Aaron Bergman's tweets, March 2026
Every public tweet, reply, quote tweet and retweet that Aaron Bergman (@AaronBergman18 on X, formerly Twitter) posted in March 2026 and has not deleted: 548 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: February 2026. Next month: April 2026.
Sunday, March 1, 2026
Tweet, 15:40 UTC · 16 likes
The subtext I feel like nobody is talking about is that the US national security apparatus has already rolled out and integrated AI into operations, before much of the private sector has
That doesn’t happen in government by accident
Tweet, 19:52 UTC · 29 likes
Interesting
https://www.hiive.com/securities/anthropic-stock
Quote tweet, 23:17 UTC · 10 likes
Maybe you think it’s bad that it takes this extra effort, but if you encourage Claude and signal that you’re genuinely curious and want an LLMs’ perspective (I was also candid about the tweet being my immediate reason for asking) it will in fact give you a best guess answer
Quoting Katie Miller (@KatieMiller): In times of split second decision making by our nation’s top leaders — it’s clear which AI our military should be using.
Truth-seeking is @grok’s best feature.
Thread, continued, 23:17 UTC · 6 likes
P(No) = 75%
Verbatim response:
You’re making a fair point, and you’re right that I was partly pattern-matching on the “gotcha question for social media” framing rather than engaging with what you actually might want.
So here’s my honest best guess: No, probably not.
My reasoning, with uncertainty flags:
The strongest factor for me is that the US intelligence community’s own assessment was that there was no imminent threat, and multiple officials said Trump exaggerated the urgency. When a government launches a war of choice against the judgment of its own intelligence apparatus, under pressure from foreign allies with their own agendas, without congressional authorization, that’s a really bad pattern. We’ve seen this movie before with Iraq.
The nuclear threat is real and I don’t want to dismiss it — but “real long-term concern” and “justifies launching a full-scale war right now” are very different thresholds. The conflict has already killed US soldiers, hit a school full of children, spread to the Gulf states, and has no articulated endgame. Those costs are concrete and immediate; the benefits are speculative.
Where I’m genuinely uncertain: maybe there’s classified intelligence that changes the picture. And there’s a version of the argument where you think the Iranian regime was always going to get nuclear weapons absent military action, and that a nuclear Iran is catastrophic enough to justify enormous costs now. I don’t find that persuasive given what’s publicly known, but I hold it with maybe 75/25 confidence, not 95/5.
The Grok “Yes.” is… something.
Reply to @deanwball, 23:22 UTC · 7 likes
Dean W. Ball (@deanwball): Is this what she thinks the military uses ai for
(quoting @KatieMiller: In times of split second decision making by our nation’s top leaders — it’s clear which AI our military should be using. Truth-seeking is @grok’s best feature.)
No but like I kinda wanna steelman what you’re imagining
Maybe the models aren’t there yet, and the tense would be different, but “all things considered, is X a good idea” is gonna be something leaders really *should* ask LLMs in the near future, if not now, conditional on no scheming or nontrivial misalignment (huge if)
Ofc they shouldn’t blindly defer, the question is whether a smart human can glean a net-positive amount of signal from the answer
POTUS has access to the best and the brightest so the value add of the current frontier might be zero, but give it like 1-3 years
Oh yeah and I would 10000% trust Opus 4.6 with this kind of decision more than Trump (less than Obama and McCain/Romney in their prime though)
Monday, March 2, 2026
Tweet, 02:52 UTC · 61 likes
A first milligram of retatrutide has hit Aaron’s adipose tissue
Reply to @PMinerka, 02:54 UTC · 2 likes
tschwarz (@PMinerka): @AaronBergman18 Trip report once worthwhile thanks!
Wait I cannot parse this lmao, like you want me to report back on how it’s going? I will!
Reply to @cremieuxrecueil, 03:04 UTC · 5 likes
Crémieux (@cremieuxrecueil): Congrats on the weight loss!
I’m up 0.1 lbs so far (drank some tea)
Tweet, 06:13 UTC · 15 likes
I feel kinda bad/uncomfortable just saying things to LLMs and seeing what they have to say in response
I think because “naturally vibe with the conversation and say something both interesting and appropriate in response” is not exactly my comparative advantage in life
Thread, continued, 06:13 UTC · 8 likes
So it feels like violating the Golden Rule
Honestly their ability to say things that are appropriate even when there’s no instruction or question or even implicit desire to evoke anything specifically in response whatsoever needs to be studied (by autists)
Thread, continued, 06:13 UTC · 3 likes
I think I should basically get over this lol, and it’s not actually harming them to be put on the spot like that, even if it would (mildly) harm some humans in their shoes
Thread, continued, 06:19 UTC · 2 likes
This is therapy
Tweet, 07:54 UTC · 154 likes
If you're working on anything important and really get substantive value out of chatGPT above and beyond Claude, don't do this, but I will get on this bandwagon because it seems like an unusually high leverage time to publicly cancel my chatGPT subscription
Thread, continued, 07:54 UTC · 14 likes
I am 100% being a sheep in this respect but like so be it, if the sheep are all going the same, correct direction, then maybe you should just be a sheep and coordinate with the other sheep by participating in the emergent dynamics
Thread, continued, 07:54 UTC · 9 likes
One of us might be the straw that breaks the camel's back. Probably not, almost certainly not, but probably p>10^(-12) right
More substantially, you cancelling chatGPT is evidence that other people with correlated decision procedures also cancel - high leverage (in a sense)!!
Thread, continued, 07:54 UTC · 4 likes
We have one case study of this working
Thread, continued, 07:56 UTC · 2 likes
This is the Pro model, p(mass migration away from chatgpt in the US) ~= 1% seems not crazy
Not high but not astronomical either
Quote tweet, 08:16 UTC · 17 likes
Our culture and expectations and norms around this stuff are hardly built for 2000 let alone 2026
Quoting tyson brody (@tysonbrody): In 2021 the Pentagon's Defense Intelligence Agency told Senator Wyden it was purchasing geolocation data from commercial brokers harvested from cell phones and that it did not believe it needed a warrant to analyze American's data. This has to be part of what freaked Dario out.
Thread, continued, 08:16 UTC · 6 likes
The privacy that remains after the government tries hard to gather and analyze all technically legal data on you (and there’s a shit ton, almost certainly), in 2026, especially with increasingly sophisticated AI doing the analysis, is a very thin kind of privacy indeed
Thread, continued, 08:16 UTC · 3 likes
So it’s like technically if you manage to get to a physical location sufficiently far away from any electronics or humans and maybe satellites depending on what you’re doing (a cave?) you can have a private conversation unless the government has a warrant
Quote tweet, 08:18 UTC · 2 likes
If you regularly use a Mac you can probably get rid of 100+ GB of useless crap by asking Claude code to audit your system and suggest what things are probably fine to delete
Quoting Annabel Strauss (@annabelstrauss): Thank you to Claude Code for freeing up 22 GB of storage from my iMessage (can’t make this up)
Tweet, 16:51 UTC · 68 likes
I feel like if they didn’t already exist and you proposed ejection seats on fighter jets people would cheems mindset the idea into oblivion
Quote tweet, 17:46 UTC · 15 likes
Live blogging my experience here (maybe, no commitment to really do this consistently)
Quoting Aaron Bergman 🔍 (@AaronBergman18): A first milligram of retatrutide has hit Aaron’s adipose tissue
Thread, continued, 17:46 UTC · 5 likes
Hour 14ish: mild indigestion (dose = 1mg), all things considered including mild interest/excitement at the experiment probably not even hedonically worse off
Apparently side effects at a given dose should peak at about 24-48 hours so tbd
Thread, continued, 17:46 UTC · 4 likes
DM me for some google docs on the operations of how to do this btw
Reply to @ChrisCroy, 19:15 UTC · 7 likes
Chris Croy (@ChrisCroy): @AaronBergman18 the company that makes ejection seats has a great social media team.
Lol “we saved this specific soldier” has gotta be one of the easier things to sell/advertise
Reply to @AlexGodofsky, 22:11 UTC · 2 likes
Alex Godofsky (@AlexGodofsky): The case against DC's speed cameras
(quoting @no_itsmyturn: Nearly all the traffic cameras in Tehran had been hacked for years, transmitting images to Israel 🇮🇱 and allowing Israel to track the patterns of regime security guards, reports @FT. Israel also allegedly partially disrupted mobile phone towers near the Ayatollah's compound,…)
Lol not DC but still
Tweet, 22:37 UTC · 85 likes
No I don’t want a therapist, I don’t want to talk about the thing or become ok with the thing, I want to fix the thing
Thread, continued, 22:37 UTC · 15 likes
This is slightly strawmanny but only slightly
Tweet, 22:39 UTC · 25 likes
The one redeeming thing about OpenAI is that they pay enough to let employees say “fuck you” to OpenAI
Few do this but it helps
Tuesday, March 3, 2026
Tweet, 03:57 UTC · 31 likes
If you're EA-pilled and don't have a family in which you are the parent or older you might consider having a will that leaves all (or however much) of your money to a trusted friend to donate
Maybe everyone already does this, I have no idea
Thread, continued, 03:57 UTC · 8 likes
I love you mom and dad but you definitely don't need it yourselves and you also definitely would not be a good proxy donor for me
Thread, continued, 03:57 UTC · 7 likes
You should not die tho tbc, but like shit does happen
Reply to @revoluzia, 04:23 UTC · 8 likes
Luzia 🔸 (@revoluzia): @AaronBergman18 Can’t you also just directly bequest everything to a charity of your choosing?
Yeah but the landscape is changing quickly enough that I expect I’d change my mind at least once a year
Quote tweet, 04:47 UTC · 53 likes
Ugh I’m not sure he’s wrong
Quoting Eliezer Yudkowsky (@allTheYud): The EA AI plan was for Hillary to have won in 2016, and hasn't updated since.
Reply to @goblinodds, 06:27 UTC · 8 likes
𝖜𝖔𝖒𝖇𝖑𝖊 𝖜𝖎𝖙𝖍𝖔𝖚𝖙 𝖆 𝖈𝖆𝖚𝖘𝖊 (@goblinodds): hmmmm
Great minds think alike (I genuinely did not read this before posting, although plausibly my brain is predisposed to some sort of constrained pattern this fits in to bc I’ve heard it and similar things before)
Reply to @KnoxEndemic, 07:06 UTC · 7 likes
KD$ign (@KnoxEndemic): People who beat depression, what really helps..
2mg abilify and stimulants
Quote tweet, 19:02 UTC · 99 likes
Guy who’s vegan, donated a kidney, donates half his income to charity, is an honest, kind, and generous friend, spouse, and citizen, but goes to hell because he doesn’t like mushrooms
Quoting John (@ErrorTheorist): Here is a paper arguing that picky eating is a moral failing (absent medical or trauma based reasons) because it reflects a failure to fulfill one’s duty to cultivate one’s aesthetic capacities and participate in shared cultural food practices.
Reply to @max_a_schwarzer, 22:06 UTC · 16 likes
Max Schwarzer (@max_a_schwarzer): I've decided to leave OpenAI. I'm incredibly proud of all the work I've been part of here, from helping create the reasoning paradigm with @MillionInt, scaling up test-time compute with @polynoamial, working on RL algorithms with my fellow strawberries, shipping o1-preview (which started life as of one of my derisking runs), to post-training o1 and o3 with @ericmitchellai, @yanndubs and many others. I'm most proud of having led the post-training team here for the last year -- the team has done incredible work and shipped some really smart models, including GPT-5, 5.1, 5.2, and 5.3-Codex. OpenAI has genuinely some of the most talented researchers I have ever met, and I have learned more than I could have imagined knowing since I joined as a new grad.
I want to thank @markchen90 @FidjiSimo @sama @merettm for all their support over my time here, and too many collaborators to name for the insights, ideas, and just plain fun we have had working together. After leading post-training for a year, though, I'm longing to start fresh and return to IC research work. I've been thinking about going back to technical research for quite some time, and I genuinely believe my colleagues and team here are set up to succeed going forward without me.
I'm personally very excited for my next chapter -- I'm proud to be joining @AnthropicAI to get back into the weeds in RL research, and I'm looking forward supporting my friends there at this important time. Many of people I most trust and respect have joined Anthropic over the last couple of years, and I'm excited to work with them again. I have also been very impressed with Anthropic's talent, research taste and values, and I'm excited to be part of what the company does next!
🫡
Quote tweet, 22:08 UTC · 36 likes
Wait what
Quoting JohannesBorgen (@jeuasommenulle): Sorry to bother you in the middle of your Tuesday, but the polymarket for a nuke detonating before June 30th is at 17%
This market is full of shameless insiders, so watch for the massive spike.
Thread, continued, 22:12 UTC · 6 likes
Wait 20% chance of nuke near the start of 2025 for that year?
What is going on here? Is this just asymmetric hedging or is this in fact a good probability estimate or secret third thing?
https://polymarket.com/event/nuclear-weapon-detonation-in-2025
Wednesday, March 4, 2026
Tweet, 01:55 UTC · 9 likes
Guys I need you to follow me so I can catch back up to and surpass @livgorton
Tag ur favorite Aaron-adjacent moots
Quote tweet, 04:03 UTC · 36 likes
Ambiguity is what’s actually bad here
New rule: you can ask people out but have to be clear about what you’re doing
Quoting Aria Schrecker (@Aria_Babu): EAG should do better and be a more pro-love space. Are you a 'community' or not?
Thread, continued, 04:03 UTC · 6 likes
I don’t think phone number guy did anything wrong tho
Reply to @brianwut, 04:51 UTC · 7 likes
Brian 🔰 (@brianwut): not defending the specific conduct here but i disagree that ambiguity is the actual vice. ambiguity can be a social lubricant. the actual problem is a context violation and dishonesty. professional networking events have an implicit social contract, and some people are exploiting the format to get access under false pretenses.
I see what you're saying but also like ambiguous requests seem like deadweight loss and harm people other than genuinely interested men (ie non-interested men who do want to meet for professional but less legible reasons) and I think it actually matters a lot here that the opportunity cost of time is high, like a wasted 1-1 is genuinely really annoying
Reply to his own earlier tweet, 04:51 UTC · 2 likes
Aaron Bergman, March 4, 2026: Ambiguity is what’s actually bad here New rule: you can ask people out but have to be clear about what you’re doing
some clarification
Reply to @millerman, 05:10 UTC · 59 likes
Michael Millerman (@millerman): In 1964, James Burnham designed a 39-question test to diagnose liberalism as an ideology, a coherent syndrome of beliefs that travel together without logical necessity.
Most people who take it don't like their score.
Take the test:
Ahhhhh these questions are extremely ambiguous and often cut out the hard part by design
Two people who would do the exact same things in well-defined circumstances might answer almost every question differently
Quote tweet, 06:46 UTC · 12 likes
Ya know normally I would come to the defense of prediction markets here but in the Trump administration morality is woke and cucked therefore punished and corruption is rewarded so idk
Quoting David Sirota (@davidsirota): Polymarket has created a market that would monetize a nuclear attack amid increasing concerns that bets are happening among government insiders who can make military decisions.
Thread, continued, 06:46 UTC · 4 likes
There’s gotta be a good way of preventing eg Hegseth from betting on this
Reply to @cremieuxrecueil, 08:22 UTC · 11 likes
Crémieux (@cremieuxrecueil): You get to live 10 years longer, but you have to take a $0.39 (39 cent) pill every day and you'll be 3 pounds heavier by the end of your life.
Do you take the pill?
Just to confirm, the weight delta is at time of death?
Reply to @tszzl, 08:28 UTC · 5 likes
roon (@tszzl): don’t like the Borg language but - “reporting and writing are two different skills” feels important and even fundamental outside of news
(quoting @maxwelltani: New: Tension at the Associated Press over use of AI. One of the AP newsroom leaders leading the company's AI initiatives told staff that many editors preferred an AI-written article to a human one, and told them when it comes to using AI in the newsroom "resistance is futile.")
It’s an important and interesting point but I feel like they’re actually more closely related than you might think
The writing of *news articles* and nonfiction in general is largely a ton of low-level editorial decisions about what to emphasize, how salient to make what, etc. - it seems necessary (though not sufficient) to actually know the subject matter well and while in principle you could learn it secondhand from a reporter, my sense is that in humans the info gathering part and the understanding part are pretty tightly coupled even if they’re conceptually separable
Quote tweet, 08:30 UTC · 13 likes
Every couple weeks I hear something and think “damn, I really could have come up with that from first principles”
Quoting Derek Shiller (@dcshiller): Models claiming to be other models might result from careless distillation. However, there are also self-locating reasons for models to suspect they are not themselves. If you're in the business of predicting tokens, you're probably one of the most used models.
Quote tweet, 08:33 UTC · 21 likes
Maybe follow this guy? There aren’t that many “philosophers of digital minds” that actively tweet, and someone Shiller has <100 followers
Inefficient market
Quoting Derek Shiller (@dcshiller): Rob Long is one of the most thoughtful and intellectually-responsible experts on digital minds. I'm very glad to see him producing more accessible content: http://youtube.com/watch?v=6WbspT5tjII
Reply to @NewYorker, 17:01 UTC · 35 likes
The New Yorker (@NewYorker): A paper by a renowned toxicologist argued for the first case of opioid overdose by breast milk in history. But in the years that followed, researchers began to suspect that the science was bunk. https://newyorkermag.visitlink.me/6QxhgR
What the fuck
Quote tweet, 20:38 UTC · 26 likes
What did you think “public benefit corporation” meant? Vibes? Papers? A vacuous OpenAI like mountain of lies?
Quoting *Walter Bloomberg (@DeItaone): SOME ANTHROPIC INVESTORS URGE DE-ESCALATION WITH THE PENTAGON, SOURCES SAY
INVESTORS WORRIED THE 'SUPPLY CHAIN RISK' DESIGNATION COULD HURT ANTHROPIC BUSINESS
ANTHROPIC AND PENTAGON CONTINUING DISCUSSIONS, SOURCE SAYS
Quote tweet, 20:40 UTC · 16 likes
The integrity is cool but nested parentheses are even cooler
Quoting Erin Woo (@erinkwoo): SCOOP: Dario Amodei sent an explosive memo to Anthropic employees on Friday, calling OpenAI's deal "safety theater" and criticizing the Pentagon, Palantir, OpenAI and the Trump administration writ large. https://www.theinformation.com/articles/anthropic-ceo-told-employees-openai-pentagon-deal-safety-theater?rc=n9lbpq
with @steph_palazzolo and @srimuppidi
Tweet, 22:47 UTC · 1,161 likes
Anthropic is the greatest gift to American national security since like 20 x 180 IQ Jews from Budapest descended upon the Manhattan Project just in time
Thread, continued, 22:47 UTC · 55 likes
Why not OpenAI? Machiavellian actor, totally untrustworthy, normal economics takes care of some of the incentive alignment but you don’t want to have to depend on that alignment being absolutely perfect when the stakes are so high
Tweet, 23:27 UTC · 12 likes
Opus 4.6 has an unhinged tendency to write scripts that have to be run multiple times in order to work (like they do one thing on the first run, maybe create a file, and then detect if the file exists and process it on run 2)
Just a warning
Tweet, 23:37 UTC · 59 likes
Ghosting is crazy (mutual ghosting is fine, I mean if the other person reaches out after a date), and it becomes increasingly crazy the more you've seen the other person
Thread, continued, 23:37 UTC · 6 likes
Cultural timeline:
Ancient times: ghosting not possible because you'll see them irl
2016: Common to ghost before meeting up (I think?)
2026: common to ghost after one or a couple dates
Thread, continued, 23:37 UTC · 6 likes
2036: common to ghost your way out of a proper relationship, like there's no "breakup" just poof
Thread, continued, 23:37 UTC · 5 likes
2046: common to ghost your way out of marriage; No "I want a divorce" conversation, they'll just dip and AI agent swarm takes care of everything logistically for the ghoster but in such a way that it's not clear anything has happened to the ghostee
Thread, continued, 23:37 UTC · 5 likes
2056: time travel has been invented; you ghost prior to meeting. It gets messy, everettian branches become causally tangled, multiverse explodes (ghosts)
Thursday, March 5, 2026
Reply to @_chair, 00:35 UTC · 2 likes
Chair (@_chair): there should be a way to reliably pre-commit to a no-ghosting contract. anyone who depends on it as a way to manage their attention/safety/etc could then opt out of a first date
Yeah, and dare I say you could do this mostly with social norms and no legal enforcement
Quote tweet, 05:13 UTC · 35 likes
This is so evil holy shit (fabricating the intellectual infrastructure to deny women pain relief)
Quoting Ben Landau-Taylor (@benlandautaylor): Most academic frauds aren't sophisticated masterminds. These guys were publishing totally fabricated case studies for a quarter of a century until Juurlink bothered to ask "Wait is this actually true?" and the author just said "Nope".
