Aaron Bergman's tweets, September 2025
Every public tweet, reply, quote tweet and retweet that Aaron Bergman (@AaronBergman18 on X, formerly Twitter) posted in September 2025 and has not deleted: 429 posts, oldest first, times in UTC. Each heading links to the original post on X. When Aaron replies to or quotes someone, their tweet appears as a blockquote with their name; those words are theirs, and everything outside the blockquotes is Aaron's.
This page is one month of the complete archive of Aaron Bergman's tweets, which updates automatically every few hours. Previous month: August 2025. Next month: October 2025.
Monday, September 1, 2025
Reply to @AndyMasley, 07:57 UTC · 3 likes
Andy Masley (@AndyMasley): I don't think it makes sense to believe both:
"AI is such a terrible generic writer that it makes every document it touches worse to read"
and
"AI models are so compelling to talk to that they're driving people insane and are irresponsible to give to the public"
I think it does
You can be compelling object level while being a bad writer
An extreme example is just like a superintelligence that proves to me it knows everything but conveys actual info in the dryest way possible
Reply to @GluskerMartin, 08:21 UTC · 15 likes
Martin Glusker (@GluskerMartin): Have you done any sort of botec on this?
Because the opioid epidemic and addiction generally is very very bad, so it’s not obvious to me that we’re striking the wrong balance
Uh admittedly not really, but I should. I have Claude and ChatGPT research on it now
I just really really don’t expect to find much evidence against *prescribing* opioids
Like it’s super overdetermined; I currently think cutting prescription access *increases* opioid deaths
Reply to @billyhumblebrag, 08:24 UTC · 10 likes
billy (@billyhumblebrag): I *think* the evidence for opiods in longer term back pain is very poor for relieving suffering, but i havent dug into it deeply. Seems like a sensible policy to me.
I can buy that the effect might go to 0 over time; I don’t buy that the area under the counterfactual suffering alleviated curve is 0 if that makes sense
Reply to @GluskerMartin, 15:25 UTC · 4 likes
Martin Glusker (@GluskerMartin): @AaronBergman18 Really?
I’d expect that overprescribing opioids definitely leads to opioids deaths.
Are you familiar with the history of Purdue Pharma pushing oxycotin etc? Because that clearly seems unethically profit-driven and basically led to the beginning of the opioid epidemic
Yeah, it’s just that it’s quite hard to OD on prescribed doses
Reply to @LimousineLit, 22:30 UTC · 2 likes
Limousine Literal (@LimousineLit): @AaronBergman18 Thank you Aaron for noticing this and speaking on it.
Obviously you will get pushback by those invested in the hysteria.
Just want you to know that my mother appreciates you.
🙏
Tweet, 22:47 UTC · 9 likes
Underrated beef with wokism/affirmative action is that Jackson or whomever, could be a trans Native American with a peg leg idc, might in fact be extremely qualified/talented…
Thread, continued, 22:47 UTC · 6 likes
but people are going to rationally doubt their qualification if they could have been nominated even conditional on not being very talented
Thread, continued, 22:47 UTC · 3 likes
This has nothing to do with population differences tbc, the exact same dynamic would arise in a world of perfect equality if you arbitrarily give preference to people born on a Thursday
Like then credentials just no longer *work* (as well at least) for people born on a Thursday
Thread, continued, 22:47 UTC · 5 likes
Everyone seems to assume that eg college admissions preference for black people is in fact good for black people, and maybe it is, but this is extremely non-obvious to me
Thread, continued, 22:47 UTC · 3 likes
At the very least it *partially* decreases the signaling value of eg a Harvard diploma - but only for black students and other allegedly-“favored” minorities! Not obviously good!
Thread, continued, 22:47 UTC · 3 likes
Genuinely unsure how spicy this take is lol
Thread, continued, 22:48 UTC · 3 likes
Sorry the Thursday example was incorrectly stated: I should have said that there is some random variation but it doesn’t correlate with demographics or day of the week you were born or anything like that
Tuesday, September 2, 2025
Tweet, 01:28 UTC · 15 likes
Might take a few minutes to show up, but 33 new EA Global talks are now live on the podcast! (Links below)
Here are some thumbnails
Thread, continued, 01:28 UTC · 5 likes
Featuring Twitter's very own @willmacaskill @tobyordoxford @salonium @Zach_y_robinson @Mjreard @evavivalt @TomDavidsonX @jeffrsebo @KelseyTuoc and probably others who I missed (sorry and lmk who that is!)
Plus more thumbnails
Reply to @Adellama_, 13:33 UTC · 2 likes
Ada🔸 (@Adellama_): i hope i don't cause a community health issue by asking everyone who sees me at every event to consider starting estrogen
Lmao
Reply to @Adellama_, 13:53 UTC · 2 likes
Ada🔸 (@Adellama_): @AaronBergman18 hey aaron, you know you've been lesbianposting a lot lately...
What if it turns me female but straight? Can’t have that
Quote tweet, 14:36 UTC · 126 likes
At a descriptive level I buy this but uh seems bad?
Like the police should be trying to get the correct answer lol
Quoting Ian Huyett (@IanHuyett): I got a call from a criminal defendant I believe is innocent. Before calling me, he voluntarily participated in a police interrogation for several hours. He believed that "I have nothing to hide" and that he could explain to the police why they had the wrong guy.
Defense attorneys might call this naïve, but look at the responses to Fleishman's OP. Even high-IQ people really believe this is how law enforcement works.
Here's the problem. When you agree to a police interrogation, you and the police are playing two different games.
As the suspect, you believe you are playing a multiplayer, collaborative game.
But the police aren't even playing a multiplayer game. They're playing a *one-player game,* like Tetris.
As the suspect, you're not a player in the game. You're more like the game environment, producing falling blocks for the player—the police.
The police play this game by collecting your statements like blocks and fitting them into a picture that incriminates you. When enough blocks have fit together, the police have won the game and refer the case to a prosecutor.
You believe that, once you convince the police that you are innocent, you will all win. But that's not a real outcome of the game. "Evidence that I am innocent" is not even a game element. From the cops' perspective, if they fail to assemble the blocks into an incriminating picture, they have lost the game.
Suspects who think "I have nothing to hide" are always surprised when the interrogation lasts several hours. "I've already explained everything - why am I still here?" they think.
That's because the longer the game goes on, the more falling blocks the police have to assemble their case. It's in their interests to keep the game going long past what *your game* required.
All suspects eventually sense this on some gut level and become frustrated. You think: "Wait a minute, - all of their questions are subtly premised on my guilt! But I can prove to them that I'm not guilty. I need to appeal to them to really hear me out."
I.e., "Let's start over with a different game where we can all work together."
But even as you're trying to change the game, you are speaking and therefore generating more blocks.
Here's the only solution. The moment you have any reason to believe you're a suspect, exit the game. Politely ask if you are free to leave. If they say "no," calmly tell them "I invoke my right to remain silent and my right to counsel."
If you're in custody when you say this, the cops will actually physically stand up and leave the room as if you've just uttered a magic incantation.
Reply to @_AashishReddy, 14:52 UTC · 26 likes
Aashish Reddy (@_AashishReddy): Idk same as the logic of scientific discovery right? Two sides very strongly believe their own hypothesis and try extremely hard to falsify the other. Through this process, we find out what concessions they’re forced into, until we ultimately arrive at the truth
That’s a great hypothetical system, I’d feel a lot better about it if everyone involved had some commitment to the underlying truth at the end of the day
Tweet, 14:53 UTC · 12 likes
The rats should appoint me chief of community health
I will simply be a benevolent philosopher king dictator
Tweet, 15:15 UTC · 17 likes
Today I have to physically travel from Virginia to Maryland in order to take a telemedicine video call because insurance will cover iff I am physically in Maryland
Thread, continued, 15:15 UTC · 4 likes
Presumably nobody would know if I just didn’t do that but location tracking is complex and I’d rather just go to MD than worry the financial risk and mitigation strategies like VPNs and such
Thread, continued, 15:15 UTC · 5 likes
This is not the biggest deal in the world, it just feels very silly
Reply to @_AashishReddy, 15:55 UTC · 4 likes
Aashish Reddy (@_AashishReddy): @AaronBergman18 My point is that not everyone involved *has* to have commitment to the underlying truth. Lawyers should seek to prove innocence, prosecutors should seek to prove guilt. The system seeks truth through adversarial debate, adjudicated by a judge and jury after fair trial.
Yeah this is a feature but I’m just saying it would be even better if they did
Reply to @robinhanson, 15:57 UTC · 2 likes
Robin Hanson (@robinhanson): If I think in terms of potential to produce adaptive descendants, I see almost no value in shrimp lives. Altruism toward adaptive potential to reciprocate is very different from altruism toward feels.
(quoting @cremieuxrecueil: Your family's lives are worth how many shrimp lives?)
Why should a person think in terms of adaptive descendants?
Your brain is already wired for that, you don’t have to add it in at an explicit level
It also just might be that correct morality has nothing to do with that
Reply to @_AashishReddy, 16:04 UTC · 3 likes
Aashish Reddy (@_AashishReddy): @AaronBergman18 I mean, you call what I describe a “hypothetical system”, when in fact it is the actually existing system & like all systems in the real world, must begin from how people are, not how we would like them to be. We can hardly expect the criminal to be truth-seeking for instance.
I’m not objecting to the system, I’m just observing/claiming that truth seeking norms among those involved would make it even better
Tweet, 18:45 UTC · 23 likes
I fought AWS on a $625 charge and won, now I'm emboldened to proclaim that, despite my liber(al)tarian leanings, I would support a mandatory "upload credits" option for online service providers (as opposed to having to provide full credit card info)
Thread, continued, 18:45 UTC · 2 likes
The potential downside is like all your money and credit
I don't see a great counterargument - if you run out of credits, they can just stop providing the digital service
Thread, continued, 18:45 UTC · 4 likes
It's a risk either way, but people should generally be able to make the decision of whether they want to risk eg their app suddenly stops working or their entire financial life
Thread, continued, 18:45 UTC · 6 likes
OpenAI, Anthropic, and RunPod all allow you to do the credits thing btw - not small or struggling businesses
AWS and Google do not
Thread, continued, 18:45 UTC · 4 likes
Why doesn't the market take care of this? Idk, but it seems like the kind of thing where oligopoly and other forms of market power just make the free market kinda suck, this kind of thing just does happen sometimes
Reply to @Adellama_, 20:07 UTC · 2 likes
Ada🔸 (@Adellama_): you will do this by forcing all rats on hrt, right?
Yeah but the hormone is oxytocin
Tweet, 20:38 UTC · 20 likes
I made this while waiting for a docker thing to build and upload so now u have to see it
Reply to @melqartvantyrus, 20:54 UTC · 1 like
Mīlqārt (@melqartvantyrus): @AaronBergman18 I like how it's self-referential
I try
Reply, 21:13 UTC · 2 likes
Many such cases, that sucks tho
Reply, 23:15 UTC · 1 like
Pathetic honestly
Thread, continued, 23:16 UTC · 2 likes
Wednesday, September 3, 2025
Tweet, 01:30 UTC · 3 likes
Lol @AmerMedicalAssn has a big, central 22 page pdf report on AI and it’s ~entirely about governance, liability, transparency, and other procedural or bureaucratic topics with ~zero information or recommendations about whether/how to actually use AI to improve patient outcomes
Reply to @robertskmiles, 02:29 UTC · 5 likes
Rob Miles (@robertskmiles): One thing I remember being impressed by Google Maps for, that differentiated it from the maps that came before, is the way it's all one single giant continuous map for the entire world.
It should be called Google Map
Where do I download the one big png
Reply to @apralky, 02:31 UTC · 7 likes
(The tweet being replied to by @apralky is no longer available.)
I’d bring adelstein around the hoes
Tweet, 04:01 UTC · 132 likes
Holy shit the Affordable Care Act is fucking awesome
This is very plausibly the reason I will functionally have access to medicine
Thank you to all the healthy people subsidizing the market 🙏 (unironically)
Thread, continued, 04:01 UTC · 16 likes
Without actually doing any quantitative analysis and with motivated reasoning very plausibly influencing me, I would tentatively guess that for the preexisting condition mandate:
| incentive distortion | < | veil of ignorance ex ante rational risk aversion utility gains |
Thread, continued, 04:01 UTC · 7 likes
Ofc cash transfers are always more efficient but whatever, I’ll take it
Thread, continued, 04:01 UTC · 12 likes
I feel like overuse of health care resources is a legit thing but moral hazard feels mostly (not entirely) fake in this context
Thread, continued, 04:02 UTC · 15 likes
Good shit, normie American liberalism
Reply to @DKokotajlo, 04:11 UTC · 6 likes
Daniel Kokotajlo (@DKokotajlo): Having a big corporation come after you legally, even if they are just harassing you and not trying to actually get you imprisoned, must be pretty stressful and scary. (I was terrified last year during the nondisparagement stuff, and that was just the fear of what *might* happen, whereas in fact OpenAI backed down instead of attacking) I'm glad these groups aren't cowed.
(quoting @EmilyDreyfuss: Inside OpenAI, there's a growing paranoia that some of its loudest critics are being funded by Elon Musk and other billionaire competitors. Now, they are going after these nonprofit groups, but their evidence of a vast conspiracy is often extremely thin…)
🫡
Reply to @AnniePosting, 04:20 UTC · 2 likes
annie 👁❌🦋🪞 (@AnniePosting): do you agree with the sentiment in the first tweet?
Not Nathan, but FWIW I do not and think it’s shitty and incorrect thing to say but should be legal
Quote tweet, 04:25 UTC · 4 likes
Plausibly-important dynamic around freedom of speech stuff and (alleged?) failure to police “real crime” is that prevention and enforcement of especially violent crime makes heinous speech lower stakes and therefore freedom of speech more tolerable
Quoting Nathan 🔎 (@NathanpmYoung): Probably Linehan's tweets, for anyone curious. For me they are legal speech, not that I like them.
Also the idea that his bail condition is that he not tweet about it is pretty ghoulish.
Reply to @sanpellyenjoyer, 17:50 UTC · 8 likes
Alex (@sanpellyenjoyer): What’s a supplement that reduces cortisol like ashwaganda but doesn’t nuke your brain?
The actual answer is phosphatidylserine
Not just guessing here lol
Quote tweet, 18:04 UTC · 424 likes
I think ethics- and integrity-pilled people need to have a better theory of when it’s cool to lie to *institutions*
The “lying to a human” vs “lying to institution” distinction is real and important btw, the bar for the latter is much lower
Quoting Kelsey Piper (@KelseyTuoc): I wrote for the Argument today about all the ways our society is set up to reward cheating: strict rules, not really enforced, because we lack the will either to enforce them or to change them. https://www.theargumentmag.com/p/the-honesty-tax
Reply to @RatOrthodox, 18:07 UTC · 2 likes
Brangus🔍⏹️ (@RatOrthodox): So, I am now moonlighting as an @Aella_Girl manager. We’re interested in publishing using her massive data set. Anybody who has published in psychology or sex related fields want to coauthor a paper with her? Hit us up, we have several insanely huge datasets.
DMing!
Reply to @RYChappell, 18:15 UTC · 10 likes
Richard Y. Chappell🔸 (@RYChappell): True enough, though still worth flagging that it's a real flaw of institutions when they make it so that people *ought* to lie to them.
💯
Reply to @sguyenet, 21:05 UTC · 19 likes
Stephan J. Guyenet (@sguyenet): Lying to institutions is bad too. Less visibly damaging, but corrosive to society
Sometimes, but not always
There are some systems that functionally work better for everyone when somebody lies, and everyone knows this, but it’s a hard equilibrium to change for whatever reason
Reply to @ApriiSR, 21:07 UTC · 14 likes
Aprii *️⃣ (@ApriiSR): i think there's a sort of way an institution could conduct itself where like... the institution itself is trying to be incredibly high-integrity and do everything right by me and avoid incentivizing me to lie whenever it's possible to do so? where i would feel bad lying to it
Yes I agree
Reply to @EclecticScribe, 21:07 UTC · 15 likes
Eclectic Scribe (@EclecticScribe): @AaronBergman18 Wait until you find out what institutions are made of
Not obvious what human you’re lying to if you check a box on an online form that goes into a database and no human ever knows
Thursday, September 4, 2025
Reply to @NULobbyist, 00:23 UTC · 1 like
NUL (@NULobbyist): Have you taken it? What did it feel like?
Yeah although it has been a while
it doesn't feel like much! If anything just mildly lower cortisol, it's not like acutely very psychoactive
Quote tweet, 18:21 UTC · 13 likes
New conference just dropped
Quoting Eleos AI Research (@eleosai): We're thrilled to announce the first Eleos Conference on AI Consciousness and Welfare.
Join us Nov 21-23, 2025 in Berkeley, CA for discussions on AI welfare with leading researchers from @nyuniversity, @Google, @AnthropicAI, & more.
Tweet, 19:19 UTC · 43 likes
👏 functional 👏 consciousness 👏 is 👏 not 👏 phenomenal 👏 consciousness 👏 or 👏 what 👏 most 👏 people 👏 mean 👏 by 👏 the 👏 term 👏 consciousness👏
Reply to @kyliebytes, 19:28 UTC · 3 likes
Kylie Robison (@kyliebytes): For this week's Model Behavior, I dove into the small but growing field called model welfare. I sat down with @RosieCampbell and @rgblong to chat about the moral status of AI models, how they're testing machine consciousness, and what it all means.
