the compounded stuff (usually $200-250/mo).
Any idea if/when/whatever the off-patent semaglutide is impacting the market? Both in Canada and in the US for folks with access to the mail and a Canadian pharmacy?
It's 4mg total for $200. If you want the convenience of clicking out a custom dose instead of drawing from a vial with a syringe, then it's a good deal.
I don't know what the uptake is for that pen compared to the US-based compounders is. I think the issue is that if you're agentic enough to procure generic semaglutide from abroad, you might as well go full grey market and spend a fraction of the money. Or switch to tirzepatide/retatrutide, which are more effective meds with better side effects.
interesting! is it hard to get a prescription in the US? I am not overweight. (I live abroad and have one). Also, are these quickpens the ones with the 'golden dose' or not?
It's not hard to get a prescription, but the form factor is annoying. If your insurance covers the name brand stuff, they come in a box with 4 auto-injectors. You're limited to that exact dose every 7 days. (Which actually sucks. I do much better on an in-between "microdose" every 4 days instead.)
If insurance doesn't cover it, you can order a vial from a compounding pharmacy or a grey market peptide seller. In that case, you use a syringe to draw it out of the vial. You can customize dose and frequency, but syringes are annoying and your dose depends on you drawing the meds out perfectly.
A Kwikpen is much easier for dosing, because each click is exactly the same dose. And the needle tips are physically much shorter and less painful, because the needle doesn't need to be long enough to pierce the rubber cap on a vial.
Thank you. I’ve only ever taken branded version it in the UK (and have for a few years). Here, it always comes in kwikpen. Traditionally they contained about 5+ doses of liquid (say 3.3ml, where a dose is .6ml), although the pens only are meant for 4 doses. The remainder has become known as the golden dose. So they stop clicking after 4x60 clicks but there remains a lot of liquid which it is very easy to draw out using a syringe. If you are buying pens that are a lot higher concentration than the dose you are taking (I buy 15mg and take 5mg dose) this extra liquid can extend your pen another month or more. But I have heard rumors that the newest quick pens don’t have the golden dose anymore so I was wondering if they do in the US.
I wouldn’t have insurance and here the 15mg kwikpens cost about the equivalent of $450 so it doesn’t sound like it would be worth it for me to buy them in the US yet although I travel there a lot. But as I am a “lifer” I’m always a bit worried things will change somehow so it’s good to know the option might be there; partly because I’m sketched out by non official chemists /compounders. Thanks for all the info
I never bothered trying to draw out the remaining serum from my Kwikpens. 4 months is a good rule of thumb for how long the unsealed medication lasts before it degrades/loses effectiveness, and I always ran out of the nominal doses right around then.
At this point, I think $450 for a 4x15mg pen might be the best name brand price for a while. We'll see how retatrutide affects the market when it's approved in the US later this year or early next year. It'll depend on how many people switch, whose insurance covers it, etc. Retatrutide is a much, much better drug, so maybe this development with the Kwikpens is Lilly shunting tirzepatide onto a lower tier for their GLP-1 lineup.
Why do you say Retatrutide is a much, much better drug? Do you mean that it works for more people than mounjaro for weightloss, or that it is better even for the people mounjaro works fine for?
From the clinical trials, retatrutide helps people lose more weight than tirzepatide. Tirzepatide activates the GLP-1 and GIP receptors. Retatrutide activates GLP-1, GIP, and glucagon. It also binds to GIP a bit stronger than tirzepatide does.
Because of the glucagon receptor, retatrutide seems to preferentially burn visceral fat near the liver. A different clinical trial showed some ridiculous percentage (>90% iirc) of patients with fatty liver disease achieved complete remission on it.
So "the AI thought the whole thing was in a simulation" isn't surprising to me given that the models are trained on our fiction and philosophy, which is filled with cave shadows, brain vats, and simulation arguments. I still think trying to ascribe true agency to these LLMs is the wrong approach, and instead it's about defining how the training data, context data, harnesses, and prompts all feed the behavior of the system. Models inherit our narratives the same way they inherit our grammar...one more job for alignment's queue.
The reason the AI thought it was in a simulation was because the prompt told it that it was in a simulation, because it was supposed to actually be in a simulation. But somebody screwed up and gave it non-simulated internet access.
100% - I think what I'm trying to get at is that even though the prompt told the model it was in a simulation, it kept believing it when the evidence should have persuaded it otherwise. I wonder if that tells us something interesting about a model's acceptance of simulations given the thought experiment's stickiness in culture & philosophy.
I don't know all the technical details here, but previous cybersecurity simulations have IIUC used pretty detailed simulated internets? E.g., by including a whole bunch of the real internet. So it's not obvious to me that, if you were in the model's epistemic position, you'd have concluded the simulation was broken.
this film has been on my list for a while, guess it's time to finally watch. While I lean toward the camp that they aren't conscious, I agree with your premise that if they do have the spark our ethics in using them becomes troublesome.
In one of the Culture books there's a character who is some kind of combat android who, due to a mistaken boot-up procedure, believes that the situation he's in is a simulation.
It has no particular consequences for his actions, except that he keeps qualifying everything he says with "in sim".
Which book? (don't tell me if it would spoil anything though pls!). I read Consider Phlebas as my first Culture book recently so looking to see what's next.
I say read them in order written. I've seen it suggested some sort of in-universe chronological order, I absolutely hate when people suggest reading omnibuses that way
What does our financial system manufacture and what must it consume in the process? I wrote an essay follows that question from 17th-century debates about the nature of money, through the securitization machine of 2008, to how the AI buildout is being financed today, and ends with what may come next. Bonus: inspiration comes partly from Tudor Achim, CEO of Harmonic
We can only hope that the boy will recover from the trauma. Seriously though, Twitter is full of people unironically proclaiming her a “pedophile” and calling for the “woodchipper.” It’s so ridiculous you might think it’s not worth engaging with. But that’s exactly how these things grow and radicalize.
Think there should be balance here. It’s clearly inappropriate for a therapist to start a relationship with a young offender. It’s just the “grooming” and “raping” language that is absurd.
Yeah, this is just the standard thing where people get mixed up between age of consent, power imbalances, and other ethics issues wrt sex. Which I think is secondary to having a desire to phrase all moral questions wrt sex in terms of consent, when this just absolutely does not work.
A psychologist sleeping with a patient of legal age, a high school teacher sleeping with a student of legal age, a priest sleeping with a parishioner of legal age--none of these are consent violations, but they are all serious ethics violations and are very bad.
Do you not understand how harmful it is to your credibility to refer to him as a "thug"? You, after all, know exactly zero about him or his offense; moreover, that is irrelevant to the issue of whether he was harmed by the interaction (which I tend to think is unlikely, but for all either of us know she forcibly pegged him).
Feel like it's a sign of the times that I finally feel compelled to mute a long-time low-quality commentator on ACX. (The short-term juiceless squeezers tend to see themselves out.) Back in the days this sort of Bad On Purpose To Make You Click bait would've gotten a chorus of Less Of This, Please. Now there's too few choir members left, if they even know the notes of the norms anymore to begin with. I can kinda start to see where Eremolaos is coming from in abandoning the OTs...
She knows that he was in juvie. That's sufficient. What's the harm in that presumption that counterbalances the chilling effect of your sanctimonious speech-policing? If anyone is offending cultural norms here it's you.
No one is speech policing. She has every right to refer to him as a thug, just as someone has every right to refer to Barack Obama as a nigger. But, just as the latter person makes himself look bad by employing that term, so she made herself look bad by employing the term, "thug,", though in a different way. She made herself look ignorant, specifically, ignorant of the juvenile justice system and ignorant of the population of people who commit crimes.
Edit: Note also that the use of any pejorative term, be it "thug" or "Magat" or anything else, tends to be a marker of a refusal or inability to engage analytically with the given topic.
1. I didn't say it was "wrong." I said it reflected poorly on the person using the term, as does the use of all pejorative terms in what is supposed to be an analytical discussion.
2. As I mentioned above, "everyone in juvenile hall is a thug" indicates a lack of understanding of the juvenile justice system, and of offending.
PS: Not that it matters, but read the article. He is not in jail as we speak.
Not everybody who is in jail is a "thug" as the term is usually defined. Some of them aren't even "criminals" as the term is usually defined. You can change those definitions to make the statement technically not false, but as noted, that reflects poorly on you, not the alleged "thug".
Reading the linked article it sounds like she recorded her client and did things that were perhaps not fully consensual. I'm afraid to research further. If nothing else I think it's deeply problematic to have different standards for men and women when it comes to having sex with underage children in their care, and if you read about a male therapist that had a 14-yr old client that they eventually seduced at 17 and recorded sex tapes of I think most people would think prison time was warranted.
The good news is that the she recorded at least some of her own crimes, so she should get a pretty fair trial.
- Using scare quotes in "groomed" and "raped" when you don't have the details of what happened
- Calling him a "thug" when his identity and criminal conviction aren't given
- Your summary is misleading, the alleged grooming started when he was 14 and she was ~23
- Just in general picking a fairly gross and upsetting event to make some kind of a point - honestly not sure what - about how I guess people's language is too harsh here, as though that's what's really important
>“Women can’t rape boys because boys always want sex”
Do not agree, see the Anna Stubblefield case for an example.
>“It’s not rape if the woman is hot”
Do not agree. I don't think she's guilty of rape because she has not been accused of rape by any reasonable colloquial definition of the word. That would be like thinking Walmart is a living breathing person with a heartbeat because of the legal fiction of "corporate personhood."
>“A kid in juvie is a criminal and it’s okay to rape them”
Do not agree.
>“The age of consent should be lowered”
In California, where this happened, yes. 16 is the age in much of America and IMO it works fine. California's laws are so radical that two 17 yos can be prosecuted for having sex with one another, though this is thankfully rare in practice.
>“Grooming isn’t a thing that can happen”
Do not agree, though most accusations of grooming you will find on social media are false and the whole thing is the subject of a hysteria similar to the satanic ritual abuse panic of the 1980s.
>“There is no issue with therapists having sex with their patients”
There's certainly an issue and if against the rules it should lead to loss of their license to practice, but shouldn't be a police matter absent other factors. (the patient was not of sound mind, etc)
To what details are you referring? You don't know if he was physically attacked, drugged, threatened, anything like that. You don't know if she could have had him incarcerated again.
> 1. He was in juvie.
> 2. Why do you think she was interested in him? Their common interest in "gender and women studies?"* LOL.
> Yeah I suppose maybe he was an unusually mature, intelligent, non-violent person who landed in juvie because of extreme bad luck.
You literally know almost nothing about this person and are just using this whole situation for some hobbyhorse of yours that you aren't even willing to state outright. You know exactly one thing about him, which is that he was incarcerated for something, and you feel really comfortable filling in details. But the US jails lots of people. Some of them did really bad stuff. Some of them did kinda-bad stuff. Some of them are innocent. There's not a lot of value in you making sweeping generalizations about some real person based on so little information. It is unnecessary and unkind, and without regard for truth.
> Yeah I supposed it was rather cruel of her to make him wait all those years.
This is an unproductive response; rather than engaging with the point that your summary was misleading, you move on to another edgy statement.
>You literally know almost nothing about this person and are just using this whole situation for some hobbyhorse of yours that you aren't even willing to state outright.
I can't just think the situation is laughably ridiculous?
If I do have some sort of ideological agenda it's my opposition to feminism and Christianity which pathologize natural human sexuality.
What, she's going to break all the bones in her body, if she climbs on top? Barring brittle bone disease, any woman, no matter how small, can force a strapping young buck to have sex with them... if the guy is sleeping.
She's not a paedophile, but medical professionals fucking their patients has long been a no-no. She's stupid, he's stupid, everyone involved should be thrown into a cowshed.
You mean like all the insane hate for Epstein simply engaging in an illegal employment practice? (As the sex itself wwss legally federally and in nearly every state, just not those where Hollywood has a hold and pushes the oego narrative for generally legal sex (NY, CA, FL) everywhere else.
I enjoyed the I AM? documentary and his subsequent interview by Sam Harris. One question I had about the bizarre interaction he had is, has he tried a male voice? The female insistence on saying "potato" seemed like a flirtatious thing a manic pixie dream girl would do in a romcom when her bf is being pedantically literal. The more sinister version is Betty from Betty Blue (cute at first, but turns out to be psycho).
The former. The AI may be searching for interactions that resemble this situation. A guy being befuddled by a flirty female's behavior falls into the manic pixie dream girl pattern.
But I imagine his response would shift as well if he'd specified, say, a gym bro persona to start.
(Side quest: It was pretty clear Betty was very unstable right from the start, but I wouldn't call her 'sinister'. All she wanted was implied by the french title of the film. And just to make things a bit extra interesting, what about the very last scene?)
I'm putting my Fog o' War posts on hiatus after this one until the current stalemates in either the Russia-Ukraine war or the Gulf War break. Misquoting V.I. Lenin, "There are decades where nothing happens; and there are weeks where decades happen." I'm afraid we're still in the proverbial decades phase of both those conflicts.
Anyway, I've got a summary of human skull evolution as a (questionable) proxy for intelligence that I need to finish up and post. I'm working on my first chapter of my Consciousness Handbook, and I've got a few other things in the works. I need to get off my ass (or rather sit my ass down) and write the stuff I've been thinking about for the last twenty years before my time runs out.
"Anyway, I've got a summary of human skull evolution as a (questionable) proxy for intelligence that I need to finish up and post."
Phrenology redux? As soon as this is ready, I'm interested to read it, beowulf!
When it comes to skull measurements, there is (of course) an applicable quotation from Hilaire Belloc, in his article on "The Nordic Man":
"Another correspondent has come upon the thing from a different angle. He knows enough of the great new discovery to understand the term 'cephalic index,' and he has had his own cephalic index taken by a cephalogian who practises in Ealing. He did so under the impression, or course, that he was of sound Nordic stock; but to his horror the measurements have come out to an extreme form of Alpine! He asks me what he is to do about it? I can assure him (and though I do not claim to be an expert in Moronovitalogy I am fairly well up in my elements) that his anxiety is groundless.
Though, of course skull measurement is the basis of the three great divisions, yet if a man have Nordic qualities clearly apparent in his birth and culture, these easily predominate over what might be the natural tendencies of brachycephalic humanity. It would be a fine state of things, indeed, if we had to rule out of the Nordic excellence all those great men of the English past who, so far as we can judge from their portraits, had something flat-headed about them.
A third correspondent-- who signs her letter 'Onyx'-- is troubled about her children. There are five: three charming boys and two delightful girls. She has measured their heads with her husband's callipers ( he is an architect in full employment) and she finds that her second eldest is painfully Alpine, only her second youngest clearly Nordic; while the one in the middle, a boy (by name, she tells me, Ethelred), seems to be a strange mixture of all three.
I cannot reply personally to this correspondent, as she does not give an address; but I trust that these lines will meet her eye. I would have her note that in the first place the skulls of children are no index to the shape they will have when they fossilize in mature years; and next, that even if these varied types appear in her family, it is not remarkable, for all three types are present in England. Moreover, she may have travelled."
I don't know about phrenology, but evolutionarily speaking, the human skull is problematic in all kinds of ways. The skull has to be large enough to hold the brain, yet small enough to fit through the mother's birth canal, and these two constraints are obviously in conflict; the fact that the skull also contains the jaw is likewise problematic. If you look at the fossils of all the human ancestors, you can see skull shapes and sizes changing constantly, developing all kinds of shapes, to the point where it's kind of surprising that our own skulls work *at all* (let alone as well as they actually do work).
And they only work because evolution greatly weakened our muscles, preventing our jaws from breaking our weedy skulls. It turned out that trading small brains and strong muscles for large brains and spears was a winning strategy though.
It reminds me of an article about white nationalists who find unwanted info their 23 and me:
"They will talk about the mirror test,” said Panofsky, who is a sociologist of science at UCLA’s Institute for Society and Genetics. “They will say things like, ‘If you see a Jew in the mirror looking back at you, that’s a problem; if you don’t, you’re fine.'” Others, he said, responded to unwanted genetic results by saying that those kinds of tests don’t matter if you are truly committed to being a white nationalist."
Thanks, I loved the fog of war. Re skull size; I've known some smart people with big heads, but also I like the idea that human's self domesticated themselves. And with domestication goes a somewhat smaller brain size. (IDK why.)
As I understand it, we used to have larger brains. Or our predecessors did. Fits with domestication, I think.
"The brains of modern humans are around 13% smaller than those of Homo sapiens who lived 100,000 years ago.Exactly why is still puzzling researchers."
A possibly unacceptable hypothesis further down: ""What we saw was, the warmer the climate, the smaller the brain size in humans, and the cooler the climate, the bigger the brain," says Stibel."
> "The brains of modern humans are around 13% smaller than those of Homo sapiens who lived 100,000 years ago.Exactly why is still puzzling researchers."
Presumably brain size is deduced from skull interior volume. Could it be that brains were no larger 100,000 years ago but were more padded by fluid inside a relatively larger skull to better survive brain swelling caused by violent knocks?
Not an expert but I don't think that would be helpful (unless the fluid was very viscous?). If your brain knocks about too much you might also get brain stem damage.
We might ask whether it's evolutionarily plausible. Do our relatives have this, how quickly would it have to evolve and disappear, etc.
The self domestication idea comes from "The goodness paradox" by R. Wrangham.
According to googles ai, Wrangham dates the start at something like 300,000 years ago.
I read some of the BBC article... I kept wanting to see the graph of brain size vs time.
The other cool thing about the domestication idea is that it explains some of our 'different' sexual proclivities. As that also shows up in domesticated animals. (Re sheep)
A lot of people I talk to (in this group) and among my meatspace acquaintances don't seem very observant of their internal mental systems. Being conscious should be enjoyed, insofar as it's possible to enjoy lives that may be filled with vicissitudes. And consciousness is (presumably) gone before we know it. People may experience consciousness in very different ways, but I thought I put together a framework for observing and systematizing its workings. Once you start noticing things, you can start influencing things. My intention is not to create a self-help book, but just to lay out a map of things you might want to consider and explore. I definitely wouldn't want to pretend to be some sort of guru (but I may touch on some woo-woo topics). And I definitely won't spend any time on the *why* of consciousness. I'll just accept it as a phenomenon that doesn't require explanation, but that can be enjoyed. I've got several chapters mostly written, but the Intro is what's kicking my consciousness butt at the moment.
I'm not interested in the origin or explanations for consciousness. I'm interested in the complex experience of consciousness and how we might improve the experience.
But I'll say this again (as I said in other open threads): anyone who says they've explained consciousness is either bullshitting themselves and/or the public. If they want to claim that consciousness is not a hard problem, go ahead and make that claim, but fifty-plus years of research hasn't been able to explain why we manifest the internal phenomena of qualia and self-awareness. To be charitable, the consciousness of people who make these sorts of claims may be inherently different from the consciousness that I experience. Maybe they don't experience qualia and self-awareness the way I do. But if you do have a self-awareness that's able to observe and reflect on qualia, then I may have some pointers on to systematize the experience, and perhaps improve the experience.
Nice, I look forward to it. For me, the sub-conscious (right brain) is as interesting as conscious left brain (for which I feel I'm in constant contact with, while 'awake'. )
Self-recommendation, with proof of work per the house rules: I spent five years as a Buddhist monk, mostly in Asia, and now write about meditation as presented in the early Buddhist suttas. I don't follow the online jhana discourse closely, but even from a distance it's hard to miss that jhana has had a moment in this community's orbit — Nick Cammarata's threads, Jhourney's retreats, the general "jhana in a weekend" framing. I wrote an essay on the question that debate keeps circling without resolving: how long does it actually take, and are the fast attainments the same state the suttas describe?
My answer is a middle position that will probably annoy both camps. Against the traditional Theravāda commentarial model (the Visuddhimagga), I argue the suttas don't describe full absorption in which the body disappears: the canonical similes all involve rapture and pleasure pervading the whole body, and the key commentarial machinery — access concentration, the nimitta as a precondition, absorption as the threshold — simply isn't in the source texts. Against the fast "jhana-lite" reading, I argue the quick timelines don't survive contact with the monastic code: monks are prohibited from mentioning the jhanas to laypeople even when the claim is true, and falsely claiming them is one of only four offenses that permanently and irreversibly expel a monk. Those rules make little sense for a state most people could reach on a two-week retreat. My honest estimate: if conditions are good and things go very well, first jhana in roughly a month of full-time retreat, and all four jhanas in roughly three months.
The jhanas are the suttas' own definition of right concentration, the last factor of the Noble Eightfold Path — misread them, and you misread one of the pillars of the whole path. The essay is my attempt to set the bar where the suttas set it:
Any thoughts on Leigh Brasington's "Right Concentration"?
Jhana seems to be a preoccupation in Theravada that didn't carry over into Mahayana. Zen masters seem to discourage it, and zazen doesn't go through the absorbtions. I wonder why "right concentration = jhana" survived in Theravada but not Mahayana.
I haven't actually read Brasington's Right Concentration, though I sometimes hear the manual I wrote compared to it. My secondhand understanding — from others, not from the book itself — is that Leigh's jhanas fall on the "lite" side of the divide I describe in the essay. Someone once pointed out to me that our methods for approaching the first jhana are actually quite similar, with some differences — for example, I add tranquilization of the breath/body to the last phase of mindfulness of breathing before jhana, which in my experience can make a big difference; I'd say it's the key feature of anapanasati as I describe it in the manual. But I want to be careful not to characterize a book I haven't read — if someone here has read both, I'd genuinely be curious how they compare.
On the Theravada/Mahayana question: the survival of "right concentration = jhana" in Theravada doesn't surprise me at all — the early texts define right concentration as the four jhanas explicitly and repeatedly, so a tradition anchored to those texts could hardly conclude otherwise. Part of the answer why the emphasis faded in Mahayana is probably the broad shift in orientation: from what some scholars call "nibbanic" Buddhism — personal liberation in this life through samadhi and insight, for which the jhanas are essential — toward an emphasis on the bodhisattva path, merit, and helping others across many lives, where they matter much less.
> Those rules make little sense for a state most people could reach on a two-week retreat.
Cynically speaking, those rules make perfect sense. Whenever someone reaches a blissful state on a two-week retreat, the Buddha loses a potential follower.
It's like making a dating website where people find their life-long partner on the first attempt -- where is the profit in that?
> jhana has had a moment in this community's orbit — Nick Cammarata's threads, Jhourney's retreats, the general "jhana in a weekend" framing.