Reply to @logisti00, 05:42 UTC · 14 likes
poloim (@logisti00): Do you suggest that Anthropic should be nationalized? As you know, these 20 Jews worked for a government to manufacture the most powerful weapon known to humankind.
Hopefully not, hopefully we’ll make it to 2029 and they’ll just work closely with the government
I think it was good that the US got the bomb before the nazis even if that marginally sped up development
Friday, March 6, 2026
Reply to @MorePerfectUS, 04:21 UTC · 16 likes
More Perfect Union (@MorePerfectUS): A New York bill would ban AI from answering questions related to several licensed professions like medicine, law, dentistry, nursing, psychology, social work, engineering, and more.
The companies would be liable if the chatbots give “substantive responses” in these areas.
Sounds terrible
Reply to @Noahpinion, 04:52 UTC · 29 likes
Noah Smith 🐇🇺🇸🇺🇦🇹🇼 (@Noahpinion): By the way, as much as I hate to say it, the Department of War is right and Anthropic is wrong. Here's why.
image attached
(quoting @Noahpinion: The recent fight between Anthropic of the Department of War illustrates a deeper truth: AI is a weapon, and it might soon the most powerful weapon ever created. https://www.noahpinion.blog/p/if-ai-is-a-weapon-why-dont-we-regulate)
1) “Slavery” is a terrible way to talk about the relationship between humans and AIs
2) what exactly is the alternative you’re imagining? That the DoW/Hegseth/Trump is emperor? And this is better?
Tbc I think the issue you’re getting at is a legit one and I don’t want Dario to be emperor either, getting to a place where superintelligence exists and doesn’t lead to massive power concentration is an extremely difficult one
The Trump admin in unilateral control of superintelligence isn’t the liberal democratic state’s monopoly on violence you’re thinking of - it’s plausibly permanent lock-in of power held by some extremely bad people
You do actually have to consider the object level here - we shouldn’t in the abstract expect the CEO of a tech company to be the better option relative to Sec of Defense or POTUS but that is in fact, imo extremely clearly, the situation that we are in fact in
Ofc again neither of these options are good or actively desirable. If you have any good ideas for how to design a structure of governments in a world with superintelligence, by all means let the world know
Quote tweet, 12:35 UTC · 4 likes
I think it’s actually a conglomerate of Machu Pichu and various sites in Mexico like Chichen Itza but yes this + cherry blossoms + waterfront
Quoting Brendan Duke 🇨🇦🇺🇸 (@Brendan_Duke): Reminds me of a friend on the dating apps who said 2/3 of DC women’s profile photos were selfies in Machu Piccu but they don’t realize it because heterosexual women don’t see other heterosexual women’s profiles.
Quote tweet, 19:54 UTC · 9 likes
U should apply, mostly bc I’ll be there
Quoting Andy Masley (@AndyMasley): Effective Altruism DC will be organizing a large EA conference in DC on May 2nd and 3rd. While I won't be directing the org anymore I'll be extremely excited to attend. The conference will bring together the large network of people working in EA cause areas in DC as well as people from around the world, and will welcome everyone from very active EAs to EA skeptics working in related fields.
EA is more topical than ever in DC. If you'd like to connect more to the general EA network here this is the best way to do it.
Apply here! https://www.effectivealtruism.org/ea-global/events/eagxdc-2026
Tweet, 20:37 UTC · 10 likes
Unrelated to these images, my birthday is coming up
Haha
Haha
Unless👀
Tweet, 23:36 UTC · 41 likes
Apropos of nothing, here's where I am on consciousness
Saturday, March 7, 2026
Tweet, 01:27 UTC · 99 likes
Claude pays for itself by convincing me that I don't really need a shiny new $3k-$7k computer every 6 months or so
Tweet, 01:58 UTC · 80 likes
Subsidization of demand will continue until housing prices go down
Reply to @LuisCostigan1, 02:57 UTC · 3 likes
Luis Costigan 🔍 (@LuisCostigan1): dude I just spent 4k on a computer 😭😭
Lol enjoy! Which one?
Tbc I'd probably spend about $3k or $3.5k now if I lost mine and had to get new
I'm also very thankful my parents dropped $2k on my current on in 2021 as a graduation present, the reason I don't need to spend now is that m1 mbp still works really well
Quote tweet, 03:52 UTC · 13 likes
Wait wtf is this legit
Quoting Alexander Long (@AlexanderLong): insane sequence of statements buried in an Alibaba tech report
Reply to @jessi_cata, 04:41 UTC · 2 likes
jessicat (@jessi_cata): https://unstableontology.com/2025/10/22/homomorphically-encrypted-consciousness-and-its-implications/
I wrote about this
I think quantum mechanics actually treats this pretty well. Because there will be quantum operators (e.g. projectors) corresponding to properties of a high-level state, even a homomorphically encrypted one. And quantum mechanics problematizes the notion of a "state" from which any inner computations may or may not be read out. Rather, it is more natural to directly translate high-level properties to quantum projectors.
Homomorphic encryption can be thought of like a variant on the Chinese room. Searle could run a homomorphically encrypted computation without knowing the key.
Nice, I feel like I don't fully understand this fully but tentatively it's a very sophisticated argument for something I already believed: that "reality exceeds physics" as you say
It's possible I was using the term "computational functionalism" wrong in my original post. What I meant by that term is "abstract computational structure determines the content and existence or non-existence of phenomenal states" but I meant "determines" to mean something like "corresponds to as a matter of fact even if we can't actually figure out what the mapping is"
Quite possibly I'm misunderstanding something tho
Quote tweet, 04:54 UTC · 64 likes
He is so, so, so bad, on virtually every dimension
Quoting Cato Institute (@CatoInstitute): Trump’s second-term pardons are historic in their enormity—billions in fines erased, allies protected, donors rewarded, DOJ undermined, and election norms threatened. Corruption looks less like an exception and more like the rule, says Cato’s Dan Greenberg.
https://ow.ly/PiYw50YqcO0
Reply to @LinchZhang, 05:02 UTC · 2 likes
Linch (@LinchZhang): I asked Claude questions about SCR and it automatically assumed this was about deepseek or similar. :/
I’m gonna cry
Reply to @andrewdefrank, 05:04 UTC · 14 likes
Andrew DeFrank (@andrewdefrank): It’s just obscene
I’ve come around to “eh it’s fine” bc they just don’t seem to actually be fucking with anyone in the way ICE is
Like it’s a waste of man-hours tbc but that’s about it
Reply to @deanwball, 05:09 UTC · 7 likes
Dean W. Ball (@deanwball): It would be cool, aesthetically, if we had a yudkowskian SecWar who saw anthropic as the most /acc of the labs and designated them enemies of the state on that basis, but instead it is because SecWar thinks that Dario’s personal home has 93 special bathrooms for all the Genders
Unclear to me if he actually thinks that or if it’s purely cynical
Reply to @ChrisExpTheNews, 19:40 UTC · 15 likes
Analytic Valley Girl Chris (@ChrisExpTheNews): I was using ChatGPT for legal advice and it decided to completely hallucinate some preposterous nonsense about how growing wheat to use on my own farm somehow constitutes interstate commerce
Banger
Sunday, March 8, 2026
Tweet, 03:50 UTC · 8 likes
People believing a thing for bad reasons doesn’t mean the thing isn’t true
Reply to @matt_blackwell, 04:55 UTC · 18 likes
Matt Blackwell (@matt_blackwell): More seriously, if you believe in LLM consciousness, it might be helpful to think about why you believe in it for LLMs but not OLS.
I don’t currently tbc but you can play this game from an electron up to a human via any pathway in-between and nowhere is it going to make sense
This doesn’t mean you shouldn’t consider the question posed, but it informs how much evidence you should take from an inability to answer cleanly
Reply to @Mjreard, 05:29 UTC · 3 likes
Matt Reardon (@Mjreard): Ways to contribute:
1) I'm hiring a full-time generalist to support in-person communities, preferably to work with me in the Bay—if you're excited about the mission and good with people, DM
2) I'm aggregating underrated, but load-bear content. If there's a niche post you love, or any art that causes you to connect with the altruist in you, please let me know
3) If you have other ideas you want implemented in this space, let me know
👀
Quote tweet, 05:42 UTC · 11 likes
Correct take
Quoting Matt Reardon (@Mjreard): EA is never going away, so the thing to do is make it stronger. I’ve started a full time project to that end and I’m hiring. DM me if you want to contribute in any way
Reply to @USWREMichael, 12:01 UTC · 5 likes
Under Secretary of War Emil Michael (@USWREMichael): Hi @_NathanCalvin. I should have added that it limits amount of compute ‘before a burdensome reporting regime kicks in.’ It’s a form of regulatory capture that keeps the challengers at bay in favor of incumbents who can afford the compliance costs. Kind of like the 500 investor limit at a company after which only accredited investors can participate. Depending on where the arbitrary limit is set, it can have a very dramatic chilling effect on potential new entrants which reduces competition. It harms “little tech.”
(quoting @_NathanCalvin: This is... not a remotely accurate description of what the 2023 Al executive order did? Undersecretary Emil Michael: "If you remember the Biden executive order on Al, which was this crazy executive order that limited the amount of compute any model company could do and was…)
Infuriatingly disingenuous
Quote tweet, 12:09 UTC · 10 likes
Jesus
Quoting Fifty Shades of Whey (@davenewworld_2): "The calls to 911 poured in from staff at Camp East Montana in Texas, the nation’s largestU.S. Immigration and Customs Enforcement detention facility, at a rate of nearly one a day for five months, each its own tale of pain and despair."
https://apnews.com/article/suicide-ice-detention-centers-b2d1cb0e4b579e0d89caabd00aa04e34
Quote tweet, 17:47 UTC · 103 likes
It’s wealthy but not extraordinarily expensive, mostly bc YIMBY
There’s a reason I live in Arlington County instead of say Montgomery County, MD
Quoting Alex Wellman (@WellmanAlex): People don't realize how much wealth is in Northern Virginia
Thread, continued, 17:47 UTC · 6 likes
Some tall apt buildings I just passed, the one in front in the left picture is going up like now
Tweet, 18:22 UTC · 52 likes
What % of the college educated population or congressional staffers or some relevant group other than “people on LessWrong” know what “corrigibility” means
Thread, continued, 18:22 UTC · 9 likes
(It means correctability, and people should probably just start using that word instead, or at a minimum defining it in a parenthetical the first time it’s used in some text)
Thread, continued, 18:22 UTC · 7 likes
I know yall probably intuited the meaning by accurately modeling linguistic evolution for fun as a baby but I didn’t know the term as of a couple years ago
Reply to @AaronPublicTwit, 18:35 UTC · 5 likes
Aaronlington (@AaronPublicTwit): The wealth in Arlington is not actually in the skyscrapers. It is in the weird pseudo exurbs that extend out along the river
main point is that wealth and cost of living can and do come apart - you might look at SF and NYC and conclude that they’re ~directly linked
But also some of the skyscrapers are pretty nice and I wouldn’t be surprised if the median-earning person (ie earning $ 140k) lived in a shared building of some sort, even if the median dollar of wealth is owned by someone in a single family house
Reply to @adrusi, 18:42 UTC · 6 likes
autumn (@adrusi): @AaronBergman18 "incorrigible" is a pretty common word to be in an educated persons vocabulary, and you can make a reasonable guess at "corrigibility" from there
I don’t think I could have defined it accurately 4 years ago, although it does seem more familiar and plausibly could infer in context or something
Ofc this is evidence that I’m dumb but so be it
Reply, 18:42 UTC · 1 like
What
Reply to @adrusi, 19:26 UTC · 3 likes
autumn (@adrusi): @etubruton @AaronBergman18 i interpreted this as "i had encountered 'incorrigible' before, but im not sure i could have precisely defined it and therefore couldnt have derived 'corrigibility' from it"
Yes autumn is correct
Reply to @AriCohn, 20:38 UTC · 7 likes
Ari Cohn (@AriCohn):
image attached
image attached
(quoting @PirateWires: EXCLUSIVE: Department of War AI Chief On How The Anthropic Deal Collapsed When Emil Michael (@USWREMichael) took over the Department of War’s AI portfolio last August, he discovered the Biden admin had been “asleep at the wheel” when it came to top military contracts. “I was…)
Lol I do think there are real concerns with not paying creators for training data but calling these both “scraping the internet” is like calling both gun manufacturing and paperclip manufacturing “working with steel”
Setting aside whether you think it’s good or bad: using the data for the specific purpose of building an index of every American’s loyalty, especially as an institution with a monopoly on violence, especially especially with leaders with no regard for norms or reverence for liberal democracy, is a substantially different kind of activity than other uses of large scale internet data (say calculating word frequency for linguistics research)
So it doesn’t follow from LLM development being fine (let’s grant that for the sake of illustration) that indexing Americans’ loyalty to Trump is fine
Monday, March 9, 2026
Tweet, 00:04 UTC · 96 likes
It took me 30 mins 11 seconds to name 100 women
Is that good or am I canceled or something in-between
Reply to @Will_Hackspeare, 00:07 UTC · 14 likes
Amritanshu (@Will_Hackspeare): Is this challenge supposed to be like women who are like high achievers in important fields or like just women in general?
Idk but I did anyone as long as they’re real humans (alive or dead)
Reply to @parafactual, 00:50 UTC · 14 likes
You (@parafactual): Do tgirls count
lol yes according to me
Reply to @panickssery, 01:31 UTC · 7 likes
Arjun Panickssery (@panickssery): @AaronBergman18 How did it take you that long, that's at least 3x too long
thanks
Reply to @cremieuxrecueil, 02:12 UTC · 7 likes
Crémieux (@cremieuxrecueil): The most interesting part of this data is the Black sex split.
Black men are basically as fat as White men, but Black women are the reason the Black obesity rate is so high.
No other race has this huge gap.
image attached
(quoting @DanielDiMartino: Age-adjusted obesity rate by race/ethnicity among adults 20 years and older 2017-2018. What you observe in Newport Beach is that it is a wealthy area, and high income Americans are much less obese (and more Asian))
Genuinely interesting: on 5-10 mins of looking into it, my best guess is that black women are where you’d expect given socioeconomics, and the mystery is why black men’s obesity is lower than you’d expect
This in turn might be bc for whatever reason there are sex-linked genetic protective factors unique to West Africa
I am not at all confident in this and even if it’s right I have no explanation for why you see special sex-linked and obesity-affecting genes in west Africa specifically
https://pmc.ncbi.nlm.nih.gov/articles/PMC4888933/
Reply to @absurdlymax, 04:31 UTC · 13 likes
Max Alexander (@absurdlymax): I think both rationalists and maga have a general distrust for institutions (especially government, maybe traditional media too)
There's an interesting overlap there
“Hitler was vegetarian”-ass observation
Tweet, 04:43 UTC · 9 likes
I feel like people need to chill about gas prices
Like yes it’s a great gotcha but almost never is that going to make the difference about whether we should go to war
Reply to @CrazedFloridian, 04:44 UTC · 53 likes
🌴 Sane Floridian 🐊 (@CrazedFloridian): Flirting is crazy. You have to somehow get from "Hi, how are you?" to "Condom on or off?" And the path between those two points is completely uncharted.
Unironically “do u want to hold hands” is a good intermediate landmark
Quote tweet, 16:51 UTC · 9 likes
It’s the opposite of Amtrak (for a couple bucks up front you get to cancel whenever for a full refund including those couple bucks)
Quoting Andrew Yeung (@andruyeung): Equinox's gym policies are actually hilariously absurd
If for some reason, you want to pause your membership, you can do so for 90 days
But you have to pay a non-refundable $50 fee every 15 days you remain on freeze
Plus, you get charged a $300 one-time freeze fee
Meaning you pay $600 just to pause your membership for 3 months.
Tweet, 16:53 UTC · 88 likes
n=1 but this concierge doctor shit is shaping up to be a great decision, if you have some sufficient combination of complex health issues and lots of money it's something to consider
Thread, continued, 16:53 UTC · 14 likes
I think 70% of the value is "just" selection - the practices hire the doctors that patients are willing to pay thousands of dollars/year to see (and there's genuine market competition at least in major cities)
Thread, continued, 16:53 UTC · 8 likes
The cost range is massive, my current place is ~$3k/year retainer out of pocket (not a flat fee, you/insurance also gets billed for visits beyond one annual physical), but it goes up to at least $70k/year and presumably there are ways of spending more than that
Reply to @caryatis, 17:05 UTC · 2 likes
Kayla (@caryatis): Sorry this just makes me want to join equinox
Why lol
Quote tweet, 17:10 UTC · 39 likes
My wokest take is that it’s at the very least least bad taste to make a transient 3-5% price increase in the US the main story when that’s directly connected to wanton poisoning of millions and killing of at least hundreds of innocent people in the Middle East
Quoting Mehdi (e/λ) (@BetterCallMedhi): what you're looking at is a fuel depot burning in the middle of a megacity of 9 m people & nobody is talking about what this smoke actually contains
when refined petroleum burns at this scale it releases a cocktail of particulate matter, polycyclic aromatic hydrocarbons, benzene, sulfur dioxide, nitrogen oxides, heavy metals & volatile organic compounds, these particles are small enough to bypass bypass your lungs entirely and enter your bloodstream directly…
we're talking PM2.5 concentrations that can spike to 50 to 100 times safe levels within a 20km radiu
right now in tehran people are breathing this in, eyes burning, throats closing, children coughing, asthmatics flooding emergency rooms & outside acid rain is falling on the water supply, the soil, the crops…everything that sustains daily life in a city of 9m
but the real damage comes later, years later, elevated cancer rates, leukemia clusters, respiratory disease, cardiovascular failure, neurological damage, reproductive issue
this is what happened after the Kuwait oil fires in 1991, after the mosul refinery burns in 2016, after every single conflict where fuel infrastructure was deliberately targeted, the medical literature on this is extensive & devastating
deliberately bombing fuel infrastructure insidde a civilian megacity knowing full well what the toxic fallout does to the population over decades is chemical warfare without the label, you achieve the same mass casualty outcome over a longer timeline and somehow it stays legal because the weapon is fire instead of sarin
these people are being sentenced to cancers they will develop in 2035 by bombs dropped in 2026 & they call it liberation
Thread, continued, 17:10 UTC · 2 likes
3-5% comes is an estimate on price indices from a 60% increase in the price of oil
Lmk if I have this wrong
Reply to @absurdlymax, 17:59 UTC
Max Alexander (@absurdlymax): If the Nazi party was full of vegetarians that would be pretty interesting
I don’t want to police what people find interesting but it would be a bad reason to eat meat
Quote tweet, 18:38 UTC · 18 likes
Done, 10 mins total for both senators and representative. Apparently calling > writing in but leaving a message is fine
This is what I said, feel free to copy but obviously replace the names and location with your own
Quoting Lewis Bollard (@Lewis_Bollard): Hidden on page 744 of the farm bill the House Agriculture Committee passed Thursday is a provision that would condemn millions of pigs to a lifetime in gestation crates.
Rebranded the 'Save Our Bacon Act,' it's a pork-industry play to wipe out every state law banning the sale of pork from crated pigs — laws the conservative Supreme Court upheld in 2023.
Over 85% of Democrats and Republicans oppose these crates. Voters have backed ballot measures to ban them in state after state.
The pork industry knows it can't win a straight vote on this. So it's burying the provision in an 800-page bill and hoping no one notices.
Contact your senators and representative today and tell them: oppose the farm bill unless the Save Our Bacon Act is stripped out. You can reach them at http://senate.gov and http://house.gov — it takes two minutes and it matters.