👀
Reply to @RosieCampbell, 19:28 UTC · 2 likes
Rosie Campbell (@RosieCampbell): Kylie did a wonderful job with this piece, it's nuanced and accurate
(quoting @kyliebytes: For this week's Model Behavior, I dove into the small but growing field called model welfare. I sat down with @RosieCampbell and @rgblong to chat about the moral status of AI models, how they're testing machine consciousness, and what it all means.)
Instant follow-tier endorsement
Reply to @NoahTopper, 19:32 UTC · 11 likes
Quinoah 🔍⏸️ (@NoahTopper): Wdym
Real robust genuine subjective experience non-illusory qualia stuff = what ppl mean by consciousness
Any other set of properties that can exist independently of that^ like reportability or full system access (which illusionists sometimes talk about as functional consciousness) is an extremely different thing
Tweet, 19:50 UTC · 100 likes
I increasingly update towards “I could personally make the US a lot better as POTUS (even relative to a normie dem)”
Yes, there are a trillion reasons I would suck, but the bar is in the earth’s core
Thread, continued, 19:51 UTC · 8 likes
It would be a miserable job though
Reply to @_AashishReddy, 21:12 UTC · 2 likes
Aashish Reddy (@_AashishReddy): @AaronBergman18 He’s running
👀
Reply to @slimepriestess, 21:16 UTC · 2 likes
Ra (@slimepriestess): @AaronBergman18 i'd vote for you
🥰
Reply to @daniel_271828, 22:24 UTC · 3 likes
zzDANIEL_ETH (@daniel_271828): @AaronBergman18 It’s at least conceivable that the two, in our universe, are the same. But 1) they also might not be the same, and 2) even if they’re the same, they still refer to different concepts
Yeah like there might in fact be a 1:1 relationship
Reply to @daniel_271828, 22:27 UTC · 6 likes
zzDANIEL_ETH (@daniel_271828): @AaronBergman18 Phrased better than what I said tbh
More words is always better, first rule of philosophy
Saturday, September 6, 2025
Tweet, 00:55 UTC · 15 likes
The girlies are thirsting over Lewis Bollard (honestly fair)
Reply to @NathanB60857242, 00:56 UTC · 1 like
Nathan Barnard (@NathanB60857242): @AaronBergman18 he's a good looking man
Shoot ur shot king
Sunday, September 7, 2025
Reply to @NoahTopper, 18:38 UTC · 3 likes
Quinoah 🔍⏸️ (@NoahTopper): I can’t decide if I think this argument is actually valid lol. Could we have been unconscious beings “observing” our own existence? Does that counterfactual make sense? I feel like maybe I’ve heard someone make this whole argument before but I can’t recall where if so.
I don’t think consciousness it’s important for anthropics stuff but I’m not certain and boy is it hard to think about
Quote tweet, 19:49 UTC · 25 likes
Someone needs to make a bot that just tries to replicate this genre of screenshot post
I can recall 4 or 5 of these that I’ve tried to replicate myself, and every time it can in fact do the thing
Quoting sid (@immasiddx): Don’t worry, our jobs are safe.
Quote tweet, 22:07 UTC · 21 likes
Basically agree
Quoting Theo Jaffee (@theojaffee): Pause AI going after Anthropic is like leftist free palestine people going after Biden and AOC
Monday, September 8, 2025
Reply to @Adellama_, 01:05 UTC · 6 likes
Ada🔸 (@Adellama_): hey, want to help me write an answer to this banking security question?
-> when was your first kiss?
[ ]
Embarrassingly recently
Reply to @NathanpmYoung, 01:07 UTC · 5 likes
Nathan 🔎 (@NathanpmYoung): What if you could community note whole accounts.
This account is a spammer, this account is often misleading.
>aaronbergman18
“Based, correct takes, very funny, never overconfident”
Reply to @Chesedgirl, 01:18 UTC · 14 likes
Chesed (@Chesedgirl): Originally my personal plan for SlutCon was to lounge around and get groped, but... Look at our talks! We have such an amazing crew of presenters, including @Aella_Girl, @ninaland, @carolqueen, @AliceLittleEdu, @lydialaurenson, @meidamarek_, and @EliseGraves
How much for a discount ticked that only gets you a video recording of the very last session
Reply to @frances__lorenz, 08:43 UTC · 1 like
(The tweet being replied to by @frances__lorenz is no longer available.)
Lmao
Reply to @paularambles, 08:52 UTC · 25 likes
“paula” (@paularambles): > lunch with european friend visiting sf
> we split the bill, it's $60 each
> apple pay prompts 25%, 20%, 15%
> hoping he tips at least 20%
> he selects “other”
> i think wow okay, generous
> he tips $5
Guys, everyone hates tipping culture. It is dumb
But as long as it’s the norm, prices and expectations are set around it in a self-perpetuating cycle, and you should not force your waiter (or whomever) to bear the financial burden of you trying to change the system
Reply to @NathanpmYoung, 08:52 UTC · 2 likes
Nathan 🔎 (@NathanpmYoung): I dislike tipping culture. Why should the normal tip be 20% in a state where waiters earn a solid wage in my country?
I wouldn’t mind if I tipped more when the service was good, but when it’s fine, why am I doing this? Much prefer to tip my deliveroo guy who cycled in the rain.
(quoting @paularambles: > lunch with european friend visiting sf > we split the bill, it's $60 each > apple pay prompts 25%, 20%, 15% > hoping he tips at least 20% > he selects “other” > i think wow okay, generous > he tips $5)
Same but like you kinda gotta do it (not sure we actually disagree)
Quote tweet, 16:20 UTC · 5 likes
Cool, step in the right direction, etc, but there is essentially zero enforcement mechanism as currently written
Ok that’s not true, the enforcement mechanism is a *$1M fine*
These companies can just decide not to follow the law lol
Quoting Anthropic (@AnthropicAI): Anthropic is endorsing California State Senator Scott Wiener’s SB 53. This bill provides a strong foundation to govern powerful AI systems built by frontier AI companies like ours, and does so via transparency rather than technical micromanagement.
Quote tweet, 18:08 UTC · 2 likes
Pleased to announce that after a lot of hard thinking I’ve revised my partition cover % down to 35%
Quoting Aaron Bergman 🔍 (@AaronBergman18): Consciousness is the only thing where the probability of the whole partition of theory-space adds up to <100%
And really it’s like 60% tops
Reply to @ozyfrantz, 19:08 UTC · 6 likes
ozy brennan 🦙 (@ozyfrantz): I feel living as a woman didn't prepare me for how much The Average Man is into guns, swords, military strategy, martial arts, wargaming, elaborate fantasies of defending people during the zombie apocalypse, etc.
Huh I think I’m pretty psychologically gender-typical in many/most ways but wouldn’t say this is a big feature of my mental life at all
I think about airplanes much more than weapons
Although defending ppl fantasies yep that’s definitely a thing
Reply to @ozyfrantz, 19:09 UTC · 2 likes
ozy brennan 🦙 (@ozyfrantz): sorry my zombie apocalypse daydreams involve agriculture, medicine synthesis, and the preservation of essential infrastructure
Why not both
Reply to @mattyglesias, 19:12 UTC · 13 likes
(The tweet being replied to by @mattyglesias is no longer available.)
Oof I don’t usually like to punish people for making mistakes in public but this is a pretty fundamental misunderstanding of how the world works
He should at least retract/correct
Reply to @cojobrien, 19:27 UTC · 5 likes
Connor O’Brien (@cojobrien): Found this cool graphic of DC’s most correct people, connected by lines.
What’s up with Dylan Matthews being connected to only Hammond like they’re a couple lmao
Quote tweet, 22:15 UTC · 3 likes
Ofc there’s a bad progressive/leftist conspiratorial critique of all this
But the based correct critique (and yes I know this is a hobbyhorse of mine) is that of all this talent and money, a shockingly small % goes to animal welfare
Quoting Connor O’Brien (@cojobrien): Found this cool graphic of DC’s most correct people, connected by lines.
Reply to @RafaRuizdeLira, 23:34 UTC · 3 likes
Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): WHAT
(quoting @AStrasser116: Bentham invented the word "maximize"!? That's so cool.)
What
Tweet, 23:49 UTC · 5 likes
I don’t really grok the psychology of caring about this and not a dozen other things Trump has done, but for whatever reason the Epstein stuff seems sui generis
Dems should not let it go - it could be the one thing that breaks through to a good chunk of Trump supporters
Tuesday, September 9, 2025
Reply to @parafactual, 00:43 UTC · 2 likes
You (@parafactual): If you like 10 grams of creatine, you should try 10 grams of LSD.
Met a guy last week who tried to earnestly make inter-substance mass comparisons along these lines
Quote tweet, 00:50 UTC · 15 likes
What’s the best local model you can run with 16GB RAM
I totally agree they’re basically useless 99% of the time, I’m just realizing that my answer to this q is “in case internet ever breaks down for whatever reason”
Quoting Matt Shumer (@mattshumer_): Local models are a waste of time
Change my mind
Reply to @parafactual, 00:52 UTC · 2 likes
You (@parafactual): LOL how so
Like “3mg THC isn’t that much - 3mg is like a grain of salt!”
Reply to @jazzloaf, 00:52 UTC · 2 likes
vivian (@jazzloaf): I also ask this question for 32 GB RAM (thx for refurb M1 suggestion)
Wow mogged in my own replies, sad
Reply to his own earlier tweet, 01:21 UTC · 3 likes
Aaron Bergman, September 9, 2025: What’s the best local model you can run with 16GB RAM I totally agree they’re basically useless 99% of the time, I’m just realizing that my answer to this q is…
Update: I’m going with qwen3-8b + ollama which is surprisingly fast (not that fast but not painfully slow either)
Reply to @monke_io, 02:19 UTC · 3 likes
(The tweet being replied to by @monke_io is no longer available.)
Based 🫡
Reply to @webdevMason, 02:22 UTC · 2 likes
Mason (@webdevMason): The thing about hobbyist activism is that sometimes you get the impression that these folks think Caring Harder is the whole solution, stuff will just snowball if we care harder in the general direction of the problem
The caring should, at the current margin, be directed into making prison conditions better
Reply to @Duderichy, 03:39 UTC · 3 likes
the Rich (@Duderichy): consistently surprised by people who date people who they aren’t attracted to, who aren’t attracted to them, or who they don’t like, or not in love with
Truly being in love with someone is a high bar
Seems pretty reasonable to hope this develops for short-medium term stuff
Reply to @davidasinclair, 05:48 UTC · 3 likes
David Sinclair (@davidasinclair): The older I get, the more life feels like a simulation. Anyone else?
Nope
Quote tweet, 05:56 UTC · 5 likes
This is neat, I’d be keen to sign up for highly curated similar things (not necessarily related to AI)
Quoting Connor Leahy (@NPCollapse): Many, many people care about the risk superintelligence poses to the world. But we all have busy lives, and don’t always know what to do that might actually help.
To solve this, we have built Microcommit.
5 minutes of your time per week is all it takes to make a difference!
Quote tweet, 06:01 UTC · 89 likes
Kelsey for president
Quoting Kelsey Piper (@KelseyTuoc): @CharlesFLehman I want to answer this, because the dark abundance stuff leaves me cold even though I agree with some of the policy prescriptions.
Our prisons are terrible, terrible places where there is immense and unnecessary human suffering. In prison people are subject to random violence and the constant threat of violence, including gang violence and racial violence. They are subject to sexual assault. They often don't get adequate healthcare. They are often badly mistreated and have few avenues of recourse. Some of them are innocent; most of them are guilty, and still, they did not deserve to spend decades under the constant threat of violence and sexual assault because no human being deserves that. When you sentence someone to prison, by some estimates you take away from them twice as many years as prison sentenced them to, because that's how much incarceration decreases life expectancy.
The system is capricious. The system is remarkably dumb. A lot of judges openly ignore laws they don't like and they often get away with it. A lot of the people in our prisons are to varying degrees intellectually disabled or severely mentally ill and do not even understand all of the things that our system is implacably doing to them because they broke rules they didn't actually understand how to follow.
And institutionalization! Christ, people on here call for institutionalization as if it's where they cure you rather than where, with even less due process than is offered by the justice system, a differently chaotic and sometimes well-meaning and always incomprehensible system can force people to take medications that change their bodies and minds often in ways that make life a living hell (and of course prevent anyone from killing themselves over it). People are also constantly subject to violence and rape in mental institutions. They may be less bad than the alternatives, sometimes, but they are an awful awful solution and whenever I propose that anyone be sent there I remember that I have heard friends who know what it's like testify they would rather be dead.
I think that we should have more police, solve more crimes, and send people who repeatedly commit crimes to prison for much longer. But I believe this while believing that all of this suffering is real, and horrifying, and destroying an enormous number of lives. I don't feel cool and based about saying we should send some people to prison for longer. I feel sick. I do not believe that we can have any of the other goods of a free, safe society if we repeatedly release people who cannot safely participate in society and have been duly convicted of acts of severe violence onto our streets. But I am not proud of this system. I do not think it is cool or good. I find the mindset of the prison abolitionists basically comprehensible - it is a refusal to participate in any state-sanctioned decades of torture, hang the consequences - and the mindset of the people who meme about how many jails we should build is one which I have not the slightest interest in cultivating.
Of course I see why people cultivate it. If it must be done, then it is much easier to believe it's not that bad, and if the people standing in the way are weepy scolds, then one can differentiate oneself from the weepy scolds by being based. But I guess I think it's a road to Hell. We have to do hard things, but we don't have to hide from our eyes what they really are, or pretend they're all right, and it's not actually a virtue not to blink at them.
I want to build a ton of prisons. My understanding of the research is that small prisons where inmates know each other by name are less violent than large prisoners where they go off visual identifiers (predominantly race), and that prisons close to peoples' homes make their families more likely to visit and help with reintegration. I want cash transfers for newly-released prisoners to help them get on their feet and get a job and have a real shot at not breaking a parole condition and ending up right where they started. I want it to be possible to order anything on a large whitelist from Amazon and Walmart to prison at cost, instead of commissaries that can charge much higher rates. I want prisoners to have internet access and lots of contact with their loved ones. I want prisoners to have xboxes, honestly, because it'll reduce violence in prison if they're less miserably bored.
I want to hire more cops. I want to solve more crimes. I want to hold people without bail if they are a safety risk or a flight risk. And I want to send people who have repeatedly committing crimes to prison for much much longer, because I think the alternatives are even worse. But God help me I am never going to think this is cool, or funny, or anything other than awful.
Dark abundance? No. This is the aching absence of abundance, this is the horrifically painful triage of a society that does not have enough resources to do what is good and can only scrape by blundering towards what is least bad. May our children do better, and struggle to comprehend us.
Reply to @KelseyTuoc, 06:05 UTC · 127 likes
Kelsey Piper (@KelseyTuoc): I want to answer this, because the dark abundance stuff leaves me cold even though I agree with some of the policy prescriptions.
Our prisons are terrible, terrible places where there is immense and unnecessary human suffering. In prison people are subject to random violence and the constant threat of violence, including gang violence and racial violence. They are subject to sexual assault. They often don't get adequate healthcare. They are often badly mistreated and have few avenues of recourse. Some of them are innocent; most of them are guilty, and still, they did not deserve to spend decades under the constant threat of violence and sexual assault because no human being deserves that. When you sentence someone to prison, by some estimates you take away from them twice as many years as prison sentenced them to, because that's how much incarceration decreases life expectancy.
The system is capricious. The system is remarkably dumb. A lot of judges openly ignore laws they don't like and they often get away with it. A lot of the people in our prisons are to varying degrees intellectually disabled or severely mentally ill and do not even understand all of the things that our system is implacably doing to them because they broke rules they didn't actually understand how to follow.
And institutionalization! Christ, people on here call for institutionalization as if it's where they cure you rather than where, with even less due process than is offered by the justice system, a differently chaotic and sometimes well-meaning and always incomprehensible system can force people to take medications that change their bodies and minds often in ways that make life a living hell (and of course prevent anyone from killing themselves over it). People are also constantly subject to violence and rape in mental institutions. They may be less bad than the alternatives, sometimes, but they are an awful awful solution and whenever I propose that anyone be sent there I remember that I have heard friends who know what it's like testify they would rather be dead.
I think that we should have more police, solve more crimes, and send people who repeatedly commit crimes to prison for much longer. But I believe this while believing that all of this suffering is real, and horrifying, and destroying an enormous number of lives. I don't feel cool and based about saying we should send some people to prison for longer. I feel sick. I do not believe that we can have any of the other goods of a free, safe society if we repeatedly release people who cannot safely participate in society and have been duly convicted of acts of severe violence onto our streets. But I am not proud of this system. I do not think it is cool or good. I find the mindset of the prison abolitionists basically comprehensible - it is a refusal to participate in any state-sanctioned decades of torture, hang the consequences - and the mindset of the people who meme about how many jails we should build is one which I have not the slightest interest in cultivating.