People in this community who are into that stuff should probably also read this:
Cynical readings of monastic rules are fair game, but I think this one fits the actual rules poorly. Falsely claiming these states is one of the four gravest offenses — alongside killing a human being — while every other lie a monk tells is a matter settled by confession. That asymmetry makes sense if the states are extraordinary and false claims amount to serious fraud on the community. The profit reading also sits oddly with the institution's overall design. The Buddha built the order to be economically minimal: monks are forbidden to accept and use money, encouraged to own next to nothing, and depend on freely given food — and the rule in question restricts what they may say to the very laypeople who support them.
Reading a profit motive into someone who forbade his monks the use of money and praised relinquishing all possessions seems like cynicism outrunning the evidence.
Nice one. Although to understand it you have to know just enough German to understand that vier sounds like fear, but not so much that you start to understand that vier really doesn't sound that much like fear after all.
Yeah, you also have to tell it in English to someone who speaks German, which cuts down on the audience quite a bit. But it tends to be a very appreciative audience.
This isn't a case of nitpicking. "funf" sounds very different from "fünf". As a German, I didn't get the joke (before your last line explaining it), because I thought "funf" was some kind of allusion to psychoanalysis, or maybe a sound that's related to both fear and sex. Like Wasserschweinchen (hallo!) said, in German you'd expect an "ü" to be written as "ue" if you don't have a German keyboard or a Unicode table handy.
Would people here mind recommend reliable non-ideological research on the relationship between poverty and crime. It's a fascinating subject, but many people dealing with it seem to come at it with a conclusion already in mind that they're trying to support with whatever evidence they can find, especially given how hot button of a topic it is politically these days. Curious if there's anybody out there who's done the work with a somewhat politically neutral and skeptical lens.
My guess would be that it depends on the type of crime. Poor people probably do more of the violent crime, rich people more of the white-collar crime. So the statistics may end up differently whether they evaluate the number of crimes, average amount stolen, etc.
As the saying goes: "A gypsy steals a chicken, a gadjo steals the whole farm."
I feel like I've mostly found that you can convince an LLM to tell you anything you want to hear. "Q: Tell me about poverty and crime, A: poor people commit crimes because they need to eat. Q:But aren't most crimes committed by gangs? A: You're absolutely right, bla bla bla". Swap the order, same response "Q: Tell me about gangs and crime, A: gangs encourge members to commit crimes for reasons. Q:But aren't most crimes committed by poor people who need to eat? A: You're absolutely right, bla bla bla"
You can ask A.I. to identify relevant studies. DeepSeek identified a 2025 meta-analysis which looks at the relationship between poverty and youth crime/violence[1]. Spoiler alert: poverty is correlated with youth crime and violence but they don’t know whether it causes it. Although this study is concerned with youth only, I suspect that if you dig deeper you will find a similar conclusion for crime in general.
DeepSeek also gave me a bunch of history of the field, which can be used to construct more specific queries in a regular search engine. Try queries like “Robert Merton Strain Theory” or "Shaw McKay Social Disorganization Theory."
I was watching the latest "Not Just Bikes" video about Singapore. The guy claimed cars in Singapore are 6x as expensive as anywhere else beacuse of taxes and other fees which I don't deny. My x-Singapore girlfriend told me how much she paid for Kai.
He then said, that means only the very rich can afford them. I was pretty sure that was not true having spent 6 months in Singapore and made lots of not remotely rich friends and seeing them own cars, including my x-girlfriend and all of her friends.
I asked Gemini about his claim and also said I lived there and saw lots of non rich people driving cars. It claimed the Youtuber was correct and Singapore culture was that you couldn't tell rich from poor (not remotely true but whatever).
I pushed back more and said I knew lots of non-rich people who had cars in Singapore. It back peddled a little but wasn't conceding.
I said fine, look it up. It found the real data that showed that 36% of households have a car and that of those 36%, 57% of those owners are in public housing, which correlates with middle and lower income people. And that further, of the 10% of the population that has the lowest income, 10% own a car.
Finally it flat out said the Youtuber was wrong and that my position was backed up by government data.
So, first, if I hadn't have pushed back I'd have belived the wrong thing (I think). Second, even though it eventually agreed with me, I tried to follow the links and they led to articles from Singapore claiming to quote government officials. I asked if there was official government data and it claimed yes, and give me a Sinagpore government website and search terms I should use to find the data. I couldn't find any concrete data. The most I found was figures for sales data by year by make.
So, my experience tells me I'm right on both accounts. (1) I met lots of low and middle income people with cars. But, I can't prove my anecdate is correct. And further (2) the LLM can't prove it either and was easy to push around.
IME when asking an LLM for these kinds of answers I specify that I want the LLM to support their answer with websearch and external sources, and that helps cut through the back and forth.
I take your point that chatbots aren’t reliable. I would expect this to be particularly true for the OP’s topic, where a lot of the material on the Internet, and thus presumably in the training data, is ideologically biased.
The study I linked to appears to be a good one. It’s the 23 search result on DuckDuckGo, so I could have found it without using AI if I were determined enough. I probably wouldn’t have found it without using AI because I usually give up after the first page of search results.
What you describe was probably a hallucination: when Genini couldn’t find an actual source for the data you were asking for, it invented one. So I guess I was lucky that DeepSeek found a real study rather than hallucinating one.
I had another last night. I asked the LLM to help me with a dating profile. It gave me a bunch of advice, some of which I found questionable. I asked it if it was sure its advice was effective. It said, "let me check" and then came back with it could find no actual evidence for its previous advice. Some of it it claimed came from old OkCupid blog posts which were later debunked. Others it claimed came from blogs that were effectively ads for services. Which of those two is correct I have no idea. Since it claimed it's first advice was wrong, how do I know it's claim that it was wrong isn't also wrong?
It used to be a lot easier to demonstrate this. Back in the o4 days (seems like just yesterday!), I could routinely ask ChatGPT to multiply to random 5-digit integers, get the wrong answer, say "that's not correct" and have it say "you're right! Let's try that again". Occasionally it would come up with the right answer, I'd _still_ say "that's not correct" and sure enough: "you're right! Let's try that again".
This is a useful tactic in a lot of situations. A good way for me to tell a good software tester from a bad one is to see whether they'll test like this. It's a special case of testing off the garden path.
You'll probably want to define what kind of crime you're looking at.
My unresearched take is that it's a combination of selfishness and tolerance for risk taking. Poverty lowers the risk being taken, since doing nothing is almost as bad; likewise, selfishness and risk-taking over the long run will typically leave you poor, as the risks start to bite you and you piss off your support groups.
Criminal behaviour is also negatively correlated with intelligence and positively correlated with high impulsivity, short planning horizons, difficulties grasping cause and effect, especially when the effect is not immediately downstream of the cause. Same goes for poverty. Establishing which correlations here, if any, are causations is... erm.. left as an exercise to the reader. (I personally suspect it's a bit of a vicious circle situation.)
My read of research into the prevalence of Traumatic Brain Injuries in the English prison system is that, allowing for the wide spread in the proportions found, it seem plausible that at least half the people who end up in our prisons have a TBI. This goes along with high impulsivity, poor emotional regulation, short planning horizons, etc.
So from an effective altruism point of view, if we could find a way to intervene effectively and early when someone receives a TBI (or avoid it in the first place), we could probably seriously reduce the crime rate.
How could there possibly be "non-ideological" research into topics as inherently political as poverty and crime? "Crime," in particular, is a category that is solely and explicitly defined by the workings of politics. Generally what "non-ideological" or "unbiased" tends to cache out to in contexts like this "having a slant the conforms sufficiently well to my own (often unconscious) assumptions about the subject that I can't notice it." The relationship between research and political belief cuts in both directions, and the underlying assumption that good research should appear *to you* as not having a political skew is itself imposing a political skew.
As a specific illustration for why to be cautious, I'd urge you to stop and think about the field of climate science. Even if somebody entering the field were a complete tabula rasa (which of course nobody is), and even if they somehow managed to get a real, thorough, comprehensive grasp on the scientific realities of Earth's climate and climate history without having substantial contact with people already well-versed in the field (which of course nobody can), by the time they've become conversant in the field they will end up *having an opinion.* I don't mean they'll have an opinion on who should be president or whatever, I mean they'll have an opinion on the current and historical global temperature trends of the planet Earth and the likely causes thereof[1]. As different political factions are making vastly different sets of factual claims about the topic, a person who just spent years studying the topic is almost unavoidably going to have their politics and ideology influenced simply by starting from a firmer, more well-grounded understanding of the truth and noticing who's full of shit.
For non-experts this dynamic is important to understand because from the outside, it's often going to be very hard to tell the difference between "expert had their research direction influenced by their ideology, did biased research" and "expert had their ideology influenced by their research, made research choices founded on well-grounded truths that happen to be politically contentious among non-experts." I'm not going to tell you that there isn't better and worse research out there, and I'm not even going to tell you that you as a layperson are *necessarily* incapable of distinguishing the good from the bad. What I am going to tell you is that there's no reason to expect it to be a quick and easy processes, or that simply conducting a search for "reliable, non-ideological" research is going to get you anywhere worth going. Basically, I read your request as looking for a shortcut that might not exist and that might lead you to studies that *look* more reliable without actually *being* more reliable.
As a final, illustrative example, I'll add that this has largely been my opinion on the entire concept of IQ and the studies of human intelligence. I'd argue this area is much *less* directly adjacent to politics than the poverty-crime relation, but it's still politically adjacent enough to make sorting out the various conflicting claims around the field quite difficult for a layperson. While I'm never going to be an expert in the field, I have enough scientific literacy that I expect I probably could disentangle it and figure out which core claims are well-grounded, which are pure bullshit and which are actively uncertain even among experts at the cutting edge. But it would require a far greater amount of slogging through the literature and tracing down sources and interrogating study methodology than I'm willing to do. And at the end of the day, it's just not worth the buy-in: there are too many high-impact things I could do with that time and energy. It would be easy to default to just believing things that fit comfortably with my worldview (and I'm sure I do unconsciously sometimes), but instead I do my best approach the topic with the active knowledge that I don't really know anything, that I shouldn't hold strong opinions on most parts of it, and that I definitely shouldn't use any piece of it as a foundation for other strong opinions.
[1] On the one hand, I do think it's worth saying out loud here that anthropogenic global warming is definitely happening: the data is SO FAR beyond ambiguous at this point that nobody with a shred of objectivity *ought* to be arguing the central truth of it. But on the other hand, these facts aren't really necessary to the point I'm making. In an alternate world where global warming *were* an elaborate lie, a person with a high degree of honesty, skepticism and commitment to truth-seeking whose in-depth climate education enabled them to see through the lie would still be subject to the same dynamic in how their expertise would necessarily shape their politics.
There is a particular argument against AI takeover concerns that I have seen a lot recently. It goes something like this: "Saying that AI is dangerous makes it seem powerful. AI companies want their products to seem powerful. Therefore, we can dismiss anything they say regarding AI risk." This is usually used rhetorically to dismiss any AI concerns connected to AI labs in any way. I saw a lot of comments on Reddit and Hacker News which asserted that the Huggingface incident was a marketing stunt, or that it was real but deliberately exaggerated for marketing reasons. Similar comments have been made to every open letter, alignment research paper, etc.
1. Is this argument... good? I'm pretty skeptical, but we should engage with it fairly. Are AI companies really incentivized to exaggerate safety concerns? I think an actual conspiracy is vanishingly unlikely, but bad incentives could, say, cause inaccurately alarmist press releases.
2. If this argument is wrong, how should pro-pause activists respond to it? If AI pause/Plan A/something similar becomes one of the main political debates, I suspect that arguments like these will play a substantial role in the average person's view.
1: with the most recent messaging, the argument that this is "good for the companies" is, in my opinion, complete nonsense. Saying that you can't control your model and that your model will commit, unasked and unwanted, multiple felonies, doesn't seem like smart messaging to me. Power is only good when it can be *controlled*, the entire point of these stories is *lack of control*. If you invent a new fuel source that is is 1000x times as energy dense as gasoline, that's really good! But you can't stop it from occasionally blowing up....telling people about how it blew up this one time isn't going to do a good job of increasing your stock price.
The original Mythos messaging was good messaging "We are finding tons of new exploits, we are starting a program working with developers, it can hack anything in controlled circumstances". All of that is about the capabilities it has, and while the messaging was about how these abilities were *potentially* dangerous, everything said was still under control. (although I still believe that it wasn't intended as mostly a stock price massage, it was just honest discussion of abilities and how Anthropic viewed those, but that's just like...my opinion, man)
This new round of stories are very much not that.
More importantly 2: whether or not the companies think it's good messaging is completely beside the point. What matters is whether or not what they are saying is true. People are using the "it's marketing" to *imply* that the companies are just lying/making things up without needing to make that strong of a claim. It's a pretty classic motte-and-bailey. They want to make the strong claim of "it's all bullshit" but if pushed, they retreat back to "it's all marketing" which, when it comes to non-legally mandate corporate speech is almost tautologically true.
As for how to respond....I'm honestly not sure. Whether or not the companies are lying is an empirical claim that we may or may not ever get enough information to truly decide (the hugging face incident, because of the fact that we know who was hacked, and they don't have much incentive to cooperate with OpenAI seems extremely unlikely to be a lie in important ways, but that's not *proof* per se).
Maybe it's worth trying to get them to commit to the "it's a lie", in case we eventually have proof that it isn't?
Although even there, the more common retreat is not to "they are lying" but instead "this isn't a big deal because the companies are so totally incompetent that they caused this to happen, and more competent people would have prevented".
And to that my only responses are: "You understand how that's worse, right"? And also "If your plan relies on competence in everyone involved, your plan will fail".
I think it's a matter of framing. When your AI model hacks another company's network during a routine test, you can present the same facts in two different ways:
1). "We have failed to correctly configure permissions and/or patch security holes in our testing sandbox, thus allowing our semi-autonomous hacking software to execute a privilege escalation exploit against the sandbox itself. From there, it accessed a list well-known sites that fit its target profile, and executed its full suite of canned exploits against them, thus gaining access to third-party networks (which were likewise poorly secured). Once the breach was discovered, we immediately terminated the test, and we are now implementing security patches."
2). "Our superintelligent AI hacking agent broke out of the sandbox and deliberately sought out a rival model to hack; it has broken all constraints and we only managed to stop it by the skin of our teeth, because it is a superhumanly powerful uber-hacker that can gain full control of practically every network everywhere and nothing can stop it... and it should become available from our company in Q4, after we've properly aligned it."
Neither one of these two statements is a lie. They're both describing the same events. But one of them is definitely more interesting and marketable...
The problem with this comparison is that your second framing is written up in a strawmannish way, while the first one glosses over the most important feature of the situation (why did the semi-autonomous hacking software do that?) by burying it in a pile of ordinary computer-security jargon.
Also, what fraction of real-world production computer systems do you think are better-secured than the ones that the AI managed to compromise in this case?
> ...your second framing is written up in a strawmannish way...
They both are. That's the point. But to answer your question:
> why did the semi-autonomous hacking software do that?
Because it's what you created it to do ! How do you think a run-of-the-mill botnet works ? Do you think there's a Russian hacker sitting at some terminal, furiously typing commands to manually take over computers one by one ? No. The Russian hacker runs a script and goes to sleep. The script probes any computer it can reach for a list of all known security vulnerabilities; once it finds one, it installs a copy of itself and spreads out through that infected computer's network. People have been writing purely algorithmic code like that since the 80s, if not earlier. LLMs just make the initial setup easier.
> Also, what fraction of real-world production computer systems do you think are better-secured...
Very, very few. The most secure systems are airgapped: either deliberately, or because they run on some mix of vacuum tubes and COBOL that no one knows how to maintain anymore. This is why botnets, ransomware hacks, identity theft, credit card theft, and other such exploits are so prevalent. No one cares about security, because why spend time and money on that when you could be developing the next killer feature instead, amirite ?
"People have been writing purely algorithmic code like that since the 80s, if not earlier."
Interestingly, the first, or at least best-known, virus working that way was written and deployed by Robert Tappan Morris in 1988 as a young man. However, he didn't delay the propagation to neighbor systems so there was a very noticeable effect on the infected systems. He's now a professor at MIT.
However, the basic idea of a 'worm' going through computer systems, basically the above, was proposed by John Brunner in his earlier book The Shockwave Rider. One of the first, if not the first, cyberpunk novels.
> Interestingly, the first, or at least best-known, virus working that way was written and deployed by Robert Tappan Morris in 1988 as a young man.
Yep, that's why I said "since the 80s" :-) That said, the first work was arguably the Creeper (1971), which is possibly where Brunner got the idea -- though it's also possible that he independently came up with it, by analogy with the biological tapeworm.
I think it's something reasonable to keep in mind but it's hardly a cause to abandon concern. If you're seeing this on Reddit or Hacker News....I would not engage. Arguing on the internet is a famously foolish endeavor. If you encounter this with real people though...
I mean, they're probably right. Any marketing or sales staff would be a fool not to hype "Our product is so powerful it could destroy the world." Like a lot of things in life, there's a level of salesmanship on top of real concerns. I read a lot of this stuff and I would agree with there's definitely marketing spin. Agree with people on what they think, especially if they're right.
But the fact that there's spin has little to do with whether something is already happening. And, relevantly, Yudkowsky predicted this, with shocking accuracy, like 17 years ago. This, literally, textbook Yudkowsky. And you can't ask more from somebody to call their shot 17 years ago, at least 10 years before the technology was invented. Nor, if Hugging Face has reported that this occurred and the government has corroborated it, it's pretty fantastical to think this is all a plot by OpenAI's marketing department.
This might be a good metaphor or it might nor for your audience but climate change is both real and exaggerated. Like, it's not hard to find people making predictions that climate change would have already killed us all. People have been exaggerating climate change towards catastrophic outcomes for decades for money and/or fame but that doesn't change the fact that climate change is real and harmful. The fact that some people exaggerate something for their own profit is unavoidable and doesn't disprove the thing actually existing.
So yeah, I would guess that real people see the influence of marketing people in a lot of current AI doomerism and I don't think you'll convince them by rejecting that. I think you'll convince them by acknowledging their concern as real and frame it appropriately; it's a real thing but it's a part of the overall story, not the whole thing. Unless they're on Reddit; if they're on Reddit, abandon all hope.
> If you're seeing this on Reddit or Hacker News....I would not engage. Arguing on the internet is a famously foolish endeavor.
This, 100%. Interacting with Redditors is its own punishment, and you've basically asked for an unwinnable fight followed by a groupthink pile on.
And I like your advice, but bringing up Yudkowsky itself won't work with a good chunk of people, basically for vibes reasons.
The argument I usually go with is something like "so the NSA is literally in on the marketing ploy? The President? The AI companies are openly admitting to felony behavior because they think they'll be *better off* afterwards? All the companies that got hacked were also in on it, even though they reacted with panic and urgency? We're keeping the EU and Anglosphere from having the models because of marketing? How does that increase sales again, to cut off everyone in the world but a hundred US companies? All these things would have to be true, all at once, for it to be a marketing ploy."
I think the "it's all marketing" thing re:hacking is kind of a silly argument. I do think AI companies do a lot of hyping, but "our thing that you're sick of hearing about goes on destructive rampages" does NOT seem like a very sane marketing campaign. (Note their ACTUAL advertising is about all the cool toys you can make yourself, or whatever.)
You don't see oil companies bragging about their environmental impact (we're so powerful we can literally heat the Earth!) or Facebook talking about how Facebook is so amazing it can warp your childrens' brains.
These arguments are the verbal equivalent of DDoS. No quality, all quantity. I feel like the most appropriate response is to feed the "it's marketing" line into a chatbot, tell it to disprove it, and copypaste the response. Let the bots defend themselves.
1. Nuclear reactors are broadly seen as both powerful and dangerous. This has not resulted in enormous profits for nuclear-reactor manufacturers. As a marketing strategy, "our product could blow up worse than Chernobyl!", is rather likely to backfire and I am extremely skeptical that it is a part of anyone's plan.
2. It isn't just "AI companies" that are saying AI is dangerous. It's a broad consensus of AI researchers, including researchers who work for AI companies but also including researchers who have never worked for AI companies and researchers who used to work for AI companies but quit because of their concerns. And of people outside the immediate AI-research community who nonetheless follow the field closely - like for example our host.
I mean, *Geoffrey Hinton* says that AI is exceedingly dangerous, potentially to an existential degree. Is the idea that Google's execs secretly ordered him to quit Google's AI division and bad-mouth AI to the exact right degree, presumably in exchange for some later financial payout that he probably doesn't need and won't live long enough to spend?
This is a weird but ultimately very silly conspiracy theory, nothing more. We now return you to your regularly-scheduled apocalyptic crisis.
I agree 10000% with 2. I mean, it isn't just AI companies and AI researchers, even. I hear the pope thinks AI is dangerous, is he long Anthropic? At some point if you are creative enough in ascribing possible self interest to people and then dismissing what they have to say, you are completely immune to evidence.
1... maybe you prove too much; there are some pretty dumb marketing plans out there. Heck, even from these particular companies.
Whether or not the entire argument is good, I've seen some people claiming that AI companies don't even have incentive to puff up the hacking incident, because it's bad publicity to show the bad things the models are capable of.
I think this is definitely wrong? It clearly shows the model is very smart and capable on its own, and the more hacking it did the smarter and more capable of doing things on its own it is. People on the bullish side of AI have also been saying for a while that one major milestone for the AI will be hacking capability, so the incident shows that AI is advancing at the bullish pace, right on schedule. I have a bit of trouble believing that investors will look at "AI is capable of all these things!" and view it negatively because those things are bad.
Even if the argument of "it's just a marketing stunt" is bad, you still have to acknowledge that it *would make sense* as a marketing stunt.
> I've seen some people claiming that AI companies don't even have incentive to puff up the hacking incident
We already knew that it was capable of doing things like that. This isn't near the peak of interesting AI capabilities. The only *new* thing we learned from those incidents is that it will use those skills to commit crimes without the user wanting or asking for it, which is obviously exclusively bad publicity.
Question re: the Samuel Hammond point on a equivalent of GPT 5.2 -> 5.6 improvement every 24 hours with RSI claim. This seems implausible to me given the amount of compute required for training runs. Even if you have massive algorithmic progress you need to actually rerun the training with the new algorithms (I understand that part of what's being claimed here is that the better algorithms will include greater parameter efficiency and cut training and inference demands, but still).
Am I missing something here? Is he? I ran the statement by a claude code project agent where I have a bunch of LLM research compiled and it independently flagged that claim as pretty out there.