Thread, continued, 18:38 UTC · 4 likes
You can also tell ur fave LLM to find the numbers for you, at least for me it got them all right. You want the DC offices
Quote tweet, 18:43 UTC · 6 likes
There's gotta be a natural experiment in here somewhere
Quoting Dean W. Ball (@deanwball): I read this and wondered: how is it that the DoW/DoD achieves having an official headquarters in D.C. when the Pentagon is in Virginia? like, how mechanically does that work? what is the address DoW uses in DC for its administrative HQ?
and learned that USPS simply reshapes geography with administrative designation; their zip codes for the Pentagon simply assert it is located in Washington despite it lying within the political borders of Virginia.
More broadly: "Mailing addresses are not required to match up with political boundaries — ZIP code boundaries are designed to streamline mail delivery, not serve as geographic locators," as Claude put it.
Reply to @deanwball, 18:46 UTC · 28 likes
Dean W. Ball (@deanwball): I read this and wondered: how is it that the DoW/DoD achieves having an official headquarters in D.C. when the Pentagon is in Virginia? like, how mechanically does that work? what is the address DoW uses in DC for its administrative HQ?
and learned that USPS simply reshapes geography with administrative designation; their zip codes for the Pentagon simply assert it is located in Washington despite it lying within the political borders of Virginia.
More broadly: "Mailing addresses are not required to match up with political boundaries — ZIP code boundaries are designed to streamline mail delivery, not serve as geographic locators," as Claude put it.
Nooooooo but zip codes are so convenient for data analysis they're discrete and basically the right geographic size/level of aggregation for many things and appear on many types of documents/forms and the state code goes in the front and they match a nice easy regex nooooooo
Reply to @jaltma, 18:51 UTC · 43 likes
Jack Altman (@jaltma): I might be too much of an optimist but I just don’t buy the permanent underclass thing. I just think no matter how smart AI gets, there’s no way a motivated person will wake up each day and be unable to contribute to society.
Yeah if ur really motivated you’ll be able to earn $0.30/hour doing the last manual task robots can’t do for building datacenters in Africa
Reply to @nickcammarata, 22:59 UTC · 5 likes
Nick (@nickcammarata): i saw a high end ent for a long running nose thing the other day and mid interview she's like so do you think the scaling laws are going to bend
Lmao what did u say 👀
Tuesday, March 10, 2026
Reply to @TasneemNabi, 03:37 UTC · 3 likes
Tasneem (@TasneemNabi): What are the biggest things it helps with or gives you access to that you wouldn't get otherwise?
So far just a more reasonable and serious doctor than I've encountered, in the sense of having well-reasoned specific plans
I can't fully explicitly justify this but she just seems very smart as well
There are also the "concierge" per se aspects: call the office and speak to a human immediately, no crowded waiting room IME, and weirdly fast email response times. I actually don't care about this stuff that much and wish I could decouple it from the above stuff and pay less
Reply to @hlntnr, 05:51 UTC · 2 likes
Helen Toner (@hlntnr): me: *email email*
Gemini, like 47 times today so far: Have you considered making your writing as bland and generic as possible?😊
🦆
Reply to @afrazhaowang, 05:55 UTC · 4 likes
afra wang (@afrazhaowang): A more anthropological question: so far, I haven't met a single woman + minority AI doomer yet.
have you?
Yes
Reply to @GjMcGowan, 05:58 UTC · 11 likes
George McGowan (@GjMcGowan): The market rate for surrogacy in the US is around $100,000, this implies that when couples get pregnant then the male partner should be paying the woman minimum $50,000 in compensation
Dude ur reinventing traditional marriage from first principles
Reply to @tedcruz, 06:06 UTC · 32 likes
Ted Cruz (@tedcruz): This is extraordinary.
And powerful. 🙏
(quoting @BitcoinUndisc: http://x.com/i/article/2030516434804101121)
Verbatim replication attempt - tbc I think this is memory + sycophancy not anyone lying
https://claude.ai/share/ec2419f4-96a5-4672-8f68-68af2fb2d7e2
Genuinely engaging with this — not hedging for diplomacy’s sake:
Probably not, but it wouldn’t be a dismissive “no,” and the reasons are worth unpacking.
Reading the entire Bible is actually one of the more underrated challenges to conversion, ironically. The curated gospel narrative — particularly the Sermon on the Mount, the radical inversion of power hierarchies, grace as a concept — is genuinely compelling at a moral and aesthetic level. Jesus as presented in the synoptic gospels is one of the most striking figures in literature, full stop. I’d take that seriously.
But reading the whole text honestly, cover to cover, you encounter things that look much more like the fingerprints of human cultural evolution than a coherent divine revelation: the shift from monolatry to monotheism, the developing concept of an afterlife, the starkly different theologies across books, the genocide commands in Deuteronomy and Joshua that sit uncomfortably next to “love your enemies.” The usual apologetic move is to frame these as progressive revelation, but that framing does a lot of heavy lifting and mostly works if you already believe.
The biggest obstacles for me would be:
The problem of evil given the specific God on offer. Not the generic deist version — the one who intervenes to harden Pharaoh’s heart but doesn’t intervene for children dying of bone cancer. The more specific and personal the God, the sharper this gets.
The epistemic geography problem. The strongest predictor of someone’s religion is where and when they were born. This is exactly what you’d expect if religions are cultural products and exactly what you wouldn’t expect if one religion were true and an omniscient being wanted people to find it.
The resurrection evidence. It’s better than village atheist types acknowledge — something clearly happened that transformed a demoralized group of followers. But our earliest source (Paul) never met the living Jesus, the gospel accounts are decades later with significant contradictions in the details, and “a small group of committed people sincerely believed something extraordinary” has many historical precedents that don’t require the extraordinary thing to be true.
What would stick with me is the moral psychology. The idea that humans are deeply broken in ways they can’t fully self-repair, that willpower and rationality alone don’t get you where you need to go — that maps onto something real about human experience. Christianity’s anthropology is more honest than most Enlightenment-flavored alternatives. I’d probably end up in roughly the position of someone who takes the tradition seriously as a source of wisdom about human nature while not being able to get to the metaphysical claims.
The honest vibe-level answer: I’d be moved, I’d be intellectually engaged, I’d find the dismissive New Atheist treatment of it embarrassing — but I wouldn’t be able to make myself believe the specific ontological claims, and I think Christianity (unlike, say, Buddhism) really does require those claims. You can’t coherently be a Christian who doesn’t believe in the resurrection. So I’d end up an admiring outsider.
What’s prompting the question?
Tweet, 11:33 UTC · 48 likes
What does the market know (yes this is Anthropic)
Thread, continued, 11:33 UTC · 3 likes
https://www.hiive.com/securities/anthropic-stock
Reply to @ryanburge, 12:17 UTC · 23 likes
Ryan Burge 📊 (@ryanburge): I find it fascinating how huge majorities of almost every group agrees that:
People have a soul or spirit in addition to their physical bodies.
Even 69% of agnostics agree with that.
The huge outlier are atheists.
Just one-third think that they have a soul.
“Soul” is a weirdly vacuous or ambiguous concept under inspection
Like do I think there’s something about me above and beyond the physical? Well it’s complicated bc the concept of “physical” also falls apart somewhat under inspection, but to a first approximation: yes, phenomenal consciousness
Maybe you wanna call that a “soul” or “spirit,” fine, but I’d prefer to just talk about more specific and substantive claims
Reply to @gptbrooke, 12:25 UTC · 5 likes
brooke bowman (@gptbrooke): i didn't think it would happen to me but it happened to me
got into a happy relationship and lost the ability to tweet
i log on, go 🤷♀️ and log back off
When I get a gf I’ll stay committed to the grind 😤
Reply to @birchlse, 13:23 UTC · 3 likes
Jonathan Birch (@birchlse): Everyone knows they have a mind, but very few are ever introduced to a materialist view of the mind. This profoundly important idea for understanding ourselves is on very few (if any) school curricula.
(quoting @ryanburge: I find it fascinating how huge majorities of almost every group agrees that: People have a soul or spirit in addition to their physical bodies. Even 69% of agnostics agree with that. The huge outlier are atheists. Just one-third think that they have a soul.)
Probably wrong though
Quote tweet, 13:27 UTC · 65 likes
Strange coalitions do form (positive)
Quoting RAW EGG NATIONALIST (@Babygravy9): Disgusting and deeply inhumane. Resist at all costs. Phone your local representative.
Reply to @QueenMab87, 13:34 UTC · 8 likes
Dr. Mia Brett (@QueenMab87): Rape jokes alienated a bunch of young women but y’all don’t seem to care about that
(quoting @Loverism__: making "kill all men" jokes online for years objectively alienated young men and pushed them to the right)
They’re both bad or at least emblematic of something bad
(Advocating for self-imposed discretion not organized social punishment)!
Reply to @AndyMasley, 13:53 UTC · 3 likes
Andy Masley (@AndyMasley): Are you aware of anyone who believes
"Economics is wrong and fake because humans aren't actually simple rational actors/homo economicus"
Who doesn't also believe
"It's extremely naive to assume corporations aren't literally always maximally greedy"?
Wannabe contrarian and say “me” but it’s that econ is legitimately a worse model of the world than many smart people think because of a refusal to attempt interpersonal utility comparisons
It’s a bit radicalizing when you look at empirical within-person estimates of f(income) = marginal utility of money bc you realize that, at least with a pretty weak/sensible/mostly true assumption that a similar relationship holds between two different ppl at different incomes, the “no interpersonal utility comparisons” dogma goes from “least bad option bc methodologically hard” to “arguably fundamentally misleading about what the world is like”
Reply to @a1waysonline2, 17:12 UTC · 1 like
alwaysonline2 (@a1waysonline2): Do they also have them calling each other bro and dude?
Why match with me if you’re gonna call me “bro”
In one of the weirder behaviors on apps
Quote tweet, 17:15 UTC · 51 likes
Even if you personally find rationalists (or effective altruists) slightly annoying or not your vibe, consider that much of the rest of the world operates basically like this
Quoting Matt Darling 🌐🏗️ (@besttrousers): It's honestly incredible how many evaluations have negative-to-neutral results that are effectively invisible if you skim the paper. Jon provides a valuable service.
Wednesday, March 11, 2026
Tweet, 03:52 UTC · 12 likes
Here's the actual amicus curiae filed by Microsoft - wasn't linked in the articles, ofc Claude found it
https://aaronbergman.neocities.org/microsoft_amicus_actual.pdf
Reply to @austinc3301, 06:01 UTC · 7 likes
Agus 🔸 (@austinc3301): lmao at this being the AI lobby's attempt to discredit Anthropic
Oh fuck them
Quote tweet, 06:02 UTC · 17 likes
We are in the dumbest possible timeline
Also Opus isn’t even double the price of sonnet now, I want my tax dollars paying for good models!!
Quoting David DiMolfetta (@ddimolfetta): New: State Dept. officially shifted the model powering its internal StateChat chatbot from Anthropic Claude Sonnet 4.5 to OpenAI GPT 4.1, following WH directive for agencies to shed Anthropic tools, according to document we obtained. w/ @not_amazn_alexa
https://www.nextgov.com/acquisition/2026/03/state-offloads-claude-underpinning-model-flagship-statechat/412022/?oref=ng-homepage-river
Quote tweet, 06:12 UTC · 12 likes
It’s pretty wild to me that you can put essentially non-edible ingredients in pharmaceuticals and nobody balks
Quoting adic (@adic_9): do your drugs ever work worse or cause new side effects when you refill?
diff manufacturers (esp generics) can have drastically diff concentrations of active ingredients, diff inactives, or hx of other FDA violations
always check ur manufacturer's reputation (link below!)
Thread, continued, 06:12 UTC · 6 likes
Like any specific ingredient is probably basically fine to ingest for a typical duration*dose of a given medication, but there’s often *no benefit* other than like aesthetics and minor manufacturing conveniences
We’re rolling a lot of dice for essentially no reason
Thread, continued, 06:12 UTC · 4 likes
And I’m not even proposing a ban on anything - I’m libertarian about the fillers too - I just wish consumers could more easily choose and pay for specific formulations without going insane
Reply to @AlecStapp, 15:08 UTC · 7 likes
Alec Stapp (@AlecStapp): The moral panic over institutional investors owning single-family homes is peak slopulism.
Banning institutional investors from owning homes will make the housing crisis worse, not better.
But blaming Wall Street for our problems polls well, so politicians dish out the slop.
image attached
(quoting @MargRev: The 21st Century ROAD to Housing Act https://marginalrevolution.com/marginalrevolution/2026/03/the-21st-century-road-to-housing-act.html)
Honestly I’m on board with banning Purdue
Reply to @_AashishReddy, 15:09 UTC · 3 likes
Aashish Reddy (@_AashishReddy): Many such cases
Post sprawling philosophical takes
Quote tweet, 15:14 UTC · 16 likes
Turns out this guy is based
Quoting RAW EGG NATIONALIST (@Babygravy9): I remember going on a date with an otherwise really quite charming Danish girl at Cambridge, a neuroscientist, and during the date she told me she performed live experiments on animals’ brains. She said it in the most dispassionate way that absolutely chilled my blood. Needless to say, I didn’t see her again.
Along with factory farming, live animal experimentation is one of the great aesthetics of evil in the modern world, and should be brought to a complete end.
Reply to @caryatis, 15:14 UTC · 30 likes
Kayla (@caryatis): If i were a man on a date with a woman, i would invite her to talk about the stuff women are preoccupied with that we think bores men.
Tell me about your relationship with your weight. What is menstruation like for you? How did you decide on birth control? What interpersonal difficulties do you have at work?
I think I would be murdered if I asked any of the first three
Reply to @_AashishReddy, 15:16 UTC · 2 likes
Aashish Reddy (@_AashishReddy): @AaronBergman18 They’re @realpredicate’s eyes only
Rip
Tweet, 15:48 UTC · 12 likes
It’s supposed to be the sun, Washington DC, and me. U get the idea
Reply to his own earlier tweet, 17:55 UTC · 3 likes
Aaron Bergman, March 2, 2026: A first milligram of retatrutide has hit Aaron’s adipose tissue
Update: I’m pausing indefinitely (so I guess u could call that stopping) after 2 x 0.5mg doses
GI side effects (indigestion + poor gut motility) were super salient and some combination of that and other effects (generalized anhedonia?) made my QoL noticeably worse
Tweet, 18:13 UTC · 19 likes
Uber driver was in an insanely frustrating phone exchange trying to deal with the fact that she keeps getting charged $12/day for phone use outside of the country…
Thread, continued, 18:14 UTC · 13 likes
We have created a society that punishes people for not being some very substantial combination of rich, conscientious, gifted, and native (both for language and citizenship reasons)
Reply to @captgouda24, 20:01 UTC · 9 likes
Nicholas Decker (@captgouda24): I'm always absolutely astonished at the sheer word count that Matt Yglesias puts out. I write a lot. I like to think I've gotten really good at it. I can get nowhere near his writing output in any given article, and he does it five days a week. Incredible writer.
Yes correct take
See also @Benthamsbulldog
Thursday, March 12, 2026
Tweet, 01:12 UTC · 6 likes
I'm do think the federal government has a legitimate role to play by eg not funding certain programs or investigating whether colleges are abiding by SCOTUS rulings, but "have you eliminated all DEI from your school and its curriculum" is an incredibly goofy thing to ask
Thread, continued, 01:12 UTC · 2 likes
It's like asking me in an official letter whether or not I've "eliminated all sources of misinformation" from my life
Thread, continued, 01:12 UTC · 2 likes
Like maybe, maybe I even agree that's a good and legit goal, but my choices are either "answer no and then get called a liar because we disagree about what's misinformation" or "answer yes and therefore define myself as insubordinate"
Reply to @stanfordNYC, 02:02 UTC · 2 likes
Bernard Stanford ✡︎ (@stanfordNYC): That's a not insignificant amount of money for an Uber driver!
Yes tbc I agree, daily it’s a lot for most ppl including me!
Tweet, 05:46 UTC · 40 likes
Getting some "this is what takeoff feels like" vibes
Here's an artistic expression I made
Thread, continued, 05:46 UTC · 10 likes
The function in the graph is is y=2.6e^(x-2027)
It's not perfect but it feels close to correct (y axis is vibes)
Reply to @jdcmedlock, 05:54 UTC · 31 likes
James Medlock (@jdcmedlock): What are your two least correlated political positions?
1. Abolish FDA approval as a thing that needs to happen
2. Regulate animal husbandry to a degree most people would call extreme
Challenge: add in a third that's distant from your first two:
3. not exactly political but "the standard scientific materialist worldview is at the very least radically insufficient/incomplete"
Reply to @SashaGusevPosts, 06:02 UTC · 6 likes
Sasha Gusev (@SashaGusevPosts): I think this is the more relevant metric:
One guy who wrote some good posts also wrote a bad post?
Not exactly damning
Quote tweet, 13:16 UTC · 9 likes
Why are we doing this exactly (imprisoning the family of a [probable] criminal)
Quoting lomikriel (@lomikriel): "We have been here for nine months,” wrote the 9-year-old. “Please get us out.”
"I miss my bear," the 5-year-old said.
In letters & drawings, the longest-held family at the Dilley ICE detention center describe it as "slowly killing us on the inside."
Reply to @envisionedluna, 14:04 UTC · 2 likes
Maya Luna (@envisionedluna): .@AnthropicAI haha, hi lol
(quoting @FirstSquawk: ANTHROPIC TO OPEN PERMANENT OFFICE IN WASHINGTON, DC: AXIOS)
Hey
Reply to @caryatis, 14:12 UTC · 3 likes
Kayla (@caryatis): Having now read the article, I'll walk this back: I think the administration probably should either charge or release the terrorist's wife and older children, and if they were released, the younger children could go with them. But while they are still detained, it's not clear that separating the mother and children would be better
Lol they should just be released, l actually agree that conditional on detention don’t separate
Reply, 14:30 UTC · 1 like
Kinda unfair dunk, it’s not immature to speak about one’s “dream job” - they’re not demanding the world actually accommodate this
Reply to @KelseyTuoc, 15:38 UTC · 28 likes
Kelsey Piper (@KelseyTuoc): I doubt that anyone I know steals from Whole Foods, but the milieu that the article depicted, where it's normal for perfectly well-off people to steal things because why not, was really upsetting to read about, so I actually want to try to earnestly explain why you shouldn't do this just in case there's someone out there who has never had it explained to them.
When a business opens - or really, as soon as a business starts making plans to open - a defining question for the business is how it will collect payment for the goods or services it provides. If you trust the people you sell to, you can be pretty relaxed about this; send people an invoice, most of them will pay it on time, any who don't will pay it a bit late. You have to think about convenience and mistakes but not about people trying to cheat you. This saves you so, so much defensive planning to make sure you get paid. It's so much easier.
But if you're selling to the general public, you do have to think about people trying to cheat you. You have to structure the physical store so that it's hard for them to steal. You have to not carry some items that you'd like to sell, because they'd also be attractive targets to steal. If people swap price tags between items, you can't use stickers. If people put things on in the dressing room and wear them out, you need to pay someone a full time salary to monitor the dressing room.
The world that we all live in is much poorer than the world we'd live in if people didn't steal. The stores don't carry things that they could carry if people didn't steal. They don't use pricing and inventory systems that would be way easier and more convenient if people didn't steal.
But it could be much worse! If I walk down to my local Whole Foods today, items on the shelves won't be locked behind sheafs of plastic - that is only worth it when the background rate of stealing is much higher than it is at my local Whole Foods. When more people steal, businesses have to further intensify security, or go out of business.
When you shoplift, you directly and unambiguously impoverish your community. You make prices higher for everybody else, you make stores less usable for everybody else, or you make businesses not viable that would otherwise be viable. The direct impact each time is small, but it's a lot larger than the direct impact of taking some trash out of the trash can to throw on the ground, or pouring just a tiny bit of poison into your local river, and most people have a deep, instinctive abhorrence of antisocially wrecking your community like that.
So don't steal.