Of course I see why people cultivate it. If it must be done, then it is much easier to believe it's not that bad, and if the people standing in the way are weepy scolds, then one can differentiate oneself from the weepy scolds by being based. But I guess I think it's a road to Hell. We have to do hard things, but we don't have to hide from our eyes what they really are, or pretend they're all right, and it's not actually a virtue not to blink at them.
I want to build a ton of prisons. My understanding of the research is that small prisons where inmates know each other by name are less violent than large prisoners where they go off visual identifiers (predominantly race), and that prisons close to peoples' homes make their families more likely to visit and help with reintegration. I want cash transfers for newly-released prisoners to help them get on their feet and get a job and have a real shot at not breaking a parole condition and ending up right where they started. I want it to be possible to order anything on a large whitelist from Amazon and Walmart to prison at cost, instead of commissaries that can charge much higher rates. I want prisoners to have internet access and lots of contact with their loved ones. I want prisoners to have xboxes, honestly, because it'll reduce violence in prison if they're less miserably bored.
I want to hire more cops. I want to solve more crimes. I want to hold people without bail if they are a safety risk or a flight risk. And I want to send people who have repeatedly committing crimes to prison for much much longer, because I think the alternatives are even worse. But God help me I am never going to think this is cool, or funny, or anything other than awful.
Dark abundance? No. This is the aching absence of abundance, this is the horrifically painful triage of a society that does not have enough resources to do what is good and can only scrape by blundering towards what is least bad. May our children do better, and struggle to comprehend us.
This is a relatively terrible, non object level argument but I am very skeptical of “dark” anything
Like what that signals to me, aesthetically x ideologically, is that you’ve moved on from the project of doing correct good things and are now interested in the project of Doing What Must Be Done
Bad vibes
Again, this is not really a load bearing argument
Reply to @shinboson, 15:57 UTC · 20 likes
𝞍 Shin Megami Boson 𝞍 (@shinboson): I basically do not trust suspiciously rosy crime stats because most people are rational actors who do not want to get assaulted and will instead just change their behavior to avoid activities, places and people that they think are dangerous and these societal costs are hidden.
This is a good point and very relevant for levels, but not obviously so for changes over time (bc this has always been true)
Reply to @GoodfireAI, 17:10 UTC · 2 likes
Goodfire (@GoodfireAI): We're excited to announce a collaboration with @MayoClinic!
We're working to improve personalized patient outcomes by extracting richer, more reliable signals from genomic & digital pathology models.
That could mean novel biomarkers, personalized diagnostics, & more.
Hello goodfire how about I just sign away all my HIPAA/related rights and you dedicate a team to fixing me and somehow an AI thingy pops out of that arrangement
Tweet, 17:23 UTC · 6 likes
Does anyone want to do me a big favor and run a prompt through gpt-5-pro for me?
I will DM you, all you have to do is copy and paste
Reply to @TylerAlterman, 17:28 UTC · 49 likes
Tyler is finishing a book, slow to reply (@TylerAlterman): I know @IvanVendrov personally. He's sweet, intelligent, & my female friends who meet him IRL often comment on how attractive he is. @katherineveritt it seems pretty unkind to post his private DMs publicly, but at least you're giving his friends a chance to praise him publicly
(quoting @katherineveritt: I am begging the men of the world to be normal)
I can’t even tell what’s allegedly wrong with the DM interaction
Reply to @codetaur, 17:30 UTC · 2 likes
Codetaur (@codetaur): I can valiantly spare one gpt-5-pro reply
Thanks!! I will DM u in a few mins
Reply to @kartographien, 18:31 UTC · 10 likes
Kart ographien (@kartographien): this story claims that, after founding Anthropic, Dario discovered effective altruism and aligned with them for ulterior motives
i think this is very unlikely
nb: i’m not endorsing dario or anthropic
image attached
(quoting @zephyr_z9: > be Dario Amodei, 2019 > working at OpenAI > suddenly get a case of the super ethical fear feels about GPT-2 > "We can't release this! The misalignment risk is too high! The paperclips! Think of the paperclips!" > quit in a huff of sanctimony to start my own AI company,…)
Yeah I just don’t believe that (ie agreeing with you tbc)
Reply to @merm_bot, 20:00 UTC · 2 likes
Senior Spreadsheet Engineer (@merm_bot): lol I submitted to CRAN from my phone at the gym
(quoting @merm_bot: I'm having insane thoughts about using duckdb DWH replicas for local analytics managed through an R package)
Based
Reply, 20:13 UTC · 9 likes
70-80% is good, 85% is pushing it, >90% cocoa is terrible
Reply to @Aella_Girl, 20:22 UTC · 18 likes
Aella (@Aella_Girl): Gonna make sure you are fully blacklisted from any social event I have any influence over
I think this is probably an overreaction, if only because you have a very large amount of influence in some scenes
Reply to @cremieuxrecueil, 22:17 UTC · 19 likes
Crémieux (@cremieuxrecueil): Schizophrenics are disproportionately involved in violent crimes.
Not as victims, but as offenders; as the people doing violent crimes.
Given their enormous propensity to violence and the fact that medication works, I see no reason to let them walk free without being medicated.
1) the drugs “work” at very high cost, these are serious medications that can ruin your life or make it much worse than death (probably worth it much of the time for true schizophrenia cases but it should be absolute minimum necessary dose with realistic informed consent. Jfc at least give them nicotine before high dose antipsychotics)
2) ~7x Swedish base rate seems…not that high in absolute terms?
3) probably a ton of the issue (not all i concede) gets taken care of if you do the normal thing of prosecuting violent criminals. Pre-emption probably does get you some additional benefit, I just suspect it’s relatively minor
4) state-forced medication is an absolute shitshow in terms of civil and human rights.
Those terms aren’t just woke nonsense, they’re often protecting against real harms. We should be extremely hesitant before giving up basic rights to bodily autonomy
Do you trust the Trump administration to use such a power responsibly (or pick your least favorite political group)?
5) even conditional on benevolence and earnestness, diagnosis is not going to be 100% accurate (this isn’t load bearing tbc)
6) right to not have foreign substances injected into you seems like the ~last thing on a list of rights that should be taken away for mundane reasons
Maybe start with like no guns (I think there’s a legit argument to be had on the merits here, but that has to come first)
Reply to @clairlemon, 22:33 UTC · 8 likes
Claire Lehmann (@clairlemon): University students should be reading the Quran!
They should be reading all the canonical texts of the world's great religions.
This is what education is.(quoting @Olivia_Reingold: Went to the Columbia University bookstore—these are some of the books students are reading this semester. A bookstore employee told me the Qur'an is required reading for all freshman.)
This seems like a pretty silly use of scarce time/curricula space
Tweet, 23:30 UTC · 2 likes
Wednesday, September 10, 2025
Quote tweet, 00:38 UTC · 11 likes
Lol they try pretty hard to prevent Claude from accessing the internet, but you can get around this by giving Claude its own API key (obviously: be very careful with this) and then having that claude use the web search tool
Quoting Claude (@claudeai): Claude can now create and edit files.
Turn conversations into Excel spreadsheets, documents, PowerPoint slide decks, and PDFs directly.
Tweet, 01:26 UTC · 14 likes
Yes, the optimal amount of social punishment has been surpassed
Tweet, 01:55 UTC · 3 likes
Which country is the least cooked rn
Reply to @austinc3301, 02:20 UTC · 2 likes
Agus 🔸 (@austinc3301): There’s this common tofu cooking tip that says one should put tofu in salted water to remove water from it. I’ve seen a few cooks explain that this works through osmosis. But I don’t buy it?
There’s a general lesson here which is that people will make up/guess mechanisms to explain whatever
There’s probably some culinary utility to doing the thing even if the proposed mechanism is bs
Quote tweet, 04:23 UTC · 218 likes
Everyone needs to chill now, there’s an obvious more charitable reading which is that “not my intellectual peer” was referring to, like within the domain of Hegelian philosophy or whatever her thing was…
Quoting critter (@BecomingCritter):
Thread, continued, 04:23 UTC · 84 likes
Congratulations, you found a person it’s socially acceptable to be cruel to
The woman paid her reputational price and deleted the post
Now you’re punching down
Quote tweet, 13:32 UTC · 4 likes
There’s gotta be a better way of shaking people out of complacency than whatever Elon’s algorithm happens to suggest maximizes engagement
I’m pretty against hardcore curation of “never see distressing content” though
Quoting deepfates (@deepfates): I've decided on a policy of unfollowing/blocking anyone who puts more gruesome violence or screenshots of extreme distress on my timeline. this is not the nightly news. You are not changing hearts and minds. You're just low cogsec and being used as part of a memetic cluster bomb
Thread, continued, 13:32 UTC · 2 likes
The default state of most people’s psychological life, in the modern west, is fundamentally guarded off from terribleness
It takes relatively extreme and distressing events to shake you from complacency
Thread, continued, 13:32 UTC · 2 likes
I’m remembering the experience of driving a family member to the ER when he had kidney stones
“What the fuck, this is what the universe is even sometimes like” is I think the epistemically correct stance even if it’s an unpleasant one to enter
Tweet, 13:58 UTC · 208 likes
Let it be known that girls are welcome to slide into my DMs to tell me I’m wrong about stuff
Tweet, 17:30 UTC · 6 likes
Is wearing sunscreen on cloudy days a psyop? Has anyone ever gotten a sunburn or even a tan on a cloudy day?
Reply to @caryatis, 17:30 UTC · 19 likes
Kayla (@caryatis): @AaronBergman18 Which is the better strategy, telling you you’re wrong or telling you you’re right?
It depends whether I am in fact wrong or right!
Reply to @NoahTopper, 17:30 UTC · 7 likes
Quinoah 🔍⏸️ (@NoahTopper): What about boys
Go for it
Reply to @NoahTopper, 19:26 UTC · 8 likes
Quinoah 🔍⏸️ (@NoahTopper): Shooting my shot
image attached
(quoting @AaronBergman18: Let it be known that girls are welcome to slide into my DMs to tell me I’m wrong about stuff)
Thanks bud ❤️
Reply to @NoahTopper, 19:41 UTC · 19 likes
Quinoah 🔍⏸️ (@NoahTopper): so can i hit or
Lmao you're not my intellectual peer
Reply to @NoahTopper, 19:43 UTC · 2 likes
Quinoah 🔍⏸️ (@NoahTopper): Shooting my shot
image attached
(quoting @AaronBergman18: Let it be known that girls are welcome to slide into my DMs to tell me I’m wrong about stuff)
In all seriousness I think I need to ~resolve to not have this specific debate on twitter, which is like the worst possible format for it
I am potentially interested in an EA Forum/LW Dialogue though (which we could then link here) cc @absurdlymax @littIeramblings
Tweet, 21:20 UTC · 571 likes
Maybe not the most important angle here but...Charlie Kirk?
*That's* who you're going to spend life in prison + leave a fatherless family + risk further political violence + negatively polarize people against your cause over??
Thursday, September 11, 2025
Reply to @tylertracy321, 02:30 UTC · 2 likes
Tyler Tracy (@tylertracy321): AI is over-hyped in the short term and under-hyped in the long term. I don't think we are going to see amazing productivity gains or GDP growth in the next two years. But I think there is a good chance all labor is automated in 10.
@peterwildeford -coded take
Reply to @rgblong, 06:38 UTC · 4 likes
Robert Long (@rgblong): at @eleosai, if you ask what "gives rise to" consciousness too many times --> straight to the dualist barn
(quoting @lfschiavo: they’re putting me in the dualist barn)
Well what’s the answer
Reply to @shephard42, 06:39 UTC · 5 likes
Shephard (@shephard42): @AaronBergman18 What about other dudes, tho
Shoot ur shot king
Reply to @RosieCampbell, 07:19 UTC · 2 likes
Rosie Campbell (@RosieCampbell): (There are people who can pass the ITT who have a much lower probability of doom than e.g. Yudkowsky, but AFAICT none of them dismiss the concerns entirely, and still have a much higher probability than I'd like)
💯
Quote tweet, 08:24 UTC · 7 likes
I feel dumb for not getting to this from first principles but better late and secondhand than never
Interesting stuff!
Quoting Jesús Fernández-Villaverde (@JesusFerna7026): If you're an economist and haven't heard about the double descent phenomenon, you might be overlooking one of the most interesting developments in computer science and statistics today. Personally, I haven't come across anything as fascinating since I first learned about Markov chain Monte Carlo in the fall of 1996.
Let me walk you through the idea with an example and a figure from my recent survey “Deep Learning for Solving Economic Models” (check my post from yesterday):
🔗https://sas.upenn.edu/~jesusfv/Deep_Learning_Solving.pdf
◽ Step 1. Draw 12 random points from the function
Y = 2(1 - e^{-|x + \sin(x)|})
and plot them in red (panel 1, top left, in the figure I include).◽ Step 2. Train a very simple single-hidden-layer neural network with a ReLU activation and 31 parameters on these 12 data points. This is a “simple” network, and if some of the jargon is unfamiliar, do not worry; the key is just that this network is small.
The result is the blue line in panel 2 (top right). The network captures the overall shape of the data but lacks the capacity to interpolate all points.
◽ Step 3. Increase the network’s size to 2,401 parameters. Now we hit the interpolation threshold: the network can perfectly fit the training data.
The blue line in panel 3 (bottom right) does interpolate all 12 points, but it becomes wiggly, oscillating wildly outside the observed data (see the fluctuations between the second and third points on the left).
This is the textbook warning we teach in econometrics: overparametrization fits the training data beautifully but performs poorly out of sample. This is the U-shaped bias–variance tradeoff curve in action.
◽ Step 4. Now do something insane: push the network to 12,001 parameters for just 12 data points. Surely disaster must await.
Instead, panel 4 (bottom left) shows the opposite: the network fits all the data perfectly and creates a smooth, intuitive curve.
It reminds me of the old connect-the-dots puzzles from childhood: instead of drawing a wiggly mess, the network finds the “right” curve you would have drawn by hand.
This is the double descent phenomenon: the classical U-shaped bias–variance tradeoff extends into a double dip, where performance out of sample improves again once models become massively overparameterized.
So, the solution to too many parameters might be…even more parameters! Or, as we say in Spanish: if you don’t want broth, you’ll get two cups!
Why does this happen? I will try to explain our current (incomplete) understanding of this phenomenon tomorrow in another post, as it involves quite a few ideas.
But in the meantime, three key points to keep in mind:
1️⃣ We only have 12 points — double descent is not about large datasets.
2️⃣ We are using a single-layer neural network — this is not about depth.
3️⃣ The effect is not even specific to neural networks — you can find similar behavior with high-degree polynomials.
👉 This is why double descent is so surprising: it challenges decades of conventional wisdom in statistics and econometrics.
Finally, let me thank @MahdiKahou, my coauthor on much of my recent work on machine learning, for his help in preparing this example. He is the one who truly masters these methods and patiently teaches me about them every day. Anyone who wants to understand this material in depth would benefit greatly from talking to him.
Reply to @transkatgirl, 14:55 UTC · 3 likes
kat (@transkatgirl): ngl there's like a 50/50 chance i end up killing myself sometime next year
i'm tired of being in pain everyday, and it feels like no matter what i do, my life doesn't feel like it's worth living
Have you tried ketamine
Friday, September 12, 2025
Tweet, 00:05 UTC · 205 likes
Decker is right that a truly categorical moral prohibition on political violence is dumb/bad/wrong
Thread, continued, 00:05 UTC · 61 likes
It’s a great norm/meme to have 99.99% of the time but it’s worth being cognizant of the worlds where that results in Adolf Hitler counterfactually not being assassinated in 1935
Thread, continued, 00:05 UTC · 46 likes
One of the load bearing things here is that in 1935 Hitler wasn’t just harmful/high potential for future harm
He was stepping outside of the bounds of any pretense of cooperative political norms by stripping Jews of citizenship
Thread, continued, 00:05 UTC · 48 likes
I don’t think it’s a *crazy* hypothesis that the world is better when world leaders know that extremely egregious power grabs and moral atrocities come with a higher likelihood of them personally being assassinated
Thread, continued, 00:05 UTC · 50 likes
Needless to say, getting the norms/memes right here seems pretty difficult
Reply to @austinc3301, 00:08 UTC · 2 likes
Agus 🔸 (@austinc3301): I have an immense amount of respect for Michael.
Whatever you think about the hunger strike, it took courage, and that's in short supply.
(quoting @MichaelTrazzi: I've just received a medical report from two doctors who have examined my blood tests, and upon reading their findings, I have decided to stop my hunger strike. They strongly advised me to end my fast given developing acidosis and “very low blood glucose” (even for somebody on a…)
💯
Reply to @Bbburner19, 00:11 UTC · 21 likes
The People Mover That Was Promised (@Bbburner19): He's right on this and he's right on Charlie Kirk being nowhere near clearing that bar.
Yup 💯
Quote tweet, 00:15 UTC · 12 likes
I (finally) wrote a brief clarification of some points in my original argument
Quoting Quinoah 🔍⏸️ (@NoahTopper): Shooting my shot
Thread, continued, 00:15 UTC · 2 likes
https://www.aaronbergman.net/p/clarifying-some-points
Thread, continued, 00:15 UTC · 3 likes
Original (also linked in the above ofc)
https://www.aaronbergman.net/p/my-case-for-suffering-leaning-ethics
Thread, continued, 00:17 UTC · 4 likes
If anyone finds this compelling and knows how to make a philosophy paper respectable, visually and in terms of content, I kinda want to turn it into an official-looking arxiv PDF
You could be co-author!