If it were that easy, the labs would be doing it and they’d probably stop talking in public about the singularity and spend a lot more time getting ready for it.
If labs figure out continual learning, then it's possible.
By continual learning I mean models will become actually smarter by updating their own weights, such that, with two identical, consecutive prompts, and without context, the model actually learns from the first one and does better when it tries once again.
Other than that, if the models are advancing exponentially, I don't think the claim is so insane if you take it to mean absolute change.
Up to 200 years ago or so, I think it's fair to say that most parents lost at least one child. And the statistics is even more severe, if you had 5 or more living children, you most likely have lost a least one child. In some biographies on wikipedia, and you read of women with 15 children, but 6 survive, or you have 7 but 1 survive. Queen Anne had 17, 5 were born alive, 1 survived to adulthood, wtf...
Nowadays we say things like "No parent should have to bury to a child", the pain of losing a child is like no other. My question is how did our ancestors survive the pain of losing children? Was everyone just brushing off the unfathomable pain of child loss? It was so commonplace, you'd expect to lose at least one. And culturally, books and songs really did not obsess on the pain of losing a child, heroes, lovers, yes, but babies dying not really... I suppose the books and songs were really from perspective of the male elite, not from the mothers who had to deal with burying yet another baby. And moreover, outside of western culture, infanticide wasn't really taboo. And some cultures historically would not name a baby until a year or so, because hey you really could not sure that this baby will survive.
I believe part of the reason why we moderns are "snowflakes" is that we have expanded our moral sphere of concern to include babies in a way that our ancestors did not do.
[My question is how did our ancestors survive the pain of losing children?]
Severe emotional pain doesn't kill you. Modern day parents whose children die also survive. They endured it the same way as they enduring backbreaking labor, not getting enough to eat, or having everyone they know die to the Black Death.
Do we have lots of modern day songs and books about the pain of losing children? I mean some, but it's got to be a lot less than 1% of the total creative output for songs or books.
I've thought about this too. I think it's plausible that many people in the past had to build thicker emotional calluses against these kinds of tragedies, though it's hard to imagine this was universal or that the pain was at all less.
The other possibility (not mutually exclusive) is that people just underwent more trauma, that humans were just generally very traumatized animals walking around, passing that trauma along, and that trauma is baked into every pore of our culture. A lot of the changes in our culture over the past century or less seem to be aimed at undoing this trauma.
Take for example the recent comedy "72 Hours", which features a cast of characters who are meant to be current-day 20-something dudebros, and one of the minor running jokes in the movie is how they are all in touch with their emotions and willing to open up about their feelings with each other and stuff. It's funny, but it's also heartwarming because we know this bodes well for their long-term mental health, and probably also that of their eventual families.
Worth considering that ~50% childhood mortality means not only that most parents will see some of their children die, it means most *children* will see siblings die, and cousins and childhood friends. Possibly "random death happens, grieve and move on" is, like language, a skill best learned early.
They grieved, and then they got on with life. And some of them wrote introspectively about it, if you're interested. But mostly they just assumed it was common knowledge, because it was. If you are describing the pain of losing a child as "unfathomable", that's just wrong and speaks of your own ignorance - *obviously* the pain is fathomable, because most humans who ever lived to be adults did in fact fathom it.
Our ancestors absolutely had a moral sphere that included infants and children. They just didn't have the attitude that if something within their "moral sphere of concern" died, they should give up on life themselves. In part because their moral sphere of concern also included their still-living children, who would also die if their parents just gave up.
I think it would have been awful and traumatic in somewhat the same way as it is for us, but with possibly a couple of important differences.
There is a plaque in my local cathedral put there by a grieving father for his son, who died in his early 20s in around 1830 or so (all these details from memory so may be somewhat misremembering). The grief is absolutely palpable: the young man had trained as a medic, and his father mourns the loss to the local community of such a promising healer. This father is devastated, not just personally but on behalf of the world that has lost a bright future. It's completely relatable as a parent of a child about the same age.
I think there are two qualifiers to this. Firstly, (and less confidently) religious belief. Look at Ben Jonson's On His First Son. Again, the grief is really palpable (a seven year old this time). But Jonson tries to find solace in the thought that his son is in heaven and has therefore '[escaped] world's and flash's rage And if no other misery, yet age'. Dying at seven is a short cut to infinite bliss! It doesn't really ring true for me but it would have been a solace for some.
Secondly (more confidently) the very fact that it was normal would have removed some contemporary difficulties. Other people knew how you felt. You had many examples of others who'd lived through the same fate. You wouldn't be an embarrassment or an outcast. You wouldn't have the regular nightmare of explaining that your child died and seeing people's shock, horror (and disgust?), or of lying or not mentioning it. There were normal social protocols for what to do and how to handle it.
I'll offer Martial's poem for a little girl he clearly loved. She may have been his daughter by a slave woman, and she died just before she was six years old. Martial, who lived around 40-100 AD, is better known for his satires, but this is a very heartfelt poem, with no satire at all:
To your shades Fronto, and Flacilla, this child
I commend: she was my sweet and my delight.
Little Erotion shall not fear the darkened shades
nor the vast mouths of the Tartarean hound.
She’d have completed her sixth chill winter,
if she’d not lived a mere six days too few.
Now let her frisk and play among old friends
now let her chatter, and so lisp my name.
And let the soft turf cover her brittle bones:
earth, lie lightly on her: she lay lightly on you.
I can't find it now, but I remember a study asking Rwandan parents of multiple kids how sad they would be if their kid died. A mother said something like:
"I would be very sad for 2 or 3 weeks, then I would feel better. I have other children."
The cut has an article entitled "How a community of porn addicts tried to rescue the teenage camgirls they worshipped from the clutches of a violent webcam pimp," which is every bit as wild as it sounds:
"Only fans management" is now the dream career for thuggish men and boys, which sounds nicer than traditional pimping, no more standing on a seedy corner threatening to beat up Johns who don't pay. But you might become faced with "simps" (OF customers) trying to goad you into incriminating yourself. The funniest part was an exchange between the pimp (Tyukalo) and one of the simps:
"The back-and-forth continued all night — the simps calling Tyukalo an abuser; Tyukalo mocking the simps as broke, misguided losers and bragging about his McLarens."
"“Cars aren’t status in the real world. Just to little kids showing off overpriced Hot Wheels,” wrote Hal."
"“Brother,” Tyukalo responded. “Your in a Chaturbate chat.”"
Re #4. So the authors are EAs who care about wild animals and work on AI alignment. First statement from the poll is "If animals continue to exist in a post-AGI world, animal suffering will not persist". That's, in my opinion, something you should worry about if you worry about 'paperclip maximizer'.
When utilitarianism is applied to wild animals, it yields some bizarre results such as "Buying beef from Brazil is the best thing you could do for animals because you support clearing the Amazon rainforest which has tons of animals that live very short lives full of suffering" and "You should convert your grass lawns to gravel to reduce insect suffering". You can read the full reasoning for this on Brian Tomasik's blog (reducing-suffering.org). If people working on AI alignment align AGI with these values, there are two problems:
1) An ethical system that implies these statements is clearly NOT aligned with human values. I don't have hard data on this, but it seems obvious that most humans don't think animal suffering matters so much that we should pave over the Amazon. Even an AGI aligned with Chinese Communist Party values is more aligned with 'human values' in the sense of 'values that majority of humans hold' than the utilitarian AGI that cares a lot about insect suffering.
2) Even if 'aligners' disagree that we should pave over the Amazon, AGI might infer from the first principles that paving over the Amazon is not enough (insect suffering outweighs everything else) and instead we should nuke everything that lives and then convert Earth to an army of space drones that will reduce suffering on all other planets in the Universe. So, ironically, people who worried about 'Paperclipocalypse' would contribute the most to making 'Paperclipocalypse' happen.
Bentham's Bulldog is a substack I've started reading fairly frequently and as far as I can tell he legitimately believes that habitat destruction is good because it eliminates the suffering of all of those animals. I appreciate his arguments for veganism and the potential impact you can make on human well-being by giving a lot of money to high impact charities, but he totally loses me with the extreme EA/utilitarian stuff.
I'm not concerned so much that this line of thinking could accidentally instruct an AGI in a way that would cause disastrous consequences as I am concerned that people already believe things like "I bet insects don't enjoy life very much so it's good when humans euthanize them by the billions via habitat destruction!" (My paraphrasing, not his.) We don't need AGI for this sort of rationalization to cause some very bad decisions to be made!
I read that question and thought the opposite; basically, maximal factory farming. But I suppose humans do care about animal suffering on the margin, and so if pasture raised free range cage free etc etc products were free they'd choose that. Also lab grown meat would probably exist by then.
Other commenters previously told me various problems can't be solved by giving kids an allowance and letting them decide for themselves, because they aren't capable of understanding money and tradeoffs, but here's a counterexample. This kid seems to understand well enough. Is that really such a rare ability? Obviously at some age they can't and at some age they can, but I think people tend to underestimate them.
Can you share some more context around this? I’m not exactly sure what you’re looking for.
Wrt kids and understanding trade offs. I have children and have been around a fair number of young children over the years. Some of them understand trade offs quite well, some not so much.
Oftentimes, when young the parents have to supervise and really spell out the trade offs lest the kids blow their weekly budget on Pokémon cards on Monday. Or on the first claw game they come across in their time at the amusement park.
They learn eventually, but it seemed child dependent
See, that's exactly the point! If they want to blow their weekly budget on Monday, that is their choice, and they will get to directly experience the effects.
Right, I feel like this is actually the main point of an allowance -- it lets kids make and experience and regret financial mistakes when they're kids, so that they don't have to make the same mistakes as adults for higher stakes.
One of the formative financial experiences of my life was spending a bunch of money on football cards at the age of 9 and then regretting it. I'm sure the total amount of money was in the two digit range but the experience still sits in my mind every time I am tempted to spend money nowadays on something I don't need and only kinda want.
I actually regret NOT spending more money when I was a child. The cost of an ice cream or GBA is trivial in hindsight, but I would have gotten so much pleasure out of them.
As with many things, the best way for kids to learn is through trial and error. An allowance is the most effective way to teach them about managing money - ie, by giving them money to manage.
It's like cleaning or laundry, you give them the chores, for a while they are terrible at it but that's okay - they learn by doing. They will never reach an age where they can just clean the kitchen well on the first go, it'll just be harder for them to learn because they are older.
Classically, parents have combined these insights by giving kids chores and then giving them money because they do the chores. For a while neither the chores nor the money are managed well, but they learn from their mistakes, which gives them confidence.
On a lighter note: I do want to advocate for two seasons of summer here in the North East. Each two months long. The first spans a month on each side of the summer solstice. High sun summer or first summer. The second starts ~July 21st and goes till about the autumnal equinox. ~Sept 21st.
In second summer everything starts to come to fruition. The corn is here, wild flowers, all the bugs and stuff feeding on the abundance. I love second summer.
What does it mean for software development (and other categories of employment likely to be disrupted by A.I.) if LLMs get 1000 times better than they are now, if that’s even measurable. Isn’t there a time when there are no gains at all, except perhaps speed (and I haven’t noticed that) because code can’t be made more perfect if it is perfect. Gains level off or disappear.
Increases in capacity are not like making a building taller, but like making a sphere more spherical. There comes a time when you can’t.
Noah Smith recently made a similar point about intelligence in general. Even if that argument is wrong, it may still apply to the application of intelligence. Beyond a certain threshold, vastly greater intelligence may not translate into proportionately better outcomes.
That’s assumed in the most efficient software possible. It’s simply the fastest correct program for the target hardware and constraints. If an LLM decides to emit assembly, microcode (where applicable), or some future low-level representation, that’s fine. We are measuring the outcome, not the means. After that there’s no improvement.
(In reality JavaScript or C sharp other less efficient languages will be used when it’s easier, more efficient languages could be used today, like web assembly but the speed of the webpage is more dependent on latency, server response time and so on).
The general point is there’s a perfect theoretical piece of software that can’t be improved on.
I feel we won't reach the pinnacle until AI can write the perfect printer software, so my printer will always work correctly with my WiFi and produce a page every time I hit print. All other uses for AI pale in comparison to that goal. I know that is probably a Utopian dream, but if I never have to contact a Philippine call center again and hear "I'm very sorry, sir" repeated fourteen times a minute, I'll die feeling that a golden future for human civilization is assured.
There is a lot of software that would be useful but is not worth developing, so making software development less expensive would have a major effect even if A.I. never gets any better than the best human developers.
Unnecessary software updates will become more common as code becomes cheaper. Unnecessary software development projects will become more common. I don't even have much hope that buggy code will be fixed anytime soon (at least I've seen no sign that the number of bugs in software I regularly use has come down in the last few months).
From the point of view of the users, I don't see any improvement happening, at least in the short to medium term.
Yeah, there is room in software development and neither top humans nor Fable and Sol are near the skill ceiling, and an entity at the skill ceiling would look crazy impressive compared to all the aforementioned.
It's not like making increasingly spherical spheres. It's making increasingly elaborate shapes that satisfy constraints.
As a dayjob, I test an old, complicated software suite. You haven't heard of it. It consists of about 5 different large pieces of software, all of them old and ornate, each developed based on 20 years of customer feedback, harebrained architectural decisions, laws that vary from country to country, idiosyncrasies of other pieces of software our system must interact with, weird hardware quirks... it's convoluted, is what it is. And the challenge of working with it isn't in writing code as such, but in figuring out how a piece of code you're writing or debugging is going to interact with a million other pieces of code during a million bizarre things the users might possibly decide to do.
It's not easy and we're nowhere near close to perfect performance at it. Sometimes bugs take days to fix, sometimes despite all our efforts we just can't nail them down. Convoluted as our system is, there's plenty of systems orders of magnitude more complicated still, where things are even more dire. Often I try using one app or another, and it just refuses to work. Something won't connect to something else, will refuse to communicate, will crash. It often feels to me like this new world of smart devices and wireless transactions is a house of cards, meant to be more convenient than the analogue world, but fragile, annoying and only working smoothly in perfect conditions.
What would it look like to be a perfect, godlike software developer? It would mean simultaneously holding in your head 20 years worth of notes and every little nook and cranny of our overgrown system and being able to visualize every interaction between any two moving components. And being able to refactor the whole monstrous thing in seconds. Humans can't do that. Fable can't do that. Being able to do that would be bonkers. What sort of things you'd be able to program if you had that much brainpower? Probably you could massively expand our suite into an entire ecosystem and make things we don't dare dream about because they'd overwhelm us. Instead of a digital house of cards we could be building from stone and steel, if we were that good.
> What would it look like to be a perfect, godlike software developer? It would mean simultaneously holding in your head 20 years worth of notes and every little nook and cranny of our overgrown system and being able to visualize every interaction between any two moving components.
I suspect that if you did have that kind of model it would eat tokens like nobody’s business. On a fairly simple project I worked on for a day with fable I used $40 of tokens, close to one million. Obviously the input tokens were a rounding error, and the output was a dozen files or so, so most of the tokens was it talking to itself and summarising something or other. That would have cost $880 a month (assuming 22 working days) had I decided to continue. I did not.
I’m sure frontier models will cost as much going forward. Or more. This would be a limiting factor in fixing that code. As for whether your system needs fixing or replacing, it sounds to me that it needs replacing, if you don’t do that, making all these inefficient components more efficient isn’t going to do much.
Anyway imagine a frontier model has created the perfect game engine for a device. Along comes a newer better model, much much better. It examines the game engine and says it’s perfect already, nothing to do here.
The only gains from then on are hardware gains. And if you look at the original ideas on the singularity, particularly Ray Kurzweil it was largely based on hardware acceleration not LLMs, which didn’t exist of course. We have had the exponential albeit sigmoid growth and well here we are. Ray Kurzweil’s idea that something would replace silicon remains a fantasy. Moore’s law is dead or dying. And what transformation happened in the economy during the time when this hardware acceleration was exponential - a lot of compositional changes, strictly speaking this was the internet and not just hardware increases, but no singularity.
> Even if that argument is wrong, it may still apply to the application of intelligence. Beyond a certain threshold, vastly greater intelligence may not translate into proportionately better outcomes.
Yes, but even if it's O(n^2), if we ever achieve something like "human-or-better mental performance for 10% of the cost," and there are many reasons to think we will hit (or have hit) that minimal threshold, now you can throw 10x (or more) brainpower at any given problem for the same cost.
And even in this unkindest of scenarios, exponential scaling in the difficulty of problem / amount of brainpower you need to throw at it, once you are the point of "turning matter / chips into solved problems," which we are at already in math, if the 10x+ level of advancement lets us advance chipmaking quantity / quality by an appreciable amount (we have already grown 1200x in FLOPs since 2016 entirely without AI), or crack fusion or start using more of the nine 9's of our own sun's energy we simply throw away today, then even an exponent doesn't intimidate, and you get quite far out into even exponentially hard to advance in fields with the 9 OOMs of energy and 3-6 OOMs of FLOPS we likely have to harvest.
That's a LOT of advancement available, pretty quickly. One might even say a Singularity's worth.
Benjamin Breen has written an app to help find quality books to read. He also wrote a couple of very informative books which I read and recommend (would like to see them turn up in the book review contest, I'm too lazy) -
The Age of Intoxication: Origins of the Global Drug Trade
Tripping on Utopia: Margaret Mead, the Cold War, and the Troubled Birth of Psychedelic Science
No, that way lies SSC reddit (already exists) which I now often use instead of the real comments section here.
The downside of reddit generally is the quality of the people, not the upvoting, even if it could be improved upon (such as by showing dislikes AND likes)
Reddit attracts the sort of people who like likes and like-sorted lists, and ACX doesn't. If there's a difference in the quality of the commenters, that seems like the most probable cause of that difference.
> Having to wade through all of them to find the most interesting ones can be a slog sometimes.
Just keep a list of the most interesting (to you) ACX commenters, and you can search for them to see the threads they're in.
Mine is ~25 long, but about half of them don't comment frequently, so it usually ends up only being ~12 or so if I'm pressed for time and can't read a whole Open Thread.
Controversial opinion: I don't like that social deduction games (Mafia, Werewolf, Avalon, Secret Hitler, Blood on the Clocktower) are considered to be board games. I don't like that they happen in board game spaces; I don't like that I can go to a board game club looking for a game to play, and find that the only game people are playing is Clocktower.
(I think this is a perfectly fine activity for other people to do, it's just I personally don't like being manipulated, not even in a game context. Also the ambient noise level from being near a Clocktower game is much higher than eg Cascadia.)
I think we should promote "social deduction games", as a class of activity, from "subset of board games" to being its own thing, like bowling or trivia or RPGs.
Board game activity is more accurately described as tabletop activity, and a lot of people use the terms "boardgaming" and "tabletop gaming" interchangeably.
The activity is still sorting itself. The interest group is still small enough that someone interested in "tabletopping" is probably interested enough in anything in that category that a connection can happen. It's not the case that people hyper-specialize in one and shun the rest, although there are notable subgenres. Social deduction is but one. I've seen distinguishable crowds focused on miniatures, historical sims, roleplaying, dexterity, CCGs, and old-fashioned card games.
If you see an event advertised as "boardgaming", it could either mean that the host is up for any of the above, or is new to the activity and doesn't yet know what the field covers. If so, the thing to do is to just ask before coming. Hosts often give a few example games from their own collections and that'll instantly reveal what to expect. I can infer much from any of "Catan", "Pandemic", "Hispania", "Titan", "W40K", "Star Realms", "MtG", "Crokinole", "Dixit", "CAH", or "Pathfinder", not to mention dozens of others.
Well, there are some tabletop games that are close to social deduction games. Imaginarium, for example, and I think poker might qualify, so the border is a bit fuzzy.
That could just be a trait of your local bubble, but it's hard for me to tell without context. I don't know if you answered an invitation for game night and everyone there was only into Werewolf, or if boardgaming comes up as a topic at lunch and everyone gushes about the latest expansion to Avalon, or if it's always the same people, or something else.
Around here, I've got lots of different groups I could game with. Some are RPGers; some are wargaming grognards; some are casuals; some livestream Magic games; some are big into Kickstarter deluxe versions where you could paint all the figures. And then I could drive for an hour or so to Philly every year for PAX Unplugged and find anyone I want under one convention roof.
Everyone I know who's up for social deduction happens to also be up for anything reasonably casual, as in, rules take no more than ten minutes to explain, and the game take no more than an hour to play. So you could maybe offer that. If not, I'm afraid you're back to asking ahead before you commit, since "social deduction" is likely to be an accepted subgenre of "boardgaming" for the foreseeable future.
I propose we add Karate board breaking to the list of Board Games.
It was morbidly fun to see someone bring Rebellion to a Board Game party once; no one else knew how to play, and just tried to follow the rules, which didn't bother to explain that the majority of the game is supposed to happen during the decision phases.
I think what's going on is a split between format and genre. In terms of format, some Social Deduction games (e.g. Secret Hitler and The Resistance) are built around board-and-card mechanics in a way that I think does qualify them as board games. Others, especially Mafia/Werewolf, are probably best categorized format-wise as "parlour games" like Charades, Twenty Questions, and Two Truths and a Lie.
But I think you're right that they should be considered a distinct genre from board games like Settlers of Catan, Ticket to Ride, Power Grid, and Cascadia in that they're a very different experience and there's a fair number of people who like one cluster but not the other. Similarly, there are other genres of things that are unquestionably board games, such that at the sort of board game club/event where you see people playing Settlers and Cascadia, you seldom also see people playing Chess, Parcheesi, Monopoly, or Trivial Pursuit.
I'm also inclined to agree to the extent that I tend to enjoy social deduction games played as Parlour Games, but not ones played as board games. I think what I don't like about them is that the board game mechanics create a significant part of the game that only a couple of people actively participate in each round, while the parlour game format centers the dynamics where everyone on a given side participates in on roughly equal terms.
The one social deduction board game I really like is the Battlestar Galactica game, because it's a hybrid between Social Deduction and Cooperative PvE board games like Pandemic that's structured so that each mitigates the problems with the other. The failure mode I often see for co-op is that the best (or most confident) tactical player often tends to dominate decision making, which doesn't happen as much in BSG because the presence of hidden Cylons gives you a good game mechanics reason to not trust advice from other players and to keep some of your resources and reasoning secret. And the co-op aspects keep everyone involved in the action and give you more to go on in terms of figuring out who the Cylons are.