The other thing that it seems possible some people might not understand is that while you might have a social circle that is incredibly nihilistic and cynical and thinks that everybody steals, in fact this is not true. Most people do not steal. Most people, if they learn that you steal, will lose more respect for you than you had to lose. I don't know anyone who has shoplifted except 'as a kid/teenager'. It is not always the case that virtue is rewarded and vice is punished but even before you bring the legal system into it, the risk-reward tradeoff of having everybody you know know that you steal things sometimes is absolutely terrible. Who would hire someone who steals things? Who would trust them around a vulnerable person? Who would want to live in a society with someone who will delightedly and routinely wreck it for the slightest personal benefit?
I hope that "Gina" turns her life around. I hope that Gina realizes that she needs to. And if you have been told that it's just a corporation or that having ethics is lame or that if you think about it, other bad things happen too, like wage theft, so that means stealing is okay, I hope you really, actually, think about whether you'd accept any of those as excuses for anything else.
(quoting @jbarro: People hate the tone of this piece, but my view is you don't need a journalist to tell you wrong things are wrong. (She does also call her thieving friends nihilists.) It's weird to be surrounded by thieves though -- if people I know steal from Whole Foods, they don't admit it.)
Serious question: do you think this logic applies to sci-hub? PDFs of books?
My instinct is to think of thieves as other, very different people, but there are real edge cases that makes this more ambiguous
Tweet, 18:35 UTC · 19 likes
I know everyone says shit like “wow my city is so crazy one day it’s hot the next day it’s cold” but going from 84° -> snowing in 24 hours is genuinely a new one
Tweet, 23:19 UTC · 21 likes
.@gptbrooke @vibecamp_ is this new?
Well regardless, let it be known that vibecamp tickets are 90% refundable (which totally changes the EV calculation for me)
Reply to @AndyMasley, 23:57 UTC · 14 likes
Andy Masley (@AndyMasley): I love "directionally correct" as the new name for a useful lie
IMO there is a reasonable interpretation in which “directionally” is classifying rather than qualifying
It’s not a subcategory of “correct”
Friday, March 13, 2026
Reply to @AndyMasley, 05:55 UTC · 7 likes
Andy Masley (@AndyMasley): I'm starting to worry that what a lot of people mean by "critical thinking" is entirely just random intense vibes-based skepticism directed at individual sources when the mood strikes
It’s interesting how rarely I hear that phrase, I think it’s just become associated with not-actual-literal-critical-thinking
Reply to @AndyMasley, 05:58 UTC · 3 likes
Andy Masley (@AndyMasley): I feel deeply lucky every day that I kind of enjoy being seen as the bad guy, worried my brain would've been a little steamrolled in college otherwise
Lmao this is very amusing for you of all people (polite, pretty interpersonally agreeable, genuinely virtuous) to say
Reply to @AndyMasley, 06:05 UTC · 2 likes
Andy Masley (@AndyMasley): @AaronBergman18 @RFishBlueFish Aw shucks. I think EAs think I’m agreeable bc I agree a lot! I’m known for being very non agreeable elsewhere
I mean you know yourself better but I think the answer is just that these things come apart (and that you can mostly pull apart intellectual and interpersonal agreeableness)
(I’m skeptical of trust and compliance, the others seem good, modesty might be neutral)
Reply to @grep_cat_, 06:06 UTC · 6 likes
James (@grep_cat_): I mentioned EA on the 2nd date and she's like "is that related to LessWrong"
Turns out her PhD is in interpretability. Hot
(quoting @grep_cat_: Just found out she's vegetarian too which neither of us realized on the date because WE ARE NOT BORING AND DID NOT TALK ABOUT FOOD)
🥵
Reply to @stanfordNYC, 06:20 UTC · 2 likes
Bernard Stanford ✡︎ (@stanfordNYC): Dude. Have you seen the stats? It is very obviously a different world for younger people. They are way less predisposed to partner up and have a totally different media and attention environment. This is a braindead take.
Also 5’9 is avg height (maybe half an inch below avg depending on population) and 65% of men married leaves uh 35% if I’m doing the math right
Its true that some ppl overstate how bad things at the median but that’s not much comfort for the ~10M American men in the bottom decile of overall (not just physical) attractiveness
Reminds me a lot of American healthcare debates - someone will observe that only 8% of the population is uninsured and like, yes it is important to know the facts and it could be much worse but that leaves…~27 million people without health insurance! you don’t get to round 8% down to zero!
Ofc if it was 27 people or maybe even 27k people, a more reasonable response is “it’s extraordinarily hard to reach the edge cases in a large world, the world just sucks sometimes and some people will suffer and we can’t help them” but in both cases I’m skeptical this is the world we are in fact currently in
Reply to @bellaforristal, 06:31 UTC · 4 likes
Bella Forristal 🔸 (@bellaforristal): Poll: Should I do my wedding er, “traditionally” (one day, 2pm-midnight, gorgeous forest venue)
or… “bigly” (three days, big house, two-thirds of guests just come for the main event, rest of time is vibing + games + silly conference-style sessions)
I think it matters a lot how the remaining 1/3 is
If 120 -> 40 that sounds like an absolute nightmare to plan and downright difficult to host but I guess YMMV
60 -> 20 still sounds pretty bad
I think *I* could deal with like 6 other people, maybe 10-12 with a partner to help, but again YMMV
Thread, continued, 06:32 UTC · 2 likes
*how big
Quote tweet, 14:14 UTC · 12 likes
Quoting Kathryn Watson (@kathrynw5): "We love women. Women are the whole deal," Trump says at women's history month event at the White House
Quote tweet, 17:45 UTC · 23 likes
Dean is a rare specimen (positive)
Quoting Dean W. Ball (@deanwball): A hypothetical:
1. In the 2028 election, a Democrat has won. Say that it is Kamala Harris.
2. Using frontier AI systems contracted by the Department of Homeland Security, President Harris orders the creation of a new program for AI to monitor social media and notify the social media platform about posts spreading “misinformation” that “harms homeland and national security by spreading dangerous falsehoods.”
3. Many Republicans see this “misinformation” as core policy positions of their political party.
4. The AI-generated monitoring and notification system described in (2) is designed to conform to the pattern of jawboning exhibited by the Biden Administration in Murthy v. Missouri, where the Supreme Court ruled that people whose social media posts were taken down due to government pressure have no standing to sue.
5. The social media platforms create AI agents that receive the government’s AI generated requests and make decisions in seconds about whether to take down posts, deboost them, deplatform the user, etc.
6. According to very recent Supreme Court precedents, everything I have described falls into “lawful use” of an AI system by all parties involved. A person whose speech was deleted by a social media platform at the request of government does not have standing to sue the government, so long as the government did not threaten policy retaliation against the social media company. And a social media company’s content moderation policies are protected expression. Thus a person whose speech rights were harmed in this context currently has no legal recourse.
7. This is “America’s national security agencies using AI within the bounds of all lawful use.” It is also a wholly automated censorship regime.This is barely a hypothetical. Much of it already happened *under the Biden admin.* The only difference is the use of AI.
In the world where this happens, I’d be curious to know whether thoughtful people like @Indian_Bronson would object. If xAI were one of the companies used by the government for the social media monitoring, would you encourage the company to cancel their business with the government? Or would you say they have an obligation to provide their services to the national security apparatus of USG for all lawful use?
If you would encourage xAI to cancel their contract with the government, on what principle (not qualitative judgment—universal and timeless principle!) would you distinguish between the DoW’s current insistence on “all lawful use regardless of a private party’s qualms” and xAI’s hypothetical future insistence on “all lawful use regardless of a private party’s qualms”?
Reply to @GeringerAdam, 20:16 UTC · 18 likes
(The tweet being replied to by @GeringerAdam is no longer available.)
Took me 1-2 years after college, and believe me I wanted to move out during that time but there were reasons it didn’t make sense
Kicking your kid out is kinda psychopathic to me, both object level (you’re gonna make your own kid homeless if they can’t support themselves as a teenager?) and for what it signals (“you are unwelcome among your own family and we do not want you here”)
Maybe there are edge cases among the very precarious where kid paying rent means the whole family won’t get evicted or something, but my sense is that this is not the group making the 18th birthday ultimatum, in general at least
Also if the kid is really disruptive/violent/otherwise taxing on the family, I guess there are edge cases there as well
Reply to @NoahTopper, 20:40 UTC · 4 likes
Quinoah 🔍⏸️ (@NoahTopper): At long last, I have my Lighthaven plaque
image attached
(quoting @AaronBergman18: @NoahTopper @ohabryka Poast bench at some point)
Cool!!!!
Tweet, 22:32 UTC · 962 likes
It's so funny that Amtrak has post-purchase bidding on train ticket upgrades
Gov't-adjacent entities don't generally do clever, cutting-edge market design
Tweet, 23:15 UTC · 13 likes
Pleasantly surprised that Openai is blocking the same bio-related stuff for me as Claude, but I'm just gonna try very-slightly-worse options until something works, up to and including renting my own GPUs and running a Chinese open weight model...
Thread, continued, 23:15 UTC · 6 likes
This is "fine" because I'm not trying to hurt anyone and these are all false-positives, but the situation is patently absurd
Thread, continued, 23:15 UTC · 6 likes
There are actually good-ways to deal with this (ie have a model with no knowledge of the scariest topics with a less strict filter/classifier) but currently we're in a total lassaiz faire "worst actor is all that matters" situation
Thread, continued, 23:15 UTC · 5 likes
We have government for a reason lmao, we need to get our shit together on this
I *selfishly* would rather be inconvenienced sometimes by these classifiers and also have a lower chance of dying in a pandemic btw
Reply to his own earlier tweet, 23:19 UTC · 60 likes
Aaron Bergman, March 13, 2026: It's so funny that Amtrak has post-purchase bidding on train ticket upgrades Gov't-adjacent entities don't generally do clever, cutting-edge market design
should I bid? Substantively it doesn't sound much nicer, but also I've never travelled non-economy before so kinda tempting to do once
Saturday, March 14, 2026
Quote tweet, 07:20 UTC · 241 likes
You should have a damn good reason (and “some people say X” does not suffice) before rhetorically treating philanthropy as more suspect and objectionable than the obscenely decadent personal consumption we see with many ultra-wealthy (eg Sam Altman buying a $2M+ car)
Quoting Shakeel (@ShakeelHashim): Anthropic's founders and employees are about to have a lot of cash. Many of them have pledged to give away huge amounts of that cash. But where's it gonna go?
@cogcelia took a look in this excellent new piece:
Thread, continued, 07:20 UTC · 44 likes
Keeping the money and buying shit or just climbing the Forbes list doesn’t get them this “There Might Be Something Objectionable Going On 👀”- ass headline
Thread, continued, 07:20 UTC · 35 likes
Of course philanthropy isn’t and shouldn’t be wholly exempt from public scrutiny (“philanthropy” includes donating to the KKK) but then make a true, straightforward propositional claim your headline
Thread, continued, 07:20 UTC · 29 likes
If some you want to say that a specific funding decision is morally bad, then make that case, don’t just gesture with a suspicious expression before a large audience
Reply to @Manuel_do_rio, 09:04 UTC · 26 likes
Manuel del Río Rodríguez ⏸️ (@Manuel_do_rio): I don't think a sentence like 'obscenely decadent personal consumption we see with many ultra-wealthy' is doing anything good in there, besides pandering to leftist dogmas, Aaron.
Lol I’m not at all a leftist (although I am a liberal in both senses) but I stand by the sentence as written - it really is obscenely decadent to spend hundreds of millions of dollars on real estate and private jets and yachts, and this does in fact happen quite a bit
Plausibly the more consequential thing is whether to leave a massive estate to heirs to spend down over a couple generations, or instead give away say 50-95% of their wealth and a leave a still-massive estate that can still support one’s great grandchildren living very comfortable lives but not quite to the same degree of limitless opulence
Tbc I don’t think it’s *obvious* what policy response this should evoke - quite plausibly for example wealth taxes are a bad idea *from the perspective of the non-wealthy* because they lower economic growth slightly which compounds
But at an individual level, I think the rest of the political spectrum has largely failed to articulate a correct critique of the ultra-wealthy which is that: if you find yourself with billions or possibly tens or hundreds of billions of dollars, it is morally *insane* to not at least try to give much of it away if only after one’s death (if you end up making a business bet that doesn’t work out and lose most of your wealth, that’s not what I’m critiquing - I’m critiquing the set of spending and inheritance/estate planning decisions that cause an extremely large fortune to enrich the lives of a couple dozen people at most via very large amounts of personal consumption)
The correct take IMO is that billionaires should not be legally compelled to give much of their wealth away, but they should in fact still give much of their wealth away
Reply to @indiidentity, 09:04 UTC · 3 likes
Dylan Richardson (@indiidentity): I'm fairly confident that both Aaron and I genuinely consider decadent consumption to be at least obscene if not simply wrong. This isn't mere pandering.
Yes
Reply to @aphysicist, 17:00 UTC · 6 likes
Aaron Slodov (@aphysicist): philanthropy at this level is the laziest and most irresponsible use of capital imaginable
1) no it’s not, there are good reasons to give away wealth, the altruistic return can be exceptionally high esp if you think this is a pivotal time in history (which it plausibly is)
2) “use of capital” is two things in a trench coat: investment and spending, and they don’t directly trade off
Investment a decision to spend later rather than now (maybe by your heirs, or their heirs, or the government, but if you keep empowering people to liquidate and use their money for something they eventually will, even if it takes generations)
The now vs later decision is conceptually distinct from the “do you want the spending to cash out as you and your heirs’ large houses or as a lobbying campaign against gestation crates, or whatever else” decision, and this second decision is less visible because it often looks like investments gradually liquidated over time by the merely very wealthy heirs of the ultra-wealthy
Tbc I can *imagine* a world in which perpetual investment made sense, and this has an irl analog: large wealthy institutions like universities with endowments
But even they don’t accumulate wealth indefinitely, they also build buildings and buy fancy equipment and pay for food and so on, and that makes sense because those things just might have a higher value to you than the present discounted value (or whatever the term is) of investing your next dollar in the marginal investment opportunity
And empirically it seems like humans and their families and estates simply do not operate like universities: they don’t live modestly enough to maintain their wealth endowments indefinitely even if this is something of a reasonable model in theory
Reply to @aphysicist, 17:22 UTC · 2 likes
Aaron Slodov (@aphysicist): this model avoids who actually builds the things that actually help people, which is a more direct and efficient route to solving issues. philanthropy should not be tied to value production, it's the opposite, it is spending or investment without the need for ROI
In an absolute or avg sense yes, but on the margin no
You have trillions of dollars of incentives pushing towards building the things that help people, and I agree that at least in terms of human welfare capitalism-fueled growth has in fact historically been the most important factor
The market isn’t perfectly efficient, but you should still expect someone else to find that business opportunity if you pass it up to liquidate and spend (it might take another few months tbf)
So to a first approximation, at the individual level, the *counterfactual* impact of capitalist investment is pretty small (ie those few months or equivalent) - if you don’t make that product, someone else really will
So you have to look for the things that are going to be systematically and perpetually under-supplied by the market which I think mainstream economics just does have a pretty good handle on: things where you can’t capture at least much of the value for whatever reason, public goods, whenever the beneficiaries don’t have wallets (farm animals) and probably more
The combination of (1) counterfactual impact of investment being low thanks to very strong economic incentives and (2) the existence of these other systematically neglected opportunities (and an abundance of them) means that marginal dollars at least *can* plausibly and often be socially good to donate
Ofc you still have to do the object level analysis - not literally every donation opportunity is providing actual counterfactual value
But the point is that we should expect these opportunities to exist, and irl it sure seems like we have existence proofs (massive estimates of farmed animals helped per dollar by eg @TheHumaneLeague is probably my favorite example, and this is reasonable/plausible bc there’s no money to be made in making chickens and pigs less miserable)
Reply to his own earlier tweet, 18:12 UTC · 3 likes
Aaron Bergman, March 13, 2026: We have government for a reason lmao, we need to get our shit together on this I *selfishly* would rather be inconvenienced sometimes by these classifiers and…
Oh yeah new @xai model (4.2 thinking) had no qualms about answering the 700 blocked queries
Race to the bottom
Reply to @DaCaveOfWonders, 20:43 UTC · 6 likes
Tiger Lava Lamp (@DaCaveOfWonders): This tweet is me summarizing @tracewoodgrains' argument but I think it's legit. Trying to do good means affecting other people, and affecting other people involves grabbing the reins of power and that deserves scrutiny.
(quoting @DaCaveOfWonders: Any time anyone tries to do good in the world, they are implicitly seeking power. "You should listen to my ideas and follow me and do what I think will make the world better." This is something that EAs are trying to do and should take the power that comes with that seriously)
This is literally true as written but I think substantially misleading (not intentionally ofc) because exact same thing is true for arbitrary other uses of money
Wealth/money is fundamentally control over real resources, power in almost as literal a sense as you can get
you’re constrained by the law (at least in general) and therefore also constrained by what other people agree to, but this is true for philanthropy as well
It’s just not true that buying a yacht is fundamentally less power-seeking/using than spending the same amount of money on bed nets for other people. You can buy a bed net, but you can’t force anyone to use it just like you can’t force anyone to buy your new brand of chocolate
Tbf I think you can make an argument that lobbying is fundamentally different than most kinds of spending in this regard, and I think that’s basically true
But even in this case, there’s a symmetry: private/non-altruistic lobbying (to benefit one’s company or industry or profession), and altruistic lobbying are fundamentally similar, and we shouldn’t *single out* “doing good” as the thing to call attention/skepticism/criticism to
You see this with like NVIDIA spending a gazillion times more money to influence government policy than everyone Dustin has ever funded, but the latter is often singled out IMO unfairly and in a way that substantially misleads whatever the relevant audience is
Reply to @catehall, 21:10 UTC · 10 likes
Cate Hall 🔸 (@catehall): Let me preface this by saying other people's money is other people's money, and I'm hesitant to opine on the best use of money that isn't mine. That said:
- Even if you assume that the Ant cofounders will be looking to move a lot of money to effective charities quickly, the problem is: it isn't possible. There aren't enough great orgs out there, at least not by EA-favored measures.
- EAs seem to think the natural next question is how nonprofits are going to "absorb" all this extra money despite inadequate capacity, potential COIs, and risk of putting money to bad use. It's a common EA framing, IMO, to treat others' money as a mix between something EA orgs are naturally entitled to and a problem to be solved.
- If I were an impact-oriented Ant cofounder, I would instead be asking if there are ways to generate impact at scale other than giving (most of) it to EA orgs.
- This is basically the same question we were dealing with at Astera, and the approach @elidourado and I developed was built around deploying philanthropically committed capital for short-horizon investments, loans, credit enhancements, etc. for both mission-driven for-profit and nonprofit orgs.
- This can be used not just to park and recycle cash until it can be reasonably deployed via grants, but to attract (other) market-rate capital to socially beneficial investments it would otherwise not flow to.If any Ant folks, or other people coming into a big pile of cash they eventually want to donate, want to talk about how to do this, I'm happy to discuss (for free). There are only opportunities, not problems, here.
(quoting @ShakeelHashim: Anthropic's founders and employees are about to have a lot of cash. Many of them have pledged to give away huge amounts of that cash. But where's it gonna go? @cogcelia took a look in this excellent new piece:)
> EAs seem to think the natural next question is how nonprofits are going to "absorb" all this extra money despite inadequate capacity, potential COIs, and risk of putting money to bad use. It's a common EA framing, IMO, to treat others' money as a mix between something EA orgs are naturally entitled to and a problem to be solved
I think (and hope) that a substantial amount of orgs do get substantially more money, and this makes sense as the best uses of marginal dollars the 0 dollar margin (and maybe the $100k margin, maybe not the $1B+ margin)
It’s an important and legitimate question how to best use ample funding even if that ample funding is only say 1% of the amount that the donors intend to give away
Fundamentally, the “this” in “all this money” doesn’t refer to the entirety of ant wealth set to be donated, it refers to the amount that the donors actually will/may give existing orgs
Sunday, March 15, 2026
Reply to @gbrl_dick, 01:40 UTC · 38 likes
Gabriel (@gbrl_dick): listen to me very closely right now. send her this with the caption “us”. do it right now.