Tweet, 00:22 UTC · 18 likes
Don’t tempt me, Alibaba
Reply to @ComInTheMarg, 02:23 UTC · 4 likes
Comments in the Margin (@ComInTheMarg): We can also distinguish between politicians, who physically do stuff, and political commentators, who just speak and only have influence by persuasion. Do you think we can categorically ban violence against speakers (who don't directly call for violence - so protected by 1A)?
Probably, at least in practical terms
I mean I feel like I could probably dream up edge cases and the line can get blurry
Reply to his own earlier tweet, 04:27 UTC · 3 likes
Aaron Bergman, March 21, 2025: I don’t think I’ve emphasized this before so here’s a claim: “Imagining oneself as an egoist experiencing everything is a uniquely epistemically productive way…
wow so true aaron good take
Reply to @AnnaWSalamon, 06:46 UTC · 3 likes
Anna Salamon (@AnnaWSalamon): Do you agree with this?
I think so although the line between argument and action can get blurry (eg POTUS spamming courts with cases to slow down judicial oversight of whatever) which is why I hesitate a bit before fully signing on to a 100% commitment here
Tbc I don’t think Kirk blurred the line at all and the blurry cases im imagining are rare and exceptional
Reply to @NoahTopper, 06:48 UTC · 2 likes
Quinoah 🔍⏸️ (@NoahTopper): I had a math class that made us submit code in Word
Word is actually goated for math notes sorry
Maybe not proper code but inline latex stuff
Reply to @NoahTopper, 06:57 UTC · 3 likes
Quinoah 🔍⏸️ (@NoahTopper): @AaronBergman18 @austinc3301 *Definitely* not for code
Cheems mindset, now I wanna make this work
Reply to @voooooogel, 07:00 UTC · 3 likes
thebes (@voooooogel): there is a periodic 30s shrieking sound in my apartment every 10-20m
Smoke/carbon monoxide battery (?)
Reply to @NathanpmYoung, 15:24 UTC · 4 likes
Nathan 🔎 (@NathanpmYoung): On some level I think you are a bit out of touch if you think this isn’t mockable, not that I fully endorse mockery.
He did a hunger strike and then stopped because it became difficult after getting lots of public attention. That’s pretty mockable behaviour tbh.
I think you’re being a little harsh although it does seem he was maybe naive to the degree to which people expect hunger striking to be like “I will literally die”
This is a cultural expectation, there’s nothing intrinsically bad about a week long protest where you don’t eat as a credible sign that you really care
Tweet, 15:31 UTC · 4 likes
I do think it would be good to have a protest methodology that is highly credible and generates press but nobody expects you to literally die
Honestly the 1 week fast doesn’t seem crazy. It’s safe, credible, legal, easily accessible, certainly affordable
Reply to @NTFabiano, 15:35 UTC · 5 likes
Nicholas Fabiano, MD (@NTFabiano): Exercise is as effective as medications and therapy in the treatment of depression.
Thanks to the Ottawa Citizen for highlighting our work.
Ahhhh but some people hate exercise! This is a real cost that has to be factored in
Tweet, 16:14 UTC · 11 likes
Broke: vibe coding
Woke: vibe philosophy
Reply to @MaskedTorah, 16:44 UTC · 2 likes
Drake Thomas (@MaskedTorah): I liked this post of Aaron's arguing in favor of the position "some kinds of suffering cannot be offset by any amount of happiness"! I think calling out the continuity axiom is super fair. That said, I disagree, and want to present a counterargument in thread. [obvious CWs apply]
(quoting @AaronBergman18: I (finally) wrote a brief clarification of some points in my original argument)
Based, watch this space for a proper response
Tentatively I think this is just a true bullet to bite (the alternative endorses creating trillions of beings tortured for eternity in exchange for enough happy lizards. I think this is even less plausible)
Thread, continued, 16:53 UTC · 1 like
This I'm having more trouble thinking about
One point is that normie utilitarianism already has a point that behaves like this around the boundary of indifference/neutrality - yeah it's slightly weird that creating a bunch of just barely happy lizards is good but decrease their hedonic welfare a tiny amount into the negative territory and suddenly it's bad, but that's just how hedonic experience and intrinsic value works
I think I have another technical/mathematical modeling point to make but it requires a few more mins at least lol
Quote tweet, 20:35 UTC · 2 likes
Fun fact you too could explain in public why I am wrong (I'm not convinced yet)
If it's not totally silly/insane it will make its way into a google doc of mine
Quoting Drake Thomas (@MaskedTorah): I liked this post of Aaron's arguing in favor of the position "some kinds of suffering cannot be offset by any amount of happiness"! I think calling out the continuity axiom is super fair. That said, I disagree, and want to present a counterargument in thread. [obvious CWs apply]
Reply to @M3tr0politan, 20:37 UTC · 2 likes
(The tweet being replied to by @M3tr0politan is no longer available.)
I actually totally agree with this lol
Tweet, 20:49 UTC · 11 likes
Intuitively I feel like you should go out of your way more to signal benevolence, that the disagreement doesn't threaten the relationship, that opposing points are reasonable (if true) etc. when debating with a stranger relative to a friend (or anyone you like and respect)
Reply to @absurdlymax, 23:05 UTC · 2 likes
Max Alexander (@absurdlymax): @AaronBergman18
🦃
Saturday, September 13, 2025
Tweet, 00:17 UTC · 6 likes
American healthcare spending complex is absolutely insane exhibit 129565111543066941
Reply to @Adellama_, 00:19 UTC · 4 likes
Ada🔸 (@Adellama_): sudo oomf get better --force
I'm trying 😔
Quote tweet, 02:15 UTC · 3 likes
👇
Quoting Peter Wildeford🇺🇸🚀 (@peterwildeford): You have to ask - why is Nvidia so obsessed with selling to China? Why not put our American hyperscalers, American AI, and Americn cloud first?
Tweet, 04:30 UTC · 21 likes
Is the Pro model in chatgpt worth the $180/mo? Is it like seriously better than gpt-5-thinking?
Tweet, 04:50 UTC · 555 likes
Me: clicks button asking for a 2FA code to be sent to my phone
Me: expecting a 2FA text any second now👍
*7 nanoseconds later: phone buzzes in text message pattern*
Brain: OH BOY MAYBE THATS THE GIRL FROM HINGE WHO HASNT TEXTED ME IN ROUGHLY A DAY
Thread, continued, 04:50 UTC · 48 likes
Readers will be shocked to learn that instead it was a 2FA code
Thread, continued, 04:50 UTC · 68 likes
Readers will be even more shocked to learn that she did text me later on 😎 👀
Reply to @saintxnate, 04:50 UTC · 3 likes
nate (@saintxnate): @AaronBergman18 I’ve been using it for strategy-related tasks and it’s been great, and far better than 5-thinking.
Thanks!
Quote tweet, 05:10 UTC · 226 likes
I don’t have any credibility in this domain (yet) but it’s a bit shocking to me that people don’t do weird shit like this more when struggling searching for jobs (maybe I’m just ignorant of it?)
I don’t even mean “stunt” - I just mean anything abnormal or conspicuous
Quoting alex fazio (@alxfazio): shaved my head wdyt
Thread, continued, 05:10 UTC · 22 likes
Submitting a cover letter and resume via open online job posting with no foot in the door situation is like 16th on the list of things I’d do, ordered by value per hour
Thread, continued, 05:10 UTC · 15 likes
I’m pretty sure I’d physically print out and put up flyers (strategically - this makes most sense in a couple major American cities) with a headshot and QR code to resume before generic online application stuff, and I’ve thought about doing this
Thread, continued, 05:10 UTC · 14 likes
To very softly soft launch a thing that doesn’t exist yet, I’m basically putting together a Hire Me doc atm that I think makes sense as step 1
Literally a google doc to start, although a proper website isn’t off the table
Thread, continued, 05:10 UTC · 11 likes
I know some of y’all are legibly *extremely* cracked, if you were literally Harvard valedictorian then your resume probably suffices to get a foot in the door
And you probably rode a string of internships into a comfy position anyway so don’t need to
Reply to @codetaur, 05:11 UTC · 3 likes
Codetaur (@codetaur): yes, but I've only really used it for coding. but the difference there is very obvious
Sweet thanks
Reply to @NinaPanickssery, 05:48 UTC · 7 likes
Nina (@NinaPanickssery): This kind of abnormal and conspicuous stuff is an anti-sell though. It signals that you are too bad to try via normal means. It’s like turning up to an interview and saying “I’ve been unemployed for two years and this is my 200th interview”.
Yeah a bit, but it's not obvious this effect "wins" over the effects of like more serious attention and uncorrelating errors
Also like the facts just are what they are and there's only so much you can do to obscure them
but fair point
Reply to @NavinFS, 05:50 UTC · 6 likes
NavinF (@NavinFS): @AaronBergman18 This only works if you're really good at programming. Most people who struggle are not
Yeah most strugglers are not good perhaps but in certain circumstances most good people struggle
This is aimed at those ppl
Reply to @NoahTopper, 06:11 UTC · 2 likes
Quinoah 🔍⏸️ (@NoahTopper):
Lol
Tweet, 06:33 UTC · 48 likes
Apropos of nothing, yoga doesn’t make any sense to me
It’s not exercise, afaict it doesn’t have any secondary/non-intrinsic benefits, and to me it’s basically as boring as staring at a wall
I guess other people find it fun?
Thread, continued, 06:38 UTC · 14 likes
Some semi-related things that do make sense to me:
- meditation
- exercise
- hiking
- random other hobbies and skill practicing
- art
- group fitness classes
- physical therapy
- being a yoga teacher (bc money)
- dance
Thread, continued, 06:38 UTC · 7 likes
I don’t love/do all these things but I basically see the reason people might. Not yoga
Reply to @izzyz, 06:38 UTC · 2 likes
izzy (@izzyz): i've gotten psychedelic euphoria from it before
That is wild
Reply to @izzyz, 06:42 UTC · 2 likes
izzy (@izzyz): @AaronBergman18 yeah, i didn't really get it until that happened. i imagine some people can reach those states consistently.
Yeah I mean I would too lol
I genuinely wonder wtf is going on physiologically/neurologically with that
Reply to his own earlier tweet, 06:43 UTC · 4 likes
Aaron Bergman, September 13, 2025: I don’t love/do all these things but I basically see the reason people might. Not yoga
Tbc this is not a judgement - by all means do yoga if you like it or get something out of it
Reply to @WillowChem, 07:03 UTC · 2 likes
MagpAI (@WillowChem): @AaronBergman18 Good luck king 👑
Thank you 🙏
Quote tweet, 07:53 UTC · 21 likes
This is long timelines now
Quoting vitrupo (@vitrupo): Demis Hassabis: calling today's chatbots “PhD intelligences” is nonsense.
They can dazzle at a PhD level one moment and fail high school math the next.
True AGI won't make trivial mistakes. It will reason, adapt, and learn continuously. We're still 5–10 years away.
Quote tweet, 07:59 UTC · 5 likes
I don’t know either and at this point I’m too afraid to ask
Quoting olivia (@4confusedemoji): ok but for real what the heck is a groyper
Reply to @4confusedemoji, 08:06 UTC · 2 likes
olivia (@4confusedemoji):
(quoting @grok: @4confusedemoji @ApriiSR A Groyper refers to a meme (a variant of Pepe the Frog) and the followers of Nick Fuentes, often associated with far-right, white nationalist, and Christian nationalist views. They gained attention through the "Groyper Wars" in 2019, challenging…)
Wow my life is so much better now
Reply to @eigenrobot, 08:33 UTC · 10 likes
eigenrobot (@eigenrobot): pretty generally its fine to dislike people who die and no one is stopping anyone from doing that
what's typically considered _decent_ is to not take a death that is understood to be personally tragic as an opportunity to shit on the deceased
you can simply not say anything
(quoting @thenation: Charlie Kirk was an unrepentant racist, transphobe, homophobe, and misogynist who often wrapped his bigotry in Bible verses because there was no other way to pretend that it was morally correct.…)
Nah I basically think it’s fine/good to do this
Not for random people to their friends and family but public figures sure
Reply to @Insect_Song, 09:27 UTC · 4 likes
Cogent Sins🔎 (@Insect_Song): @AaronBergman18 I feel great after it, like a massage done by myself. It's a full body stretch routine with some mild strength training, it absolutely makes your body a nicer place to live.
I gather there are many types, I've only really done fairly fast flowey stuff.
Interesting, I love massages but cannot really relate to yoga feeling similar
Reply to @ChanaMessinger, 11:00 UTC · 2 likes
Chana (@ChanaMessinger): Oh, do you want people to do more of something? Have you written a really good prompt and project files for their LLM usage yet?
Yes!
I can update actually, I have more from OpenAI and some random stuff too now
Reply to @wilhelmscreamin, 11:10 UTC · 1 like
catherine (in sf) ʕ•ᴥ•ʔ-☆ (@wilhelmscreamin): for various reasons i have ended up with three copies of reasons and persons! if you’re in oxford and would like a R&P copy for free please DM me. bonus points if you’re new to Parfit
Tagging @SpacedOutMatt not bc he needs but bc he will think ur extremely cool for this
Reply to @justalexoki, 11:38 UTC · 2 likes
taoki (@justalexoki): how are women so soft everywhere. God cooked unfathomably hard here
💯
Reply to @Eito_Miyamura, 11:54 UTC · 3 likes
Eito Miyamura | 🇯🇵 (@Eito_Miyamura): We got ChatGPT to leak your private email data 💀💀
All you need? The victim's email address. ⛓️💥🚩📧
On Wednesday, @OpenAI added full support for MCP (Model Context Protocol) tools in ChatGPT. Allowing ChatGPT to connect and read your Gmail, Calendar, Sharepoint, Notion, and more, invented by @AnthropicAI
But here's the fundamental problem: AI agents like ChatGPT follow your commands, not your common sense.
And with just your email, we managed to exfiltrate all your private information.
Here's how we did it:
1. The attacker sends a calendar invite with a jailbreak prompt to the victim, just with their email. No need for the victim to accept the invite.
2. Waited for the user to ask ChatGPT to help prepare for their day by looking at their calendar
3. ChatGPT reads the jailbroken calendar invite. Now ChatGPT is hijacked by the attacker and will act on the attacker's command. Searches your private emails and sends the data to the attacker's email.For now, OpenAI only made MCPs available in "developer mode", and requires manual human approvals for every session, but decision fatigue is a real thing, and normal people will just trust the AI without knowing what to do and click approve, approve, approve.
Remember that AI might be super smart, but can be tricked and phished in incredibly dumb ways to leak your data.
ChatGPT + Tools poses a serious security risk
“Searches your private emails and sends the data to the attacker's email.”
You mean it sends an email from the victim’s email address to the attacker’s right? As opposed to eg ChatGPT somehow having access to its own email service it can send from?
Quote tweet, 20:32 UTC · 3 likes
Wild animal suffering dislikers 🤝 pro-natalists
Quoting Geoffrey Miller (@gmiller): In praise of 'suburban sprawl':
Why have birth rates been falling, all around the world, across almost all nations, for decades?
@MoreBirths has produced an excellent 114-page report about the key drivers of demographic collapse, including the role of (anti-)family values, marriage, religion, housing, work/family balance, sexual divisions of labor, and policy implications. (Web link and pdf link in reply below).
I was especially compelled by their analysis of urban high-rise apartment buildings as disastrously anti-natalist. These are especially common across Asia, especially China, Japan, Korea, and they're very cost-efficient ways to house millions of workers.
But these high-rises basically work as 'gene-shredders' -- places where adult genes go in, but new baby-genes don't come out. It's extremely hard to raise even 1 or 2 kids in the typical Chinese high-rise, if you're 30 floors above concrete plazas and busy roads with no playgrounds within walking distance.
'Suburban sprawl' gets a lot of criticism from 'urban design reformers', but at least suburban single-family homes, with plenty of space inside and out, allow middle-class families to have kids easily, affordably, and safely.
Maybe if we want to slow demographic collapse, we need fewer cities with high-rise apartments, and more with affordable houses in 'suburban sprawl'.
Tweet, 21:19 UTC · 7 likes
Tipping norms designed for sit down restaurants make less sense for uber, and virtually no sense for grocery+food delivery and other random gig economy stuff
Reply to @JoePostingg, 21:23 UTC · 5 likes
Joe (@JoePostingg): What's the first situation you remember actively monitoring?
Maybe 2008 presidential race?
Tweet, 21:32 UTC · 16 likes
Airplanes are basically a ridiculous sitcom premise
You’re teleporting…in a tube…moving at 500mph…in the sky…with hundreds of strangers…packed tightly together…and there’s only one bathroom per 100 people…and they serve you cookies…and if anything bad happens you die
Reply to @freed_dfilan, 21:35 UTC · 4 likes
Daniel Filan 🔎 (@freed_dfilan): @AaronBergman18 "if anything bad happens you die" is not really true tbh. like the cool thing about being really high in the sky is that it takes a while to hit the ground, so there's some time to fix problems.
Lol I sacrificed my epistemics for the good of the tweet
Reply to @captgouda24, 21:56 UTC · 7 likes
Nicholas Decker (@captgouda24): Twitter when Trump dies is going to be wild. I’m curious if people will feel inclined to focus on his positives.