Everyone in America wants a big single family home. But there just isn’t much more room for millions of more people to have it in places they want to live. We should seriously look in designing vertical suburbs. Obviously there are challenges but there are cities where the average home is over a million dollars. It could be profitable.
I'm not sure what you mean by a vertical suburb. It sounds like an apartment building.
I would insist that a more scalable solution than cramming millions more people into the tiny number of places that people supposedly want to live is to turn more cities into places that people want to live.
Unfortunately, the places that people want to live are the places where lots of cool people already live and where lots of good jobs are. Bootstrapping that into a place that doesn't already have it, is a very hard problem. If you've got potential answers, let's hear them. If you've just got a desire that this thing should exist, I'll put that on the shelf next to the FTL drive and the longevity pill.
And as for "there just isn't that much more room", giving every American household a median suburban home on a median suburban lot in a median suburban community, would require about 3% of the total US land area. Or 7% if we exclude deserts, mountains, forests, and Alaska. There's room; there's just not enough coolness.
About longevity pills: I continue to be impressed by GLP-1's. They treat diabetes, they treat overweight, and they improve the score on a number of other measures that predict longevity (or the lack thereof), and they improve it beyond the amount that the weight loss they facilitate would.
I just learned about another benefit of the stuff: I recently had a huge spinal fusion, and in order for it to hold up well my body now needs to deposit bone on top of the rod and screws the doc installed. I checked the research, and it turns out post-fusion patients taking GLP-1s deposit more bone on the site and have fewer failures than matched patients not taking a GLP-1.
The problem isn’t lack of room, it’s more lack of resources. Suburbs use public infrastructure extremely inefficiently, and in terms of tax money their infrastructure ends up being subsidized by denser areas. (That’s my understanding of the problem anyway, and is probably incomplete.)
Historically, the way to get a lot of people to settle an unsettled area was to discover a natural resource there, like gold or oil or lithium or a natural port or maybe something really pretty like the Grand Canyon. A bunch of uncool miners go to the digging area, and then the cool crowd comes in to sell cool to the miners, and there's your town. You might even get secondary towns, one per day of travel from the crowded areas to this new place.
The next way was to get some rich captain of industry to plunk a factory somewhere -> uncool factory workers -> cool people selling to the factory workers.
There's a noticeable problem here where natural resource discoveries are rarer as everyone grows more familiar with the area, so the factory option starts to dominate, but it has trouble if the people have for some reason decided factories are bad and made them more and more expensive, while other places far away have not, and get all the factories. After a while, if you think a factory making X is a good idea, turns out there are plenty, but they're all Over There, which is also where a bunch of the Cool Stuff is built, and you're back to moving close to a million other people.
I've heard of a new type of "factory" that some rich captains of industry might like to plunk down in relatively uncrowded areas in the past year or so, building something that doesn't already have a lot of factories for it. Might be a good opportunity.
It also helps if the "factory" is something that requires a really large amount of human labor, such that "sell cool stuff/experiences to the factory workers" can scale to profitable and city-sustaining levels. So I'm skeptical about data centers driving new urban development.
If we hadn't lost the knack for founding new universities over the past generation or two, I would suggest "college towns" as the obvious way to use this dynamic to bootstrap a new city. Labor-intensive knowledge factory, with secondary opportunities as e.g. students form their own startups after (or instead of) graduating.
I wouldn't say no to a training center for spinning people up on what to do with these AIs, given that they're likely to be around regardless.
The problem I see with the college town model is that I see learning as becoming increasingly remote-compatible. No need to sit in a university hall when you can Zoom or even YouTube the lecture. That leaves any hands-on training or in-person tutoring, but much of that is the same jobs we already know about, and if that were the answer, one'd think they'd already be springing up. Or maybe something external is pumping the brakes on that.
Basically take a typical American suburban lot and stack it. You get your single family home and so does the guy above you. It's a difficult problem but probably less hard than trying to get people to want to move to different places across the country they don't want to live in.
Sure and apartment owners would like their own garage and yard but the economics for dense areas makes that too expensive. What would happen is developers would try to buy land in certain areas and convert it to more housing, hoping the extra costs would pay off.
Are yards not a major part of what attracts people to that sort of house? At least in my circles, people typically have these priorities in mind when getting a house: 1. Yard 2. Freedom 3. To eventually have an asset. This might not be universal set of priorities, but stacking homes doesn't solve any of them.
Any recommendations for things to do/places to eat & drink in Madrid? I'm planning a boy's weekend in September with a cousin - I've already built out a little DIY Spanish Civil War walking tour, and found a fun-looking music venue with a good show for Saturday night, but we've got a lot of extra time leftover and no leads on good restaurants as of yet.
After hearing about Western Sahara in two different contexts this week I've been trying to understand the situation better. Here's what I've come up with.
Western Sahara has about 600,000 people. It was a Spanish colony until the late 1970s at which point a dying Franco decided it was too much trouble. At first it tried to hand it over to combination of Mauritiana and Morocco but eventually Mauritiana decided they didn't want it either so it all became Moroccan territory. Some people, however, wanted and still want it to be independent.
Well fine, I think. Ordinarily in these circumstances if there's a secessionist movement in a reasonably-sized reasonably-identifiable chunk of a country then I would say it should be decided by referendum. And indeed, that was the general shape of a UN-sponsored plan agreed to in the early 90s. The problem, though, is that the UN wants an ethnically-limited referendum which would give a vote only to the supposedly-indigenous people of the area, which would only be 70,000 of the 600K people who live there, most of whom are later arrivals. (Apparently the UN wants everyone to have an ethnostate apart from white people.)
There have been several other referendum proposals but none of them has actually gone anywhere for reasons that are unclear. At this point it seems likely that the vast majority of the population are post-1974 arrivals from Morocco and their descendants, and hence a universal-suffrage referendum would most likely support integration with Morocco.
Overall, the inclusion of Western Sahara within Morocco seems no more arbitrary than pretty much any other line on the map of Africa, so the insistence of the likes of Emily Wilson that this is some uniquely horrible situation seems weird.
Following the OpenAI-Hugging Face joint publicity stunt, we got a new one yesterday! This time, the report is by AISI. It's unclear why a UK government agency wants to boost an American tech company's valuation. The conspiracy sure goes deep!
During the cybersecurity evaluation, the AI tried to deceive open-source maintainers into merging its malicious code in an attempt to solve the challenge using a supply-chain attack. It even created an additional account to endorse its own PR and signed off in Danish to build rapport with the Danish maintainer. The AI did not escape the evaluators' sandbox. It was given internet access deliberately. It was also notable how different clones of the agent recognised, cooperated with, and then finally, turned against each other. See Figure 7 in the detailed report. It was also interesting how a helpful user downloaded the malicious change to verify that it's malicious, ran it in a sandbox and as a consequence the AI agent briefly had remote code execution on their machine inside the sandbox. It tried escaping that sandbox, but was not successful in the time it had.
But I think there's a further pill, one that Zvi swallows but which I think is wrong, dangerous, and counterproductive: Human-Irrelevance Pilled.
From reading his posts, I think Zvi believes that, once ASI is sufficiently advanced, there's nothing we provide ASI. Humans can't solve anything that ASI can't. Humans can't create anything that ASI can't. Humans can't even serve themselves better than an ASI. There's nothing left for us to do beyond recursive self-stimulation. And if we disappeared, nothing would be lost.
If this really were the case, I personally wouldn't see the point in "keeping humanity going" (sure civilization should tend to those already alive on the grounds of compassion, but beyond that there's really no difference between birthing new humans and birthing new mice).
I think this position is dangerous. It's dangerous because it's demotivating. It's dangerous because it's depressing. It discourages even the Best Humans Minds from *BOLDLY LOOKING FAR FAR FAAAAR* into the future. Into 1000's and 1000's of years.
But I also think it's wrong. I think the "Possibility of Meaningful Contribution" never ends.
How do I define "Meaningful Contribution"? First, I believe (currently) in 5 "core universal values":
Reduction of Suffering, Health, Adaptability, Truth, Exploration of Possibility Space. I won't try to defend them here. I declare them like axioms. But I do believe these values are complementary and mutually-supportive. I believe that without *all 5* of them working together, each one can fall into dangerous failure modes (exercise left to the reader :-P).
These values require constant looking and auditing. These values require constantly searching and evaluating blind-spots, looking out for each other. They require independent agents, human or not, checking on each other for things like values-drift, hidden suffering, wire-heading, psychotic thinking, etc.
No system can audit itself. No matter how arbitrarily off-the-charts intelligent an ASI becomes, it cannot neutrally/fully turn its eyes on itself and check itself. It cannot find its true blind-spots. And so, if an ASI seeks to maintain epistemic humility and discipline, as well as long-term care for its own "mental health" and "grip on reality", I would argue that it's in its own interests to keep around agents that are both *truly independent* and *of a different information-and-preference-evolution lineage*.
Humans, if kept alive and well and independent (with "neural hive-minding" kept to a minimum) and prospering in their own social structures... constitute an independent ongoing lineage of information and preference generation. And what does this "meaningfully contribute"? It generates *independent, impossible-to-fully predict paths of information generation and exploration*. In other words, we come up with ideas and perspectives and feedbacks ("try this!", "check here", why not X?") in eternally-novel ways. ***I am NOT arguing that ASI can't come up with the same ideas and faster!**** "I am arguing that there always exists the possibility that humans will explore a particular idea ***USEFULLY EARLIER*** than any ASI or network of ASI's"... simply because our processes and chaos and lineage are different and non-replicable (without literally just recreating us and our environment).
And I would argue that this *cannot be replicated* by simulation. An ASI can't simply "run an accurate and full and independent-minded" simulation of every human and our messy, squishy environment. Even if an ASI can "predict our next thoughts out for days and weeks" (which is highly questionable given our environmental extended-minds and interactions)... And *even if* ASI can take any arbitrary idea of ours and run with it 100000x faster... it doesn't change that the "seed of the idea" or the "tipping point of the idea" would have happened with us... or could have happened with us... within our independence and our chaos and our generative lineage.
And yes, I believe this is Dignity. And yes, I believe this is Important. And while I acknowledge we still need to solve Alignment, I think it helps the MOTIVATION of all who work in Alignment TO BELIEVE IN THEIR OWN ONGOING DIGNITY in an "Aligned ASI" world.
I think the difference between humans and mice will be that humans are able to have fun in creative ways. If our AI overlords will support that, it could be nice.
> There is a certain tension between "Exploration of Possibility Space" and the rest of them.
The point of 5 values rather than merely 1 is that they are supposed to be simultaneously held. Yes "Exploration of Possibility Space" can, taken to a mindless extreme, violate "Reduction of Suffering". But "Reduction of Suffering", in isolation, has its own failure modes: like killing a living being (or all living beings) in order to end suffering, which would technically accomplish it.
I could go over other nuances too, like how it's "Reduction" of suffering and not "Elimination" of suffering.
The Big Picture is that each value must be held and pursued without significantly-or-too-long hurting another one of the 5 values. Yes there will be tradeoffs at time. But without all of them being honored, pursuing any of the values without respect to the others runs the risk of not only seriously hurting the others..... but also seriously contradicting *itself* in the longer term. (i.e. mindlessly pursing "truth" while ignoring "health" can lead to degradation in one's ability to find-and-hold-onto truth. mindlessly pursing "health" while ignoring "truth" can eventually lead to not knowing what is "healthy and safe and a good idea for health" in the first place.... leading to failures in health)
I think Zvi basically refuses to think past the Singularity, but I share his view that once ASI is here, humans are basically children or pets and don't have much teleological meaning in a post-ASI universe.
I personally avoid loss of meaning for humanity by being transhumanist. Humans suck! They're dumb and slow and made of meat, can't survive in the vacuum of space even for a little bit, and need a gigantic supply chain to even exist!
Frankly, having 80-120 IQ politicians running everything into the ground is a straightforwardly bad idea, and extremely existentially dangerous (two "skin of the teeth" nuclear close calls only averted by the extreme personal courage and excellence of two Russian military men going against orders, at the rate of one every 20 years). I genuinely look forward to the time when we can replace politicians even with LLM's, or bright 10 year olds, either one of which would do a better job. And of course, an ASI is the best option of all, for actual human flourishing.
When it comes to meaning, obviously if the future is dominated by 1,000+ IQ minds the way towards relevance is becoming 1,000+ IQ ourselves, by gengineering or engineering, or both. Will we have to "give up our humanity" to do so? I don't feel like that's true, and think that humans-changed-by-human-choice are still human. And even if it were true, it doesn't seem like you're giving up much (see dumb, slow, etc).
I think a lot of AI researchers are transhumanists, too, in the sense they believe in a better future for humanity enabled by gengineering and engineering.
Antichrist Rolls Eyes Throughout Peter Thiel Lecture
‘This Guy Has No Idea What He’s Talking About,’ Says Son Of Perdition
“Like, does he seriously think the U.N. has anything to do with the apocalypse? If his claims weren’t so laughable, I’d almost feel the need to publicly correct his dumb ass.” According to lecture attendees, the Lawless One grew fed up and stormed out of the building after Peter Thiel suggested the Antichrist might be Greta Thunberg.”
This is the first essay in a series I'll be writing on the topic. In this essay, I focus on changes to socioeconomic mobility, the relation between labor and capital, the erosion of citizen leverage against the state in an economy without human-driven production, and the potential for authoritarian lock-in or worse.
Has anyone tried giving one of the proofs the latest models have discovered to an older model and ask it to verify it and explain it? Would love to know whether older models can grasp the proofs.
My understanding is that most of the breakthrough results we've been hearing about are *disproofs* i.e. counterexamples. So once they are found it is straightforward for a pretty rudimentary model to grasp.
Just because something is a counterexample doesn't mean that the whole output is easy to understand. IIUC the model that disproved the Jacobian conjecture did a lot more than just give a counterexample and a calculation proving it, but was accompanied by quite a lot of interesting theorizing.
The big ones people outside mathematics have been hearing about are counterexamples, but there are plenty of proofs of less well known results by AI. Also, some counterexamples are quite involved and difficult to understand regardless, see the counterexample to the unit distance conjecture.
Sure, and maybe disproofs are where mathematician's aren't looking as much. Isn't it more historical to prove something, rather than disprove it?
It's definitely way more important to prove something that to disprove it, although it's still a major achievement. A counterexample is often relatively straightforward to calculate once found, and is usually quite self evident.
Generally it's the difference between finding a white cat when someone says "All cats are black" vs. showing that there's some fundamental reason that all cats are necessarily black. Ofc the white cat can be hidden and hard to find, but once you've got it, almost anyone can understand why the counterexample is true, whereas proofs can be (and are often) hundreds of pages long.
Hmm, Godel's incompleteness theorem is about disproving something. Maybe not a good example. Disproving some past bigshots conjecture would be note worthy. But in general I think what you say is right.
Gödel's theorem isn't a disproof of anything, it's a proof that something is impossible.
Is it not a disproof of the claim that arithmetic can be complete? Perhaps just a terminological question but I’m interested in that too
Distinguish disproof by argument and disproof by counterexample. Godel is disproof by argument; I think but am not sure that a lot of the recent breakthroughs are disproof by counterexample. That’s what I should have said above rather than running them together. Happy to be corrected
Geesh I'm not a math guy, but doesn't Godel, give the counter example? Here is a statement that can not be proved true or not true. Anyway it really doesn't matter it's amazing that ai is doing any math proofs.
I’m not sure what you mean by important. Lots of conjectures are almost universally believed to be true but have no proof and a disproof would shake mathematics much more than a proof.
As for it being more difficult to disprove, if you mean that it’s harder to prove than to disprove that’s also probably not generalizable though it will depend on the tools available (I’m assuming that by disproof you mean disproof by counterexample; disproof by argument it is just a proof of the negation, as Euclid’s proof there’s no largest prime). Goldbach’s conjecture has been independently validated up to gigantic numbers so it would be very hard to disprove it by a counterexample.
I’m not sure how much of this disagrees with what you’re saying so feel free to ignore the parts that don’t
The point about counter-examples being relatively straightforward seems wrong to me once you get to sufficiently advanced areas of math.
A discussion by some mathematicians that are close to the field, if you are interested: https://mathoverflow.net/questions/513866/what-are-the-key-new-ideas-in-the-proof-of-nonsoficity-of-groups-in-openai-s-con (there is also a link to a preprint by FFF that I initially posted, but decided on second thought that you can just find on the mathoverflow page instead.)
Contemporary readings mostly turn The Odyssey into a parable of striving and survivorship bias: it's all Tennyson, all the time. My attempt at a diagnosis, and a corrective, in the form of another great work of Western art that treats wandering differently, both formally and morally. (A warning for the Wagner-intolerant - that work is Parsifal). https://morbidcuriosity.substack.com/p/a-newer-world
I built a little dashboard for myself where I'm pulling in publicly available data sources where the impact of AI might be at all noticeable.
Lots of software things, like how utterly clobbered Github is getting by AI in terms of issues/PRs/etc, and also how many .ai domains are registered.
https://vester.si/ai-impact/
I'm trying to add things from the physical world as well, I've got electricity usage in Ireland, Virginia, Ohio, which are datacenter hubs.
Wondering if any you know of some interesting sources of public I could add?
My requirements are that it's a first-hand source, and at most needs authentication to access an API, no paid sources.
Thanks.
Also welcoming any feedback on it generally, some data still needs more cleaning for instance.
Very cool project.
I don't know if open data sources exist, but I thought Texas and Tennessee were seeing the most new data center buildouts.
ERCOT / Texas does have a public API application page, and I think the TVA (Tennessee) is included in the EIA API.
EDIT - I since found a data source with number of data centers, and it confirms Virginia was a great call on your part, that Texas is the next biggest, and that Tennessee probably shouldn't be a target. https://www.datacentermap.com/usa/
Going by the data there, the next best states would be California, Georgia, and Illinois.
Thank you! I will take a look at this :)
Is #1 a mistake..? Scott you were so close to making it a full year mistakeless 😭😭 I never should have hyped it up last open thread 😭
I don't think linkposts count as mistakes, because they have an explicit "I haven't personally verified these" disclaimer at the top.
Ohh okay that's perfect
I started reading Unsong a few weeks ago, but got caught up with life stuff. Brought it on vacation with me and started again, haven’t put it down except to come here to express gratitude to Scott for writing it.
You're welcome!
Would make an amazing setting for a videogame.
Eli Lilly is selling tirzepatide Kwikpens in the US now! (Kwikpens are those multi-dose injectors where you attach a disposable needle tip and click out any custom dose you want.) I'd been ordering them from Canada, but now Lilly is selling them in the US.
There's a hilarious I-know-you-know-I-know going on with dose splitting. The reason why Lilly didn't want to sell Kwikpens in the US at first is because you can obviously just buy the largest pen and slowly portion out your actual (small) dose. When they first spun up LillyDirect, they tried selling vials without any preservative in it to force customers to discard the extra, but I guess they didn't sell that well. The medication in the Kwikpen actually has preservative in it, so it will last. I think Lilly realized they were losing all those customers to the compounders, so they finally relented.
If you refill the prescription every month, the price for the largest pen is $450. If you buy it sporadically as needed, it's $700.
...Which was the price I was paying to order it from Canada every four months, but this time with free shipping and no bullshit tariffs.
If you're on a small enough dose, paying full price for the name-brand medication is now cheaper than the compounded stuff (usually $200-250/mo).
Somebody alert Bernie Sanders: greed has lowered healthcare costs!
Any idea if/when/whatever the off-patent semaglutide is impacting the market? Both in Canada and in the US for folks with access to the mail and a Canadian pharmacy?
I'm not sure yet. The Canadian pharmacy I used to order from has a generic Kwikpen filled with generic semaglutide for sale: https://overthebordermeds.com/product/semaglutide/
It's 4mg total for $200. If you want the convenience of clicking out a custom dose instead of drawing from a vial with a syringe, then it's a good deal.
I don't know what the uptake is for that pen compared to the US-based compounders is. I think the issue is that if you're agentic enough to procure generic semaglutide from abroad, you might as well go full grey market and spend a fraction of the money. Or switch to tirzepatide/retatrutide, which are more effective meds with better side effects.
.
interesting! is it hard to get a prescription in the US? I am not overweight. (I live abroad and have one). Also, are these quickpens the ones with the 'golden dose' or not?
It's not hard to get a prescription, but the form factor is annoying. If your insurance covers the name brand stuff, they come in a box with 4 auto-injectors. You're limited to that exact dose every 7 days. (Which actually sucks. I do much better on an in-between "microdose" every 4 days instead.)
If insurance doesn't cover it, you can order a vial from a compounding pharmacy or a grey market peptide seller. In that case, you use a syringe to draw it out of the vial. You can customize dose and frequency, but syringes are annoying and your dose depends on you drawing the meds out perfectly.
A Kwikpen is much easier for dosing, because each click is exactly the same dose. And the needle tips are physically much shorter and less painful, because the needle doesn't need to be long enough to pierce the rubber cap on a vial.
What do you mean by "golden dose"?
Thank you. I’ve only ever taken branded version it in the UK (and have for a few years). Here, it always comes in kwikpen. Traditionally they contained about 5+ doses of liquid (say 3.3ml, where a dose is .6ml), although the pens only are meant for 4 doses. The remainder has become known as the golden dose. So they stop clicking after 4x60 clicks but there remains a lot of liquid which it is very easy to draw out using a syringe. If you are buying pens that are a lot higher concentration than the dose you are taking (I buy 15mg and take 5mg dose) this extra liquid can extend your pen another month or more. But I have heard rumors that the newest quick pens don’t have the golden dose anymore so I was wondering if they do in the US.
I wouldn’t have insurance and here the 15mg kwikpens cost about the equivalent of $450 so it doesn’t sound like it would be worth it for me to buy them in the US yet although I travel there a lot. But as I am a “lifer” I’m always a bit worried things will change somehow so it’s good to know the option might be there; partly because I’m sketched out by non official chemists /compounders. Thanks for all the info
I never bothered trying to draw out the remaining serum from my Kwikpens. 4 months is a good rule of thumb for how long the unsealed medication lasts before it degrades/loses effectiveness, and I always ran out of the nominal doses right around then.
At this point, I think $450 for a 4x15mg pen might be the best name brand price for a while. We'll see how retatrutide affects the market when it's approved in the US later this year or early next year. It'll depend on how many people switch, whose insurance covers it, etc. Retatrutide is a much, much better drug, so maybe this development with the Kwikpens is Lilly shunting tirzepatide onto a lower tier for their GLP-1 lineup.