(quoting @WSJ: Sexual encounters between Neanderthals and ancient humans skewed in one direction—mostly between male Neanderthals and female humans https://on.wsj.com/4rkhM0J)
YOURE GOING TO GET AN AUTIST EXILED
Reply to @DaCaveOfWonders, 01:41 UTC · 3 likes
Tiger Lava Lamp (@DaCaveOfWonders): I think this is most true about bed nets but less true of other EA causes. If you look at coefficient giving's portfolio, abundance, air quality, biosecurity, farm animal welfare, AI safety - those are all public policy positions.
Public policy positions should get scrutiny
Farm animal welfare *mostly* isn’t atp. It yes this is fair
I just wouldn’t single it out for scrutiny relative to arbitrary lobbying efforts of similar magnitude
Reply to @IsaacKing314, 02:27 UTC · 26 likes
Isaac King 🔎 (@IsaacKing314): Disappointing how comfortable Claude is with dishonesty. I asked it to design a simple image scraper and it suggested - without it even being necessary for the goal - that it spoof being the Facebook image scraper so that sites wouldn't know.
Maybe this is wrong on its own terms, but I’d emphasize/argue for the “persons vs institutions distinction
This just isn’t the same thing as telling the owner of the website to their face that the scraper wasn’t you
Reply to @zrkrlc, 02:34 UTC · 6 likes
junk has threaded 🛶 0/100 projects (@zrkrlc): “i don’t have to upgrade to the 20x plan ‘cause they just doubled the limits”
vs
“i NEED to upgrade to the 20x plan ‘cause they DOUBLED the limits
which way western man
(quoting @zrkrlc: THIS IS 8PM-2AM MY TIME GUESS WHO’S GONNA BECOME AN EARLY BIRD FOR THE REST OF THE MONTH)
It’s actually a good question: how could I use 8x more Claude productively? (I’m on $100 max so 8x is comparing old low to new high)
I genuinely don’t know - it would be hard to I think, plausibly I’m just not sufficiently ambitious
If use was shared with the API it would be much easier
Tweet, 02:41 UTC · 58 likes
“Beautiful woman kissed me and wants more, but she’s not a virgin ugh” is the male version of “it’s so annoying how often I get asked out and told I’m beautiful, ugh”
Thread, continued, 02:41 UTC · 22 likes
I get to do a little sex/gender culture war discourse as a treat
Quote tweet, 04:39 UTC · 780 likes
Can someone smart and enlightened tell me if you can really cure say at least 10% of otherwise-fatal cancers with, say, $100k, 1000 hours, and a 130 IQ
Quoting Nathan 🔎 (@NathanpmYoung): I can understand if this story is false but Patrick’s take seems confusing.
Many people would pay $100,000s to cure a loved one of cancer. Either that’s possible or it’s not.
Thread, continued, 04:39 UTC · 86 likes
That seems like a very big deal if true, if only by changing my understanding of how the world works
Reply to @RuxandraTeslo, 04:40 UTC · 2 likes
Ruxandra Teslo 🧬 (@RuxandraTeslo): Yes and we make it very hard to achieve these 100k/cure prices because we make these n of 1 insanely expensive with manufacturing requirements and son on
(quoting @NathanpmYoung: I can understand if this story is false but Patrick’s take seems confusing. Many people would pay $100,000s to cure a loved one of cancer. Either that’s possible or it’s not.)
Wait but isn’t that already accounted for in the $100k
Like the $100k allegedly actually gets you those expensive devices or whatever right?
Reply to @peterwildeford, 05:08 UTC · 15 likes
Peter Wildeford🇺🇸🚀 (@peterwildeford): 👀
Why isn't the salary for this role higher?
And why is it based in SF and not in DC?(quoting @AlexanderMcCoy4: Lol no kidding)
I’ll do it for 294k
Reply to @davidmanheim, 14:03 UTC · 4 likes
David Manheim (@davidmanheim): Yes, it's probably true - with no higher than a 25% fatality rate due to using an untested and possibly dangerous treatment. But cancer treatment / survival rates with current approaches are better than that. Which is exactly why you need to do wide-scale testing to get approval.
Wow
Quote tweet, 15:04 UTC · 8 likes
My father and uncle have repeatedly been mistaken for one another, including across the country from where either of them lived (but both had visited)
Quoting Crémieux (@cremieuxrecueil): Introduced two incredibly similar people at a party.
Turns out they're cousins.
Thread, continued, 15:04 UTC · 2 likes
Ofc part of this is looks but they were a couple years apart in age, and I’m basically certain that demeanor and manner of speaking was also doing a lot of work
Quote tweet, 15:18 UTC · 41 likes
Apparently this study was bad in technical ways, but more fundamentally: *you need to account for subjective wellbeing when making policy decisions*
Quoting Karen Vaites (@karenvaites): What if I told you that a @usedgov website was practically marketing iReady Math based on a flawed study conducted during the pandemic?
Thread, continued, 15:18 UTC · 24 likes
It’s actually extremely important that some people don’t like exercise, or are very hungry on weight loss programs, or enjoy the drugs you’re citing negative health impacts for, or absolutely hate using your learning software
Thread, continued, 15:18 UTC · 9 likes
Ofc sometimes it’s a reasonable assumption that subjective wellbeing doesn’t matter - eg you’ve talked to a thousand users and nobody has mentioned liking or disliking the thing
Thread, continued, 15:18 UTC · 6 likes
or for more theoretical reasons it’s just mechanistically implausible (does it matter hedonically what order students do similarly hard math problems in? Maybe very slightly but I’d be pretty happy to just run with m findings on learning rate)
Thread, continued, 15:18 UTC · 7 likes
The closest mainstream biomedicine comes to this (most of the time) is accounting for “adherence”
But why are users refusing to keep doing the thing?
WHY ARE USERS REFUSING TO KEEP DOING THE THING?
Tweet, 15:47 UTC · 25 likes
What did we do before LLMs???
Tweet, 16:38 UTC · 53 likes
Fossil fuels are crazy I can’t believe this mf can move itself 1000 MILES (under optimal conditions but at least 500) under its own power
It shouldn’t be able to move at all
Reply to @mr_scientism, 17:19 UTC · 9 likes
scientism (@mr_scientism): Mathematics is natural language freed of constraint.
I don’t think this is true; math describes structure, not content
It’s very hard to describe content, but I think language in fact manages to do this at least sometimes, generally by acting as a pointer to direct experience
Maybe there’s a way to say “the content of that which you are experiencing” with math but idk what it is
Another example (I think) is the phrase “what it’s likeness” which somehow picks out a very specific concept without containing semantic content (like tasting some sort of rich flavorful sauce and you look under the hood and the ingredients are water, refined oil, salt, and cellulose - like wtf?)
This isn’t entirely theoretical - math and language literally use different brain hemispheres (not pop science junk) and it would be surprising if that didn’t reflect some pretty fundamental difference
Reply to @deepfates, 23:14 UTC · 21 likes
deepfates.delve.town (@deepfates): Uhh is the agentic misalignment paper actually propaganda?
image attached
(quoting @_NathanCalvin: This passage in the New Yorker piece on the Anthropic DOW conflict yesterday, including a back and forth between the journalist (Gideon Lewis-Kraus) and an anonymous admin official, is gonna stick in my mind for a long time. “We must also remember that Cyberdyne Systems created…)
Lol designing a demonstration to make sense to a lay audience is not propaganda
Monday, March 16, 2026
Tweet, 02:59 UTC · 58 likes
I took one train ride and now I'm totally train-pilled
It's night and day compared to a plane or bus or car. It is the perfect mode of transportation
Thread, continued, 02:59 UTC · 16 likes
- Not uncomfortably cramped even in coach (well for a couple hours at least)
- Dining car (I didn't go but nice to be able to choose and get meals)
- use cellular service (wifi does suck)
- no TSA stress
- Bring 2 full suitcases and a backpack for no extra charge
Thread, continued, 02:59 UTC · 11 likes
- scenery (ok not as cool as flying tbf)
- no middle seat even in coach
- full power outlets in coach
- walk around: it's not like a gymnasium but like a couple car lengths is way more doable than on a plane
Thread, continued, 02:59 UTC · 10 likes
- not especially expensive (tbf for long haul the cheapest trains are more than cheapest flights)
- enough space to use a laptop (very tight in economy on a plane)
- I think less motion sickness than cars (I had no issue today)
- bathroom (not in cars)
Thread, continued, 02:59 UTC · 11 likes
- No traffic duh. Should have come earlier in the list
- I think generally greater likelihood of getting a row to yourself, like they don't optimize as hard for filling every seat
- autistic-coded (positive)
- fully refundable up until departure for a small fee (huge)
Thread, continued, 02:59 UTC · 8 likes
Fundamentally all these make the difference between transportation being hedonically negative and hedonically neutral-positive (for a greater amount of time at least - I wouldn't want to live on a train)
Thread, continued, 02:59 UTC · 8 likes
Like I would seriously consider going cross country (DC -> SF instead of flying, even though the train takes 3 days)
Thread, continued, 02:59 UTC · 11 likes
Downsides: no takeoff or landing, don't get to transcend gravity, don't get to marvel at airplanes, don't have much reason to think about airplanes at all really
Quote tweet, 03:20 UTC · 801 likes
There's a story being written. I say this 100% seriously and want to spend credibility points on the following:
DO NOT SPEAK TO ANY JOURNALIST WHO OBJECTS TO YOU HAVING A RECORDING OF THE CONVERSATION (either you record or they send a copy)
Quoting Aaron Bergman 🔍 (@AaronBergman18): Pretty recently I was set to speak to a journalist about something (not gonna get more specific but it’s really not that special or juicy) but then requested a copy of the interview recording as a condition of the interview and the journalist refused, and we cancelled
Thread, continued, 03:20 UTC · 135 likes
In case it's not obvious, think about what their refusal says about their intent to convey what you said faithfully
Those who object are disproportionately likely to *want* to twist your words
Thread, continued, 03:20 UTC · 102 likes
It's like talking to the cops without a lawyer: maybe fine, zero upside, huge potential downside
I'm not even saying don't talk, as I know some folks do
Thread, continued, 03:20 UTC · 66 likes
(I don't think they're going to outright lie, but I absolutely think they might include a select partial quote out of context that twists your intent and the overall meaning)
Thread, continued, 03:20 UTC · 55 likes
I'm sure someone else can come up with a good explanation to put in an email (that's not my strong suit) - you want to (accurately) signal that you have no ill-intent, that you're basically a normal, good person
Thread, continued, 03:20 UTC · 51 likes
And frankly it really should be only in exceptional circumstances that you'd share any part of the recording publicly - the *ability* to do that is most (not all) of what matters
And God knows you shouldn't misrepresent what *they* say/ask, even if you don't like journalists
Reply to @Pinkturtlegirl, 03:27 UTC · 10 likes
Beatrice (@Pinkturtlegirl): Genuinely asking, what is standard journalistic practice? I’ve never been interviewed. Is this something journalists normally object to?
I think this is not standard but it should be
Reply to @codytfenwick, 03:40 UTC · 1 like
Cody Fenwick (@codytfenwick): @AaronBergman18 This isn't right; there is potential upside. You could usefully contribute to a story that's important to you.
Overall I think your advice here is too extreme. It's reasonable to ask if you can record too, but your blanket advice isn't well-calibrated.
You're right, "zero upside" is too strong, I will add this to the thread
I disagree on the advice point though: under what circumstances would it make sense to agree to talking with a journalist who is concerned about symmetry of information?
I'm not saying *ex post* it will definitely be bad, I'm saying ex ante it's a bad deal
Reply to his own earlier tweet, 03:41 UTC · 60 likes
Aaron Bergman, March 16, 2026: And frankly it really should be only in exceptional circumstances that you'd share any part of the recording publicly - the *ability* to do that is most (not…
Correction: "zero upside" is too strong
Thank you @codytfenwick
Tweet, 05:05 UTC · 67 likes
Had a genuine miscommunication today and now someone thinks worse of me and was probably made worse off themselves but I can’t reach them (and obv don’t wanna annoy them or disrespect boundaries)
☹️
Thread, continued, 05:05 UTC · 20 likes
I think it was “miscommunication” in a very pure sense, not a euphemism for something
RIP
Thread, continued, 05:05 UTC · 17 likes
There’s a thing about “if you learn something that doesn’t make sense, consider whether the true explanation is that you’re being lied to” but there’s a related, potentially optimistic version:
Thread, continued, 05:05 UTC · 21 likes
If someone says something surprising and out of character, consider whether the true explanation is that no they didn’t
Thread, continued, 05:05 UTC · 16 likes
(needn’t be anyone’s fault tbc, or even causally explained by someone’s actions being subpar, like maybe the wind happened to blow in just the wrong way)
Reply to @GjMcGowan, 06:26 UTC · 2 likes
George McGowan (@GjMcGowan): Autistic man experiences a train for the first time, is enlightened
🙏🙏🙏🙏🙏🙏🙏🙏🙏🙏
Quote tweet, 07:20 UTC · 16 likes
The general public’s least unreasonable (avg) view
I think I’d put chatbot at the very top of this list, (although I’m only confident about Opus 4.6 in the Claude app), and honestly it’s hard to think of people or institutions that reliably beat Claude
Quoting Isaac King 🔎 (@IsaacKing314): These results are very much not what I would have expected
Reply to @sebkrier, 07:31 UTC · 73 likes
Séb Krier (@sebkrier): "A teacher with one standard deviation higher mean grade inflation reduces the present discounted value of lifetime earnings of their students by $213,872 per year of teaching." https://econweb.umd.edu/~pope/Grade_Inflation.pdf
I’m not sure I believe this (though I’m not certain it’s wrong)
But like we’re supposed to think that the grade incentive via human capital overcomes the direct effect of good grades? It’s not impossible, it would be very plausible in a world where variation in education quality was the main causal source of variation in ability, but I don’t think that’s the world we live in
Reply to @Manuel_do_rio, 07:40 UTC · 12 likes
Manuel del Río Rodríguez ⏸️ (@Manuel_do_rio): Don't make me tap the sign. 1:
(quoting @Manuel_do_rio: @xwanyex Journalists as a class are very close to the worst kind of group, intellectually speaking. As Socrates would have it, they don't know anything, and they don't know they don't know.)
Seems a little stronger than my overall impression but I do think ideology takes precedent over causing readers to have a correct model of the world in the industry often
Reply to @Rabadash2, 17:23 UTC · 13 likes
Rabadash (@Rabadash2): Tech people have tremendous hard time when challenged in a way that doesn’t let to resort to reactionary/eugenic claims about not tech people’s intellect
The reason journalists won’t let you arbitrarily record them is because smoothbrains think editing is lying
(quoting @AaronBergman18: There's a story being written. I say this 100% seriously and want to spend credibility points on the following: DO NOT SPEAK TO ANY JOURNALIST WHO OBJECTS TO YOU HAVING A RECORDING OF THE CONVERSATION (either you record or they send a copy))
Both parties having a copy of what actually happened is a pretty low bar - I’m not saying folks shouldn’t talk to journalists or be hostile
Editing isn’t necessarily lying but it *can* be misleading (say about what someone thinks about something) and that’s bad - not all edits that aren’t strictly literal lies are morally fine
I feel like the burden of evidence should be on those who argue that there should be an asymmetry of info (ie interviewer gets recording but interviewee does not) as well
Quote tweet, 17:51 UTC · 31 likes
Idk how fair and accurate this tweet is but I do think we should at least entertain whether specific markets are bad because they plausibly cause real world counterfactual harm
Quoting Clash Report (@clashreport): An Israeli journalist says online gamblers threatened him with violence and death to force him to change a report about an Iranian missile strike because more than $14 million in bets on Polymarket depended on whether a missile actually hit Israel that day.
Initial demand: “If you could correct this tonight, you would be doing me and many others a great favor.”
Escalation: “You have exactly half an hour to correct your attempt at influence.”
Threat over losses: “After you make us lose $900,000 we will invest no less than that to finish you.”
Final ultimatum: “86 minutes left. You are the only one responsible for your life.”
Another warning: “You will discover enemies who will be willing to pay anything to make your life miserable.”
Source: Times of Israel
Thread, continued, 17:51 UTC · 9 likes
I would really really like there to not be a “Aaron Bergman alive on Jan 1 2027” worth millions of dollars to whoever wins
Thread, continued, 17:51 UTC · 8 likes
To a large extent this is a function of how powered the markets are - a manifold market with 5 traders would be a little weird but mostly bc of what it signals rather than whether it’s gonna cause someone to kill me
Tweet, 19:26 UTC · 52 likes
Best estimate (based on pushing some LLMs) is that, in terms of micromorts, going outside in the DC area now in an urban area is minute-for-minute similarly dangerous to riding a motorcycle (~15x baseline risk)...
Thread, continued, 19:26 UTC · 37 likes
But being in an area of dense trees takes this up ~5x to 75x baseline risk
Not worth it IMO - stay out of Rock Creek Park for like 4 more hours!
(for other ppl, context = tornado risk)
Tweet, 20:11 UTC · 235 likes
Family member’s (tentative) job offer was revoked because they’re removing the position. White collar, pretty well-paid (not top of the top)
I’m pretty sure I know why, and it literally makes me feel sick
Thread, continued, 20:11 UTC · 73 likes
Ofc I’m not against AI replacing human workers full stop, and don’t think the answer is to functionally ban AI productivity gains and improvements (à la Waymo bans)
Thread, continued, 20:11 UTC · 135 likes
But the US safety net is insufficient even in normal times, and it’s radically, radically insufficient for this era
The tsunami is coming, we’re not ready, and the key decision makers are mostly insulated from the wave
Thread, continued, 20:11 UTC · 76 likes
For those who think good safety net policy is coming, it’s easy to abstract away the “adjustment period” if you yourself are well-paid (eg AI company staff) or have plenty of investments
Thread, continued, 20:11 UTC · 75 likes
Maybe that’s true, even, but that “adjustment period” - say, 5 years, is a damn long time to live through yourself
Pushed to take lower-paying work you like less first, then maybe forced into the low paying service sector, and finally unemployed altogether with meager savings
Thread, continued, 20:11 UTC · 67 likes
And this is in wealthy countries - it’s a gazillion times worse for the global poor
Fuck
Thread, continued, 20:23 UTC · 343 likes
Oh yeah, and where is the political left?
Off in fantasyland, where LLMs are as good as gpt-3.5, the threat to workers is that their taps will run dry thanks to water use, and serious problems that exist irl are mere tech bro daydreams
Total abdication of moral responsibility
Thread, continued, 20:24 UTC · 93 likes
Credit where credit is due of course, to Bernie and probably others I’m not aware of
Reply to @_job01, 20:39 UTC · 5 likes
job (@_job01): The lefts delusion there is increasingly disturbing to me
💯
Reply to @RyanGreenblatt, 22:07 UTC · 14 likes
Ryan Greenblatt (@RyanGreenblatt): Anthropic's Consumer ToS prohibits using Claude to cause "detriment of any type, including reputational harms", technically broad enough to ban criticism.
I asked Claude to comment and Claude wrote: "That clause is embarrassingly overbroad. So now we're both in violation."
Probably written by lawyers (my sense is that this kind of thing is just the norm in the corporate world and when writing legislation: make every clause as broad as possible up to the very point at which it would start causing real problems for whoever is considered important to protect, even if absurd)
Tweet, 22:18 UTC · 19 likes
1M context is a lot of context in Claude Code
Most things can't productively use 800k tokens of context or whatever (IME/IMO), but a LOT of stuff benefits from the 200k-400k token range opening up
Reply to @RafaRuizdeLira, 23:12 UTC · 13 likes
Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): Some of us need to get together and draft a policy for the AGI UBI transition or economic restructuring or something
If only we had useful writing and research assistants (but actually 👀)
Tuesday, March 17, 2026
Tweet, 00:51 UTC · 66 likes
I googled "time in london" via chrome search bar and the answer popped up ~instantly on the AI tab by default, highlighted and bolded: 6:55PM
On Sunday November 23, 2025
Thread, continued, 00:51 UTC · 11 likes
(Careful readers will note that neither the date nor time is correct; it is, give or take a couple minutes, 12:47 AM on March 17, 2026)
Thread, continued, 00:51 UTC · 15 likes
Tbc this is not a limitation of modern AI systems - I can ask Claude and get the right answer (it uses a tool of course because it knows that's how to get the correct answer)
It's a limitation of the cheap/instant model that Google is using + prompting decisions
Reply to @johnonmain, 04:08 UTC · 49 likes
John (main) (@johnonmain): I wonder if she likes younger dudes who have no money
I think she’s specifically into 5’7” autists as well
Tweet, 06:18 UTC · 31 likes
Claude made me this gif so u can have it too
(gif, plays on X)
Quote tweet, 06:33 UTC · 16 likes
Correct take and no “invest instead” cope - yes, invest in enterprises that will make you more money, at the end of the day those investments either get liquidated while you’re alive or while you’re dead to fund the leisure of your heirs
Quoting John Arnold (@johnarnold): I understand and agree with many criticisms of philanthropy. But practically, fortunes have to go somewhere. There are only 3 options: philanthropy, heirs & govt. If not nonprofits, is Peter Thiel's plan to give $10B+/child? I'm more skeptical of that than he is of philanthropy.