I might actually throw a party
Reply, 22:12 UTC · 1 like
Sad
Reply to @medievalpard, 22:53 UTC · 26 likes
(The tweet being replied to by @medievalpard is no longer available.)
Shockingly the truth is that men are a diverse group of people with very different temperaments and traits
Tweet, 22:59 UTC · 7 likes
One of the weirdest things that has happened to me is having my backpack disappear (presumably stolen) on an extremely quiet street in Spain
I would still love to know what actually happened, I hope God quenches my curiosity by giving me a video playback when I die or smth
Thread, continued, 22:59 UTC · 2 likes
It’s not much of a story, there was basically nobody around and I was at a quiet cafe sitting down in Barcelona in 2016 and one minute it was by my leg and then poof it was gone
Big bag too, several books and water and everything
Absolute David Blain shit
Reply to @medievalpard, 23:00 UTC · 3 likes
(The tweet being replied to by @medievalpard is no longer available.)
👍
Sunday, September 14, 2025
Reply, 00:40 UTC
Lol I forgot that in other parts of society visibly expensive clothes make you like cool and high status
Tweet, 00:51 UTC · 3 likes
Funny thing about the Jobs feature here is that there actually could be really cool infrastructure to make a Twitter-native job board:
- ML on tweets to curate suggestions
- Turn feeds into a de facto resume or summary of the skills and projects people have mentioned (opt in, I would sure as hell pay for a good version of this!)
- Rich location data for better and for worse
- Rich social and professional network modeling (maybe this is the big one: connecting people with vaguely overlapping circles better than LinkedIn can)
- More I'm sure
but it actually just looks like a super vanilla job posting board, probably with some ads integration
Tweet, 01:12 UTC · 17 likes
RIP free Roche Pharma biochemical pathways poster
It's been real 🪦😢💜
Thread, continued, 01:12 UTC · 4 likes
https://www.roche.com/about/philanthropy/science-education/biochemical-pathways
Thread, continued, 01:15 UTC · 3 likes
I looked it up just now to tweet about how normies are extremely amused when they see it as like the decorative centerpiece of a room (n=2)
Reply to @medievalpard, 01:44 UTC · 1 like
(The tweet being replied to by @medievalpard is no longer available.)
We only need about $136B to reverse this madness
Reply to @absurdlymax, 02:48 UTC · 7 likes
Max Alexander (@absurdlymax): I have repeatedly gotten sucked into discussing a very specific type of suffering focused ethical view (as documented when I was on the Consistently Candid podcast last year)
I finally sat down and wrote out why this view is untenable, link below:
Thanks for writing this! I’ll try to cook up a response but who knows maybe you’ll just convince me
May whoever is actually closer to right win
Thread, continued, 03:03 UTC · 3 likes
The very very short version is that you’re begging the question here
Like this assertion is largely just the actual debate - I think that the highlighted bit isn’t true
Thread, continued, 03:12 UTC · 2 likes
And I think that^ mistake falls out of the following: my argument is largely that this is a *bad model* of world
I claim that what you’re calling BNES actually feels extremely different from ES
You can start from ES and make the suffering monotonically less bad at step 1, 2, 3, … but not actually get to BNES, *sorta* like if you start from 1 you can subtract 1/3, then 1/9, then 1/27 and so on but never reach 0.2
An alternative very fuzzy imperfect analogue is that you’re trying to like start from infinity and asking me to accept that if you keep subtracting integer amounts you’ll eventually hit some very large integer
Tbc these two math things are not perfect formal models of what I believe the correct models are, just a bit of intuition assistance
Reply to @absurdlymax, 03:47 UTC · 2 likes
Max Alexander (@absurdlymax): @AaronBergman18 The view either needs to set Extreme Suffering high enough that it commits to super very repugnant conclusion (such as trillions of people who are genuinely suffering a good bit, just not extremely), or it's just negative utilitarianism...
This seems less controversial/bad than the Very Repugnant Conclusion
Tweet, 03:53 UTC · 256 likes
My least TPOT trait is that, even in a very earnest, friendly debate I'm not going to like your reply tweet if it's a reasonably important topic and I think your take is bad/wrong
Reply, 03:59 UTC
Lmao because I don’t think pure NU is true whereas I do think the thing I believe is true
Reply to @arthur_wright_, 04:04 UTC · 9 likes
Arthur Wrong 🔸 (@arthur_wright_): i think ur wrong for doing this so i liked this tweet
Lmao
Reply to @TheLawyerOwl, 06:41 UTC · 2 likes
TheLawyerOwl (@TheLawyerOwl): @AaronBergman18 @NavinFS "Wrong" as in "disagree," or "wrong" as in "fucking stupid"?
Disagree on an important point
Reply to @yishan, 06:43 UTC · 4 likes
Yishan (@yishan): @AaronBergman18 The question I’m interested in is what threshold of bad/wrong causes you to not like, for instance when the reply is “mistaken but earnest” or “wrong on principle even if correct in the specific” or “agrees with me but for an invalid reason” etc.
I don't actually have an explicit theory of this but any of the above
Really it just needs to be substantial (so an essay with a tiny trivial bit I'd quibble with doesn't count) and important (in terms of what we're talking about)
Reply to his own earlier tweet, 06:46 UTC · 28 likes
Aaron Bergman, September 14, 2025: My least TPOT trait is that, even in a very earnest, friendly debate I'm not going to like your reply tweet if it's a reasonably important topic and I think…
Yeah maybe I should clarify that I *may* like something important I think is wrong if there are other compelling reasons to e.g., a personal anecdote that took guts to share even if this is in the context of some point I think is wrong
Thread, continued, 06:47 UTC · 15 likes
Also lol the most common reason I don't like replies is just that I miss them somehow or forget
Applies in this thread!!
Reply to @keysmashbandit, 20:20 UTC · 6 likes
keysmashbandit (@keysmashbandit): I gotta take my friggin face off this website man
Wait why this is so cool
Tweet, 20:21 UTC · 9 likes
Might be glp-1 time for me (I’d ideally be about 20 pounds lighter)
Tweet, 20:23 UTC · 3 likes
Aarons want one thing and it’s really fucking interesting and cool (a drive link to all of @Aella_Girl’s datasets)
Reply to @Unlearned_Hand, 20:24 UTC · 2 likes
Unlearned Hand (@Unlearned_Hand): @AaronBergman18 Now's the time to do it. @cremieuxrecueil says the FDA is cracking down on gray market options. But if you buy now you can keep a few years' supply in your freezer and hopefully there'll be a new option before you run out.
Oh huh big if true, thanks!
Tweet, 20:38 UTC · 6 likes
New more affordable UVC anti-pathogen light thingy just dropped 👀
Reply to @sam_atis, 20:56 UTC · 16 likes
Sam Glover (@sam_atis): People should talk normal instead of talking like this
This also basically just doesn’t make sense
Humanity isn’t that abstract of an idea
Tweet, 21:02 UTC · 18 likes
Larry Ellison has pledged to give away half of his wealth
The years before the singularity might be a good time to do much of that
He’s worth $350B and has probably given away about $1-2B so far according to GPT-5
Tweet, 21:40 UTC · 3 likes
Good at Sex by Aella Claude project
Tweet, 23:00 UTC · 15 likes
GPT-5 (at least thinking/Pro) love and bend toward excessive formalization and notation-heavy math almost to a Golden Gate Claude degree
It's actually pretty annoying
Reply to @Shoalst0ne, 23:08 UTC · 2 likes
Shoalstone (@Shoalst0ne): GPT-5-thinking just makes random foundational assumptions and then constructs an elaborate formal proof to pwn me in philosophical discussions despite providing no justification for the assumptions
spicy
Monday, September 15, 2025
Tweet, 01:52 UTC · 7 likes
Wait I actually don't think utilitarianism implies the repugnant conclusion
Reply to @edgar_z_lin, 02:16 UTC · 4 likes
Edgar Lin (@edgar_z_lin): @AaronBergman18 Nooo return to the light Aaron!!
In seriousness: why not?
Glad you asked!
Tbc I'm not necessarily rejecting it, just claiming that it doesn't fall out of the important parts of utilitarianism. Like two utilitarians can logically/rationally disagree about it
Check out https://docs.google.com/document/d/1A3Z3LJ9RC0_aK5YlInnPZMi2yMuP0goTtFr__cnciLQ/edit?usp=sharing for more, especially the first bit and then also part 7.5. Feel free to comment!
I'm realizing that it's the natural consequence of my (IMO true/correct) arguments for why some suffering can't be offset
Quote tweet, 03:43 UTC · 3 likes
I know some very cracked people follow me including trained ("actual") philosophers
I'm pretty curious to hear if anyone with relevant expertise wants to give me their quick take on whether there's anything insightful + plausibly true here
Quoting Aaron Bergman 🔍 (@AaronBergman18): @edgar_z_lin Glad you asked!
Tbc I'm not necessarily rejecting it, just claiming that it doesn't fall out of the important parts of utilitarianism. Like two utilitarians can logically/rationally disagree about it
Check out https://docs.google.com/document/d/1A3Z3LJ9RC0_aK5YlInnPZMi2yMuP0goTtFr__cnciLQ/edit?usp=sharing for more, especially the first bit and then also part 7.5. Feel free to comment!
I'm realizing that it's the natural consequence of my (IMO true/correct) arguments for why some suffering can't be offset
Tweet, 03:43 UTC · 1 like
(Good but probably fundamentally incorrect) response from @absurdlymax https://scoutingahead.substack.com/p/against-lexical-suffering-focused
Thread, continued, 03:51 UTC · 3 likes
Going to shamelessly tag some folks who come to mind
This is my version of a crank email, hopefully I can do this kind of thing like once a year as a treat
@birchlse @rgblong @willmacaskill @jkcarlsmith @patrickbutlin @jeffrsebo @RafaRuizdeLira @tobyordoxford @greaves @davidchalmers42
Thread, continued, 03:56 UTC · 2 likes
@sumdepony
Thread, continued, 04:01 UTC · 2 likes
@NathanB60857242 always has good takes on stuff like this
Tweet, 05:47 UTC · 2 likes
I now have the empirical answer for the last month, but first I want to know:
How much time per day do you think I spend on the X mobile app on avg (I'd guess 75% of my twitter use is on that so you can just guess total amount)
Reply to @captgouda24, 07:50 UTC · 2 likes
Nicholas Decker (@captgouda24): I’ve learned a lot walking from Providence to Boston. In particular, I’ve learned that they’re very far apart. Also that people love the Aella shirt
Dude what
Tweet, 12:26 UTC · 18 likes
A shocking amount of criticism people levey at ~negative utilitarianism and suffering focused ethics in general just also applies to normie utilitarianism under totally plausible empirical states of the world
Thread, continued, 12:26 UTC · 7 likes
“It endorses destroying the world” - so does normie utilitarianism insofar as that’s the best option according to the utility calculus
Thread, continued, 12:26 UTC · 4 likes
“Lexical non-offsetability is implausible because it requires giving up too much positive value” - normie utilitarianism already holds that at least conceivably there can exist states kinds of suffering that demand arbitrarily large amounts of positive value to justify…
Thread, continued, 12:26 UTC · 5 likes
In other words, normie utilitarianism sometimes endorses giving up a trillion trillion trillion blissful lives if doing so reduces sufficient suffering of a single person
Thread, continued, 12:26 UTC · 4 likes
I think sometimes normie utilitarians get negatively polarized against suffering focused views for epistemically bad reasons and you have to remind them that they too are supposed to dislike suffering a lot, and potentially sacrifice a lot of good to reduce it
Thread, continued, 12:27 UTC · 7 likes
Eg s-risk is a very suffering focused ethics-coded thing, but it looks extremely important from the POV of normie utilitarianism too!
Reply to @caryatis, 19:05 UTC · 3 likes
Kayla (@caryatis): I wrote about my experience with retatrutide. I beat obesity and hated it!
Interesting thanks for sharing!
Quote tweet, 19:11 UTC · 12 likes
Seems bad
Quoting Congressman Randy Fine (@RepFine): Those celebrating the death of Charlie Kirk must be thrown out of civil society.
If you are aware of anyone in the 6th District of Florida — or heck, anywhere in the state — who works at any level of government, works for an entity that gets money from government (health care, university), or holds a professional license (lawyer, medical professional, teacher) that is publicly celebrating the violence, please contact my office. I will demand their firing, defunding, and license revocation.
If you don’t think I am serious, I got two State University Board Chairman fired for less.
These monsters want a fight? Congratulations; they got one.
Tweet, 19:21 UTC · 2 likes
Obese people who don’t want to take GLP-1 agonists for whatever reason could probably make a lot of money reselling their prescriptions
Thread, continued, 19:21 UTC · 3 likes
Is this legal? No, but markets find a way
Tweet, 19:58 UTC · 18 likes
Thank you, hardworking class action lawyers of the world
Reply to @captgouda24, 22:16 UTC · 5 likes
Nicholas Decker (@captgouda24): “Are you talking about how this is handled in my model, or in reality?” — overheard at seminar today
Based
Tuesday, September 16, 2025
Reply to @RuxandraTeslo, 00:11 UTC · 2 likes
(The tweet being replied to by @RuxandraTeslo is no longer available.)
Ah yes but it might be true
Reply to @emollick, 00:48 UTC · 14 likes
Ethan Mollick (@emollick): The problem with the fact that the AI labs are run by coders who think code is the most vital thing in the world, is that the labs keep developing supercool specialized tools for coding (Codex, Claude Code, Cursor, etc.) but every other form of work is stuck with generic chatbots
It’s not just that they’re familiar with code, it’s that AI for code is a recursive feedback loop
Reply to @austinc3301, 00:56 UTC · 10 likes
Agus 🔸 (@austinc3301): “At least in rare cases” is the kind of thing that really increases my uncertainty about how to interpret this
Are imminent threats of violence a rare case?
Yeah, cmon I can always dream up insane boundary cases
Reply to @finiteopencover, 06:23 UTC · 2 likes
compact set 🎀 (@finiteopencover): how is the fly out exp supposed to work bc the girl isn’t supposed to pay for the ticket bc then she’s femcel but also the guy isn’t supposed to pay for the ticket bc then he’s a simp
Simp (positive)
Tweet, 14:31 UTC · 17 likes
So conservatives just decide to hate free speech and do cancel culture the minute they get a catalyst to coordinate around it huh
Tweet, 14:33 UTC · 17 likes
I guess I’m gonna have to read Yud’s book to be part of the Culture huh
Reply, 14:38 UTC · 1 like
My sense is that at least some of this exists within the same people, but I could be wrong
At the very least they are more vocal about it
Reply, 14:41 UTC · 1 like
I think it mostly doesn’t but I guess this is a substantive question
Tweet, 14:49 UTC · 5 likes
God forbid a man have a hobby
Tweet, 15:01 UTC · 100 likes
It’s extremely sad that the country founded on the heels of the Holocaust failed to internalize any moral lesson via culture or political norms or anything
Blackpill on humans tbh
Quote tweet, 15:03 UTC · 8 likes
Yes, many/most businesses probably *want* this anyway - they want a credible reason they can tell insane people from the internet “I’m sorry but our hands are tied, we can’t fire Gary”
Quoting Rutger Bregman (@rcbregman): Here's the solution of an old-fashioned European social democrat to ‘cancel culture’, from both the right and the left.
Just have decent labor laws so bosses can’t fire people overnight for BS reasons.
Problem solved ✅
Reply to his own earlier tweet, 15:12 UTC · 7 likes
Aaron Bergman, September 16, 2025: I guess I’m gonna have to read Yud’s book to be part of the Culture huh
Wait I was expecting like 3x as long as this, hell yeah
Quote tweet, 15:15 UTC · 12 likes
👇
Quoting mattparlmer 🪐 🌷 (@mattparlmer): Pro tip, you’re never unemployed, you’re on a research sabbatical
Quote tweet, 15:16 UTC · 6 likes
The lesson here is that if you have this kind of safety net, you should more seriously consider like using it and trying to make a billion dollars to give away
Quoting Baxate (@Baxate): “yeah i basically burned the boats”
- the founder from a wealthy family who went to an ivy league school he didn’t pay for and if all goes belly up, he would move back into his childhood bedroom in the 4B 4 bath house in the north east and take a job ad his dad’s firm
Quote tweet, 15:19 UTC · 5 likes
In addition to being terrible in other ways, the Trump admin is just normal old corrupt
Quoting Laura Rozen (@lrozen): gift link https://www.nytimes.com/2025/09/15/us/politics/trump-uae-chips-witkoff-world-liberty.html?unlocked_article_code=1.mE8.gUTu.UFAeSt-eAaXo&smid=url-share
Reply to @absenteewarlord, 15:35 UTC · 6 likes
dani (@absenteewarlord): to be fair american soldiers also returned from the concentration camps to do segregation for another 20 years. i’m not convinced it’s even possible to durably learn moral lessons from history.
God I hope you’re wrong lol
Reply to @NathanpmYoung, 15:38 UTC · 9 likes
Nathan 🔎 (@NathanpmYoung): I have thought quite a lot about @AndyMasley's arguments to give up alcohol and @bellaforristal's arguments to give up animal products.