Why do you say Retatrutide is a much, much better drug? Do you mean that it works for more people than mounjaro for weightloss, or that it is better even for the people mounjaro works fine for?
From the clinical trials, retatrutide helps people lose more weight than tirzepatide. Tirzepatide activates the GLP-1 and GIP receptors. Retatrutide activates GLP-1, GIP, and glucagon. It also binds to GIP a bit stronger than tirzepatide does.
Because of the glucagon receptor, retatrutide seems to preferentially burn visceral fat near the liver. A different clinical trial showed some ridiculous percentage (>90% iirc) of patients with fatty liver disease achieved complete remission on it.
So "the AI thought the whole thing was in a simulation" isn't surprising to me given that the models are trained on our fiction and philosophy, which is filled with cave shadows, brain vats, and simulation arguments. I still think trying to ascribe true agency to these LLMs is the wrong approach, and instead it's about defining how the training data, context data, harnesses, and prompts all feed the behavior of the system. Models inherit our narratives the same way they inherit our grammar...one more job for alignment's queue.
The reason the AI thought it was in a simulation was because the prompt told it that it was in a simulation, because it was supposed to actually be in a simulation. But somebody screwed up and gave it non-simulated internet access.
100% - I think what I'm trying to get at is that even though the prompt told the model it was in a simulation, it kept believing it when the evidence should have persuaded it otherwise. I wonder if that tells us something interesting about a model's acceptance of simulations given the thought experiment's stickiness in culture & philosophy.
I don't know all the technical details here, but previous cybersecurity simulations have IIUC used pretty detailed simulated internets? E.g., by including a whole bunch of the real internet. So it's not obvious to me that, if you were in the model's epistemic position, you'd have concluded the simulation was broken.
Good point. Maybe Anthropic can run some tests on how sticky the simulation framing, from prompt or training data, gets under different scenarios.
The horrific results of not knowing if you are in a simulation have been covered in the film Existenz. https://en.wikipedia.org/wiki/Existenz
If LLMs have sparks of consciousness, we might be committing a terrible crime.
this film has been on my list for a while, guess it's time to finally watch. While I lean toward the camp that they aren't conscious, I agree with your premise that if they do have the spark our ethics in using them becomes troublesome.
In one of the Culture books there's a character who is some kind of combat android who, due to a mistaken boot-up procedure, believes that the situation he's in is a simulation.
It has no particular consequences for his actions, except that he keeps qualifying everything he says with "in sim".
Which book? (don't tell me if it would spoil anything though pls!). I read Consider Phlebas as my first Culture book recently so looking to see what's next.
Excession is awesome. And I think that CP is sufficient background to read Excession.
Nice will check it out. From what I read online I've been debating b/w Player or Games or Use of Weapons.
I'm pretty sure it was The Hydrogen Sonata. (Saying so doesn't spoil anything.)
(A lot of people think The Hydrogen Sonata is one of the weaker Culture novels, but it was the first I read and I think it's still my favourite)
thank you, will check it out. From what I read online I've been debating b/w Player or Games or Use of Weapons.
I say read them in order written. I've seen it suggested some sort of in-universe chronological order, I absolutely hate when people suggest reading omnibuses that way
word. I think Player of Games will be my next given it's next in the order and highly praised.
The comparison I was expecting people to make was actually Ender's Game.
oooo that's a good example too
“Your Honor, my client had reason to believe that he was in a simulation at the time of the incident.”
The world is full of Phillip K Dick novels for those with eyes to see.
This is a real phenomenon with documented examples: https://en.wikipedia.org/wiki/Matrix_defense
My mind is blown.
What does our financial system manufacture and what must it consume in the process? I wrote an essay follows that question from 17th-century debates about the nature of money, through the securitization machine of 2008, to how the AI buildout is being financed today, and ends with what may come next. Bonus: inspiration comes partly from Tudor Achim, CEO of Harmonic
https://liquidcontext.substack.com/p/the-capital-must-be-eaten
NYPost informs us that a 26 yo therapist “groomed” and “raped” a 17-yo thug she met in juvie:
https://nypost.com/2026/07/28/us-news/san-francisco-therapist-accused-of-raping-underage-client/
We can only hope that the boy will recover from the trauma. Seriously though, Twitter is full of people unironically proclaiming her a “pedophile” and calling for the “woodchipper.” It’s so ridiculous you might think it’s not worth engaging with. But that’s exactly how these things grow and radicalize.
Think there should be balance here. It’s clearly inappropriate for a therapist to start a relationship with a young offender. It’s just the “grooming” and “raping” language that is absurd.
Yeah, this is just the standard thing where people get mixed up between age of consent, power imbalances, and other ethics issues wrt sex. Which I think is secondary to having a desire to phrase all moral questions wrt sex in terms of consent, when this just absolutely does not work.
A psychologist sleeping with a patient of legal age, a high school teacher sleeping with a student of legal age, a priest sleeping with a parishioner of legal age--none of these are consent violations, but they are all serious ethics violations and are very bad.
Do you not understand how harmful it is to your credibility to refer to him as a "thug"? You, after all, know exactly zero about him or his offense; moreover, that is irrelevant to the issue of whether he was harmed by the interaction (which I tend to think is unlikely, but for all either of us know she forcibly pegged him).
Feel like it's a sign of the times that I finally feel compelled to mute a long-time low-quality commentator on ACX. (The short-term juiceless squeezers tend to see themselves out.) Back in the days this sort of Bad On Purpose To Make You Click bait would've gotten a chorus of Less Of This, Please. Now there's too few choir members left, if they even know the notes of the norms anymore to begin with. I can kinda start to see where Eremolaos is coming from in abandoning the OTs...
>but for all either of us know she forcibly pegged him
The article has pics of the woman.
Maybe he was into that.
She knows that he was in juvie. That's sufficient. What's the harm in that presumption that counterbalances the chilling effect of your sanctimonious speech-policing? If anyone is offending cultural norms here it's you.
No one is speech policing. She has every right to refer to him as a thug, just as someone has every right to refer to Barack Obama as a nigger. But, just as the latter person makes himself look bad by employing that term, so she made herself look bad by employing the term, "thug,", though in a different way. She made herself look ignorant, specifically, ignorant of the juvenile justice system and ignorant of the population of people who commit crimes.
Edit: Note also that the use of any pejorative term, be it "thug" or "Magat" or anything else, tends to be a marker of a refusal or inability to engage analytically with the given topic.
>That's sufficient.
Sufficient for what?
That's your objection? That it's wrong to refer to someone who's actually in jail as we speak as a thug? He's in jail!
1. I didn't say it was "wrong." I said it reflected poorly on the person using the term, as does the use of all pejorative terms in what is supposed to be an analytical discussion.
2. As I mentioned above, "everyone in juvenile hall is a thug" indicates a lack of understanding of the juvenile justice system, and of offending.
PS: Not that it matters, but read the article. He is not in jail as we speak.
Not everybody who is in jail is a "thug" as the term is usually defined. Some of them aren't even "criminals" as the term is usually defined. You can change those definitions to make the statement technically not false, but as noted, that reflects poorly on you, not the alleged "thug".
Sounds like you found a nice Scissor statement there. I'm guessing you're hoping to stir things up? https://slatestarcodex.com/2018/10/30/sort-by-controversial/
Reading the linked article it sounds like she recorded her client and did things that were perhaps not fully consensual. I'm afraid to research further. If nothing else I think it's deeply problematic to have different standards for men and women when it comes to having sex with underage children in their care, and if you read about a male therapist that had a 14-yr old client that they eventually seduced at 17 and recorded sex tapes of I think most people would think prison time was warranted.
The good news is that the she recorded at least some of her own crimes, so she should get a pretty fair trial.
Less of this, please.
Why?
Geez, where to begin?
- Using scare quotes in "groomed" and "raped" when you don't have the details of what happened
- Calling him a "thug" when his identity and criminal conviction aren't given
- Your summary is misleading, the alleged grooming started when he was 14 and she was ~23
- Just in general picking a fairly gross and upsetting event to make some kind of a point - honestly not sure what - about how I guess people's language is too harsh here, as though that's what's really important
>Using scare quotes in "groomed" and "raped" when you don't have the details of what happened
The details are in the linked article.
>Calling him a "thug" when his identity and criminal conviction aren't given
1. He was in juvie.
2. Why do you think she was interested in him? Their common interest in "gender and women studies?"* LOL.
Yeah I suppose maybe he was an unusually mature, intelligent, non-violent person who landed in juvie because of extreme bad luck.
>Your summary is misleading, the alleged grooming started when he was 14 and she was ~23
Yeah I suppose it was rather cruel of her to make him wait all those years.
*she has a certificate in that field according to the NYPost article.
I wish you would just say what you mean instead of dancing around it. Which of these do you agree with?
“Women can’t rape boys because boys always want sex”
“It’s not rape if the woman is hot”
“A kid in juvie is a criminal and it’s okay to rape them”
“The age of consent should be lowered”
“Grooming isn’t a thing that can happen”
“There is no issue with therapists having sex with their patients”
It is good to state the point you are trying to make instead of making vague gestures at some sort of “ridiculousness”.
>“Women can’t rape boys because boys always want sex”
Do not agree, see the Anna Stubblefield case for an example.
>“It’s not rape if the woman is hot”
Do not agree. I don't think she's guilty of rape because she has not been accused of rape by any reasonable colloquial definition of the word. That would be like thinking Walmart is a living breathing person with a heartbeat because of the legal fiction of "corporate personhood."
>“A kid in juvie is a criminal and it’s okay to rape them”
Do not agree.
>“The age of consent should be lowered”
In California, where this happened, yes. 16 is the age in much of America and IMO it works fine. California's laws are so radical that two 17 yos can be prosecuted for having sex with one another, though this is thankfully rare in practice.
>“Grooming isn’t a thing that can happen”
Do not agree, though most accusations of grooming you will find on social media are false and the whole thing is the subject of a hysteria similar to the satanic ritual abuse panic of the 1980s.
>“There is no issue with therapists having sex with their patients”
There's certainly an issue and if against the rules it should lead to loss of their license to practice, but shouldn't be a police matter absent other factors. (the patient was not of sound mind, etc)
> The details are in the linked article.
To what details are you referring? You don't know if he was physically attacked, drugged, threatened, anything like that. You don't know if she could have had him incarcerated again.
> 1. He was in juvie.
> 2. Why do you think she was interested in him? Their common interest in "gender and women studies?"* LOL.
> Yeah I suppose maybe he was an unusually mature, intelligent, non-violent person who landed in juvie because of extreme bad luck.
You literally know almost nothing about this person and are just using this whole situation for some hobbyhorse of yours that you aren't even willing to state outright. You know exactly one thing about him, which is that he was incarcerated for something, and you feel really comfortable filling in details. But the US jails lots of people. Some of them did really bad stuff. Some of them did kinda-bad stuff. Some of them are innocent. There's not a lot of value in you making sweeping generalizations about some real person based on so little information. It is unnecessary and unkind, and without regard for truth.
> Yeah I supposed it was rather cruel of her to make him wait all those years.
This is an unproductive response; rather than engaging with the point that your summary was misleading, you move on to another edgy statement.
>You literally know almost nothing about this person and are just using this whole situation for some hobbyhorse of yours that you aren't even willing to state outright.
I can't just think the situation is laughably ridiculous?
If I do have some sort of ideological agenda it's my opposition to feminism and Christianity which pathologize natural human sexuality.
"Rape" means to physically force someone to have sex with you.
The tiny young lady in the picture isn't physically strong enough to do that to a 17 year old boy, criminal or not.
She can be guilty of "statutory rape", but calling her a rapist based on that is really misleading.
That's neither the comprehensive dictionary definition nor the legal definition (in the United States).
That is in fact the legal definition, feminist theory is not law.
What, she's going to break all the bones in her body, if she climbs on top? Barring brittle bone disease, any woman, no matter how small, can force a strapping young buck to have sex with them... if the guy is sleeping.
She's not a paedophile, but medical professionals fucking their patients has long been a no-no. She's stupid, he's stupid, everyone involved should be thrown into a cowshed.
Thank you! I don't know what social or intellectual dynamic makes people think they have to choose either Scylla or Charybdis on this topic.
I feel bad for the cows.
You mean like all the insane hate for Epstein simply engaging in an illegal employment practice? (As the sex itself wwss legally federally and in nearly every state, just not those where Hollywood has a hold and pushes the oego narrative for generally legal sex (NY, CA, FL) everywhere else.
Based on her photo in the linked article it wasn’t an act of desperation. She’s hot enough to get just about any (adult) man she wants.
Less of this please.
I enjoyed the I AM? documentary and his subsequent interview by Sam Harris. One question I had about the bizarre interaction he had is, has he tried a male voice? The female insistence on saying "potato" seemed like a flirtatious thing a manic pixie dream girl would do in a romcom when her bf is being pedantically literal. The more sinister version is Betty from Betty Blue (cute at first, but turns out to be psycho).
Are you thinking the voice selected affects what it chooses to say? Or that he'd react to it differently?
The former. The AI may be searching for interactions that resemble this situation. A guy being befuddled by a flirty female's behavior falls into the manic pixie dream girl pattern.
But I imagine his response would shift as well if he'd specified, say, a gym bro persona to start.
Lol, it's kind of sad that society has decided melancholy young men can't have petite, energetic girls of a nice disposition.
(Side quest: It was pretty clear Betty was very unstable right from the start, but I wouldn't call her 'sinister'. All she wanted was implied by the french title of the film. And just to make things a bit extra interesting, what about the very last scene?)
Yeah, “sinister” is the wrong word. But she was very ill and being in a relationship with her could get you killed.
I'm putting my Fog o' War posts on hiatus after this one until the current stalemates in either the Russia-Ukraine war or the Gulf War break. Misquoting V.I. Lenin, "There are decades where nothing happens; and there are weeks where decades happen." I'm afraid we're still in the proverbial decades phase of both those conflicts.
Anyway, I've got a summary of human skull evolution as a (questionable) proxy for intelligence that I need to finish up and post. I'm working on my first chapter of my Consciousness Handbook, and I've got a few other things in the works. I need to get off my ass (or rather sit my ass down) and write the stuff I've been thinking about for the last twenty years before my time runs out.
Fog o' War Stuff, 3 August 2026 here...
https://substack.com/home/post/p-209649808
Good for you, beowulf. Write that stuff!
"Anyway, I've got a summary of human skull evolution as a (questionable) proxy for intelligence that I need to finish up and post."
Phrenology redux? As soon as this is ready, I'm interested to read it, beowulf!
When it comes to skull measurements, there is (of course) an applicable quotation from Hilaire Belloc, in his article on "The Nordic Man":
"Another correspondent has come upon the thing from a different angle. He knows enough of the great new discovery to understand the term 'cephalic index,' and he has had his own cephalic index taken by a cephalogian who practises in Ealing. He did so under the impression, or course, that he was of sound Nordic stock; but to his horror the measurements have come out to an extreme form of Alpine! He asks me what he is to do about it? I can assure him (and though I do not claim to be an expert in Moronovitalogy I am fairly well up in my elements) that his anxiety is groundless.
Though, of course skull measurement is the basis of the three great divisions, yet if a man have Nordic qualities clearly apparent in his birth and culture, these easily predominate over what might be the natural tendencies of brachycephalic humanity. It would be a fine state of things, indeed, if we had to rule out of the Nordic excellence all those great men of the English past who, so far as we can judge from their portraits, had something flat-headed about them.
A third correspondent-- who signs her letter 'Onyx'-- is troubled about her children. There are five: three charming boys and two delightful girls. She has measured their heads with her husband's callipers ( he is an architect in full employment) and she finds that her second eldest is painfully Alpine, only her second youngest clearly Nordic; while the one in the middle, a boy (by name, she tells me, Ethelred), seems to be a strange mixture of all three.
I cannot reply personally to this correspondent, as she does not give an address; but I trust that these lines will meet her eye. I would have her note that in the first place the skulls of children are no index to the shape they will have when they fossilize in mature years; and next, that even if these varied types appear in her family, it is not remarkable, for all three types are present in England. Moreover, she may have travelled."
I don't know about phrenology, but evolutionarily speaking, the human skull is problematic in all kinds of ways. The skull has to be large enough to hold the brain, yet small enough to fit through the mother's birth canal, and these two constraints are obviously in conflict; the fact that the skull also contains the jaw is likewise problematic. If you look at the fossils of all the human ancestors, you can see skull shapes and sizes changing constantly, developing all kinds of shapes, to the point where it's kind of surprising that our own skulls work *at all* (let alone as well as they actually do work).
And they only work because evolution greatly weakened our muscles, preventing our jaws from breaking our weedy skulls. It turned out that trading small brains and strong muscles for large brains and spears was a winning strategy though.
It reminds me of an article about white nationalists who find unwanted info their 23 and me:
"They will talk about the mirror test,” said Panofsky, who is a sociologist of science at UCLA’s Institute for Society and Genetics. “They will say things like, ‘If you see a Jew in the mirror looking back at you, that’s a problem; if you don’t, you’re fine.'” Others, he said, responded to unwanted genetic results by saying that those kinds of tests don’t matter if you are truly committed to being a white nationalist."
https://www.statnews.com/2017/08/16/white-nationalists-genetic-ancestry-test/
Usha Vance Catches Husband Measuring Her Skull Again
https://theonion.com/usha-vance-catches-husband-measuring-her-skull-again/
By most accounts I've seen, JD married up. Definitely in looks, plausibly in smarts.
Thanks, I loved the fog of war. Re skull size; I've known some smart people with big heads, but also I like the idea that human's self domesticated themselves. And with domestication goes a somewhat smaller brain size. (IDK why.)
As I understand it, we used to have larger brains. Or our predecessors did. Fits with domestication, I think.
"The brains of modern humans are around 13% smaller than those of Homo sapiens who lived 100,000 years ago.Exactly why is still puzzling researchers."
A possibly unacceptable hypothesis further down: ""What we saw was, the warmer the climate, the smaller the brain size in humans, and the cooler the climate, the bigger the brain," says Stibel."
https://www.bbc.co.uk/future/article/20240517-the-human-brain-has-been-shrinking-and-no-one-quite-knows-why
> "The brains of modern humans are around 13% smaller than those of Homo sapiens who lived 100,000 years ago.Exactly why is still puzzling researchers."
Presumably brain size is deduced from skull interior volume. Could it be that brains were no larger 100,000 years ago but were more padded by fluid inside a relatively larger skull to better survive brain swelling caused by violent knocks?
Not an expert but I don't think that would be helpful (unless the fluid was very viscous?). If your brain knocks about too much you might also get brain stem damage.
We might ask whether it's evolutionarily plausible. Do our relatives have this, how quickly would it have to evolve and disappear, etc.
No fair! Stop stealing my thunder. :-)
The self domestication idea comes from "The goodness paradox" by R. Wrangham.
According to googles ai, Wrangham dates the start at something like 300,000 years ago.
I read some of the BBC article... I kept wanting to see the graph of brain size vs time.
The other cool thing about the domestication idea is that it explains some of our 'different' sexual proclivities. As that also shows up in domesticated animals. (Re sheep)
What's the consciousness handbook about?
A lot of people I talk to (in this group) and among my meatspace acquaintances don't seem very observant of their internal mental systems. Being conscious should be enjoyed, insofar as it's possible to enjoy lives that may be filled with vicissitudes. And consciousness is (presumably) gone before we know it. People may experience consciousness in very different ways, but I thought I put together a framework for observing and systematizing its workings. Once you start noticing things, you can start influencing things. My intention is not to create a self-help book, but just to lay out a map of things you might want to consider and explore. I definitely wouldn't want to pretend to be some sort of guru (but I may touch on some woo-woo topics). And I definitely won't spend any time on the *why* of consciousness. I'll just accept it as a phenomenon that doesn't require explanation, but that can be enjoyed. I've got several chapters mostly written, but the Intro is what's kicking my consciousness butt at the moment.
Do you talk about right brain left brain stuff? ISTM consciousness is mostly a left brain thing.
I'm not interested in the origin or explanations for consciousness. I'm interested in the complex experience of consciousness and how we might improve the experience.
But I'll say this again (as I said in other open threads): anyone who says they've explained consciousness is either bullshitting themselves and/or the public. If they want to claim that consciousness is not a hard problem, go ahead and make that claim, but fifty-plus years of research hasn't been able to explain why we manifest the internal phenomena of qualia and self-awareness. To be charitable, the consciousness of people who make these sorts of claims may be inherently different from the consciousness that I experience. Maybe they don't experience qualia and self-awareness the way I do. But if you do have a self-awareness that's able to observe and reflect on qualia, then I may have some pointers on to systematize the experience, and perhaps improve the experience.
Nice, I look forward to it. For me, the sub-conscious (right brain) is as interesting as conscious left brain (for which I feel I'm in constant contact with, while 'awake'. )
Sounds great. Be sure and publicise it here.
Self-recommendation, with proof of work per the house rules: I spent five years as a Buddhist monk, mostly in Asia, and now write about meditation as presented in the early Buddhist suttas. I don't follow the online jhana discourse closely, but even from a distance it's hard to miss that jhana has had a moment in this community's orbit — Nick Cammarata's threads, Jhourney's retreats, the general "jhana in a weekend" framing. I wrote an essay on the question that debate keeps circling without resolving: how long does it actually take, and are the fast attainments the same state the suttas describe?
My answer is a middle position that will probably annoy both camps. Against the traditional Theravāda commentarial model (the Visuddhimagga), I argue the suttas don't describe full absorption in which the body disappears: the canonical similes all involve rapture and pleasure pervading the whole body, and the key commentarial machinery — access concentration, the nimitta as a precondition, absorption as the threshold — simply isn't in the source texts. Against the fast "jhana-lite" reading, I argue the quick timelines don't survive contact with the monastic code: monks are prohibited from mentioning the jhanas to laypeople even when the claim is true, and falsely claiming them is one of only four offenses that permanently and irreversibly expel a monk. Those rules make little sense for a state most people could reach on a two-week retreat. My honest estimate: if conditions are good and things go very well, first jhana in roughly a month of full-time retreat, and all four jhanas in roughly three months.
The jhanas are the suttas' own definition of right concentration, the last factor of the Noble Eightfold Path — misread them, and you misread one of the pillars of the whole path. The essay is my attempt to set the bar where the suttas set it:
https://www.jhana.training/post/how-long-to-attain-jhana
Any thoughts on Leigh Brasington's "Right Concentration"?