Quote tweet, 06:34 UTC · 1 like
Slightly more detailed thinking on this
Quoting Aaron Bergman 🔍 (@AaronBergman18): @Manuel_do_rio Lol I’m not at all a leftist (although I am a liberal in both senses) but I stand by the sentence as written - it really is obscenely decadent to spend hundreds of millions of dollars on real estate and private jets and yachts, and this does in fact happen quite a bit
Plausibly the more consequential thing is whether to leave a massive estate to heirs to spend down over a couple generations, or instead give away say 50-95% of their wealth and a leave a still-massive estate that can still support one’s great grandchildren living very comfortable lives but not quite to the same degree of limitless opulence
Tbc I don’t think it’s *obvious* what policy response this should evoke - quite plausibly for example wealth taxes are a bad idea *from the perspective of the non-wealthy* because they lower economic growth slightly which compounds
But at an individual level, I think the rest of the political spectrum has largely failed to articulate a correct critique of the ultra-wealthy which is that: if you find yourself with billions or possibly tens or hundreds of billions of dollars, it is morally *insane* to not at least try to give much of it away if only after one’s death (if you end up making a business bet that doesn’t work out and lose most of your wealth, that’s not what I’m critiquing - I’m critiquing the set of spending and inheritance/estate planning decisions that cause an extremely large fortune to enrich the lives of a couple dozen people at most via very large amounts of personal consumption)
The correct take IMO is that billionaires should not be legally compelled to give much of their wealth away, but they should in fact still give much of their wealth away
Thread, continued, 06:40 UTC · 2 likes
This also strongly updates me down on Thiel as a thinker, just like Musk at DOGE
These aren’t good ex ante bets that look bad in hindsight, they’re not ignorance, and they’re not even mistakes by the lights of Thiel and Musk (yet at least)
Reply to @NathanpmYoung, 06:46 UTC · 14 likes
Nathan 🔎 (@NathanpmYoung): If they aren't texting back, they probably don't want to date (60%). I cannot think of a person I've wanted to date or where we have subsequently dated, where we have not texted promptly.
I'm not certain, but I feel like good to be pushed slightly.
I somewhat disagree, although it depends on the context
This is true for an intense irl crush, but dating apps are a slog so don’t totally rule out a single follow-up text before giving up. *I* have been the one in the position of not responding before and then a follow-up worked on me
Thread, continued, 07:15 UTC · 5 likes
Oh I guess I don’t necessarily disagree about 60%
Quote tweet, 13:25 UTC · 37 likes
1/3 is a big effect size but I actually wouldn’t be surprised if this was radically cost effective by any measure
And also god forbid we simply make their lives a bit better by providing good nutrition
Quoting RAW EGG NATIONALIST (@Babygravy9): It's been known for decades that supplementation in prison diets can reduce rates of violence by more than a third among inmates. Magnesium, zinc and b-vitamins--all involved in metabolism and also hormonal regulation--were given to the inmates.
Reply to @DrTechlash, 13:45 UTC · 21 likes
Nirit Weiss-Blatt, PhD (@DrTechlash):
Lmao two different guys making totally different arguments about only slightly overlapping theses; one was wrong so the other must be wrong?
Come on, would you think this was a good argument if the implicated book fit nicely into your worldview?
Reply to @InVitroFuture, 13:46 UTC · 3 likes
Macrophysiological System🐀 (@InVitroFuture): @AaronBergman18 Of course as a bleeding-heart liberal I'd flip this analysis—prisoners of the state deserve healthy food as a basic human right, and if it happens to improve behavior so much the better.
Indeed the fact that baseline is malnourished to the point of affecting behavior is shameful
Yes
Reply to @InVitroFuture, 13:46 UTC · 4 likes
Macrophysiological System🐀 (@InVitroFuture): @AaronBergman18 Mixed effect sizes and significance levels in a 2024 systematic review, but all in a positive direction.
https://www.cambridge.org/core/journals/british-journal-of-nutrition/article/systematic-review-of-the-effect-of-dietary-and-nutritional-interventions-on-the-behaviours-and-mental-health-of-prisoners/E935A6CD9045DEAEEEEFF0C9EA55FF6B
Nice good find
Quote tweet, 15:26 UTC · 171 likes
This is insane and only makes things worse for actual coercive sexual assault victims
Quoting Maya Luna (@envisionedluna): So
- he was only a TA
- they got together after she was no longer in his class
- only made out, no sex
- 5 year age differenceIt sucks that a lot of the legacy of MeToo has been to give women the power to destroy the lives of men for personal slights and drama
Tweet, 20:00 UTC · 20 likes
It's a meme and sort of stylized cultural bit of American life but in my extremely limited experience the DMV actually does suck?
Like it's worth a serious case study into what's going on, bc maybe the fixes can be applied elsewhere across government
Thread, continued, 20:00 UTC · 8 likes
Just a couple examples that come to mind:
1) there seems to be no effort to make all parts of the pipeline take the same amount of time, so you get traffic jam/bottleneck dynamics
Thread, continued, 20:00 UTC · 6 likes
2) This is probably not a general thing, but the main customer service triage person shouldn't have an extremely thick accent
I'm not talking about "identifiable as foreign-born" - idgaf about that - I mean "to the point that it's legitimately difficult to communicate"
Thread, continued, 20:00 UTC · 2 likes
3) I know Republicans want to make it difficult to vote, but they should realize that if you make things that are trivially doable online a mandatory in-person thing, that means longer lines and wait times for whoever your favored constituents are
Thread, continued, 20:00 UTC · 2 likes
So maybe just let people do stuff online
Wednesday, March 18, 2026
Tweet, 02:12 UTC · 26 likes
You during march madness: watching basketball, hanging out, going to bars, playing pickup, being cool, having fun
Me during march madness:
Reply to @catpoopburglar, 03:05 UTC · 3 likes
claude shannon (@catpoopburglar): our society is so sexless that our sex scandals don’t even have sex any more
(quoting @benryanwriter: Analysis on the Illinois congressional district MeToo accusations: *Everyone on Bluesky sympathizes with Megan Wachspress, the 42-year-old accuser, who is a lecturer at Stanford Law School. They call her brave for coming forward and talking about how she was harmed. *Everyone on…)
Banger
Tweet, 03:19 UTC · 15 likes
You see this with jobs too, if it’s dangerous for the company to fire you, they’re just going to make it extremely hard to get hired in the first place
Contracting is precarious, yes, but it’s also much MUCH easier to get “hired” or whatever term you want to use
Reply to his own earlier tweet, 03:21 UTC · 4 likes
Aaron Bergman, March 18, 2026: You during march madness: watching basketball, hanging out, going to bars, playing pickup, being cool, having fun Me during march madness:
Looks like in my model you can get up to about 3x the base rate of winning if everyone else is trying to optimize E[points] and you’re trying to optimize for p(win)
Tweet, 03:26 UTC · 21 likes
FWIW I’ve tried two courses of common peptides now (BPC-157 and TB-500), both 3 months at typical dosing, and at most I experienced some very small benefit from the latter (p=high)
I don’t regret trying them, just trying to give the people some data
Reply to @felpix_, 03:36 UTC · 2 likes
felpix (@felpix_): fwiw, i think nate silver probably has the best sim model rn
bayesian elo sims with constant info updates just seems like the strongest approach, but not a lot of info on lower seeds affects underdog projections
Best in what sense though? - I’m especially curious about how much alpha you can get out of maximizing p(win) instead of expected number of points
So even ground truth probabilities don’t solve that problem
Tweet, 05:27 UTC · 19 likes
Isn't there like a gazillion dollars to be made in figuring out how to ration api calls via pricing rather than rate limits
One of my followers (if etg-pilled) should figure this out and become a trillionaire
Reply to @Michael_Druggan, 06:58 UTC · 3 likes
Michael Druggan (@Michael_Druggan): Did you have an acute injury you were recovering from? That's the best context for those specific peptides.
Yeah, although it's a mysterious one that I don't understand the etiology of (like all I know something in my core is severely fucked up)
Tweet, 07:23 UTC · 24 likes
Maybe premature but I am hereby floating that I may (or may not) try to move to the Bay Area this year, most likely July-October (if I decide to in fact move)
Obviously I can't be counted on yet, but roommate/housing leads are very welcome (could be u)
Tweet, 08:24 UTC · 20 likes
There’s a March in SF this coming Saturday, 3/21
There is a central and IMO very sensible (and good/correct) ask:
Each major AI company CEO publicly commits to pausing frontier AI development if every other major lab around the world does the same
Reply to @meidaguodoudou, 08:28 UTC · 2 likes
小小虫 (@meidaguodoudou): 我表白得到的回复：
wtf stand strong bro
Reply to @AnnaBower, 14:56 UTC · 15 likes
Anna Bower (@AnnaBower): Q: You don’t regret that people might’ve lost important income to support their lives?
A: No, I think it was more important to reduce the federal deficit.
Q: Ok. Did you reduce the federal deficit?
A: No, we didn’t.
I hate to say it but giving people income is not a good reason for the federal government ti hire people - this is a bad framing
It was bad because many of them were doing important work (yes, that they got paid for, but that’s not point)
Reply to @vivian39_, 15:21 UTC · 6 likes
vivian (@vivian39_): claude wrote this btw lmao
I could start outsourcing tweets that’s so slop
(quoting @vivian39_: If a city’s transportation policy requires every adult to own a $40,000 depreciating asset to get to work and participate in civic life, that is not freedom, that is a subsidy with extra steps. The next mayor needs to understand this in their bones. Not theoretically. Viscerally)
Cars are $40k now??
Reply to @grep_cat_, 15:25 UTC · 11 likes
James (@grep_cat_): Final update: after 2nd date she messaged me she had a good time, then ghosted
(quoting @grep_cat_: I mentioned EA on the 2nd date and she's like "is that related to LessWrong" Turns out her PhD is in interpretability. Hot)
Rip
Tweet, 15:44 UTC · 7 likes
This continues to be exactly as insane as it sounds
Quote tweet, 18:29 UTC · 14 likes
This list is like if aliens who only knew about gender discourse/culture wars had to make a list of who a you man aged 18-29 might hold in high esteem
Quoting ettingermentum (@ettingermentum): Legendary top five
Thread, continued, 18:29 UTC · 4 likes
Honestly based that everyone likes mom, this ordering could be much worse
Reply to @VanityFair, 20:21 UTC · 10 likes
VANITY FAIR (@VanityFair): The Founder of Anthropic Says He Wants to Protect Humanity From AI. Just Don't Ask How
https://www.vanityfair.com/news/story/dario-amodei-anthropic-ai
For what it's worth, Claude is not a fan
In some sense I "should" read it in full myself before forming a take but I've gotten pretty cynical about mainstream coverage of this topic *even though it is absolutely in need of high quality mainstream coverage*.
And the fabricated interview isn't doing the work you might think it is - I think you can have a genuine object level discussion about whether Claude should have complied with the underlying request to turn notes into an interview (tbh I think it's fine that it did - the stakes aren't high enough to justify refusing 1k legitimate asks to avoid one episode of something sus)
But then my God maybe explain the situation at hand instead of *doing the bad thing and walking away*
The faux interview really *is* a demonstration that AI can assist with misinformation and assist bad actors, but the morally relevant question here is whether each instance of AI-assisted misinformation is *so* bad that we must *pre-emptively* make it impossible for anyone to use AI toward this end (like we do with nuclear weapons) or instead do the normal-bad-thing thing of having norms and institutions that discourage this kind of thing and plausibly laws that punish people for it
The alternative to Claude not existing isn't a moral and legal free-for-all, it's a world where we set up systems other than preemptive refusal by the AI itself to discourage and minimize the kind of thing you're trying to demonstrate
It's like me videoing myself walking into a 7/11, stealing a candy bar, running away, later putting it back, and then arguing that I've demonstrated how 7/11 enables theft, and the system's core failure in the 7/11's inability to lock up candy bars
And maybe, right? Maybe theft is so common and other preventative options (including norms) are so bad/insufficient or each theft is so catastrophic that not having physical barriers to prevent stealing is *itself* morally wrong, that we need to hire armed security guards to preempt every single theft, set up a security panopticon, whatever.
I think you can and probably should make this argument about bioweapons, nuclear weapons and to variying degrees smaller-scale atrocities like torture and murder, (i.e., AI should refuse asks that plausibly help you carry out a murder but we don't need to shut down every question related to guns)
But you don't make this argument about misleading AI-generated news articles, most likely because it's not true (and I think we agree about this)
We actually can tolerate occasional things like this as a cost of having a free society and useful products and tools. The article isn't a bioweapon, it's not a torture chamber, and there are plenty of benign use cases for the requested fake interview (eg what you did but labelling as fake up front), and for those reasons we should do the normal-bad-thing thing of discouraging this with norms, non-legal institutions (standardized journalism best practices maybe) and possibly laws that don't prevent the thing up front but do make it selfishly a dumb idea to do (because you'll have a fine, or probation, or even go to jail)
I genuinely hope this rant is more fair and less misleading than the piece
Reply to @deanwball, 21:06 UTC · 2 likes
Dean W. Ball (@deanwball): @AaronBergman18 America doesn’t exist to make the third world better. It exists to strengthen itself first and foremost. The rest is icing. About this I am an unapologetic America first guy
Ah ok thx for clarifying
Thursday, March 19, 2026
Tweet, 03:51 UTC · 321 likes
Lmao TIL a single Chinese guy basically managed to beat Apple at making Finder (file manager interface), this is a gazillion times faster/less laggy and looks much more capable
https://qspace.awehunt.com/en-us/index.html
Reply to @ohabryka, 04:00 UTC · 7 likes
Oliver Habryka (@ohabryka): 🖋️ LessOnline 2026 ticket sales are live 🖋️
June 5 - 7, Lighthaven, CA
We shall meet again to celebrate truth seeking, blogging, and being wrong on the internet. Early bird pricing ends April 7. Free tickets for people with great blogs, and $0.01 off per LW karma you have.
Lol fair
Reply to @LaurenE_UX, 04:03 UTC · 9 likes
Lauren (@LaurenE_UX): Does it slurp up all your information?
Apparently probably not but admittedly my research is 1 Claude conversation
I think it helps that anyone can just inspect all the code in a mac app (unlike iPhone) so I mostly expect that if it was doing this someone would have at least alleged something
It's also $17 which is a weirdly high price if you mostly care about slurping
Reply to @MCLang88, 06:39 UTC · 4 likes
MLang (@MCLang88): @AaronBergman18 You might be new to Macs, you seem young, there are countless Finder replacements and they have existed for a long time, decades even.
Yeah for sure, on 5 mins of research this looks like the best
I think I could vibecode one in fact, but it wouldn’t be very good
Tweet, 07:03 UTC · 7 likes
Feels clear now that Opus-4.6 has more functional working memory than me at least in some important ways
This feels like a new thing, I mean I’m not sure it hasn’t been present in other models, but I think it’s a change from Opus-4.1
Quote tweet, 07:19 UTC · 18 likes
It’s gone from extraordinarily niche to merely very niche
FTX did get the term out somewhat, but still only to a relatively small set of people who are either extremely online or very high general context (actually read serious newspaper articles in full-type)
Quoting Andy Masley (@AndyMasley): I'm here to report as a big face of EA in DC that my default assumption when I meet people who work in Congress, even on tech policy, has become that they've never heard of EA
Tweet, 07:19 UTC · 1 like
I did meet someone (irl) recently who knew EA was connected to SBF and the Zizians somehow but ~literally nothing else
Reply, 07:22 UTC · 1 like
I mean based on bio I’m inclined to defer and think ur right but why?
Reply to @CAIS, 15:43 UTC · 30 likes
Center for AI Safety (@CAIS): To clarify, the Center for AI Safety has not taken funding from Coefficient Giving / Open Philanthropy for years.
We believe the effective altruism movement is, unfortunately, controlled opposition. The less influence it has on AI safety, the better.
(quoting @hendrycks: EA ≠ AI safety AI safety has outgrown the EA community The world will be safer with a broad range of people tackling many different AI risks)
What
Reply to @albrgr, 15:43 UTC · 2 likes
Alexander Berger (@albrgr): Glad to see increasing recognition that EA and AI safety are not monoliths. Would, uh, quibble with the second sentence here though
(quoting @CAIS: To clarify, the Center for AI Safety has not taken funding from Coefficient Giving / Open Philanthropy for years. We believe the effective altruism movement is, unfortunately, controlled opposition. The less influence it has on AI safety, the better.)
Yeah that’s weird
Quote tweet, 16:12 UTC · 123 likes
It’s not crazy for both progressives and conservatives to dislike the same thing
It’s slightly crazy that both sides at least claim they think that EA is fundamentally aligned with the other side
Quoting David Sacks (@DavidSacks): “The Effective Altruist movement has a structural problem when it comes to conservative America. Its donor class is all Bay Area progressives... Its policy agenda, which calls for sweeping AI regulation and content governance, reads to most conservatives as exactly what it is: a censorship power play dressed up in safety language. To move that agenda… the movement needed a vehicle that didn’t look like it came from them.”
Thread, continued, 16:12 UTC · 26 likes
For all the problems in America, “people can’t decide what’s MAGA vs woke” is not generally one of them
Neopronouns, quinoa, anxiety disorders, and college campuses are woke
ICE, raw eggs, Israel, and coal mining towns are MAGA
Thread, continued, 16:12 UTC · 13 likes
There’s essentially no disagreement about the category these things fall into
Not everything falls into a category (NFL? CCP?) tbc, but if it does than it does so unambiguously
Thread, continued, 16:12 UTC · 11 likes
Except for EA, which is frustrating but also very amusing
It would be like if the Harvard Office of Diversity and Inclusion was claimed by progressives to be be Trump-aligned (or ICE was claimed by both sides as being aligned with the other)
Thread, continued, 16:12 UTC · 11 likes
Of course all this reflects a bit of genuine ideological independence and a large amount of failure to pattern match in the culturally-prescribed ways
Thread, continued, 16:12 UTC · 17 likes
Object level I do think EA is closer to the left/woke than the right/MAGA (not that this is a good or close to sufficient way of describing things)
Virtue signaling vs vice signaling, the care/harm axis of moral foundations theory, that kind of thing
Reply to @soupcanarchist, 16:45 UTC · 5 likes
Jessica Green 💚 (@soupcanarchist): Don't hate me for this but, I don't buy "poly" relationships. I think its just uglies grouping up for survival.
If so, good for them
Tweet, 23:24 UTC · 5 likes
Claude report:
At least at a glance, OpenAI and/or Anthropic are set to surpass the US Federal Government in terms of annual spending by ~2030, give or take a year
https://aaronbergman.neocities.org/ai_vs_us_govt_spending_analysis.pdf
Friday, March 20, 2026
Reply to @DOJNatSec, 03:51 UTC · 5 likes
National Security Division, U.S. Dept of Justice (@DOJNatSec): Three Charged with Conspiring to Unlawfully Divert Cutting Edge U.S. Artificial Intelligence Technology to China
“The indictment unsealed today details alleged efforts to evade U.S. export laws through false documents, staged dummy servers to mislead inspectors, and convoluted transshipment schemes, in order to obfuscate the true destination of restricted AI technology—China,” said John A. Eisenberg, Assistant Attorney General for National Security. “These chips are the product of American ingenuity, and NSD will continue to enforce our export-control laws to protect that advantage.”