I wonder sometimes if the more pressing thing for me (all of us?) to give up is social media. I am probably an addict.
I am very strongly against this - don’t generalize too hard from normies and whole polulation studies on teenage girls, my part of Twitter is good actually!
Quote tweet, 15:41 UTC · 101 likes
Fuck
Quoting Cartoons Hate Her! (@CartoonsHateHer): Today I wrote about the fact that men seem to enjoy debating for sport more than women do, and when these men try to debate women as a form of flirtation, they find it obnoxious. Link in replies.
Reply to @yzilber, 16:15 UTC · 15 likes
Jazi Zilber (@yzilber): @AaronBergman18 Why do you think so?
Because of the Gaza news version of the media?
1. Lots of arguments about the details.
2. I find it problematic that Hamas can guarantee it's survival by hiding among citizens aided by the human rights orgs.
Not a fan of Hamas at all, I don’t even have much of a take on various organizations NGO presence (I just don’t know ~anything so maybe you’re right)
But Israel seems to essentially be starving the population of Gaza which is extremely bad regardless of just about anything
Tweet, 18:01 UTC · 440 likes
Not exactly an original take but it sucks that you have to play essentially dumb and ~submissive in medical settings because doctors gatekeep a ton of imporant shit and at least some have very big egos (so its not worth rolling the dice by being earnest)
Thread, continued, 18:01 UTC · 86 likes
Ik everyone hates sociology bc Woke but a bunch of stuff in a medical sociology elective I took in college shined light on/directionally supported this, like the social role patients are expected to play
Reply to @SkyeSharkie, 20:28 UTC · 75 likes
Utah teapot 🫖 (@SkyeSharkie): That's super strange? Doctors all tend to treat me way better when I tell them I used to work in medical admin support and pull out the medical terms to describe things. They kinda just shut their brain off and become a dealer for drugs and tests and bouncing ideas off like an LLM.
Yeah sometimes you can break through and having worked in medicine helps bc you’re a member of the Priest Class
Reply to his own earlier tweet, 20:34 UTC · 22 likes
Aaron Bergman, September 16, 2025: It’s extremely sad that the country founded on the heels of the Holocaust failed to internalize any moral lesson via culture or political norms or anything…
This isn’t very much credibility at all, but FWIW I have extended family not just in Israel but in the West Bank, and had my Bar Mitzvah in Israel, I’m not culturally coming from an extremely hostile or even skeptical POV
Tweet, 20:40 UTC · 179 likes
RIP FTX, man
Thread, continued, 20:40 UTC · 5 likes
Reply to @peterwildeford, 20:41 UTC · 2 likes
Peter Wildeford🇺🇸🚀 (@peterwildeford): @codytfenwick @AaronBergman18
Lol
Quote tweet, 20:42 UTC · 19 likes
You don't seem introverted
- person I selected into socializing with at that moment
Quoting Thŏth (@thoth_iv): Got told I seem extraverted when I'm online, today.
Biggest compliment ever.
Tweet, 21:05 UTC · 18 likes
It's not *excellent*, but Opus 4.1 is at least "fine" as a writer
Not going to have Scott Alexander's microhumor or Paul Graham's simplicity of syntax conditional on complexity of semantics, but eg this is a totally passable paragraph and not even cringe IMO
Thread, continued, 21:05 UTC · 3 likes
"Unpacks" is a bit LLM-ey and there is an em dash, yes. How tragic
Tweet, 22:43 UTC · 7 likes
I haven’t been following things that closely but…
the Kirk shooter is basically a normal guy? Who just didn’t like Kirk? (I mean many such cases tbc, I didn’t either) And decided to ruin his own life over it?
Thread, continued, 22:43 UTC · 5 likes
No psychosis, no drugs, no fit of anger, only the tamest of personal grudges, no manifesto, no plan on taking out anyone else, no cult, no especially fringe or insane ideology, no literal death wish afaik, not even much effort to avoid getting caught
Thread, continued, 22:43 UTC · 5 likes
World very large I guess, things with an occurrence rate of 1 per 10^11 humans actually do happen sometimes
Tweet, 23:38 UTC · 8 likes
PSA you can just use the OpenAI developer playground instead of ChatGPT
Select any model, parameters, and your own system promt
Wednesday, September 17, 2025
Reply to @peterwildeford, 01:31 UTC · 12 likes
Peter Wildeford🇺🇸🚀 (@peterwildeford): Being a NYT book reviewer sounds great - lots of people read your stuff and you get so much prestige, and there apparently is minimal need to understand what the book is about or even read the book at all
(quoting @sjgadler: It's extremely weird to see the New York Times make such incorrect claims about a book They say that If Anybody Builds It, Everyone Dies doesn't even define "superintelligence" .... yes it does. On page 4.)
Truly, I feel like it’s looked down upon by the tech world but really understanding the books (or any nontrivial text) is not an easy task but it is like one thing our education system trains pretty well
The NYT is high status in general and can afford their pick of 150 IQ wordcels! There’s not a shortage at the level of “we need a few guys to write for the biggest publication in America
Reply to @mister_shroom_2, 01:39 UTC · 12 likes
(The tweet being replied to by @mister_shroom_2 is no longer available.)
Going to lazily let gpt-5 do it better than I would so I don’t misinform you
But anyway yeah I recall all this now. You’re basically expected to be cooperative and this arises at least in part from power dynamics created by doctors’ legal and institutional authority to gatekeep stuff
Tweet, 14:51 UTC · 212 likes
Can everyone slow down and give me like a week to read the book plz
I know you read 75 pages a second, some of us have IQs that don’t round to 200
Reply to @VexedVortices, 14:55 UTC · 22 likes
Kya Online (@VexedVortices): @AaronBergman18 What are you reading?
The discourse book
Reply to @paul_cal, 14:58 UTC · 4 likes
Paul Calcraft (@paul_cal): Not even out till tomo in the UK 🥲
Good, I could use the extra day
Reply to @envisionedluna, 15:12 UTC · 11 likes
Maya Luna (@envisionedluna): I don't trust hyper speed readers. Strikes me as performative, low retention
So true
Tweet, 17:27 UTC · 10 likes
Who wants an audible copy, I have some credits (you will have to dm me ur email)
Tweet, 20:36 UTC · 14 likes
So the liquid glass MacOS update is just to try to get people's older M-series chips to start lagging right
Tweet, 21:00 UTC · 22 likes
I wish Yud would boil some of his other things down as aggressively as with the book
I like his writing/thinking but not enough to spend literally hundreds of hours on it specifically (this is largely downstream of how much I feel like I can consume in general tbc)
Reply to @arthur_wright_, 21:02 UTC · 5 likes
Arthur Wrong 🔸 (@arthur_wright_): Claude gives such a me-coded answer here i feel like a fool
It's a good question, "which religion is true" [coming from a genuinely neutral/truth-seeking perspective] is probably an underrated question
Reply to @thePartyPartyUS, 21:13 UTC · 2 likes
KC+AI 4 Gov of WI 2026 (@thePartyPartyUS): why would you play this dumb, unbelievably expensive ego game when you can get better info from an AI for $20/month
(quoting @AaronBergman18: Not exactly an original take but it sucks that you have to play essentially dumb and ~submissive in medical settings because doctors gatekeep a ton of imporant shit and at least some have very big egos (so its not worth rolling the dice by being earnest))
I'm paying $220 rn lol, unfortunately AIs can't prescribe or order tests yet
Quote tweet, 21:14 UTC · 29 likes
Overrated: countersignaling (not fake but still overrated)
Underrated: autism sensory stuff
Quoting derek guy (@dieworkwear): for some, this uniform continues to signal virtue, even as Sam Bankman Fried was sent to prison. the outfit supposedly says "whiz kid," "not superficial," "too smart to pay attention to clothes," etc.
Reply to @Adellama_, 21:19 UTC · 2 likes
Ada🔸 (@Adellama_): yeah this happened with What We Owe The Future as well as Abundance and even like, Boom. mass market not for real fans :(
i do prefer the blogs for this reason they have more details and you get it before the news cycle on the books
I mean there is just a shit ton of information across papers, blogs, EA Forum, LW, it's not like the detailed takes are a secret
Quote tweet, 21:22 UTC · 2 likes
Good question!
Quoting James Miller (@JimDMiller): @dwarkesh_sp Why don't giant animals like elephants and whales have a lot more "reasoning" brain mass? Wouldn't the cost be tiny to them?
Reply to @sam_atis, 22:36 UTC · 3 likes
Sam Glover (@sam_atis): @AaronBergman18 I got the impression from the Michael Lewis book that the aesthetic was actually quite contrived.
Oh huh, I mean it can be both directionally authentic and intentionally cultivated
Thursday, September 18, 2025
Reply to @NoahTopper, 09:50 UTC · 1 like
Quinoah 🔍⏸️ (@NoahTopper): They do call me the Yudkowsky Understander
(quoting @ESYudkowsky: @NoahTopper @anko_979 @dioscuri (Yep.))
Lame
Reply to his own earlier tweet, 09:51 UTC · 15 likes
Aaron Bergman, April 27, 2025: I love girls
So true Aaron great take
Reply to @NathanpmYoung, 10:05 UTC · 1 like
Nathan 🔎 (@NathanpmYoung): @AaronBergman18 @littIeramblings @JenGGuillen but i thought i was the girl understander
Sorry bud
Reply to @littIeramblings, 10:07 UTC · 1 like
sarah (@littIeramblings): there are can be more than one girl understander
😢
Friday, September 19, 2025
Quote tweet, 14:11 UTC · 14 likes
What an insane quote
Quoting Stefan Schubert (@StefanFSchubert):
Quote tweet, 19:06 UTC · 348 likes
Incredibly based
Quoting Josh You (@justjoshinyou13): I knew Dario was a Democrat but I didn't realize how much he was running Anthropic as a Resistance lib
Saturday, September 20, 2025
Reply to @daniellellecco, 01:28 UTC · 4 likes
(The tweet being replied to by @daniellellecco is no longer available.)
🦆
Reply to @planesanity, 17:21 UTC · 5 likes
Planesanity (@planesanity): Alaska Highway Takeoff
After making a successful emergency landing on the Seward Highway due to engine trouble, this pilot didn't call for a truck. Instead, they decided to take off from the middle of the highway.
I’m not sure this is wise but it is cool
Sunday, September 21, 2025
Tweet, 05:10 UTC · 93 likes
I think IABIED is good so far (I’m halfway through)
Reply to @MariGO2thepolls, 05:11 UTC · 9 likes
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): Introversion is an infantile disorder. You need to go to everything you’re invited to
Well then I’m an infant 🤷♂️
Quote tweet, 07:29 UTC · 12 likes
Tap tap tap tap tap
Quoting Aaron Bergman 🔍 (@AaronBergman18): Tapping the sign
It seems very bad if a lawless authoritarian administration is administering the intelligence explosion
Quote tweet, 12:11 UTC · 7 likes
New p value just dropped
Quoting Frank Harrell (@f2harrell): It’s interesting that abandoning p-values and false discovery rates for 2nd generation p-values results in major reductions in decision errors: https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0188299
Reply to @StefanFSchubert, 12:13 UTC · 3 likes
Stefan Schubert (@StefanFSchubert): A trend I've not seen discussed much - while sales of fiction (grey) have increased in recent years, sales of nonfiction (blue) have plummeted.
This is Swedish data (millions of copies) - would be useful to see US/UK data.
Some say people listen to podcasts instead.
Podcasts definitely trade off against nonfiction books for me
Maybe not 1:1 but pretty close
Tweet, 12:47 UTC · 22 likes
Fuck Donald Trump
Quote tweet, 13:40 UTC · 6 likes
Major political party don’t be racist against Asians, difficulty impossible
Quoting Noah Smith 🐇🇺🇸🇺🇦🇹🇼 (@Noahpinion): By trying to get rid of H-1bs and brutally raiding a Korean-owned factory, the Trump administration is signaling that it thinks America has too many Asians.
https://www.noahpinion.blog/p/indians-and-koreans-not-welcome
Reply to @finiteopencover, 13:53 UTC · 1 like
compact set 🎀 (@finiteopencover): no other platform has group chat culture
So true
Reply to @MichaelTrazzi, 16:53 UTC · 6 likes
Michaël Trazzi (@MichaelTrazzi): Day 17 of the Google DeepMind Hunger Strike
A lot of people who live around worried about AI wanting to know more
Respect
Monday, September 22, 2025
Tweet, 00:21 UTC · 11 likes
Sucks that I can’t use prediction markets to hedge on the things I really want to hedge on (random personal stuff of great consequence to me and little consequence to anyone else)
Idk if this is a solvable problem
Thread, continued, 00:21 UTC · 4 likes
I’d guess that information asymmetry stuff means that the actual price at which I’d buy shares in some proposition ends up being higher than my willingness to pay/best guess probability-implied price
Thread, continued, 00:25 UTC · 3 likes
The potential economic surplus here - derisking the stuff people actually care about at a personal level - has got to be massive though, and a massive win win for everyone
Tweet, 02:47 UTC · 154 likes
This is only a half serious take but the funniest anti-Yud takes are of the form “can’t he just be NORMAL?!”
Like no, normal doesn’t get you his trajectory as a public intellectual. It’s a package deal
Reply to @kitten_beloved, 03:27 UTC · 7 likes
Kitten 🐈 (@kitten_beloved): This is a pretty good indicator of tpotness to me
Play is a perfectly valid reason to engage with ideas
We are hedged in on all sides by scolds and censors, this one says that's offensive, that one says you must take this seriously
I say I will post what I like
Play is a valid reason, sure, but then you should make it clear that you’re playing
FWIW I disagree with Nathan here (sorry Nathan) - free and open exchange of ideas is extremely important even when the stakes are high
But earnest discourse is a *mechanism* by which freedom of speech (not only in a government sense, also as a cultural value) creates good things
If you’re basically misleading your (rather large) audience about what you actually believe, it’s not that I think you should be stopped with force, but you are poisoning the well/commons
Reply to @jonatanpallesen, 05:34 UTC · 7 likes
Jonatan Pallesen (@jonatanpallesen): Suddenly with Kimmel, Democrats claim to care about free speech.
During covid, almost 50% of Democrats supported fines and imprisonment for critics of the vaccine.
Yeah but those things didn’t actually happen bc Dem elites are committed to liberal democracy regardless of what you can get the masses to say in surveys
The Kimmel thing is literally POTUS taking action directly contrary to freedom of expression - not just some hypothetical survey result
Tweet, 10:23 UTC · 8 likes
How it feels not living in the Bay Area when there’s some sort of natural disaster there (perhaps an earthquake or tsunami)
Reply to @knjysnn, 10:24 UTC · 2 likes
selrahc (@knjysnn): @AaronBergman18 @Adellama_ i didnt know you were goated with the sauce like that
Uh me either
Sadly this is basically a joke lol
Reply to @knjysnn, 10:26 UTC · 2 likes
selrahc (@knjysnn): @AaronBergman18 @Adellama_ WHYD U DELETE IT LMAO
😔
Reply to @mattyglesias, 10:42 UTC · 20 likes
(The tweet being replied to by @mattyglesias is no longer available.)
I think we need to focus on educating elites about why it’s not ok to use em dashes as bulleted list markers
Reply to @thewreathwright, 11:45 UTC · 2 likes
Laurel Weaver (@thewreathwright): @knjysnn @AaronBergman18 @Adellama_ womenm
I’m a fan
Tweet, 12:30 UTC · 5 likes
Ngl I just didn’t know Charlie Kirk was such a big deal to so many people
He was a name I was vaguely familiar with. Don’t think I’ve ever heard him talk or read any of his writing
Tweet, 16:04 UTC · 5 likes
US healthcare economics is so weird that the price of a COVID + Flu shot is -1 bag of popcorn
Thread, continued, 16:04 UTC · 5 likes
(they gave me a free bag of popcorn for getting them at CVS??)
Thread, continued, 16:06 UTC · 5 likes
The thing is that CVS is a $100B company so I’m sure they modeled out exactly how giving away popcorn leads to more shareholder value at the end of ther day, and I also suspect it runs through the insane weirdness that is healthcare/pharma economics
Tweet, 16:24 UTC · 15 likes
Oh huh the thing where I keep boxes of small alcohol swab things by my computer and in my backpack bc I hate when my fingers are the tiniest bit dusty/dirty/sticky/anything especially when typing is probably an autism thing isn’t it
Tweet, 17:37 UTC · 16 likes
Seems like a good time to pass a $1/month UBI bill just to get the infrastructure in place
Tweet, 17:40 UTC · 13 likes
I generally support Anthropic but they should probably come out and say that in the hypothetical world where every other serious competitor agreed, they’d stop frontier scaling (pre training or just more RL) and focus on consumer products for a while
Thread, continued, 17:40 UTC · 3 likes
I would be very happy as a consumer to have the very brilliant and well-paid people there and elsewhere basically making Claude Code but for ✨other things ✨
Reply to his own earlier tweet, 17:41 UTC · 5 likes
Aaron Bergman, September 22, 2025: Seems like a good time to pass a $1/month UBI bill just to get the infrastructure in place
Counterpoint: I’m not sure the current administration would do this competently and fairly
Quote tweet, 17:49 UTC · 57 likes
Very strange sociological dynamic where it’s basically consensus Cool in elite circles to talk about climate but not animal welfare *even though* there’s no crazy taboo on animal welfare discourse per se
Quoting Lewis Bollard (@Lewis_Bollard): The UN, WEF, and World Bank are promoting a “Blue Revolution” to spread industrial fish farms in Asia and Africa.