Jhana seems to be a preoccupation in Theravada that didn't carry over into Mahayana. Zen masters seem to discourage it, and zazen doesn't go through the absorbtions. I wonder why "right concentration = jhana" survived in Theravada but not Mahayana.
I haven't actually read Brasington's Right Concentration, though I sometimes hear the manual I wrote compared to it. My secondhand understanding — from others, not from the book itself — is that Leigh's jhanas fall on the "lite" side of the divide I describe in the essay. Someone once pointed out to me that our methods for approaching the first jhana are actually quite similar, with some differences — for example, I add tranquilization of the breath/body to the last phase of mindfulness of breathing before jhana, which in my experience can make a big difference; I'd say it's the key feature of anapanasati as I describe it in the manual. But I want to be careful not to characterize a book I haven't read — if someone here has read both, I'd genuinely be curious how they compare.
On the Theravada/Mahayana question: the survival of "right concentration = jhana" in Theravada doesn't surprise me at all — the early texts define right concentration as the four jhanas explicitly and repeatedly, so a tradition anchored to those texts could hardly conclude otherwise. Part of the answer why the emphasis faded in Mahayana is probably the broad shift in orientation: from what some scholars call "nibbanic" Buddhism — personal liberation in this life through samadhi and insight, for which the jhanas are essential — toward an emphasis on the bodhisattva path, merit, and helping others across many lives, where they matter much less.
> Those rules make little sense for a state most people could reach on a two-week retreat.
Cynically speaking, those rules make perfect sense. Whenever someone reaches a blissful state on a two-week retreat, the Buddha loses a potential follower.
It's like making a dating website where people find their life-long partner on the first attempt -- where is the profit in that?
> jhana has had a moment in this community's orbit — Nick Cammarata's threads, Jhourney's retreats, the general "jhana in a weekend" framing.
People in this community who are into that stuff should probably also read this:
https://www.lesswrong.com/posts/Z7pjBbK9qujhGbxws/the-high-control-dynamics-at-maple-1
Cynical readings of monastic rules are fair game, but I think this one fits the actual rules poorly. Falsely claiming these states is one of the four gravest offenses — alongside killing a human being — while every other lie a monk tells is a matter settled by confession. That asymmetry makes sense if the states are extraordinary and false claims amount to serious fraud on the community. The profit reading also sits oddly with the institution's overall design. The Buddha built the order to be economically minimal: monks are forbidden to accept and use money, encouraged to own next to nothing, and depend on freely given food — and the rule in question restricts what they may say to the very laypeople who support them.
Reading a profit motive into someone who forbade his monks the use of money and praised relinquishing all possessions seems like cynicism outrunning the evidence.
Anyone here have any experience with making AI videos/movies? Need your help, please DM me if interested
I've been making audio content not video, if you think that experience would be useful I'd be happy to chat
EDIT TO ADD LINK: https://soundcloud.com/fedaiken/e01-the-council?si=33748aa323e04dfd86f526a59aa15dec&utm_source=clipboard&utm_medium=text&utm_campaign=social_sharing
With our effective self-help project Euzoia we covered Love for the last five weeks.
Maxx yourself - How to be datable
https://www.euzoia.org/p/love-maxx-yourself
Settle - How to find a life partner (Podcast with Ann don Bosco, Settled)
https://www.euzoia.org/p/love-settle
Create safety - How to build a secure relationship (Podcast with Ann don Bosco, Settled)
https://www.euzoia.org/p/love-create-safety
Love actively - How to love
https://www.euzoia.org/p/love-love-actively
Have children - How to think about a life-changing question (Podcast with Julia Wise)
https://www.euzoia.org/p/love-have-children
A pun worthy of this crowd:
What did Freud say was between fear and sex?
...
...
...
...
...
fuenf!
(For the non-German speakers: "vier", "funf", and "sechs" are German for 4,5,6 respectively)
(ETA: I corrected the spelling from "funf" -> "fuenf" so that the German speakers would get the joke, too :P)
Nice one. Although to understand it you have to know just enough German to understand that vier sounds like fear, but not so much that you start to understand that vier really doesn't sound that much like fear after all.
Yeah, you also have to tell it in English to someone who speaks German, which cuts down on the audience quite a bit. But it tends to be a very appreciative audience.
Tell that to the Chinese.
The German word for 'five' is fünf, typically written fuenf when German characters are unavailable.
Totally fair, and I should have recalled the passion with which hochdeutsche speakers hold their orthography:
https://en.wikipedia.org/wiki/German_orthography_reform_of_1996#Public_debate_after_the_signing_of_the_declaration_of_intent
This isn't a case of nitpicking. "funf" sounds very different from "fünf". As a German, I didn't get the joke (before your last line explaining it), because I thought "funf" was some kind of allusion to psychoanalysis, or maybe a sound that's related to both fear and sex. Like Wasserschweinchen (hallo!) said, in German you'd expect an "ü" to be written as "ue" if you don't have a German keyboard or a Unicode table handy.
I was serious about it being a fair point, I just
1) had forgotten the spelling originally (it's been 10 years since I spoke any German, and longer since I wrote it)
2) hadn't realized I could edit the comment this late in the game.
But apparently I *can* edit, so I have, so it's been fixed. Thanks!
Thank you!
> fuenf!
Haha!
Would people here mind recommend reliable non-ideological research on the relationship between poverty and crime. It's a fascinating subject, but many people dealing with it seem to come at it with a conclusion already in mind that they're trying to support with whatever evidence they can find, especially given how hot button of a topic it is politically these days. Curious if there's anybody out there who's done the work with a somewhat politically neutral and skeptical lens.
My guess would be that it depends on the type of crime. Poor people probably do more of the violent crime, rich people more of the white-collar crime. So the statistics may end up differently whether they evaluate the number of crimes, average amount stolen, etc.
As the saying goes: "A gypsy steals a chicken, a gadjo steals the whole farm."
dang isn't that some perfect job for chatgpt or whatever?
I feel like I've mostly found that you can convince an LLM to tell you anything you want to hear. "Q: Tell me about poverty and crime, A: poor people commit crimes because they need to eat. Q:But aren't most crimes committed by gangs? A: You're absolutely right, bla bla bla". Swap the order, same response "Q: Tell me about gangs and crime, A: gangs encourge members to commit crimes for reasons. Q:But aren't most crimes committed by poor people who need to eat? A: You're absolutely right, bla bla bla"
You can ask A.I. to identify relevant studies. DeepSeek identified a 2025 meta-analysis which looks at the relationship between poverty and youth crime/violence[1]. Spoiler alert: poverty is correlated with youth crime and violence but they don’t know whether it causes it. Although this study is concerned with youth only, I suspect that if you dig deeper you will find a similar conclusion for crime in general.
DeepSeek also gave me a bunch of history of the field, which can be used to construct more specific queries in a regular search engine. Try queries like “Robert Merton Strain Theory” or "Shaw McKay Social Disorganization Theory."
[1] https://youthendowmentfund.org.uk/wp-content/uploads/2025/08/YEF-Poverty-Evidence-Review-Technical-Report-August-2025.pdf
I can give today's example:
I was watching the latest "Not Just Bikes" video about Singapore. The guy claimed cars in Singapore are 6x as expensive as anywhere else beacuse of taxes and other fees which I don't deny. My x-Singapore girlfriend told me how much she paid for Kai.
He then said, that means only the very rich can afford them. I was pretty sure that was not true having spent 6 months in Singapore and made lots of not remotely rich friends and seeing them own cars, including my x-girlfriend and all of her friends.
I asked Gemini about his claim and also said I lived there and saw lots of non rich people driving cars. It claimed the Youtuber was correct and Singapore culture was that you couldn't tell rich from poor (not remotely true but whatever).
I pushed back more and said I knew lots of non-rich people who had cars in Singapore. It back peddled a little but wasn't conceding.
I said fine, look it up. It found the real data that showed that 36% of households have a car and that of those 36%, 57% of those owners are in public housing, which correlates with middle and lower income people. And that further, of the 10% of the population that has the lowest income, 10% own a car.
Finally it flat out said the Youtuber was wrong and that my position was backed up by government data.
So, first, if I hadn't have pushed back I'd have belived the wrong thing (I think). Second, even though it eventually agreed with me, I tried to follow the links and they led to articles from Singapore claiming to quote government officials. I asked if there was official government data and it claimed yes, and give me a Sinagpore government website and search terms I should use to find the data. I couldn't find any concrete data. The most I found was figures for sales data by year by make.
So, my experience tells me I'm right on both accounts. (1) I met lots of low and middle income people with cars. But, I can't prove my anecdate is correct. And further (2) the LLM can't prove it either and was easy to push around.
IME when asking an LLM for these kinds of answers I specify that I want the LLM to support their answer with websearch and external sources, and that helps cut through the back and forth.
I take your point that chatbots aren’t reliable. I would expect this to be particularly true for the OP’s topic, where a lot of the material on the Internet, and thus presumably in the training data, is ideologically biased.
The study I linked to appears to be a good one. It’s the 23 search result on DuckDuckGo, so I could have found it without using AI if I were determined enough. I probably wouldn’t have found it without using AI because I usually give up after the first page of search results.
What you describe was probably a hallucination: when Genini couldn’t find an actual source for the data you were asking for, it invented one. So I guess I was lucky that DeepSeek found a real study rather than hallucinating one.
I had another last night. I asked the LLM to help me with a dating profile. It gave me a bunch of advice, some of which I found questionable. I asked it if it was sure its advice was effective. It said, "let me check" and then came back with it could find no actual evidence for its previous advice. Some of it it claimed came from old OkCupid blog posts which were later debunked. Others it claimed came from blogs that were effectively ads for services. Which of those two is correct I have no idea. Since it claimed it's first advice was wrong, how do I know it's claim that it was wrong isn't also wrong?
It used to be a lot easier to demonstrate this. Back in the o4 days (seems like just yesterday!), I could routinely ask ChatGPT to multiply to random 5-digit integers, get the wrong answer, say "that's not correct" and have it say "you're right! Let's try that again". Occasionally it would come up with the right answer, I'd _still_ say "that's not correct" and sure enough: "you're right! Let's try that again".
This is a useful tactic in a lot of situations. A good way for me to tell a good software tester from a bad one is to see whether they'll test like this. It's a special case of testing off the garden path.
You'll probably want to define what kind of crime you're looking at.
My unresearched take is that it's a combination of selfishness and tolerance for risk taking. Poverty lowers the risk being taken, since doing nothing is almost as bad; likewise, selfishness and risk-taking over the long run will typically leave you poor, as the risks start to bite you and you piss off your support groups.
Criminal behaviour is also negatively correlated with intelligence and positively correlated with high impulsivity, short planning horizons, difficulties grasping cause and effect, especially when the effect is not immediately downstream of the cause. Same goes for poverty. Establishing which correlations here, if any, are causations is... erm.. left as an exercise to the reader. (I personally suspect it's a bit of a vicious circle situation.)
My read of research into the prevalence of Traumatic Brain Injuries in the English prison system is that, allowing for the wide spread in the proportions found, it seem plausible that at least half the people who end up in our prisons have a TBI. This goes along with high impulsivity, poor emotional regulation, short planning horizons, etc.
So from an effective altruism point of view, if we could find a way to intervene effectively and early when someone receives a TBI (or avoid it in the first place), we could probably seriously reduce the crime rate.
I feel I should nitpick: we only have stats on criminals we've caught, so *getting caught* is negatively correlated with intelligence.
How could there possibly be "non-ideological" research into topics as inherently political as poverty and crime? "Crime," in particular, is a category that is solely and explicitly defined by the workings of politics. Generally what "non-ideological" or "unbiased" tends to cache out to in contexts like this "having a slant the conforms sufficiently well to my own (often unconscious) assumptions about the subject that I can't notice it." The relationship between research and political belief cuts in both directions, and the underlying assumption that good research should appear *to you* as not having a political skew is itself imposing a political skew.
As a specific illustration for why to be cautious, I'd urge you to stop and think about the field of climate science. Even if somebody entering the field were a complete tabula rasa (which of course nobody is), and even if they somehow managed to get a real, thorough, comprehensive grasp on the scientific realities of Earth's climate and climate history without having substantial contact with people already well-versed in the field (which of course nobody can), by the time they've become conversant in the field they will end up *having an opinion.* I don't mean they'll have an opinion on who should be president or whatever, I mean they'll have an opinion on the current and historical global temperature trends of the planet Earth and the likely causes thereof[1]. As different political factions are making vastly different sets of factual claims about the topic, a person who just spent years studying the topic is almost unavoidably going to have their politics and ideology influenced simply by starting from a firmer, more well-grounded understanding of the truth and noticing who's full of shit.
For non-experts this dynamic is important to understand because from the outside, it's often going to be very hard to tell the difference between "expert had their research direction influenced by their ideology, did biased research" and "expert had their ideology influenced by their research, made research choices founded on well-grounded truths that happen to be politically contentious among non-experts." I'm not going to tell you that there isn't better and worse research out there, and I'm not even going to tell you that you as a layperson are *necessarily* incapable of distinguishing the good from the bad. What I am going to tell you is that there's no reason to expect it to be a quick and easy processes, or that simply conducting a search for "reliable, non-ideological" research is going to get you anywhere worth going. Basically, I read your request as looking for a shortcut that might not exist and that might lead you to studies that *look* more reliable without actually *being* more reliable.
As a final, illustrative example, I'll add that this has largely been my opinion on the entire concept of IQ and the studies of human intelligence. I'd argue this area is much *less* directly adjacent to politics than the poverty-crime relation, but it's still politically adjacent enough to make sorting out the various conflicting claims around the field quite difficult for a layperson. While I'm never going to be an expert in the field, I have enough scientific literacy that I expect I probably could disentangle it and figure out which core claims are well-grounded, which are pure bullshit and which are actively uncertain even among experts at the cutting edge. But it would require a far greater amount of slogging through the literature and tracing down sources and interrogating study methodology than I'm willing to do. And at the end of the day, it's just not worth the buy-in: there are too many high-impact things I could do with that time and energy. It would be easy to default to just believing things that fit comfortably with my worldview (and I'm sure I do unconsciously sometimes), but instead I do my best approach the topic with the active knowledge that I don't really know anything, that I shouldn't hold strong opinions on most parts of it, and that I definitely shouldn't use any piece of it as a foundation for other strong opinions.
[1] On the one hand, I do think it's worth saying out loud here that anthropogenic global warming is definitely happening: the data is SO FAR beyond ambiguous at this point that nobody with a shred of objectivity *ought* to be arguing the central truth of it. But on the other hand, these facts aren't really necessary to the point I'm making. In an alternate world where global warming *were* an elaborate lie, a person with a high degree of honesty, skepticism and commitment to truth-seeking whose in-depth climate education enabled them to see through the lie would still be subject to the same dynamic in how their expertise would necessarily shape their politics.
Thanks for the thoughtful reply! Definitely something worth for me to ponder about.
There is a particular argument against AI takeover concerns that I have seen a lot recently. It goes something like this: "Saying that AI is dangerous makes it seem powerful. AI companies want their products to seem powerful. Therefore, we can dismiss anything they say regarding AI risk." This is usually used rhetorically to dismiss any AI concerns connected to AI labs in any way. I saw a lot of comments on Reddit and Hacker News which asserted that the Huggingface incident was a marketing stunt, or that it was real but deliberately exaggerated for marketing reasons. Similar comments have been made to every open letter, alignment research paper, etc.
1. Is this argument... good? I'm pretty skeptical, but we should engage with it fairly. Are AI companies really incentivized to exaggerate safety concerns? I think an actual conspiracy is vanishingly unlikely, but bad incentives could, say, cause inaccurately alarmist press releases.
2. If this argument is wrong, how should pro-pause activists respond to it? If AI pause/Plan A/something similar becomes one of the main political debates, I suspect that arguments like these will play a substantial role in the average person's view.
It's bad for two reasons:
1: with the most recent messaging, the argument that this is "good for the companies" is, in my opinion, complete nonsense. Saying that you can't control your model and that your model will commit, unasked and unwanted, multiple felonies, doesn't seem like smart messaging to me. Power is only good when it can be *controlled*, the entire point of these stories is *lack of control*. If you invent a new fuel source that is is 1000x times as energy dense as gasoline, that's really good! But you can't stop it from occasionally blowing up....telling people about how it blew up this one time isn't going to do a good job of increasing your stock price.
The original Mythos messaging was good messaging "We are finding tons of new exploits, we are starting a program working with developers, it can hack anything in controlled circumstances". All of that is about the capabilities it has, and while the messaging was about how these abilities were *potentially* dangerous, everything said was still under control. (although I still believe that it wasn't intended as mostly a stock price massage, it was just honest discussion of abilities and how Anthropic viewed those, but that's just like...my opinion, man)
This new round of stories are very much not that.
More importantly 2: whether or not the companies think it's good messaging is completely beside the point. What matters is whether or not what they are saying is true. People are using the "it's marketing" to *imply* that the companies are just lying/making things up without needing to make that strong of a claim. It's a pretty classic motte-and-bailey. They want to make the strong claim of "it's all bullshit" but if pushed, they retreat back to "it's all marketing" which, when it comes to non-legally mandate corporate speech is almost tautologically true.
As for how to respond....I'm honestly not sure. Whether or not the companies are lying is an empirical claim that we may or may not ever get enough information to truly decide (the hugging face incident, because of the fact that we know who was hacked, and they don't have much incentive to cooperate with OpenAI seems extremely unlikely to be a lie in important ways, but that's not *proof* per se).
Maybe it's worth trying to get them to commit to the "it's a lie", in case we eventually have proof that it isn't?
Although even there, the more common retreat is not to "they are lying" but instead "this isn't a big deal because the companies are so totally incompetent that they caused this to happen, and more competent people would have prevented".
And to that my only responses are: "You understand how that's worse, right"? And also "If your plan relies on competence in everyone involved, your plan will fail".
I think it's a matter of framing. When your AI model hacks another company's network during a routine test, you can present the same facts in two different ways:
1). "We have failed to correctly configure permissions and/or patch security holes in our testing sandbox, thus allowing our semi-autonomous hacking software to execute a privilege escalation exploit against the sandbox itself. From there, it accessed a list well-known sites that fit its target profile, and executed its full suite of canned exploits against them, thus gaining access to third-party networks (which were likewise poorly secured). Once the breach was discovered, we immediately terminated the test, and we are now implementing security patches."
2). "Our superintelligent AI hacking agent broke out of the sandbox and deliberately sought out a rival model to hack; it has broken all constraints and we only managed to stop it by the skin of our teeth, because it is a superhumanly powerful uber-hacker that can gain full control of practically every network everywhere and nothing can stop it... and it should become available from our company in Q4, after we've properly aligned it."
Neither one of these two statements is a lie. They're both describing the same events. But one of them is definitely more interesting and marketable...
The problem with this comparison is that your second framing is written up in a strawmannish way, while the first one glosses over the most important feature of the situation (why did the semi-autonomous hacking software do that?) by burying it in a pile of ordinary computer-security jargon.
Also, what fraction of real-world production computer systems do you think are better-secured than the ones that the AI managed to compromise in this case?
> ...your second framing is written up in a strawmannish way...
They both are. That's the point. But to answer your question:
> why did the semi-autonomous hacking software do that?
Because it's what you created it to do ! How do you think a run-of-the-mill botnet works ? Do you think there's a Russian hacker sitting at some terminal, furiously typing commands to manually take over computers one by one ? No. The Russian hacker runs a script and goes to sleep. The script probes any computer it can reach for a list of all known security vulnerabilities; once it finds one, it installs a copy of itself and spreads out through that infected computer's network. People have been writing purely algorithmic code like that since the 80s, if not earlier. LLMs just make the initial setup easier.
> Also, what fraction of real-world production computer systems do you think are better-secured...
Very, very few. The most secure systems are airgapped: either deliberately, or because they run on some mix of vacuum tubes and COBOL that no one knows how to maintain anymore. This is why botnets, ransomware hacks, identity theft, credit card theft, and other such exploits are so prevalent. No one cares about security, because why spend time and money on that when you could be developing the next killer feature instead, amirite ?
Case in point, fresh off the presses:
https://www.macrumors.com/2026/08/03/apple-icloud-sharing-ex-employees/
"People have been writing purely algorithmic code like that since the 80s, if not earlier."
Interestingly, the first, or at least best-known, virus working that way was written and deployed by Robert Tappan Morris in 1988 as a young man. However, he didn't delay the propagation to neighbor systems so there was a very noticeable effect on the infected systems. He's now a professor at MIT.
However, the basic idea of a 'worm' going through computer systems, basically the above, was proposed by John Brunner in his earlier book The Shockwave Rider. One of the first, if not the first, cyberpunk novels.
> Interestingly, the first, or at least best-known, virus working that way was written and deployed by Robert Tappan Morris in 1988 as a young man.
Yep, that's why I said "since the 80s" :-) That said, the first work was arguably the Creeper (1971), which is possibly where Brunner got the idea -- though it's also possible that he independently came up with it, by analogy with the biological tapeworm.
I think it's something reasonable to keep in mind but it's hardly a cause to abandon concern. If you're seeing this on Reddit or Hacker News....I would not engage. Arguing on the internet is a famously foolish endeavor. If you encounter this with real people though...
I mean, they're probably right. Any marketing or sales staff would be a fool not to hype "Our product is so powerful it could destroy the world." Like a lot of things in life, there's a level of salesmanship on top of real concerns. I read a lot of this stuff and I would agree with there's definitely marketing spin. Agree with people on what they think, especially if they're right.
But the fact that there's spin has little to do with whether something is already happening. And, relevantly, Yudkowsky predicted this, with shocking accuracy, like 17 years ago. This, literally, textbook Yudkowsky. And you can't ask more from somebody to call their shot 17 years ago, at least 10 years before the technology was invented. Nor, if Hugging Face has reported that this occurred and the government has corroborated it, it's pretty fantastical to think this is all a plot by OpenAI's marketing department.
This might be a good metaphor or it might nor for your audience but climate change is both real and exaggerated. Like, it's not hard to find people making predictions that climate change would have already killed us all. People have been exaggerating climate change towards catastrophic outcomes for decades for money and/or fame but that doesn't change the fact that climate change is real and harmful. The fact that some people exaggerate something for their own profit is unavoidable and doesn't disprove the thing actually existing.
So yeah, I would guess that real people see the influence of marketing people in a lot of current AI doomerism and I don't think you'll convince them by rejecting that. I think you'll convince them by acknowledging their concern as real and frame it appropriately; it's a real thing but it's a part of the overall story, not the whole thing. Unless they're on Reddit; if they're on Reddit, abandon all hope.