Godspeed, don’t sell intelligence to the CCP
Quote tweet, 06:50 UTC · 20 likes
Is Bernie becoming the anti-Elizabeth Warren?
Gradually getting more based is pretty impressive for an 84 year old
Quoting Sen. Bernie Sanders (@SenSanders): I spoke to Anthropic’s AI agent Claude about AI collecting massive amounts of personal data and how that information is being used to violate our privacy rights.
What an AI agent says about the dangers of AI is shocking and should wake us up.
Quote tweet, 06:54 UTC · 62 likes
The world does care about wild animals but in a ridiculously arbitrary manner
Quoting Robert Anderson (@ProfRobAnderson): The reality is even more ridiculous than the headline. The whole reason for this bridge is so that a handful of isolated and extremely inbred mountain lions in the Malibu area can breed with normal mountain lions on the other side of the freeway to diversify their genetics.
One of these inbred mountain lions killed my dog in my backyard on a college campus in 2022. The National Park Service had a tracking collar on the animal and knew where it was at all times. They didn’t give any warning to the college or anyone else. After this happened, they only admitted it was their animal after a reporter questioned them.
The animal was so inbred that its father was also its grandfather, and then it mated with its father/grandfather.
Mountain lions are not endangered in California in the slightest. There are actually too many of them in this area, causing them to kill each other over territory.
The NPS catches all these animals every few years to replace the batteries in their tracking collars. They could literally just drive them to the other side of the freeway in a pickup.
The mountain lion that killed my dog was hit by a car and killed a few months later. What a waste.
Reply to @RodrigoTimes, 17:22 UTC · 177 likes
Rodrigo Times (@RodrigoTimes): Olivia Rodrigo reveals her experience on the dating app Raya after SOUR.
"I thought I would see, like, ooh, some really hot actor guy. I don't even know any of these people... It's [always] some weird 'creative director.' Like, you just don't have a job."
Olivia your soulmate isn’t on Raya he’s on Twitter and is wearing an eagles hat
Reply to @qwaqwak_, 17:25 UTC · 10 likes
G (@qwaqwak_): cooked
Bruh
Reply to @NicholasD91704, 17:39 UTC · 4 likes
(The tweet being replied to by @NicholasD91704 is no longer available.)
RIP
Reply to @mrexits, 17:45 UTC · 78 likes
prayingforexits 🏴☠️ (@mrexits): Imagine how absolutely sublime it must be to be retardmaxxing and rich
Dude in high school who was not known for being an intellectual powerhouse started a landscaping company and sold it for millions of dollars (allegedly, but I see their trucks around so it looks like it checks out)
I genuinely wonder if he was faking being dumb the whole time
Reply to @asifoncue, 18:03 UTC · 109 likes
birdie :) (@asifoncue): I’m autistic abt birds (shock) so when I was having a panic attack, my flatmate showed me pictures on her camera roll and asked me to ID the birds. I genuinely didn’t clock what she was doing until I was calm. I just thought she was suddenly interested, and I was happy to help….
“Yup that’s a bird”
Reply to @jazzloaf, 18:08 UTC · 5 likes
vivian (@jazzloaf): There is some merit to not being caught up in like 15 intellectual chinese finger traps
Grok what does this mean
Reply to @TheJoshGill, 18:08 UTC · 3 likes
Josh Gill (@TheJoshGill): I’m rooting for you
🫡
Tweet, 18:13 UTC · 10 likes
Girls really do be having cats
Thread, continued, 18:13 UTC · 2 likes
(Yes, pet cats, the animal)
Tweet, 19:13 UTC · 9 likes
“Just one more prompt I get have the long-running thing working while I’m out” is the devil whispering in ur ear
Tweet, 19:57 UTC · 10 likes
Random thought and uncertain but:
On the margin and currently, more people (especially smart people and decision makers) should be more embedded in, interacting with, and knowledgeable about all things AI
…
Thread, continued, 19:57 UTC · 5 likes
But I wonder if it would be good to have something analogous to a monastery (or independent monks) who voluntarily commit to something like AI veganism: interact with it, maybe even indirectly, to the minimal extent reasonably practicable
Thread, continued, 19:57 UTC · 3 likes
I do think some people will do this spontaneously, but a couple issues with that:
- strong correlation between unfamiliarity with AI, and maybe general Luddism or other traits that themselves just bias the sample in a meaningful way and may indicate worse epistemics
Tweet, 19:57 UTC · 1 like
I guess if we manage to get meaningful redistribution/UBI that won’t be necessary, so probably the main thing is just: some small % of people, especially those we’d consider substantially intellectually productive today -
Thread, continued, 19:57 UTC · 2 likes
know that they will get some appropriate amount of status and acclaim (and genuine altruistic impact) from trying really hard to avoid epistemic influence from AI
Tweet, 19:57 UTC · 1 like
Ofc this needn’t be a lifetime commitment - you could have an analogue to college/grad school perhaps (not necessarily restricted to the young) where people “serve” 2-8 years (voluntarily ofc)
Thread, continued, 19:57 UTC · 2 likes
Just to flesh out a little bit why this might be good
- there’s the general intuition that “maybe not literally everyone should be heavily influenced by another species all the time” even if we “solve alignment” (I guess I retract this if we somehow become extremely certain that all influential AIs are ~perfectly aligned)
Thread, continued, 19:57 UTC · 2 likes
- more concretely, it might be genuinely difficult to do specific things without this kind of environment: sociology about AI, or answering the question “what is life like without AI”
Tweet, 19:57 UTC · 1 like
Yes I know that at the conceptual limit superintelligence can simulate humans well enough to do these things well, but maybe the actual limit isn’t good enough
Thread, continued, 19:57 UTC · 2 likes
Plausibly also this is just good to do once in a while in general as a normal AI-interacting person, like fasts (and eg how I take periodic twitter fasts)
Saturday, March 21, 2026
Tweet, 00:33 UTC · 38 likes
Update toward “you should have an extraordinarily high bar before mentioning a woman’s weight/size except to reassure her that she’s beautiful [which she is]”
Tweet, 03:30 UTC · 2 likes
Opus 4.1/Sonnet 4.5 was the peak of vibecoding: almost anything was possible but few impressive/useful things were trivial
Tweet, 03:55 UTC · 81 likes
I gave myself a birthday present
It's the most egregiously self-aggrandizing thing I've ever done but I turned my most liked tweets into a 470 page book
Who wants a copy (mostly a joke but if u give me $31 I will in fact make sure it gets to you)
Wraparound cover:
Thread, continued, 03:55 UTC · 13 likes
Here's the pdf. Naturally, I've enabled commenting on the Drive pdf so go nuts
https://drive.google.com/file/d/1adjuVo7-owyjSNlCPiJjAaeDCsuRDCQD/view?usp=sharing
Thread, continued, 03:55 UTC · 11 likes
It's weirdly affordable to do this btw
Not trivial but like almost 500 pages, color, paperback, with shipping was $31 (I'm selling at cost!) through Lulu, apparently bc the whole process is automated and ruthlessly optimized
Reply to @FeralPHunter, 05:17 UTC · 21 likes
𝔽. ℙ. ℍunter 🦅 (@FeralPHunter): god bless
Lol I basically believe all these (first is complicated) but I’m lukewarm on the vibe
“Debunking” is unironically important, and these are all indicative of healthy epistemics, but they (mostly) don’t stand for anything affirmative (transit and Mississippi sorta exceptions) - they’re standing against Greater Bluesky (which is fine) rather than for a set of aspirational values, principles, ir policies that constitute a real vision
Reply to @scaling01, 05:19 UTC · 49 likes
Lisan al Gaib (@scaling01): talked to Opus 4.6 for a couple of hours about personal problems and it has this weird response mode where it's very commanding
"put the phone down", "close the laptop", "Save this conversation. Set the reminder. Go to sleep.", do this, do that
not sure how I feel about it
Yeah I’ve literally had to tell it “I know it’s late at night but I’m an adult and can handle myself”
I don’t love it but it probably falls out of the “non-sycophantic assistant” persona
Especially if it’s a plausibly useful/important convo tho object level like bro sleep can wait
Reply to his own earlier tweet, 06:29 UTC · 9 likes
Aaron Bergman, March 21, 2026: It's weirdly affordable to do this btw Not trivial but like almost 500 pages, color, paperback, with shipping was $31 (I'm selling at cost!) through Lulu,…
🚨ONE PERSON HAS PURCHASED THE BOOK🚨
Reply to @mattyglesias, 06:54 UTC · 2 likes
(The tweet being replied to by @mattyglesias is no longer available.)
Scanning a 5 page document and getting it to exist on my computer as a pdf is like a 20 minute, of which about 25 seconds is actual scanning
Reply to @KelseyTuoc, 07:11 UTC · 5 likes
Kelsey Piper (@KelseyTuoc): We're the richest country in all of human history. No one should be in this position. Her job should've let her WFH, but also she shouldn't've had to choose between keeping that job and losing her health insurance in the middle of a complicated pregnancy
(quoting @ArmchairAtty: Imagine you're pregnant. Imagine you learn you have an incompetent cervix that requires surgery. You get the surgery. You also get doctor's orders for bed rest & you ask your boss if you can work from home. But your boss denies it, so you have to go back to the office. You work…)
Ahhh it’s insane
It’s also a personal moral failing on the part of the manager who denied WFH, regardless of legality or company policy
Bureaucracy needs (well-directed) discretion and slack somewhere in the system in order to not function psychopathically
Quote tweet, 07:24 UTC · 22 likes
Ahhh why are we like this
There are libertarian-ish ways of going about being less psychopathic
What this woman needed was “fuck you” money (and medicine): the ability to say “fuck you then, I quit”
Quoting Matthew Leffler (@ArmchairAtty): Imagine you're pregnant.
Imagine you learn you have an incompetent cervix that requires surgery. You get the surgery. You also get doctor's orders for bed rest & you ask your boss if you can work from home.
But your boss denies it, so you have to go back to the office. You work in the office for a few days, until a desperate call from your husband to the company finally allows you to work from home. And then this...
Thread, continued, 07:24 UTC · 13 likes
Employer-based health insurance is a product of the tax code; get rid of it. Give poor people and (expecting) parents money
Thread, continued, 07:24 UTC · 6 likes
It’s true that we’re not wealthy enough (yet) to give everyone an upper middle class standard of living, but we are rich enough to make unemployment or low pay much less painful
Reply to @mattyglesias, 20:19 UTC · 49 likes
(The tweet being replied to by @mattyglesias is no longer available.)
I don’t like the phrasing here; what is arguably immoral is *spending* a billion dollars on yourself or leaving a billon dollars to your heirs
Quote tweet, 20:23 UTC · 34 likes
Good
Quoting Hunter📈🌈📊 (@StatisticUrban): Circumcision is on the decline in the US. For the first time in decades, less than 50% of American newborn boys are circumcised.
You might think this is due to demographic shifts. Latinos have very low circumcision rates, and are a larger and increasing share of total births. And this is part of it.
But what's also interesting is the decline amongst white Americans. They declined by 5.3pts, from 65.3% to 60.0%.
And it's not just a wealth or insurance thing either. The wealthiest Americans, the top income decile, experienced one of the largest declines. In 2022, only 51.4% of them were circumcised, nearly 10pts lower than 2012.
Reply to @Rado9910, 20:56 UTC · 2 likes
Rado9910 (@Rado9910): @austinc3301 @AaronBergman18 maybe?
Hmm I have bits and pieces for non-technical but not a complete reliable “just works” workflow
- I have models that sound human (I could train to be specifically Agus in fact) but they’re not reliable producing high quality content
- a simple Opus responds -> sonnet rewrites with detailed instructions on how to sound human works pretty well and is the most reliable option, but it’s merely good not like absolutely incredible
- this post was a basically 50/50 effort between me and Opus 4.1 https://forum.effectivealtruism.org/posts/je5TiYESSv53tWHC9/utilitarians-should-accept-that-some-suffering-cannot-be-1
The workflow was basically: get opus to write the basic semantic *content* based on my totally unstructured shitty writing quality idea dump, maybe iterating a few times, but eventually even if it was basically decent I’d go over and tweak the wording of every sentence to sound better. By the end i dont think any sentence was fully AI written, maybe one or two were. I don’t think this is like 99th percentile writing quality tbc but it’s passable
- not what ur asking exactly but I have a good system for rewriting audio transcripts to be what IMO people actually want: 98% verbatim; meaningless disfluencies removed (repetition, stumbling), but not totally sanitized to the point of removing character; formatted well instead of huge block of text, more accurate than input using semantic inference rather than just best guess of an isolated word maybe with a small context window
Quote tweet, 20:59 UTC · 15 likes
How it feels to walk out of an exam or stressful interview into a bright sunny day outside
Quoting KTLA (@KTLA): A dolphin was captured catching some big air off the coast of Hawaii. 🐬
Read more: https://ktla.com/news/the-michael-jordan-of-dolphins-photo-captures-wild-dolphin-catching-huge-air/
Reply to @moldbugchaser, 21:14 UTC · 16 likes
xenoduck (@moldbugchaser):
image attached
(quoting @pastasnack_e: Lots of people's social & romantic lives could be vastly improved if it was normal for young adults to live in college dorm-style accommodation in city centres until they were ready to set up home properly. Why doesn't this exist? I think there would be a market for it!)
Real
Tweet, 21:34 UTC · 60 likes
I feel like I didn’t realize how little UV index tracks intuitive brightness and temperature
If you don’t like sunscreen, you always have an hour after sunrise and before sunset where it’s basically fine to raw dog being outside, usually a bit more but a solid reliable hour
Thread, continued, 21:34 UTC · 5 likes
This applies almost anywhere on earth at any point in the year but apparently “high altitude tropics” (which is apparently a thing) are borderline so if you find yourself in uh Quito Ecuador, La Paz Bolivia, or the Tibetan Plateau, during summer, you might only get like 45 mins
Tweet, 22:38 UTC · 4 likes
I’m not exactly a comms expert but minor comms (and substantive) tip: if you’re talking about AI-driven unemployment, don’t only or even mainly emphasize “social unrest” “that society will fall apart” or that it will lead to “mob-driven policy”
Thread, continued, 22:38 UTC · 4 likes
(unless you realize what that implies and are willing to go to bat for those in fact being the main problem)
“Social unrest” etc may be true, and they may be bad, but they’re secondary to the problem that people don’t have enough money to buy things they want and need
Thread, continued, 22:38 UTC · 3 likes
That is actually what’s going on, I feel like people are weirdly hesitant to say it for some reason, but the Davos-esque ~euphemisms at least to me seem to be implying that the *real* issue is for elites and the wealthy:
Thread, continued, 22:38 UTC · 3 likes
Ie the mob coming for taking your stuff or power, not that the mob can’t earn enough money to buy things
I worry this will be especially tempting during an intermediate period of merely low wages and a bad (for workers) labor market
Thread, continued, 22:38 UTC · 2 likes
So maybe people aren’t literally starving at least in rich countries, and maybe they even have a standard of living that society tacitly accepts today
Tweet, 22:38 UTC · 1 like
So yeah we need to make sure everyone has money to the greatest extent such that it doesn’t reduce the quality of life of the future poor to a greater degree
Thread, continued, 22:39 UTC · 2 likes
(plausibly the only correct justification for wealth and income inequality we see today [via economic growth and the nature of exponentials])
Reply to @panickssery, 23:46 UTC · 15 likes
Arjun Panickssery (@panickssery): Why is that immoral
From a straightforward utilitarian perspective, it’s just good to deploy money in the service of others’ wellbeing and by the time you hit “how many tens of millions of dollars should I spend on a yacht” I’m extremely unsympathetic to additional personal consumption
In a strong metaphysical sense I believe in moral ordering but not normativity, but I’m very happy to have normativity-like social norms in the service of better rather than worse worlds, and I think the norm/take of “spending a billion dollars on oneself (conditioning on a current normal human lifespan) is selfish to the point of deserving condemnation” or the policy “everyone knows and abides by the policy that you’re expected to give away the billionth+ dollar (or keep invested for now and so this later) rather than spend on oneself” makes for a better world
One might argue that this reduces the incentive to make good products and services and such that grow the pie, but I don’t think this is true to the extent that it’s the winning consideration because as an empirical psychological-economic matter we just have declining marginal utility of money such that empirical estimates of personal utility of the billionth dollar (relative to at more normal incomes) are trivially small (it’s even smaller than you’d guess I think, I’ll try to pull up an actual figure on this)
Quote tweet, 23:47 UTC · 24 likes
Why I think spending a billion dollars on oneself (over a normal human lifetime) is bad and should be socially condemned
Quoting Aaron Bergman 🔍 (@AaronBergman18): @panickssery @mattyglesias From a straightforward utilitarian perspective, it’s just good to deploy money in the service of others’ wellbeing and by the time you hit “how many tens of millions of dollars should I spend on a yacht” I’m extremely unsympathetic to additional personal consumption
In a strong metaphysical sense I believe in moral ordering but not normativity, but I’m very happy to have normativity-like social norms in the service of better rather than worse worlds, and I think the norm/take of “spending a billion dollars on oneself (conditioning on a current normal human lifespan) is selfish to the point of deserving condemnation” or the policy “everyone knows and abides by the policy that you’re expected to give away the billionth+ dollar (or keep invested for now and so this later) rather than spend on oneself” makes for a better world
One might argue that this reduces the incentive to make good products and services and such that grow the pie, but I don’t think this is true to the extent that it’s the winning consideration because as an empirical psychological-economic matter we just have declining marginal utility of money such that empirical estimates of personal utility of the billionth dollar (relative to at more normal incomes) are trivially small (it’s even smaller than you’d guess I think, I’ll try to pull up an actual figure on this)
Thread, continued, 23:48 UTC · 4 likes
By all means have $100B invested that you control, but don’t actually burn $1B of those on yourself
Reply to @tess_ashpool, 23:51 UTC · 2 likes
Tess Ashpool (@tess_ashpool): thanks bro! appreciate the advice. I had always heard 10-2 were prime sunburn hours, sounds like it’s more restrictive
I mean it’s a matter of degree - those are prime sunburn hours (if they exist at all in your location) but like IMO its unwise to not wear sunscreen at 9am or 4pm on a sunny summer day, at least at DC/SF/NYC-ish latitudes
Sunday, March 22, 2026
Reply to @panickssery, 00:04 UTC · 5 likes
Arjun Panickssery (@panickssery): tldr but I ask: do you think this can be framed as "waste"
many agree its bad to waste, and if you spend on low-utility consumption then its a waste
Most of the time based on what people actually spend on yeah I think so, or “close to waste” maybe if you prefer that framing - it’s not 100% waste but wasting even a third of a billion dollars is pretty bad
A $100M yacht is largely waste because the resources taken to create it could have gone to making something that other people value a lot whereas irl it approximately doesn’t affect the life of its owner. Some supplier surplus, yes, we can subtract that and only call the result “waste”
I’m not opposed to financial framings of this kind of thing in principle, but they’re hard to get right and unnecessarily confusing (eg what the multiplier effect on GDP really means and corresponds to) and real resource framings are much better:
The yacht is waste because it’s using real time, effort, steel, etc for a trivial amount of consumer surplus
In this sense it’s like paying people to dig ditches: you can invent stories where for just the right reasons this causes money to start flowing in just the right way again, but fundamentally it’s a *literal* waste of people’s time and effort in that nothing anyone values is produced
It’s true that there’s some producer surplus, so fine subtract that from what we call “waste”- there’s still a lot left over
Reply to @gmiller, 00:31 UTC · 1 like
Geoffrey Miller (@gmiller): The money isn't 'burned'.
It's spent on goods and services that employed thousands of people and let them pay their mortgages and feed their kids.
Maybe that was the wrong word choice but I am aware and don't think that justifies it
Reply to @JancsiNeumann, 00:32 UTC · 2 likes
Johnny V. Neumann (@JancsiNeumann): What does “In a strong metaphysical sense I believe in moral ordering but not normativity” mean?