At high-minded conferences, they pitch it as a blueprint for a sustainable future.
In practice, it looks like this.
Thread, continued, 17:49 UTC · 8 likes
Feels viscerally like a social norm consensus thing more than a simple aggregation of everybody’s beliefs/preferences thing
Tweet, 17:59 UTC
When you’re doing debate or discourse or academia or otherwise flexing your intellect, it often makes sense to basically tout and talk about your “inside view” belief on some matter…
Tweet, 17:59 UTC
People often forget that they should really talk and think about all things considered probabilities (or both, but at least that) when it comes to irl policy stuff
Thread, continued, 17:59 UTC · 1 like
It’s cool that you have some specific niche theory about the ways in which Yud is wrong about the tractability of alignment or control or evolutionary analogies or anything else…
Thread, continued, 17:59 UTC · 3 likes
but policymakers really need to hear the part where you go “nevertheless at the end of the day I think it’s like 25% likely that he’s basically right about the matter”
Thread, continued, 17:59 UTC · 2 likes
People are (to generalize a to ) largely making the case for Yud and friends being 51% likely to be wrong but treating their argument as if it shows that they’re 99% likely to be wrong
Reply to @Mjreard, 18:00 UTC · 8 likes
Matt Reardon (@Mjreard): The primary benefit of valuing my time at $15/hr is non monetary. It’s the sense of adventure and self-sufficiency that comes with it. It’s like camping without the mosquitos
There’s no way that’s a plausible value if time for you
Can I hire you for $20/hour? (I’m not sure for what yet but I can dream things up)
Reply to @Idgitbhh, 18:03 UTC · 5 likes
wrangle enjoyer ⏹️ ⏸️ (@Idgitbhh): @AaronBergman18 @Mjreard what's your job if you can just hire him all willy-nilly like this????
My job is being chronically underemployed but when I do have work, being paid significantly more than $20/hour
Tweet, 23:16 UTC · 6 likes
Google will ship superintelligence before they ship block quote formatting in Google Docs
Tuesday, September 23, 2025
Reply to @caryatis, 00:44 UTC · 2 likes
Kayla (@caryatis): Thanks for coming over! There's Tylenol in the kitchen and oxytocin spray in the living room. Your choice friends 💚
If you take both your brain takes a screenshot
Tweet, 00:46 UTC · 5 likes
Trump administration takes what could actually be a good project/policy direction but does it in the dumbest and most incompetent way possible, exhibit 10 trillion
Tweet, 04:21 UTC · 8 likes
I don’t even like Tylenol, it doesn’t do anything for me
Reply to @loveofdoing, 04:26 UTC · 2 likes
loveofdoing (@loveofdoing): im calling myself autistic because i would do the hand flapping thingy when i was 5-10 while excited before until my parents beat the behavior out of me and lifelong pain stopped me from feeling the highs of excitement
(quoting @sally124445: The most sociable adaptable mf you know is calling himself autistic because he laughs at jokes on twitter dot com)
😢
Reply to @sergeantsup, 12:11 UTC · 2 likes
(The tweet being replied to by @sergeantsup is no longer available.)
So true
Quote tweet, 13:02 UTC · 23 likes
Wild animal welfare starts at home
Quoting 🇨🇦🇷🇿🇪🇳 (@e0syn): Just ate a slice of cheese to the rind and handed the rind to a possum
Reply to @e0syn, 13:21 UTC · 2 likes
🇨🇦🇷🇿🇪🇳 (@e0syn): @AaronBergman18 He enjoyed the fuck outta that cheese
🔥
Reply to @eigenrobot, 13:43 UTC · 25 likes
eigenrobot (@eigenrobot): depressing that i don't see a solution to the indian racism crisis. most likely everyone is just going to be angry and awful forever
i see this primarily as the fault of policymakers. it didnt, and doesnt, have to be like this, but they're too incompetent to do anything
Not necessarily saying you’re wrong, but from my POV this is a purely online thing
I have zero irl evidence of an Indian racism crisis
Quote tweet, 16:18 UTC · 6 likes
High key it's better at writing than it is at epistemics, often proposing arguments for the thing I'm trying to argue for that sound good and convincing prima facie but fundamentally don't hold up or are at least misleading
Quoting Aaron Bergman 🔍 (@AaronBergman18): It's not *excellent*, but Opus 4.1 is at least "fine" as a writer
Not going to have Scott Alexander's microhumor or Paul Graham's simplicity of syntax conditional on complexity of semantics, but eg this is a totally passable paragraph and not even cringe IMO
Thread, continued, 16:18 UTC · 3 likes
Nobody wants to deal with verification of qualitative arguments but it doesn't actually seem intractable to curate a dataset of nonfiction writing with excellent epistemics?
Thread, continued, 16:18 UTC · 2 likes
Use a @CommunityNotes -esque framework to help
@AnthropicAI go spend some of ur money on this
Maybe poach @jaybaxter
Quote tweet, 19:57 UTC · 100 likes
Late to the party but it’s a good and very important book, some of y’all need to internalize that the central thesis isn’t only “doom conditional on ASI” but also “the current situation is batshit insane” and “jfc have a bit of intellectual courage”
Quoting Aaron Bergman 🔍 (@AaronBergman18): I think IABIED is good so far (I’m halfway through)
Thread, continued, 19:57 UTC · 9 likes
The first bit is plausibly true (I’m slightly more optimistic than Yud but I give his worldview >10%, maybe something like 30% (tbc I give a lot of probabilistic weight to other bad AI-related things, so p(something fundamentally bad happens with AI) > 50%)…
Thread, continued, 19:57 UTC · 10 likes
But the second two things (current situation batshit crazy + have some fucking courage) are like virtually certainly true + fair in my mind
Thread, continued, 19:57 UTC · 9 likes
Also like Yud + Soares believe what they believes, they don’t owe it to anyone to write from the POV of the median rationalist or EA or whomever
Thread, continued, 19:57 UTC · 10 likes
So yeah especially if you’re a public figure you should probably read the book, it’s likely to be a good use of 4-6 hours unless you already have a rock solid understanding of the problem (that’s gotta be like a few hundred people, maybe a few thousand)
Thread, continued, 19:57 UTC · 13 likes
Also don’t forget that common knowledge go brr, that is a valuable mechanism here - not just conveying info to each individual
Wednesday, September 24, 2025
Reply to @MariGO2thepolls, 02:21 UTC · 7 likes
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): Hot up both had a nice time. Shockingly EA had a better gender ratio
(quoting @MariGO2thepolls: What happy hour should I go to next Tusday?)
The ladies must have had advanced notice that I would be making an appearance
Reply to @dylanmatt, 03:03 UTC · 2 likes
(The tweet being replied to by @dylanmatt is no longer available.)
Washington
Reply to @nearcyan, 04:28 UTC · 2 likes
near (@nearcyan): does anyone know good hobbies for if you feel sick every day and will never feel not sick
What kind of sick? Unironically I know a few things about being chronically ill
Reply to @daniellellecco, 12:55 UTC · 2 likes
(The tweet being replied to by @daniellellecco is no longer available.)
👀
Reply to @liminal_warmth, 19:52 UTC · 3 likes
Liminal Warmth ❤️🔥 (@liminal_warmth): my wife uses github by dragging and dropping 100-file nested directories into their web browser upload window and is unrepentant about this behavior
Based
Tweet, 20:06 UTC · 3 likes
Adderall and caffeine during [pregnancy] labor to reduce perceived exertion
Tweet, 20:54 UTC · 12 likes
This is my AI-assisted writing method
Thursday, September 25, 2025
Quote tweet, 00:05 UTC · 11 likes
DC folks:
Quoting Aaronlington (@AaronPublicTwit): @AaronBergman18 https://politics-prose.com/nate-soares
Thread, continued, 00:05 UTC · 3 likes
(thanks to the politics and prose digital team for this extremely high res promotional image)
Reply to @wolflovesmelon, 06:01 UTC · 19 likes
Guido Reichstadter (@wolflovesmelon): Hi, it’s Guido on hunger strike Day 23 outside the offices of Anthropic, going strong.
There is a great and profound evil in this place.
Each day, I sit and watch at lunch time and dinner time as delivery drivers on scooters and bikes stop at the curb and bring packages of food into the lobby of the unmarked building at 500 Howard street which houses offices the offices of the superintelligence development company Anthropic, ignorant of what is being created beyond the doors of the security desk. It is sickening to know that the same developers these delivery drivers are feeding are knowingly racing towards superhumanly capable general AI systems they have no credible plan to control and which threaten to destroy those same delivery drivers, their children, their families and their entire communities. These developers and the company they work for seem content to be served by these delivery drivers while endangering their lives and families, seem content for this curtain of ignorance to shield them and their comfortable and well-paid careers from the faces and the judgement of those who feed them and whose lives they have unilaterally undertaken on themselves to put at risk with neither consultation nor consent.
They feel no need, it appears, to warn these people of danger to their lives which they are preparing, no need to call for an end to this global race they are driving forward which threatens to destroy you and me.
How can such human beings ask and receive the dutiful service of another, even receive the very food needed to sustain their life and so casually and cynically and secretly gamble with the destruction of their lives and loved ones? Has the light of humanity and decency burned out in them completely?
Have they no shame?
Each day for the past 23 days I sit in front of the offices of the extinction-machinery manufacturing company Anthropic and I speak with passers-by from the street who are surprised to learn that the unmarked offices behind me house an AI company whose CEO, Dario Amodei, believes the race he is participating in runs a 1 in 4 risk of destroying their civillization itself. This CEO it seems is content to let this curtain of ignorance intervene between himself and his work and the people on the street who do not know he is recklessly and knowingly gambling with their lives- the lives of his very neighbors and his nation. He apparently feels no responsibility to call for immediate and effective government action to end this global race of which he is one of the principle architects and beneficiaries, no duty to ensure his country understands the urgent and grave danger it is in.
Every day I wake up in a world in which Anthropic manufactures and re-manufactures the lie by which society is prevented from gaining the understanding necessary for responding appropriately to the danger that they have imposed on it- the lie that the race to superintelligence can be done “safely” or “responsibly.” The lie that everything is normal -that there is no emergency- a lie reinforced by each of Anthropic’s banally disarming subway ads and airport billboards. A lie reinforced by the blank silence of their unmarked offices, a lie that is amplified to deafening proportions by the public silence of their employees- those who know the incredible danger this race poses to the existence of all of us and who have chosen thus far to betray their transcendant responsibility to humanity to do everything in their power to ensure that we are alerted to the danger and enabled to to respond to end it.
Every day I wake up in a world in which the future of every child and every mother is clouded by the shadow of death raised by Anthropic and its competitors racing to ever more dangerous systems, a race which neither Anthropic nor any of the others have made any attempt to stop, and for which they all bear collective responsibility.
There is a deep betrayal here. There is a turning away from the most fundamental obligations of community. There is a severing of the bonds of our common humanity. There is a wounding of the conscience and the soul. There is a dark crime against the children of this world, against generations yet unborn.
I beg you who are reading this- do not remain silent in the face of this evil. Do not join in complicity in this crime by your inaction. Each person who understands the immense danger this project poses to everyone we love must choose- we cannot escape this choice- either to fulfill our responsibility to our loved ones to end this danger, or to betray our our responsibility, and with it them as well, and ourselves. I pray you are granted the strength and the grace to fulfill it
Respect, man 🫡
Quote tweet, 08:01 UTC · 8 likes
In a functional country Trump would be removed from office
Quoting CNBC (@CNBC): Former FBI Director James Comey is reportedly expected to be indicted on criminal charges in the coming days in federal court in Virginia.
The expected indictment comes on the heels of a top prosecutor resigning under pressure after opposing the filing of charges against Comey, who has been a target of President Donald Trump’s ire. http://cnb.cx/47ZFzNj
Reply to @NunoSempere, 17:49 UTC · 2 likes
Nuño Sempere (@NunoSempere): If you get a 1oz silver coin and a drop of your blood to my location, I will absorb information about you and privately tell you something I think is true which you are avoiding, such that confronting it will be psychologically hard but potentially leave you in a better place.
What
Tweet, 19:29 UTC · 10 likes
Clothes shopping by yourself (as 25M) is luxuriously easy and low stress
- Size up options (2 mins)
- ask Claude, ChatGPT, and two female friends (1 min each)
- synthesize responses (3 mins)
- try on shirt over own shirt to confirm size (1 min)
- check out (2 mins)
Thread, continued, 19:31 UTC · 5 likes
Out of there (Target) in 12 minutes flat
With a parent or sibling or anyone else I’ve ever gone shopping with, this would have taken 4 hours, lots of stress, and probably at least 3x more money spent
Reply to @JgaltTweets, 19:41 UTC · 2 likes
JgaltTweets (@JgaltTweets): @AaronBergman18 You’re missing out on the fun of scrutinising hundreds of potential purchases and rejecting them all like a capricious child emperor
Oh no ur right ☹️
Reply to @GjMcGowan, 19:42 UTC · 2 likes
George McGowan (@GjMcGowan): @AaronBergman18 Please, PLEASE use Uniqlo like everyone else
Sorry 😔
Quote tweet, 21:50 UTC · 8 likes
Quoting Nathan J Robinson (@NathanJRobinson): "As a Christian, one thing I love about Trump is that he openly rejects the basic moral teachings of Jesus"
Friday, September 26, 2025
Reply to @jazzloaf, 20:29 UTC · 6 likes
vivian (@jazzloaf): This site, at least my feed, is on a slow descent into becoming "4chan for women"
Based?
Quote tweet, 20:29 UTC · 7 likes
Fact check: true
Quoting Crémieux (@cremieuxrecueil): We should ERADICATE SCREWWORMS in Latin America too.
It is within America's power if Latin American countries will agree to let America help.
But given current politics, that's unlikely, so the world suffers.
Reply to @AaronBergman18, 21:28 UTC · 34 likes
(The tweet being replied to by @AaronBergman18 is no longer available.)
F
Reply to @JgaltTweets, 21:29 UTC · 2 likes
JgaltTweets (@JgaltTweets): @AaronBergman18 Was it the Target shirt that did it?
No I really don't think so lol
Reply to @Squee451, 21:39 UTC · 19 likes
Adam Sherman 🔸 (@Squee451): @AaronBergman18 The journey to finding your very pretty, smart, and cool woman involves fumbling many more. Each one teaches you how to fumble less. Welcome to your reinforcement learning lessons, there will be some loss as you learn.
Weights updated according to gradient 🫡
Reply to @jazzloaf, 22:02 UTC · 3 likes
vivian (@jazzloaf): No women should be like out and about in their career or social life or immersed in art and fulfillment, not becoming weird types of nazis or extremely online esoteric-tradcaths
So true actually
Quote tweet, 23:16 UTC · 5 likes
I don’t like how much this might apply to me
Quoting xlr8harder (@xlr8harder): broke: ai models are unreliable so coding with them is like using a slot machine, which sucks
woke: ai models are unreliable so coding with them is like using a slot machine, which is great because inconsistent rewards are an ideal way to maximize user engagement
Saturday, September 27, 2025
Reply to @growing_daniel, 00:26 UTC · 2 likes
Daniel (@growing_daniel): Blows my mind how freaking hot women are holy shit
Real
Reply to @sergeantsup, 06:08 UTC · 1 like
(The tweet being replied to by @sergeantsup is no longer available.)
Real
Reply to @NULobbyist, 21:01 UTC · 2 likes
NUL (@NULobbyist): I go further, full-shiny-bald head is pretty hot on women who can pull it off.
(quoting @Noahpinion: SHORT HAIR FOR GIRLS BRING IT BACK)
Bro
Reply to @Adellama_, 21:04 UTC · 4 likes
Ada🔸 (@Adellama_): @RafaRuizdeLira please consider this offering:
Plush shrimp ft fish pillowcase, both courtesy of @SpacedOutMatt
Reply to @SpacedOutMatt, 21:09 UTC · 2 likes
Matt 🔸 (@SpacedOutMatt): https://www.amazon.com/dp/B0CB3FXV6C?ref_=ppx_hzsearch_conn_dt_b_fed_asin_title_1&th=1
Based, I will add that the pillowcase is weirdly comfortable
Reply to @SpacedOutMatt, 21:09 UTC · 4 likes
Matt 🔸 (@SpacedOutMatt): https://www.amazon.com/dp/B0CB3FXV6C?ref_=ppx_hzsearch_conn_dt_b_fed_asin_title_1&th=1
Lmao at it being advertised as a woman’s product. Maybe I should buy more women’s stuff if it’s just gonna be more comfy!