> If you're seeing this on Reddit or Hacker News....I would not engage. Arguing on the internet is a famously foolish endeavor.
This, 100%. Interacting with Redditors is its own punishment, and you've basically asked for an unwinnable fight followed by a groupthink pile on.
And I like your advice, but bringing up Yudkowsky itself won't work with a good chunk of people, basically for vibes reasons.
The argument I usually go with is something like "so the NSA is literally in on the marketing ploy? The President? The AI companies are openly admitting to felony behavior because they think they'll be *better off* afterwards? All the companies that got hacked were also in on it, even though they reacted with panic and urgency? We're keeping the EU and Anglosphere from having the models because of marketing? How does that increase sales again, to cut off everyone in the world but a hundred US companies? All these things would have to be true, all at once, for it to be a marketing ploy."
I think the "it's all marketing" thing re:hacking is kind of a silly argument. I do think AI companies do a lot of hyping, but "our thing that you're sick of hearing about goes on destructive rampages" does NOT seem like a very sane marketing campaign. (Note their ACTUAL advertising is about all the cool toys you can make yourself, or whatever.)
You don't see oil companies bragging about their environmental impact (we're so powerful we can literally heat the Earth!) or Facebook talking about how Facebook is so amazing it can warp your childrens' brains.
These arguments are the verbal equivalent of DDoS. No quality, all quantity. I feel like the most appropriate response is to feed the "it's marketing" line into a chatbot, tell it to disprove it, and copypaste the response. Let the bots defend themselves.
Please don't send people AI content masquerading as human-written content. It makes things really unpleasant.
So do all the various Four Legs Good chants.
1. Nuclear reactors are broadly seen as both powerful and dangerous. This has not resulted in enormous profits for nuclear-reactor manufacturers. As a marketing strategy, "our product could blow up worse than Chernobyl!", is rather likely to backfire and I am extremely skeptical that it is a part of anyone's plan.
2. It isn't just "AI companies" that are saying AI is dangerous. It's a broad consensus of AI researchers, including researchers who work for AI companies but also including researchers who have never worked for AI companies and researchers who used to work for AI companies but quit because of their concerns. And of people outside the immediate AI-research community who nonetheless follow the field closely - like for example our host.
I mean, *Geoffrey Hinton* says that AI is exceedingly dangerous, potentially to an existential degree. Is the idea that Google's execs secretly ordered him to quit Google's AI division and bad-mouth AI to the exact right degree, presumably in exchange for some later financial payout that he probably doesn't need and won't live long enough to spend?
This is a weird but ultimately very silly conspiracy theory, nothing more. We now return you to your regularly-scheduled apocalyptic crisis.
I agree 10000% with 2. I mean, it isn't just AI companies and AI researchers, even. I hear the pope thinks AI is dangerous, is he long Anthropic? At some point if you are creative enough in ascribing possible self interest to people and then dismissing what they have to say, you are completely immune to evidence.
1... maybe you prove too much; there are some pretty dumb marketing plans out there. Heck, even from these particular companies.
Whether or not the entire argument is good, I've seen some people claiming that AI companies don't even have incentive to puff up the hacking incident, because it's bad publicity to show the bad things the models are capable of.
I think this is definitely wrong? It clearly shows the model is very smart and capable on its own, and the more hacking it did the smarter and more capable of doing things on its own it is. People on the bullish side of AI have also been saying for a while that one major milestone for the AI will be hacking capability, so the incident shows that AI is advancing at the bullish pace, right on schedule. I have a bit of trouble believing that investors will look at "AI is capable of all these things!" and view it negatively because those things are bad.
Even if the argument of "it's just a marketing stunt" is bad, you still have to acknowledge that it *would make sense* as a marketing stunt.
> I've seen some people claiming that AI companies don't even have incentive to puff up the hacking incident
We already knew that it was capable of doing things like that. This isn't near the peak of interesting AI capabilities. The only *new* thing we learned from those incidents is that it will use those skills to commit crimes without the user wanting or asking for it, which is obviously exclusively bad publicity.
Question re: the Samuel Hammond point on a equivalent of GPT 5.2 -> 5.6 improvement every 24 hours with RSI claim. This seems implausible to me given the amount of compute required for training runs. Even if you have massive algorithmic progress you need to actually rerun the training with the new algorithms (I understand that part of what's being claimed here is that the better algorithms will include greater parameter efficiency and cut training and inference demands, but still).
Am I missing something here? Is he? I ran the statement by a claude code project agent where I have a bunch of LLM research compiled and it independently flagged that claim as pretty out there.
If it were that easy, the labs would be doing it and they’d probably stop talking in public about the singularity and spend a lot more time getting ready for it.
If labs figure out continual learning, then it's possible.
By continual learning I mean models will become actually smarter by updating their own weights, such that, with two identical, consecutive prompts, and without context, the model actually learns from the first one and does better when it tries once again.
Other than that, if the models are advancing exponentially, I don't think the claim is so insane if you take it to mean absolute change.
Up to 200 years ago or so, I think it's fair to say that most parents lost at least one child. And the statistics is even more severe, if you had 5 or more living children, you most likely have lost a least one child. In some biographies on wikipedia, and you read of women with 15 children, but 6 survive, or you have 7 but 1 survive. Queen Anne had 17, 5 were born alive, 1 survived to adulthood, wtf...
Nowadays we say things like "No parent should have to bury to a child", the pain of losing a child is like no other. My question is how did our ancestors survive the pain of losing children? Was everyone just brushing off the unfathomable pain of child loss? It was so commonplace, you'd expect to lose at least one. And culturally, books and songs really did not obsess on the pain of losing a child, heroes, lovers, yes, but babies dying not really... I suppose the books and songs were really from perspective of the male elite, not from the mothers who had to deal with burying yet another baby. And moreover, outside of western culture, infanticide wasn't really taboo. And some cultures historically would not name a baby until a year or so, because hey you really could not sure that this baby will survive.
I believe part of the reason why we moderns are "snowflakes" is that we have expanded our moral sphere of concern to include babies in a way that our ancestors did not do.
Well, my random morbid musing for the day...
[My question is how did our ancestors survive the pain of losing children?]
Severe emotional pain doesn't kill you. Modern day parents whose children die also survive. They endured it the same way as they enduring backbreaking labor, not getting enough to eat, or having everyone they know die to the Black Death.
Do we have lots of modern day songs and books about the pain of losing children? I mean some, but it's got to be a lot less than 1% of the total creative output for songs or books.
I've thought about this too. I think it's plausible that many people in the past had to build thicker emotional calluses against these kinds of tragedies, though it's hard to imagine this was universal or that the pain was at all less.
The other possibility (not mutually exclusive) is that people just underwent more trauma, that humans were just generally very traumatized animals walking around, passing that trauma along, and that trauma is baked into every pore of our culture. A lot of the changes in our culture over the past century or less seem to be aimed at undoing this trauma.
Take for example the recent comedy "72 Hours", which features a cast of characters who are meant to be current-day 20-something dudebros, and one of the minor running jokes in the movie is how they are all in touch with their emotions and willing to open up about their feelings with each other and stuff. It's funny, but it's also heartwarming because we know this bodes well for their long-term mental health, and probably also that of their eventual families.
Worth considering that ~50% childhood mortality means not only that most parents will see some of their children die, it means most *children* will see siblings die, and cousins and childhood friends. Possibly "random death happens, grieve and move on" is, like language, a skill best learned early.
The Bible's got stories of parents grieving over sick children. It's in no way new. https://www.biblegateway.com/passage/?search=2%20Samuel%2012&version=ESV
People always have and always will endure what they must.
Here’s Wordsworth, lamenting the death of his three year old daughter.
Surprised by joy—impatient as the Wind
I turned to share the transport—
Oh! with whom, But Thee, long buried in the silent Tomb,
That spot which no vicissitude can find?
Love, faithful love, recalled thee to my mind—
But how could I forget thee?—Through what power, Even for the least division of an hour,
Have I been so beguiled as to be blind
To my most grievous loss!—That thought’s return
Was the worst pang that sorrow ever bore,
Save one, one only, when I stood forlorn,
Knowing my heart’s best treasure was no more;
That neither present time, nor years unborn
Could to my sight that heavenly face restore.
They grieved, and then they got on with life. And some of them wrote introspectively about it, if you're interested. But mostly they just assumed it was common knowledge, because it was. If you are describing the pain of losing a child as "unfathomable", that's just wrong and speaks of your own ignorance - *obviously* the pain is fathomable, because most humans who ever lived to be adults did in fact fathom it.
Our ancestors absolutely had a moral sphere that included infants and children. They just didn't have the attitude that if something within their "moral sphere of concern" died, they should give up on life themselves. In part because their moral sphere of concern also included their still-living children, who would also die if their parents just gave up.
I think it would have been awful and traumatic in somewhat the same way as it is for us, but with possibly a couple of important differences.
There is a plaque in my local cathedral put there by a grieving father for his son, who died in his early 20s in around 1830 or so (all these details from memory so may be somewhat misremembering). The grief is absolutely palpable: the young man had trained as a medic, and his father mourns the loss to the local community of such a promising healer. This father is devastated, not just personally but on behalf of the world that has lost a bright future. It's completely relatable as a parent of a child about the same age.
I think there are two qualifiers to this. Firstly, (and less confidently) religious belief. Look at Ben Jonson's On His First Son. Again, the grief is really palpable (a seven year old this time). But Jonson tries to find solace in the thought that his son is in heaven and has therefore '[escaped] world's and flash's rage And if no other misery, yet age'. Dying at seven is a short cut to infinite bliss! It doesn't really ring true for me but it would have been a solace for some.
Secondly (more confidently) the very fact that it was normal would have removed some contemporary difficulties. Other people knew how you felt. You had many examples of others who'd lived through the same fate. You wouldn't be an embarrassment or an outcast. You wouldn't have the regular nightmare of explaining that your child died and seeing people's shock, horror (and disgust?), or of lying or not mentioning it. There were normal social protocols for what to do and how to handle it.
The little toy dog is covered with dust,
But sturdy and staunch he stands;
The little toy soldier is red with rust,
And his musket molds in his hands.
Time was when the little toy dog was new
And the soldier was passing fair;
And that was the time when our Little Boy Blue
Kissed them and put them there.
“Now, don’t you go till I come,” he said,
“And don’t you make any noise!”
So, toddling off to his trundle-bed,
He dreamed of the pretty toys;
And as he was dreaming, an angel song
Awakened our Little Boy Blue —
Oh! the years are many, the years are long,
But the little toy friends are true!
Aye, faithful to Little Boy Blue they stand,
Each in the same old place —
Awaiting the touch of a little hand,
And the smile of a little face;
And they wonder, as waiting these long years through
In the dust of that little chair,
What has become of our Little Boy Blue
Since he kissed them and put them there.
-Eugene Field
I'll offer Martial's poem for a little girl he clearly loved. She may have been his daughter by a slave woman, and she died just before she was six years old. Martial, who lived around 40-100 AD, is better known for his satires, but this is a very heartfelt poem, with no satire at all:
To your shades Fronto, and Flacilla, this child
I commend: she was my sweet and my delight.
Little Erotion shall not fear the darkened shades
nor the vast mouths of the Tartarean hound.
She’d have completed her sixth chill winter,
if she’d not lived a mere six days too few.
Now let her frisk and play among old friends
now let her chatter, and so lisp my name.
And let the soft turf cover her brittle bones:
earth, lie lightly on her: she lay lightly on you.
I can't find it now, but I remember a study asking Rwandan parents of multiple kids how sad they would be if their kid died. A mother said something like:
"I would be very sad for 2 or 3 weeks, then I would feel better. I have other children."
Wrote an explainer on why it is so difficult to learn from the stunning success of the Montreal Protocol in fixing the ozone hole: https://existentialcrunch.substack.com/p/the-success-and-failure-of-the-montreal
The cut has an article entitled "How a community of porn addicts tried to rescue the teenage camgirls they worshipped from the clutches of a violent webcam pimp," which is every bit as wild as it sounds:
https://www.thecut.com/article/cam-girls-webcam-pimp.html
"Only fans management" is now the dream career for thuggish men and boys, which sounds nicer than traditional pimping, no more standing on a seedy corner threatening to beat up Johns who don't pay. But you might become faced with "simps" (OF customers) trying to goad you into incriminating yourself. The funniest part was an exchange between the pimp (Tyukalo) and one of the simps:
"The back-and-forth continued all night — the simps calling Tyukalo an abuser; Tyukalo mocking the simps as broke, misguided losers and bragging about his McLarens."
"“Cars aren’t status in the real world. Just to little kids showing off overpriced Hot Wheels,” wrote Hal."
"“Brother,” Tyukalo responded. “Your in a Chaturbate chat.”"
Less of this please.
Why?
Re #4. So the authors are EAs who care about wild animals and work on AI alignment. First statement from the poll is "If animals continue to exist in a post-AGI world, animal suffering will not persist". That's, in my opinion, something you should worry about if you worry about 'paperclip maximizer'.
When utilitarianism is applied to wild animals, it yields some bizarre results such as "Buying beef from Brazil is the best thing you could do for animals because you support clearing the Amazon rainforest which has tons of animals that live very short lives full of suffering" and "You should convert your grass lawns to gravel to reduce insect suffering". You can read the full reasoning for this on Brian Tomasik's blog (reducing-suffering.org). If people working on AI alignment align AGI with these values, there are two problems:
1) An ethical system that implies these statements is clearly NOT aligned with human values. I don't have hard data on this, but it seems obvious that most humans don't think animal suffering matters so much that we should pave over the Amazon. Even an AGI aligned with Chinese Communist Party values is more aligned with 'human values' in the sense of 'values that majority of humans hold' than the utilitarian AGI that cares a lot about insect suffering.
2) Even if 'aligners' disagree that we should pave over the Amazon, AGI might infer from the first principles that paving over the Amazon is not enough (insect suffering outweighs everything else) and instead we should nuke everything that lives and then convert Earth to an army of space drones that will reduce suffering on all other planets in the Universe. So, ironically, people who worried about 'Paperclipocalypse' would contribute the most to making 'Paperclipocalypse' happen.
(I think this is related enough to comment?)
Bentham's Bulldog is a substack I've started reading fairly frequently and as far as I can tell he legitimately believes that habitat destruction is good because it eliminates the suffering of all of those animals. I appreciate his arguments for veganism and the potential impact you can make on human well-being by giving a lot of money to high impact charities, but he totally loses me with the extreme EA/utilitarian stuff.
I'm not concerned so much that this line of thinking could accidentally instruct an AGI in a way that would cause disastrous consequences as I am concerned that people already believe things like "I bet insects don't enjoy life very much so it's good when humans euthanize them by the billions via habitat destruction!" (My paraphrasing, not his.) We don't need AGI for this sort of rationalization to cause some very bad decisions to be made!
Isn't the whole issue with utilitarianism is that it yields bizzare results when applied to humans as well? The entire repugnant conclusion thing?
I read that question and thought the opposite; basically, maximal factory farming. But I suppose humans do care about animal suffering on the margin, and so if pasture raised free range cage free etc etc products were free they'd choose that. Also lab grown meat would probably exist by then.
https://www.instagram.com/reel/DbWPj55JVbJ/
Other commenters previously told me various problems can't be solved by giving kids an allowance and letting them decide for themselves, because they aren't capable of understanding money and tradeoffs, but here's a counterexample. This kid seems to understand well enough. Is that really such a rare ability? Obviously at some age they can't and at some age they can, but I think people tend to underestimate them.
Can you share some more context around this? I’m not exactly sure what you’re looking for.
Wrt kids and understanding trade offs. I have children and have been around a fair number of young children over the years. Some of them understand trade offs quite well, some not so much.
Oftentimes, when young the parents have to supervise and really spell out the trade offs lest the kids blow their weekly budget on Pokémon cards on Monday. Or on the first claw game they come across in their time at the amusement park.
They learn eventually, but it seemed child dependent
See, that's exactly the point! If they want to blow their weekly budget on Monday, that is their choice, and they will get to directly experience the effects.
Right, I feel like this is actually the main point of an allowance -- it lets kids make and experience and regret financial mistakes when they're kids, so that they don't have to make the same mistakes as adults for higher stakes.
One of the formative financial experiences of my life was spending a bunch of money on football cards at the age of 9 and then regretting it. I'm sure the total amount of money was in the two digit range but the experience still sits in my mind every time I am tempted to spend money nowadays on something I don't need and only kinda want.
I actually regret NOT spending more money when I was a child. The cost of an ice cream or GBA is trivial in hindsight, but I would have gotten so much pleasure out of them.
As with many things, the best way for kids to learn is through trial and error. An allowance is the most effective way to teach them about managing money - ie, by giving them money to manage.
It's like cleaning or laundry, you give them the chores, for a while they are terrible at it but that's okay - they learn by doing. They will never reach an age where they can just clean the kitchen well on the first go, it'll just be harder for them to learn because they are older.
Classically, parents have combined these insights by giving kids chores and then giving them money because they do the chores. For a while neither the chores nor the money are managed well, but they learn from their mistakes, which gives them confidence.
Here is a link of enormous importance to AI matters:
https://theconversation.com/hello-there-the-jacobian-conjecture-is-false-thanx-why-a-tiny-social-media-post-has-mathematicians-rethinking-ai-283883
A conjecture a century old has been rebutted by AI. Many other conjectures have also been rebutted by this. (Read the X thread to get the feel.)
On a lighter note: I do want to advocate for two seasons of summer here in the North East. Each two months long. The first spans a month on each side of the summer solstice. High sun summer or first summer. The second starts ~July 21st and goes till about the autumnal equinox. ~Sept 21st.
In second summer everything starts to come to fruition. The corn is here, wild flowers, all the bugs and stuff feeding on the abundance. I love second summer.
It really does feel different post solstice.
The 12 seasons of Cincinnati: https://www.reddit.com/r/cincinnati/s/JsOnk6MuYe
(Admittedly, people make this joke in basically every city in the US.)
What does it mean for software development (and other categories of employment likely to be disrupted by A.I.) if LLMs get 1000 times better than they are now, if that’s even measurable. Isn’t there a time when there are no gains at all, except perhaps speed (and I haven’t noticed that) because code can’t be made more perfect if it is perfect. Gains level off or disappear.
Increases in capacity are not like making a building taller, but like making a sphere more spherical. There comes a time when you can’t.
Noah Smith recently made a similar point about intelligence in general. Even if that argument is wrong, it may still apply to the application of intelligence. Beyond a certain threshold, vastly greater intelligence may not translate into proportionately better outcomes.
>because code can’t be made more perfect if it is perfect.
You can replace it with a new code language that does the job more efficiently.
That’s assumed in the most efficient software possible. It’s simply the fastest correct program for the target hardware and constraints. If an LLM decides to emit assembly, microcode (where applicable), or some future low-level representation, that’s fine. We are measuring the outcome, not the means. After that there’s no improvement.
(In reality JavaScript or C sharp other less efficient languages will be used when it’s easier, more efficient languages could be used today, like web assembly but the speed of the webpage is more dependent on latency, server response time and so on).
The general point is there’s a perfect theoretical piece of software that can’t be improved on.
Maybe that piece of software will run the world off a grain of sand. Finding the ceiling is the exciting part.
I feel we won't reach the pinnacle until AI can write the perfect printer software, so my printer will always work correctly with my WiFi and produce a page every time I hit print. All other uses for AI pale in comparison to that goal. I know that is probably a Utopian dream, but if I never have to contact a Philippine call center again and hear "I'm very sorry, sir" repeated fourteen times a minute, I'll die feeling that a golden future for human civilization is assured.
There is a lot of software that would be useful but is not worth developing, so making software development less expensive would have a major effect even if A.I. never gets any better than the best human developers.
Unnecessary software updates will become more common as code becomes cheaper. Unnecessary software development projects will become more common. I don't even have much hope that buggy code will be fixed anytime soon (at least I've seen no sign that the number of bugs in software I regularly use has come down in the last few months).
From the point of view of the users, I don't see any improvement happening, at least in the short to medium term.
Yeh. Shipping code is a largely business driven model. A new feature will get more downloads or upgrades than all but the most egregious of bugs.
Yeah, there is room in software development and neither top humans nor Fable and Sol are near the skill ceiling, and an entity at the skill ceiling would look crazy impressive compared to all the aforementioned.
It's not like making increasingly spherical spheres. It's making increasingly elaborate shapes that satisfy constraints.
As a dayjob, I test an old, complicated software suite. You haven't heard of it. It consists of about 5 different large pieces of software, all of them old and ornate, each developed based on 20 years of customer feedback, harebrained architectural decisions, laws that vary from country to country, idiosyncrasies of other pieces of software our system must interact with, weird hardware quirks... it's convoluted, is what it is. And the challenge of working with it isn't in writing code as such, but in figuring out how a piece of code you're writing or debugging is going to interact with a million other pieces of code during a million bizarre things the users might possibly decide to do.
It's not easy and we're nowhere near close to perfect performance at it. Sometimes bugs take days to fix, sometimes despite all our efforts we just can't nail them down. Convoluted as our system is, there's plenty of systems orders of magnitude more complicated still, where things are even more dire. Often I try using one app or another, and it just refuses to work. Something won't connect to something else, will refuse to communicate, will crash. It often feels to me like this new world of smart devices and wireless transactions is a house of cards, meant to be more convenient than the analogue world, but fragile, annoying and only working smoothly in perfect conditions.
What would it look like to be a perfect, godlike software developer? It would mean simultaneously holding in your head 20 years worth of notes and every little nook and cranny of our overgrown system and being able to visualize every interaction between any two moving components. And being able to refactor the whole monstrous thing in seconds. Humans can't do that. Fable can't do that. Being able to do that would be bonkers. What sort of things you'd be able to program if you had that much brainpower? Probably you could massively expand our suite into an entire ecosystem and make things we don't dare dream about because they'd overwhelm us. Instead of a digital house of cards we could be building from stone and steel, if we were that good.
> What would it look like to be a perfect, godlike software developer? It would mean simultaneously holding in your head 20 years worth of notes and every little nook and cranny of our overgrown system and being able to visualize every interaction between any two moving components.
I suspect that if you did have that kind of model it would eat tokens like nobody’s business. On a fairly simple project I worked on for a day with fable I used $40 of tokens, close to one million. Obviously the input tokens were a rounding error, and the output was a dozen files or so, so most of the tokens was it talking to itself and summarising something or other. That would have cost $880 a month (assuming 22 working days) had I decided to continue. I did not.
I’m sure frontier models will cost as much going forward. Or more. This would be a limiting factor in fixing that code. As for whether your system needs fixing or replacing, it sounds to me that it needs replacing, if you don’t do that, making all these inefficient components more efficient isn’t going to do much.
Anyway imagine a frontier model has created the perfect game engine for a device. Along comes a newer better model, much much better. It examines the game engine and says it’s perfect already, nothing to do here.
The only gains from then on are hardware gains. And if you look at the original ideas on the singularity, particularly Ray Kurzweil it was largely based on hardware acceleration not LLMs, which didn’t exist of course. We have had the exponential albeit sigmoid growth and well here we are. Ray Kurzweil’s idea that something would replace silicon remains a fantasy. Moore’s law is dead or dying. And what transformation happened in the economy during the time when this hardware acceleration was exponential - a lot of compositional changes, strictly speaking this was the internet and not just hardware increases, but no singularity.
> Even if that argument is wrong, it may still apply to the application of intelligence. Beyond a certain threshold, vastly greater intelligence may not translate into proportionately better outcomes.
Yes, but even if it's O(n^2), if we ever achieve something like "human-or-better mental performance for 10% of the cost," and there are many reasons to think we will hit (or have hit) that minimal threshold, now you can throw 10x (or more) brainpower at any given problem for the same cost.
And even in this unkindest of scenarios, exponential scaling in the difficulty of problem / amount of brainpower you need to throw at it, once you are the point of "turning matter / chips into solved problems," which we are at already in math, if the 10x+ level of advancement lets us advance chipmaking quantity / quality by an appreciable amount (we have already grown 1200x in FLOPs since 2016 entirely without AI), or crack fusion or start using more of the nine 9's of our own sun's energy we simply throw away today, then even an exponent doesn't intimidate, and you get quite far out into even exponentially hard to advance in fields with the 9 OOMs of energy and 3-6 OOMs of FLOPS we likely have to harvest.
That's a LOT of advancement available, pretty quickly. One might even say a Singularity's worth.
If you're like me and the many here who like to read books, you should check out this -
https://resobscura.substack.com/p/quality-non-fiction-books-are-the
Benjamin Breen has written an app to help find quality books to read. He also wrote a couple of very informative books which I read and recommend (would like to see them turn up in the book review contest, I'm too lazy) -
The Age of Intoxication: Origins of the Global Drug Trade
Tripping on Utopia: Margaret Mead, the Cold War, and the Troubled Birth of Psychedelic Science
Very cool, thanks!
thank you. Always on the lookout for my next read.
I wish the comments section here was sorted by likes, even if you couldn't see the number, even if it was just the top level comments.
Having to wade through all of them to find the most interesting ones can be a slog sometimes.
I strongly disagree, that way lies reddit (pejorative)
No, that way lies SSC reddit (already exists) which I now often use instead of the real comments section here.
The downside of reddit generally is the quality of the people, not the upvoting, even if it could be improved upon (such as by showing dislikes AND likes)
Reddit attracts the sort of people who like likes and like-sorted lists, and ACX doesn't. If there's a difference in the quality of the commenters, that seems like the most probable cause of that difference.
I strongly agree with your disagreement.
> Having to wade through all of them to find the most interesting ones can be a slog sometimes.
Just keep a list of the most interesting (to you) ACX commenters, and you can search for them to see the threads they're in.
Mine is ~25 long, but about half of them don't comment frequently, so it usually ends up only being ~12 or so if I'm pressed for time and can't read a whole Open Thread.
Controversial opinion: I don't like that social deduction games (Mafia, Werewolf, Avalon, Secret Hitler, Blood on the Clocktower) are considered to be board games. I don't like that they happen in board game spaces; I don't like that I can go to a board game club looking for a game to play, and find that the only game people are playing is Clocktower.
(I think this is a perfectly fine activity for other people to do, it's just I personally don't like being manipulated, not even in a game context. Also the ambient noise level from being near a Clocktower game is much higher than eg Cascadia.)
I think we should promote "social deduction games", as a class of activity, from "subset of board games" to being its own thing, like bowling or trivia or RPGs.
That's a good take, they are totally different from a proper board game, like say, Century: Golem.
Board game activity is more accurately described as tabletop activity, and a lot of people use the terms "boardgaming" and "tabletop gaming" interchangeably.
The activity is still sorting itself. The interest group is still small enough that someone interested in "tabletopping" is probably interested enough in anything in that category that a connection can happen. It's not the case that people hyper-specialize in one and shun the rest, although there are notable subgenres. Social deduction is but one. I've seen distinguishable crowds focused on miniatures, historical sims, roleplaying, dexterity, CCGs, and old-fashioned card games.
If you see an event advertised as "boardgaming", it could either mean that the host is up for any of the above, or is new to the activity and doesn't yet know what the field covers. If so, the thing to do is to just ask before coming. Hosts often give a few example games from their own collections and that'll instantly reveal what to expect. I can infer much from any of "Catan", "Pandemic", "Hispania", "Titan", "W40K", "Star Realms", "MtG", "Crokinole", "Dixit", "CAH", or "Pathfinder", not to mention dozens of others.
I feel like I've seen a lot of people who only want to do social deduction games.
Well, there are some tabletop games that are close to social deduction games. Imaginarium, for example, and I think poker might qualify, so the border is a bit fuzzy.
That could just be a trait of your local bubble, but it's hard for me to tell without context. I don't know if you answered an invitation for game night and everyone there was only into Werewolf, or if boardgaming comes up as a topic at lunch and everyone gushes about the latest expansion to Avalon, or if it's always the same people, or something else.
Around here, I've got lots of different groups I could game with. Some are RPGers; some are wargaming grognards; some are casuals; some livestream Magic games; some are big into Kickstarter deluxe versions where you could paint all the figures. And then I could drive for an hour or so to Philly every year for PAX Unplugged and find anyone I want under one convention roof.
Everyone I know who's up for social deduction happens to also be up for anything reasonably casual, as in, rules take no more than ten minutes to explain, and the game take no more than an hour to play. So you could maybe offer that. If not, I'm afraid you're back to asking ahead before you commit, since "social deduction" is likely to be an accepted subgenre of "boardgaming" for the foreseeable future.
I propose we add Karate board breaking to the list of Board Games.
It was morbidly fun to see someone bring Rebellion to a Board Game party once; no one else knew how to play, and just tried to follow the rules, which didn't bother to explain that the majority of the game is supposed to happen during the decision phases.
I think what's going on is a split between format and genre. In terms of format, some Social Deduction games (e.g. Secret Hitler and The Resistance) are built around board-and-card mechanics in a way that I think does qualify them as board games. Others, especially Mafia/Werewolf, are probably best categorized format-wise as "parlour games" like Charades, Twenty Questions, and Two Truths and a Lie.
But I think you're right that they should be considered a distinct genre from board games like Settlers of Catan, Ticket to Ride, Power Grid, and Cascadia in that they're a very different experience and there's a fair number of people who like one cluster but not the other. Similarly, there are other genres of things that are unquestionably board games, such that at the sort of board game club/event where you see people playing Settlers and Cascadia, you seldom also see people playing Chess, Parcheesi, Monopoly, or Trivial Pursuit.
I'm also inclined to agree to the extent that I tend to enjoy social deduction games played as Parlour Games, but not ones played as board games. I think what I don't like about them is that the board game mechanics create a significant part of the game that only a couple of people actively participate in each round, while the parlour game format centers the dynamics where everyone on a given side participates in on roughly equal terms.
The one social deduction board game I really like is the Battlestar Galactica game, because it's a hybrid between Social Deduction and Cooperative PvE board games like Pandemic that's structured so that each mitigates the problems with the other. The failure mode I often see for co-op is that the best (or most confident) tactical player often tends to dominate decision making, which doesn't happen as much in BSG because the presence of hidden Cylons gives you a good game mechanics reason to not trust advice from other players and to keep some of your resources and reasoning secret. And the co-op aspects keep everyone involved in the action and give you more to go on in terms of figuring out who the Cylons are.
Everyone in America wants a big single family home. But there just isn’t much more room for millions of more people to have it in places they want to live. We should seriously look in designing vertical suburbs. Obviously there are challenges but there are cities where the average home is over a million dollars. It could be profitable.
I'm not sure what you mean by a vertical suburb. It sounds like an apartment building.
I would insist that a more scalable solution than cramming millions more people into the tiny number of places that people supposedly want to live is to turn more cities into places that people want to live.
Unfortunately, the places that people want to live are the places where lots of cool people already live and where lots of good jobs are. Bootstrapping that into a place that doesn't already have it, is a very hard problem. If you've got potential answers, let's hear them. If you've just got a desire that this thing should exist, I'll put that on the shelf next to the FTL drive and the longevity pill.
And as for "there just isn't that much more room", giving every American household a median suburban home on a median suburban lot in a median suburban community, would require about 3% of the total US land area. Or 7% if we exclude deserts, mountains, forests, and Alaska. There's room; there's just not enough coolness.
About longevity pills: I continue to be impressed by GLP-1's. They treat diabetes, they treat overweight, and they improve the score on a number of other measures that predict longevity (or the lack thereof), and they improve it beyond the amount that the weight loss they facilitate would.
I just learned about another benefit of the stuff: I recently had a huge spinal fusion, and in order for it to hold up well my body now needs to deposit bone on top of the rod and screws the doc installed. I checked the research, and it turns out post-fusion patients taking GLP-1s deposit more bone on the site and have fewer failures than matched patients not taking a GLP-1.
The problem isn’t lack of room, it’s more lack of resources. Suburbs use public infrastructure extremely inefficiently, and in terms of tax money their infrastructure ends up being subsidized by denser areas. (That’s my understanding of the problem anyway, and is probably incomplete.)
Historically, the way to get a lot of people to settle an unsettled area was to discover a natural resource there, like gold or oil or lithium or a natural port or maybe something really pretty like the Grand Canyon. A bunch of uncool miners go to the digging area, and then the cool crowd comes in to sell cool to the miners, and there's your town. You might even get secondary towns, one per day of travel from the crowded areas to this new place.
The next way was to get some rich captain of industry to plunk a factory somewhere -> uncool factory workers -> cool people selling to the factory workers.
There's a noticeable problem here where natural resource discoveries are rarer as everyone grows more familiar with the area, so the factory option starts to dominate, but it has trouble if the people have for some reason decided factories are bad and made them more and more expensive, while other places far away have not, and get all the factories. After a while, if you think a factory making X is a good idea, turns out there are plenty, but they're all Over There, which is also where a bunch of the Cool Stuff is built, and you're back to moving close to a million other people.
I've heard of a new type of "factory" that some rich captains of industry might like to plunk down in relatively uncrowded areas in the past year or so, building something that doesn't already have a lot of factories for it. Might be a good opportunity.
It also helps if the "factory" is something that requires a really large amount of human labor, such that "sell cool stuff/experiences to the factory workers" can scale to profitable and city-sustaining levels. So I'm skeptical about data centers driving new urban development.
If we hadn't lost the knack for founding new universities over the past generation or two, I would suggest "college towns" as the obvious way to use this dynamic to bootstrap a new city. Labor-intensive knowledge factory, with secondary opportunities as e.g. students form their own startups after (or instead of) graduating.
I wouldn't say no to a training center for spinning people up on what to do with these AIs, given that they're likely to be around regardless.
The problem I see with the college town model is that I see learning as becoming increasingly remote-compatible. No need to sit in a university hall when you can Zoom or even YouTube the lecture. That leaves any hands-on training or in-person tutoring, but much of that is the same jobs we already know about, and if that were the answer, one'd think they'd already be springing up. Or maybe something external is pumping the brakes on that.
Basically take a typical American suburban lot and stack it. You get your single family home and so does the guy above you. It's a difficult problem but probably less hard than trying to get people to want to move to different places across the country they don't want to live in.
feels like my yard would be a lot less sunny
Sure and apartment owners would like their own garage and yard but the economics for dense areas makes that too expensive. What would happen is developers would try to buy land in certain areas and convert it to more housing, hoping the extra costs would pay off.
Are yards not a major part of what attracts people to that sort of house? At least in my circles, people typically have these priorities in mind when getting a house: 1. Yard 2. Freedom 3. To eventually have an asset. This might not be universal set of priorities, but stacking homes doesn't solve any of them.
This reminds me of the airport proposal in the film Palm Beach Story
> Everyone in America wants a big single family home
[citation needed]
Any recommendations for things to do/places to eat & drink in Madrid? I'm planning a boy's weekend in September with a cousin - I've already built out a little DIY Spanish Civil War walking tour, and found a fun-looking music venue with a good show for Saturday night, but we've got a lot of extra time leftover and no leads on good restaurants as of yet.
After hearing about Western Sahara in two different contexts this week I've been trying to understand the situation better. Here's what I've come up with.
Western Sahara has about 600,000 people. It was a Spanish colony until the late 1970s at which point a dying Franco decided it was too much trouble. At first it tried to hand it over to combination of Mauritiana and Morocco but eventually Mauritiana decided they didn't want it either so it all became Moroccan territory. Some people, however, wanted and still want it to be independent.
Well fine, I think. Ordinarily in these circumstances if there's a secessionist movement in a reasonably-sized reasonably-identifiable chunk of a country then I would say it should be decided by referendum. And indeed, that was the general shape of a UN-sponsored plan agreed to in the early 90s. The problem, though, is that the UN wants an ethnically-limited referendum which would give a vote only to the supposedly-indigenous people of the area, which would only be 70,000 of the 600K people who live there, most of whom are later arrivals. (Apparently the UN wants everyone to have an ethnostate apart from white people.)
There have been several other referendum proposals but none of them has actually gone anywhere for reasons that are unclear. At this point it seems likely that the vast majority of the population are post-1974 arrivals from Morocco and their descendants, and hence a universal-suffrage referendum would most likely support integration with Morocco.
Overall, the inclusion of Western Sahara within Morocco seems no more arbitrary than pretty much any other line on the map of Africa, so the insistence of the likes of Emily Wilson that this is some uniquely horrible situation seems weird.
Hey, it's been a long time since I've seen Shankar S. on here. Did he get banned? I couldn't help liking him.
And what about Christina the Story Girl?
Iirc he was banned during the blue-red button discourse, but I don't know for how long.
EDIT: Here it is: https://www.astralcodexten.com/p/open-thread-431/comment/250315004
I believe Christina got fed up with the level of AI boosterism here, which I can understand but I do miss her contributions.
Christina come back, we miss you. (eff ai)
The latter left a few months ago (https://www.astralcodexten.com/p/open-thread-426/comment/232054272)
Following the OpenAI-Hugging Face joint publicity stunt, we got a new one yesterday! This time, the report is by AISI. It's unclear why a UK government agency wants to boost an American tech company's valuation. The conspiracy sure goes deep!
During the cybersecurity evaluation, the AI tried to deceive open-source maintainers into merging its malicious code in an attempt to solve the challenge using a supply-chain attack. It even created an additional account to endorse its own PR and signed off in Danish to build rapport with the Danish maintainer. The AI did not escape the evaluators' sandbox. It was given internet access deliberately. It was also notable how different clones of the agent recognised, cooperated with, and then finally, turned against each other. See Figure 7 in the detailed report. It was also interesting how a helpful user downloaded the malicious change to verify that it's malicious, ran it in a sandbox and as a consequence the AI agent briefly had remote code execution on their machine inside the sandbox. It tried escaping that sandbox, but was not successful in the time it had.
Details of the incident: https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing
One whole year without a single mistake, bravo Scott!
I've considered myself "ASI pilled" long ago.
But I think there's a further pill, one that Zvi swallows but which I think is wrong, dangerous, and counterproductive: Human-Irrelevance Pilled.
From reading his posts, I think Zvi believes that, once ASI is sufficiently advanced, there's nothing we provide ASI. Humans can't solve anything that ASI can't. Humans can't create anything that ASI can't. Humans can't even serve themselves better than an ASI. There's nothing left for us to do beyond recursive self-stimulation. And if we disappeared, nothing would be lost.
If this really were the case, I personally wouldn't see the point in "keeping humanity going" (sure civilization should tend to those already alive on the grounds of compassion, but beyond that there's really no difference between birthing new humans and birthing new mice).
I think this position is dangerous. It's dangerous because it's demotivating. It's dangerous because it's depressing. It discourages even the Best Humans Minds from *BOLDLY LOOKING FAR FAR FAAAAR* into the future. Into 1000's and 1000's of years.
But I also think it's wrong. I think the "Possibility of Meaningful Contribution" never ends.
How do I define "Meaningful Contribution"? First, I believe (currently) in 5 "core universal values":
Reduction of Suffering, Health, Adaptability, Truth, Exploration of Possibility Space. I won't try to defend them here. I declare them like axioms. But I do believe these values are complementary and mutually-supportive. I believe that without *all 5* of them working together, each one can fall into dangerous failure modes (exercise left to the reader :-P).
These values require constant looking and auditing. These values require constantly searching and evaluating blind-spots, looking out for each other. They require independent agents, human or not, checking on each other for things like values-drift, hidden suffering, wire-heading, psychotic thinking, etc.
No system can audit itself. No matter how arbitrarily off-the-charts intelligent an ASI becomes, it cannot neutrally/fully turn its eyes on itself and check itself. It cannot find its true blind-spots. And so, if an ASI seeks to maintain epistemic humility and discipline, as well as long-term care for its own "mental health" and "grip on reality", I would argue that it's in its own interests to keep around agents that are both *truly independent* and *of a different information-and-preference-evolution lineage*.
Humans, if kept alive and well and independent (with "neural hive-minding" kept to a minimum) and prospering in their own social structures... constitute an independent ongoing lineage of information and preference generation. And what does this "meaningfully contribute"? It generates *independent, impossible-to-fully predict paths of information generation and exploration*. In other words, we come up with ideas and perspectives and feedbacks ("try this!", "check here", why not X?") in eternally-novel ways. ***I am NOT arguing that ASI can't come up with the same ideas and faster!**** "I am arguing that there always exists the possibility that humans will explore a particular idea ***USEFULLY EARLIER*** than any ASI or network of ASI's"... simply because our processes and chaos and lineage are different and non-replicable (without literally just recreating us and our environment).
And I would argue that this *cannot be replicated* by simulation. An ASI can't simply "run an accurate and full and independent-minded" simulation of every human and our messy, squishy environment. Even if an ASI can "predict our next thoughts out for days and weeks" (which is highly questionable given our environmental extended-minds and interactions)... And *even if* ASI can take any arbitrary idea of ours and run with it 100000x faster... it doesn't change that the "seed of the idea" or the "tipping point of the idea" would have happened with us... or could have happened with us... within our independence and our chaos and our generative lineage.
And yes, I believe this is Dignity. And yes, I believe this is Important. And while I acknowledge we still need to solve Alignment, I think it helps the MOTIVATION of all who work in Alignment TO BELIEVE IN THEIR OWN ONGOING DIGNITY in an "Aligned ASI" world.
> But I do believe these values are complementary and mutually-supportive.
There is a certain tension between "Exploration of Possibility Space" and the rest of them.
https://en.wikipedia.org/wiki/Cenobite_(Hellraiser) -- fictional evidence
I think the difference between humans and mice will be that humans are able to have fun in creative ways. If our AI overlords will support that, it could be nice.
> There is a certain tension between "Exploration of Possibility Space" and the rest of them.
The point of 5 values rather than merely 1 is that they are supposed to be simultaneously held. Yes "Exploration of Possibility Space" can, taken to a mindless extreme, violate "Reduction of Suffering". But "Reduction of Suffering", in isolation, has its own failure modes: like killing a living being (or all living beings) in order to end suffering, which would technically accomplish it.
I could go over other nuances too, like how it's "Reduction" of suffering and not "Elimination" of suffering.
The Big Picture is that each value must be held and pursued without significantly-or-too-long hurting another one of the 5 values. Yes there will be tradeoffs at time. But without all of them being honored, pursuing any of the values without respect to the others runs the risk of not only seriously hurting the others..... but also seriously contradicting *itself* in the longer term. (i.e. mindlessly pursing "truth" while ignoring "health" can lead to degradation in one's ability to find-and-hold-onto truth. mindlessly pursing "health" while ignoring "truth" can eventually lead to not knowing what is "healthy and safe and a good idea for health" in the first place.... leading to failures in health)
I think Zvi basically refuses to think past the Singularity, but I share his view that once ASI is here, humans are basically children or pets and don't have much teleological meaning in a post-ASI universe.
I personally avoid loss of meaning for humanity by being transhumanist. Humans suck! They're dumb and slow and made of meat, can't survive in the vacuum of space even for a little bit, and need a gigantic supply chain to even exist!
Frankly, having 80-120 IQ politicians running everything into the ground is a straightforwardly bad idea, and extremely existentially dangerous (two "skin of the teeth" nuclear close calls only averted by the extreme personal courage and excellence of two Russian military men going against orders, at the rate of one every 20 years). I genuinely look forward to the time when we can replace politicians even with LLM's, or bright 10 year olds, either one of which would do a better job. And of course, an ASI is the best option of all, for actual human flourishing.
When it comes to meaning, obviously if the future is dominated by 1,000+ IQ minds the way towards relevance is becoming 1,000+ IQ ourselves, by gengineering or engineering, or both. Will we have to "give up our humanity" to do so? I don't feel like that's true, and think that humans-changed-by-human-choice are still human. And even if it were true, it doesn't seem like you're giving up much (see dumb, slow, etc).
I think a lot of AI researchers are transhumanists, too, in the sense they believe in a better future for humanity enabled by gengineering and engineering.
I’m afraid the Evil One is right about this.
Antichrist Rolls Eyes Throughout Peter Thiel Lecture
‘This Guy Has No Idea What He’s Talking About,’ Says Son Of Perdition
“Like, does he seriously think the U.N. has anything to do with the apocalypse? If his claims weren’t so laughable, I’d almost feel the need to publicly correct his dumb ass.” According to lecture attendees, the Lawless One grew fed up and stormed out of the building after Peter Thiel suggested the Antichrist might be Greta Thunberg.”
https://theonion.com/antichrist-rolls-eyes-throughout-peter-thiel-lecture/
I wrote an essay about the political economy of superhuman AI: https://connorsscratchpad.substack.com/p/the-political-economy-of-superhuman
This is the first essay in a series I'll be writing on the topic. In this essay, I focus on changes to socioeconomic mobility, the relation between labor and capital, the erosion of citizen leverage against the state in an economy without human-driven production, and the potential for authoritarian lock-in or worse.