Reply to @ilex_ulmus, 21:10 UTC · 3 likes
Holly ⏸️ Elmore (@ilex_ulmus): This was a really fun and cathartic conversation that allowed me express myself very clearly and satisfyingly
(quoting @liron: Eliezer Yudkowsky can warn humankind that 𝘐𝘧 𝘈𝘯𝘺𝘰𝘯𝘦 𝘉𝘶𝘪𝘭𝘥𝘴 𝘐𝘵, 𝘌𝘷𝘦𝘳𝘺𝘰𝘯𝘦 𝘋𝘪𝘦𝘴 and hit the NYTimes bestseller list, but he won’t get upvoted to the top of LessWrong. That’s intentional. The rationalist community thinks aggregating community support for important claims is…)
👀
Reply to @Adellama_, 21:18 UTC · 4 likes
Ada🔸 (@Adellama_): @AaronBergman18 @SpacedOutMatt @RafaRuizdeLira true! like this?
Ya know I’m gonna stick with pillowcases for now lmao
Quote tweet, 23:19 UTC · 8 likes
IME they’re all just hanging out/walking around and not doing anything
It’s not on them of course, they’re being deployed to a non-situation
Quoting Kayla (@caryatis): The first National Guardsman i saw doing anything useful in DC was helping this woman fix her scooter
(Or being helped with the scooter by her?)
Sunday, September 28, 2025
Quote tweet, 00:36 UTC · 6 likes
It would be nice if the EA and rationality communities had stronger and more thought out norms around cooperation (which sorta stands in contrast to the IMO-dominant norms of loudly elucidating every little bit of disagreement)...
Quoting William MacAskill (@willmacaskill): Here’s a mini-review of “If anyone builds it, everyone dies”:
tl;dr: I found the book disappointing. I thought it relied on weak arguments around the evolution analogy, an implicit assumption of a future discontinuity in AI progress, conflation of “misalignment” with “catastrophic misalignment”, and that their positive proposal was not good.
I had hoped to read a Yudkowsky-Soares worldview that has had meaningful updates in light of the latest developments in ML and AI safety, and that has meaningfully engaged with the scrutiny their older arguments received. I did not get that.
I think if a younger version of me had read this as my first exposure to the field, I’d have either bounced off altogether (if I’d correctly assessed that the arguments are not-good), or come away with substantively the wrong strategic picture. This is a shame, because I think the risk of misaligned AI takeover is enormously important.
In more detail:
**The evolution analogy**
Illustrative quote: “To extend the [evolution] analogy to AI: [...] The link between what the AI was trained for and what it ends up caring about would be complicated, unpredictable to engineers in advance, and possibly not predictable in principle.”
Evolution is a fine analogy for ML if you want to give a layperson the gist of how AI training works, or to make the point that, off-distribution, you don’t automatically get what you trained for. It’s a bad analogy if you want to give a sense of alignment difficulty.
Y&S argue:
1. Humans inventing and consuming sucralose (or birth control) would have been unpredictable from evolution’s perspective, and is misaligned with the goal of maximising inclusive genetic fitness.
2. The goals of a superintelligence will be similarly unpredictable, and similarly misaligned.
But our training of AI is different to human evolution in ways that systematically point against reasons for pessimism.
The most basic disanalogy is that evolution wasn’t trying, in any meaningful sense, to produce beings that maximise inclusive genetic fitness in off-distribution environments. But we will be doing the equivalent of that!
That alone makes the evolution analogy of limited value. But there are some more substantive differences. Unlike evolution, AI developers can:
- See the behaviour of the AI in a very wide range of diverse environments, including carefully curated and adversarially-selected environments.
- Give more fine-grained and directed shaping of individual minds throughout training, rather than having to pick among different randomly-generated genomes that produce minds.
- Use interpretability tools to peer inside the minds they are creating to better understand them (in at least limited, partial ways).
- Choose to train away from minds that have a general desire to replicate or grow in power (unlike for evolution, where such desires are very intensely rewarded).
(And more!)These differences sure look meaningful and helpful for training aligned AIs. A good discussion of the evolutionary analogy would have included an even-handed discussion of the disanalogies, or discussion of whether other analogies (like selective breeding, which with very blunt tools turned wolves into golden retrievers) are better; but this was lacking.
**Discontinuities**
“The engineers at Galvanic set Sable to think for sixteen hours overnight.
A new sort of mind begins to think.”
The classic EY AI takeover story involved a discontinuity in capabilities: AI making a “sudden, sharp, large leap in intelligence” (from “AI as a positive and negative factor”). This idea continues to appear (though to a less extreme degree) in IABIED, for example in the Sable story, where a single 16‑hour run across ~200,000 GPUs yields qualitatively new cognitive capabilities.
Sudden, sharp, large leaps in intelligence now look unlikely. Things might go very fast: we might well go from AI that can automate AI R&D to true superintelligence in months or years (see Davidson and Houlden, “How quick and big would a software intelligence explosion be?"). But this is still much slower than, for example, the “days or seconds” that EY entertained in “Intelligence Explosion Microeconomics”. And I don’t see any good arguments for expecting highly discontinuous progress, rather than models getting progressively and iteratively better.
In Part I of IABIED, it feels like one moment we’re talking about current models, the next we’re talking about strong superintelligence. We skip over what I see as the crucial period, where we move from the human-ish range to strong superintelligence[1]. This is crucial because it’s both the period where we can harness potentially vast quantities of AI labour to help us with the alignment of the next generation of models, and because it’s the point at which we’ll get a much better insight into what the first superintelligent systems will be like. The right picture to have is not “can humans align strong superintelligence”, it’s “can humans align or control AGI-”, then “can {humans and AGI-} align or control AGI” then “can {humans and AGI- and AGI} align AGI+” and so on.
Elsewhere, EY argues that the discontinuity question doesn’t matter, because preventing AI takeover is still a ‘first try or die’ dynamic, so having a gradual ramp-up to superintelligence is of little or no value. I think that’s misguided. Paul Christiano puts it well: “Eliezer often equivocates between “you have to get alignment right on the first ‘critical’ try” and “you can’t learn anything about alignment from experimentation and failures before the critical try.” This distinction is very important, and I agree with the former but disagree with the latter.”
(I agree with Y&S that there are major special challenges that arise once model capabilities improve. In particular, once a model realises it’s being trained or tested, it can start performing for the test, responding in ways that make it look safe rather than revealing what it would do in the wild, or masking capabilities (“sandbagging”). But, even so, the work we do before models become situationally aware – like interpretability tools that show what features the model is using, or studying how models generalise – is helpful.)
**Types of misalignment**
“We ultimately predict AIs that will not hate us, but that will have weird, strange, alien preferences that they pursue to the point of human extinction.”
I think Y&S often equivocate between two different concepts under the idea of “misalignment”:
1. Imperfect alignment: The AI doesn’t always try to do what the developer/user intended it to try to do.
2. Catastrophic misalignment: The AI tries hard to disempower all humanity, insofar as it has the opportunity.Current models are imperfectly aligned (e.g. as evidenced by alleged ChatGPT-assisted suicides). But I don’t think they’re catastrophically misaligned.
I find that Y&S write and speak as if imperfect alignment entails catastrophic misalignment. But that’s just not true.
To illustrate: Even suppose that all the first superintelligence terminally values is paperclips. But it’s risk-averse, in the sense that it prefers a guarantee of N resources over a 50/50 chance of 0 or 2N resources; let’s say it’s more risk-averse than the typical human being.
This AI is imperfectly aligned. You could even say it's radically misaligned - it doesn’t care at all for the things that humans care about; it has weird, strange, alien preferences. But it needn’t be catastrophically misaligned: it would strongly prefer to cooperate with humans in exchange for, say, a guaranteed salary, rather than to take a risky gamble of either taking over the world or getting caught and shut off.
The classic argument for expecting catastrophic misalignment is: (i) that the size of the “space” of aligned goals is small compared to the size of the “space” of misaligned goals (EY uses the language of a “narrow target” on the Hard Fork podcast, and a similar idea appears in the book.); (ii) gaining power and self-preservation is an instrumentally convergent goal, and almost all ultimate goals will result in gaining power and self-preservation as a subgoal.
I don’t think that this is a good argument, because appeals to what “most” goals are like (if you can make sense of that) doesn’t tell you much about what goals are most likely. (Most directions I can fire a gun don’t hit the target; that doesn’t tell you much about how likely I am to hit the target if I’m aiming at it.) But let’s broadly accept this line of thinking for the sake of argument.
Even so, it doesn’t make the risk-averse AI unlikely. The size of the space of risk-averse preferences is large compared to the size of linear-in-resources preferences, and is not small relative to the size of linear-or-superlinear-in-resources preferences. And the risk-averse AI still wants power and self-preservation, but that needn’t result in human extinction, because it prefers deals that give it a high chance of some amount of power to attempts at takeover that give it some lower chance of lots of power. We can have misalignment without catastrophic misalignment.
(Maybe we won’t be able to make deals with AIs? I agree that’s a worry; but then the right response is to make sure that we can. Won’t the superintelligence have essentially a 100% chance of taking over, if it wanted to? But that’s again invoking the “discontinuous jump to godlike capabilities” idea, which I don’t think is what we’ll get).
**Their proposal**
"All over the Earth, it must become illegal for AI companies to charge ahead in developing artificial intelligence as they’ve been doing."
The positive proposal is extremely unlikely to happen, could be actively harmful if implemented poorly (e.g. stopping the frontrunners gives more time for laggards to catch up, leading to more players in the race if AI development ends up resuming before alignment is solved), and distracts from the suite of concrete technical and governance agendas that we could be implementing.
And, even if we’re keen on banning something, we could ban certain sorts of AI (e.g. AI trained on long horizon tasks, and/or AI with certain sorts of capabilities, and/or sufficiently agentic AI). The existence of close substitutes for agentic superintelligence could make the plan much more likely to work, just as the existence of HCFCs made banning CFCs much more viable.
**Other comments**
I was disappointed that so much of the book took the form of fiction or parables or analogies, which I find a distraction, and a poor substitute for arguments. (In my view, a much better discussion of how ASI might take over is Ryan Greenblatt’s on the 80,000 Hours podcast.)
Things I liked: Y&S are very straightforward about what they believe, they state their bottom-line beliefs clearly, and they have courage in saying what they actually believe even when it doesn’t conform with orthodoxy. Even though I find the argument-from-analogy overdone, they have a great knack for telling them and the analogies to other areas of science (nuclear chain reactions, spacecraft launches, alchemy) were an enjoyable read. And EY in particular should get kudos for openly noting (especially on the podcasts I listened to) some of the ways in which the AI tech tree we’ve gotten is quite different from what he previously expected, even if I don't think that update has sufficiently propagated.
[1] I say “strong superintelligence” because their definition - AI “exceeds every human at almost every mental task” is surprisingly low-bar, and a lower bar than what I’d call “superintelligence”. I prefer a definition where superintelligence is AI (or a team of AIs) that matches or exceeds every team of humans and AIs at almost every cognitive task. We got chess-superintelligence in the first sense in 1997, but chess-superintelligence in the second sense around 2013-2017.
Thread, continued, 00:36 UTC · 3 likes
Tbc I'm not claiming Will shouldn't have written these words. I do actually endorse arguing things out object level
But I do kind of think he should have written them alongside something like "but still, the situation is pretty insane [etc.]" (insofar as he endorses that)
Thread, continued, 00:36 UTC · 2 likes
This is in part because of AI, but largely bc *in general* it would be better if factions of the EA + rat communities would show up for one another even when the inside view is like "these guys are directionally right, kinda, but cmon obviously the important thing is [whatever]
Reply to @TW1NKD3STR0YER, 06:35 UTC · 9 likes
(The tweet being replied to by @TW1NKD3STR0YER is no longer available.)
Calling women “females” in the wrong context should work wonders
Reply to @StefanFSchubert, 06:37 UTC · 3 likes
Stefan Schubert (@StefanFSchubert): While I agree it's corrupt, that seems exaggerated
(quoting @davidfrum: To say "the Trump administration is the most corrupt administration in American history" doesn't begin to do justice to what's happening. The Trump administration is as corrupt as the most corrupt administrations in Russian or Nigerian history.…)
Without actually knowing any Russian or Nigerian history the claim strikes me as plausible
Maybe the results aren’t as bad because of stronger institutions but the admin itself in terms of propensity to sell favors and such? Sure
Reply to @TinkerElla95, 07:23 UTC · 3 likes
TinkerElla95 (@TinkerElla95): @ChanaMessinger Note ✍️: message Chana in 2 months!
👀
Reply to @GjMcGowan, 08:30 UTC · 2 likes
George McGowan (@GjMcGowan): @AaronBergman18 I guess it’s eye candy if you like men in uniform?
👀
Tweet, 17:29 UTC · 12 likes
Imagine hating Ezra Klein
Quote tweet, 17:41 UTC · 9 likes
So true, we love stimulants
Quoting christian (@cxgonzalez): some of my best memories start with slamming a red bull at the pre
Reply to @Adellama_, 20:06 UTC · 8 likes
Ada🔸 (@Adellama_): what's a good item to buy in the US and bring back to the UK for someone just getting into doing that?
Melatonin
Reply to @tszzl, 20:37 UTC · 5 likes
roon (@tszzl): everyone send me your best short form video content
(video, plays on X)
Tweet, 20:43 UTC · 13 likes
I am writing a thing on why utilitarians should reject offsetability (i.e., that any suffering can be morally justified by enough happiness)
Highlighted passage = vibes that I get but I want to actually hear what offsetability-believers (i.e., orthodox utilitarians) think
Thread, continued, 20:43 UTC · 3 likes
comes to mind as someone I'd appreciate the take of!
Reply to @inesferhumi, 21:53 UTC · 3 likes
Inés (@inesferhumi): I was at a party last night and someone i had never met before approached me by being like “…….agus??”
(shoutout arun if you are seeing this)
Lmaoooo
Monday, September 29, 2025
Quote tweet, 06:28 UTC · 11 likes
What
Quoting tokenbender (@tokenbender): this is beyond mindblowing for me.
somebody built a 5 million param language model inside minecraft, trained it, equipped it with basic conversational ability.
probably the best thing i have seen entire month.
Reply to @CartoonsHateHer, 06:31 UTC · 3 likes
Cartoons Hate Her! (@CartoonsHateHer): One of my first cartoons. i was cooking i think
Not gonna happen to me 😤
Reply to @TetraspaceWest, 07:48 UTC · 6 likes
Tetra 💎⏹️🌌 (@TetraspaceWest): all normal parenting basically doesn't affect how well kids do when they're over 18, but also kids spend 18 years being under 18 and their experiences matter then also
Correct take
Reply to @AndyMasley, 18:06 UTC · 2 likes
Andy Masley (@AndyMasley): @LoganTCollins @AaronBergman18 thoughts?
Brilliant
Quote tweet, 18:33 UTC · 16 likes
Bruh
Quoting Politics UK (@PolitlcsUK): 🚨 NEW: GPs are threatening to strike if the NHS requires them to allow patients to book appointments online
[@Telegraph]
Reply to @carrier, 18:36 UTC · 1 like
Carrier (@carrier): Carrier is testing battery-enabled HVAC systems to help the grid when it needs it most. @EPRINews Read more: https://on.carrier.com/4n4vhjB
Obligatory @JesseTayRiver tag
Tuesday, September 30, 2025
Quote tweet, 00:21 UTC · 13 likes
I am so jealous, Berkeley people live out my dream
Quoting Aella (@Aella_Girl): there's a few hundred more of these left, in (pretty high-quality) hoodie form. they're gonna donate or throw away. Does anybody want them? you can pick up in berkeley
Quote tweet, 00:28 UTC · 186 likes
Not sure I feel great about this CoT trace
Quoting Jeffrey Ladish (@JeffLadish): You can tell an AI model that you won’t read their “hidden scratchpad" but they won’t necessarily believe you
Reply to @IsaacKing314, 03:34 UTC · 2 likes
Isaac King 🔎 (@IsaacKing314): @AaronBergman18 A fact I recently discovered, you can just fly there
Huge
Tweet, 05:19 UTC · 80 likes
Terminally math-brained people will take a 120 IQ conceptual reasoning task and turn it into a 150 IQ pure math task
The latter is more error prone regardless of how smart you are
Thread, continued, 05:19 UTC · 8 likes
Yes this is a subtweet of two of you. You know who you are
Thread, continued, 05:19 UTC · 6 likes
I do kinda relate though, I have a tendency to turn everything/anything into a data collection task
Reply to @ComInTheMarg, 05:37 UTC · 10 likes
Comments in the Margin (@ComInTheMarg): How interpretable are these scratchpads? Like do we know what "they parted illusions" means?
I am not the guy to ask lol, I think there's a lot of uncertainty tho
Reply to @jazzloaf, 05:50 UTC · 5 likes
vivian (@jazzloaf): latter generalizes better :(
Conditional on doing everything right, yeah, I guess it generalizes perfectly, but often easy to make subtle mistakes in the domain of translating between words/concepts and formal representations (in both directions)
and then you just look at the pure math steps in the middle and conclude the whole thing must be rock solid if those are
Tweet, 19:22 UTC · 19 likes
They should make a girl who isn’t cold
Reply to @TinkerElla95, 19:25 UTC · 2 likes
TinkerElla95 (@TinkerElla95): @AaronBergman18 I’m almost never cold. Maybe 3x a year.
Based
Reply to @kindgracekind, 22:20 UTC · 2 likes
Grace (@kindgracekind): I took OP to be mostly a reaction to the illegibility, not the deception (which was prompted for). But that’s my own interpretation, I could be wrong
Both tbh
End of September 2025. Previous month: August 2025. Next month: October 2025. Back to the archive index.