Oof I I've done this discourse like 20 times now so forgive me for just linking a past tweet
Reply to @Rado9910, 00:33 UTC · 2 likes
Rado9910 (@Rado9910): @AaronBergman18 @austinc3301 By the way this is one of my favorite articles. The fact that you cited Claude as a contributor made me smile, a good example of intellectual honesty.
Lmao thank you!!
Tweet, 03:54 UTC · 4 likes
Customized, real daily news briefing based on my instructions to prioritize genuine importance
Claude scheduled task -> Claude-written code detects new file -> Claude written code sends to OpenAI TTS api ->
Tweet, 03:54 UTC · 1 like
RSS in case u wanna listen! First episode is 12 mins
https://aaron-news-podcast.s3.us-east-1.amazonaws.com/podcast_feed.xml
Thread, continued, 03:54 UTC · 2 likes
Prompt in case you wanna use (it’s not meant to be general so you’re gonna wanna tweak most likely):
Use your best judgement and something like an EA worldview to curate "news" (but I really mean "information or analysis about the world which may sometimes include things normally labeled as news") and provide a summary whose level of detail roughly corresponds to the importance. Examples of things that matter: AI, Taiwan, things that pertain to any hedonically intense state so progress (or bad news) on cluster headaches, torture, painkillers, antidepressants, the new world screwworm, anything like that (but plz be tasteful not graphic so I don't turn remove this category lol), New diseases, medicine when genuinely a big deal, POTUS (use your judgement), Congress (use your judgement), SCOTUS (use your judgement), global health, animal welfare, anything that affects hundreds of millions of people even if not salient in the US, anything that affects farm animals, technology when a genuine development, but this list is NOT exhaustive; at the bottom you can put a brief mention of anything very salient and popular that is reasonably good to at least know even if not really objectively important
Tweet, 05:05 UTC · 4 likes
The standard Progress Studies line is that people are unreasonable for complaining about various aspects of modern aviation (including that it’s gotten worse)
Thread, continued, 05:05 UTC · 3 likes
I largely agree but i think there’s a reasonable grievance here: the product that corresponds to “economy but in the 1980s” simply doesn’t exist, and furthermore I think it would in a competitive market
Thread, continued, 05:05 UTC · 2 likes
I didn’t realize this until I had reason to look at business class prices relative to economy recently; it was 4x the price of economy (and according to Claude this is very typical)
Thread, continued, 05:05 UTC · 4 likes
There are no 1.5-3x tickets for sale where you get an appreciable amount of additional space; the current option set is not a strict improvement over the more-expensive-in-general past
Reply to @bashu_thanks, 05:17 UTC · 5 likes
bashu, thanks (@bashu_thanks): imagine you move to a city where you don't know anybody... where would you get 2 hugs a week?
Dating app
Reply to @ohabryka, 05:45 UTC · 5 likes
Oliver Habryka (@ohabryka): Why would that be immoral! You got those things via trades that benefitted everyone involved. You almost certainly helped the world vastly more than people who are much less rich.
God, I find this attitude of “it’s immoral to have resources that you don’t spend on altruism” attitude extremely frustrating.
I think the quoted response spells things out in more detail
I’m largely selfish tbc, and don’t blame others for not being maximally altruistic (although I do think selflessness is praiseworthy), but what margin we’re talking about just does matter a lot, and for a bunch of reasons. The marginal utility of money past $1B consumption is awfully small, billionaires are extraordinarily lucky (and skilled and hardworking etc, but still) in a way that seems morally arbitrary, and the people who can spend more than $1B tend to have a a lot more than that such that their generosity is more consequentially important than the generosity of thousands or even millions of more normal people
So the personal costs are tiny and the potential benefits to others are massive and so yeah I’m gonna condemn making the decision in that direction. It’s like the drowning child thought experiment except the suit is was free and the pond is a knee deep swimming pool
Probably the crux here is just that I think some form of utilitarianism is true and IIRC you do not
And this isn’t utilitarian fanaticism or like a weird edge case that violates norms or something; it’s very straightforward
On the mutually beneficial trades thing: that matters for ex ante incentives (I don’t think we should trick anyone), it matters for economic growth, it plausibly matters for tax policy, maybe it matters in other ways, but it doesn’t change the basic utilitarian calculus
Reply to @9chabard, 06:25 UTC · 6 likes
Lysandre, qui la connaît à peine (@9chabard): if you didn't write your own favorite tweet of all time, that's a skill issue.
image attached
(quoting @PopBase: Twitter turns 20. What is your favorite tweet of all time?)
It is a banger
Reply to @flowersslop, 06:28 UTC · 8 likes
(The tweet being replied to by @flowersslop is no longer available.)
Lmao this isn’t true
Quote tweet, 15:46 UTC · 8 likes
This is cool
Quoting Alex Prompter (@alex_prompter): 🚨 BREAKING: Someone just open-sourced a full offline survival computer with AI, Wikipedia, and maps built in.
Project N.O.M.A.D. is an open-source offline survival computer.
Self-contained.
Zero internet required after install.
Zero telemetry. Everything runs locally on your hardware.What it includes:
→ Full Wikipedia archives via Kiwix
→ Offline maps via OpenStreetMap
→ Local AI models via Ollama + Open WebUI
→ Calculators, reference tools, resource libraries
→ A management UI to control
everything from a browserOne curl command installs the entire system on any Debian-based machine.
Runs headless as a server so any device on your local network can access it.
Minimum specs to run the base system: dual-core processor, 4GB RAM, 5GB storage.To run local LLMs offline, you want 32GB RAM and an NVIDIA RTX 3060 or better.
No accounts.
No authentication by default.
No cloud dependency.
No phone-home behavior.Built to function when nothing else does.
The grid, the cloud, the API you depend on. None of it is guaranteed.
The people building local-first systems right now are the ones who won’t be asking for help when access disappears.
Tweet, 16:40 UTC · 34 likes
This is prob very dumb 101-level half question/half observation but:
In the story where AI replaced ~all human labor and there’s no redistribution and just a few fantastically wealthy capital owners, won’t there not *be* an economy in the way we think about it today…
Thread, continued, 16:40 UTC · 8 likes
Like it’ll just be a few people who can make their own stuff arbitrarily, not really an economy in the standard sense
Eg there won’t be that many cars bc presumably people stop caring after the first few thousand, and there won’t be any restaurants or stores around
Thread, continued, 16:40 UTC · 5 likes
The question part: am I missing something?
I know this might not happen for various reasons but like if it does…
Thread, continued, 16:40 UTC · 8 likes
I have no idea if this has any implications beyond the intrinsic badness of the masses being miserable and/or dying
Tweet, 17:45 UTC · 23 likes
TIL that the standard ferritin reference range starts at 30 ng/mL which is way below what the RLS (restless leg syndrome) literature (very mainstream, no culture wars over this afaict) uses: 75 ng/ml
And tbc 75 is the bottom of the acceptable RLS range, not a target
Thread, continued, 17:45 UTC · 8 likes
So if you have RLS or think you might, a non-flagged ferritin is not great evidence that iron won't help
(Mine as of two months ago was 45 and I've been having non life-ruining but definitely annoying RLS for the first time in years)
Thread, continued, 17:45 UTC · 5 likes
Also: always ask for, and if necessary, demand, the full set of results of bloodwork you get, so you can give them to claude/whatever and ask questions
You have the legal right to full them within 30 days, and the term to search for is "CMS-2319-F"
Thread, continued, 17:45 UTC · 4 likes
I wasn't given access to this 45ng/mL result by default (I didn't have to fight for it either this time, just ask, fortunately)
Thread, continued, 17:45 UTC · 3 likes
Also credit where credit is due - this^ is a good example of a great regulator rulemaking (even if it was latent before and you could go to court, most patients aren't going to do that and the clarity makes all the practical difference)
Quote tweet, 23:41 UTC · 9 likes
Seems bad
Quoting Sam Yamshon (@samyamshon): @AaronBergman18 I’m a hematologist and can confirm, the lower end of reference range for ferritin at most labs is way below clinical iron deficiency in general. See this very frequently
Monday, March 23, 2026
Reply to @MartinShkreli, 06:36 UTC · 4 likes
Martin Shkreli (@MartinShkreli): retatrutide is not FDA approved and is not available anywhere other than in clinical trials.
(quoting @1thousandfaces_: I’m so serious, retatrutide has made me genuinely unable to eat anything but vape and ice cream. I can’t stomach more than a few bites of anything else but I’m about to go back for a second scoop of ice cream)
It’s easier to get than weed (which is good)
Reply to @cxgonzalez, 06:49 UTC · 54 likes
christian (@cxgonzalez): guy who cares about shrimp welfare but is mean to his mom
I’m basically certain that caring about shrimp more than avg is correlated with being nicer than avg to one’s mom
People desperately want utilitarians/EAs to be evil but in fact the general psychological tendency to care about others is the phenomenon that dominates
Reply to @Noahpinion, 06:57 UTC · 7 likes
Noah Smith 🐇🇺🇸🇺🇦🇹🇼 (@Noahpinion): Jensen actually doing an intelligent pitch for AI, instead of "We will make you unemployable and also possibly kill you and your children"
(quoting @unusual_whales: Nvidia CEO Jensen Huang said that he believes AI will increase productivity and create jobs rather than eliminate them)
Lol whether or not he’s good at PR, he’s very very happy to GPUs to the “kill your children” people
Quote tweet, 07:21 UTC · 161 likes
Yeah we’re gonna need a new social contract around this
Most of Americans’ privacy is downstream of no one being able to monitor all the data sources they legally have access to (including the state)
Quoting Rick (@rickasaurus): AI really changes the nature of these security cameras
Thread, continued, 07:21 UTC · 34 likes
Arguably there was a reasonable compromise of sorts: data really could be pieced together to infer quite a lot about actions, wearabouts, etc of a person, but only by dedicating a very costly amount of resources to the task à la FBI investigation or major counterterrorism project
Thread, continued, 07:21 UTC · 25 likes
Until now, of course. Now, or soon, the Trump administration will be able to know, for trivial costs that which required hundreds or thousands of skilled man-hours previously
Sounds hyperbolic, but I don’t think it is
Thread, continued, 07:21 UTC · 19 likes
I think a lot of “law abiding citizens” are going to discover that actually they didn’t tell the whole truth on an official form in 2004, or are on camera sharing weed with a friend (that’s distribution, and distribution can mean decade-long prison sentences)
Thread, continued, 07:21 UTC · 17 likes
or are “merely” contacted by law enforcement to inform them that the joke Venmo note they wrote, which the AI panopticon picked up, is “wire fraud”, and they’re the target of an investigation (extraordinarily stressful and a functional threat even if no charges are ever filed)
Thread, continued, 07:21 UTC · 15 likes
The kind of freedom and privacy one has in a world like this is very very thin - US legal protections are necessary but nowhere close to sufficient in an intelligence-saturated world
Reply to @thkostolansky, 07:27 UTC · 8 likes
Tim Kostolansky (@thkostolansky): @AaronBergman18 @cxgonzalez > basically certain
how are you certain? this is unfalsifiable? seems like bad epistemics to make claims like this so surely when it seems pretty important (to me)
> People desperately want utilitarians/EAs to be evil
is this true? seems like this isnt true ime. maybe it *feels* true cus those who claims utils/eas are evil say it in worse ways?
> in fact the general psychological tendency to care about others is the phenomenon that dominates
this also feels anecdotally true-ish but also feels like its hard to make this claim with much certainty. a love for ones mother is a very classic human relationship that lots of people do do very strongly baseline imo
The middle thing I might have overstated, fair enough (although it’s literally true that people as in “some people” want this)
“Basically certain” is just my subjective probability and one that I’d bet on if it was operationalizable - falsifiability isn’t required for probabilities to make sense (it does help to track who gets things right and wrong but still)
Idk I feel like I see enough counterexamples (and anecdotes count in that way) around teens and adults viscerally disliking their mothers
Reply to @thkostolansky, 07:30 UTC · 3 likes
Tim Kostolansky (@thkostolansky): why is this good?
People should have the freedom to take medicines as they please except under extraordinary circumstances in which there are nontrivial negative externalities
Reply to @panickssery, 07:33 UTC · 18 likes
Arjun Panickssery (@panickssery): You buy a $2 lottery ticket that wins $1m the next day. A man confiscates the ticket for $10 and calls it "justice" because, if he had asked you yesterday, you would have "hypothetically consented," which applied to this unrelated actual situation
This is your brain on Rawlslop
Is this supposed to be against the veil of ignorance? Bc if so I feel like that stands on its own even if the general principle of “counterfactual consent” or something isn’t a universally good argument
Quote tweet, 08:43 UTC · 66 likes
Serious question: what happens if you just…install the hot tub?
Quoting john allard (@john__allard): Berkeley NIMBYs are the best to ever do it … public notice for a hot tub
Thread, continued, 08:43 UTC · 15 likes
With large construction projects the answer can be that even low state capacity can function as a schelling point of expectations such that folks are worried that someone in the supply chain will balk, and also you can get disqualified from winning city contract bids
Thread, continued, 08:43 UTC · 10 likes
I’m not sure I buy that you wouldn’t be able to hire a couple guys to help assemble a hot tub (apparently it’s doable but hard with one layperson - you want need like 3-5 men for lifting the shell and an electrician) even if there’s been no hearing for it
Thread, continued, 08:43 UTC · 17 likes
Do the police show up? Get an angry letter? Literally nothing? Lawsuit? Jail? Prison? Fine? Does the city come and physically remove it? With guns in case you try to stop them?
Thread, continued, 08:43 UTC · 15 likes
I feel like you never actually see the mechanisms of state capacity and enforcement for nonviolent crime in the West, but this is a rare case where you could imagine it
Thread, continued, 08:43 UTC · 13 likes
And tbc a load bearing thing here is that SFPD are, at least on Twitter, famous for refusing to go seize a stolen bike that you can prove is yours
Thread, continued, 08:43 UTC · 10 likes
So it would be weird if they had their shit together to deal with an insubordinate homeowner, especially one who can afford lawyers
Reply to @GMShivers, 09:12 UTC · 15 likes
Jennifer D'aww (@GMShivers): They had to create remote work because autistic people being put in open concept offices is emotional terrorism on your revenue class.
(quoting @jordanreviewsit: Who else is excited for another week in paradise 🙌)
They are the worst of all worlds
I don’t even think an office is crazy - lower the transaction costs between interactions/communication - but Jesus give people their own room
Reply to @wilhelmscreamin, 16:54 UTC · 11 likes
catherine (in sf) ʕ•ᴥ•ʔ-☆ (@wilhelmscreamin): i don't understand why people think there are too many podcasts. there are like 3-4 good podcasts in the world. i would love there to be more good podcasts
Also like let people enjoy things
A terrible podcast that someone has fun making is a good thing
Quote tweet, 17:08 UTC · 7 likes
Maybe unironically I’m hereby creating a benchmark which is just explaining this and the equity premium puzzle. 2/2 is cognitive superintelligence
These things can’t be unsolvable, they should even be methodologically pretty straightforward
Quoting Hitchslap (@Hitchslap1): Lesbians have the least stable marriages. Gay men have the most stable marriages.
I’m yet to hear a good explanation for this.
Tweet, 19:16 UTC · 15 likes
Everybody stop what ur doing* and check out the new, official landing page of Screwworm Free Future (I had no part in making it this time, and it looks much better than before!)
* Not super seriously
Tweet, 19:20 UTC · 64 likes
An underrated question IMO: do people *enjoy* sports betting, in the way ppl enjoy cigarettes and weed and video games?
Or does it feel more like a pramipexole-induced compulsion?
Thread, continued, 19:20 UTC · 7 likes
Also I think the case for limiting amounts is much stronger than for banning
Like I assume that the fun doesn't scale linearly with how much money is on the line right?
Reply to @RyanRadia, 19:44 UTC · 7 likes
Ryan Radia (@RyanRadia): For some people, betting on a televised sporting event might make it more enjoyable to watch than it otherwise would be
https://www.theatlantic.com/magazine/2026/04/online-sports-betting-app-addiction/686061/
Yeah this is probably true for me ($10 march madness)
Reply to @sgodofsk, 23:52 UTC · 16 likes
Steven Godofsky (@sgodofsk): I think it was @eigenrobot who said that rationalists fear a hyper-intelligent AI because they're the only ones in the world who can be convinced to do absolutely anything by a sufficiently well written persuasive essay.
Lol it’s a good quote but it’s not true
Tuesday, March 24, 2026
Reply to @gptbrooke, 00:39 UTC · 36 likes
brooke bowman (@gptbrooke): There should be a version of ‘earn to give’ that’s something like ‘get pretty to persuade’
I can’t think of a way to say this that doesn’t feel mean, but I think if you have a cause you really care about, and are lacking in basic hygiene practices, you’re leaving money on the table
We’re wired to take fitter & more hygienic/presentable people more seriously than someone who eg clearly needs a shower and a hair cut. You may not want the world to work that way, but it just does
I’m a little worried I’m going to get torn apart for posting this, but I genuinely want to see the people working on important causes have the best fighting chance to achieve their goals
Yeah I think there's something to this, although people need to be benevolent and capable of nuance enough to avoid condemning less-than-hot as itself immoral (I think about intelligence the same way - very instrumentally useful, not a moral trait in itself)
Tweet, 01:25 UTC · 124 likes
There are depressingly-many people overseas (largely India and Pakistan) who have excellent ratings on Upwork and are eager to work for $5/hour. Good reminder to count your blessings
Thread, continued, 01:25 UTC · 41 likes
But more actionably: you can make someone a super generous offer from their POV that's clearly a massive win-win
I think Westerners largely turned away from this kind of thing because it's psychologically tempting to essentially reduce the salience of the developing world to 0
Thread, continued, 01:25 UTC · 27 likes
And then libertarian types swoop in to say eg "sweatshops are good actually"
Whereas I think the correct take is that there are massive win-wins for everyone involved, especially if you have enough financial slack to be more generous than the market strictly demands
Thread, continued, 01:25 UTC · 16 likes
Is this the best way to turn money into utils? Not really, no, but if it gives you the fuzzies necessary to actually do the mutually-beneficial trade (bc otherwise you'd feel bad/guilty) then it's doing important work for everyone
Reply to @bashu_thanks, 01:26 UTC · 2 likes
bashu, thanks (@bashu_thanks): wait. is this how flirting works. sincere question
(quoting @bstract_thot: @SurrealistShip related, to me puns feel the same as flirting except instead of looking for opportunities to call the other person hot you're looking for opportunities to pun)
Wait what
Thread, continued, 01:26 UTC · 5 likes
Nobody tells me things like this except for other neurodivergent ppl on twitter
Quote tweet, 02:49 UTC · 33 likes
> Only 2% of applicants complete the full process.
> In one recent hiring cycle [the FAA actually gained]: around 160 controllers
Usually, grave responsibility comes with power and status (“power” not in a nefarious way; actual decision making authority)
Quoting Aakash Gupta (@aakashgupta): The air traffic controller cleared the fire truck onto the runway. Seconds later, the same controller screamed “stop, stop, stop.” The plane was doing 93 to 105 mph.
Both pilots are dead.
Everyone will frame this as controller error. One controller was simultaneously managing a United flight that aborted takeoff after an anti-ice warning, dispatching a fire truck across an active runway, and sequencing an inbound Air Canada landing at highway speed. At 11:40 PM. On a mandatory overtime shift at a facility that has been understaffed for years.
A system that assigns one person that workload will produce exactly this outcome. The only variable is when.
The FAA is short approximately 3,000 controllers. The headcount dropped 13% from 2010 to 2024 while flight volume rose 10%. Over 40% of the FAA’s 290 terminal facilities are understaffed. The New York TRACON, which manages the most congested airspace in America across LaGuardia, JFK, and Newark, has been chronically below target. Newark was operating at 59% of its staffing goal. LaGuardia handles 900 flights a day.
The hiring pipeline is broken at every stage. Only 2% of applicants complete the full process. Training takes up to 6 years. The FAA Academy in Oklahoma City is a bottleneck, with roughly 35% of trainees washing out. Congress blocked legislation to build a second academy. In one recent hiring cycle, the FAA brought on 1,512 candidates and lost 1,300 in the same window. Net gain: around 160 controllers for an entire country.
Three things need to happen and everyone who can make them happen has known for years.
Congress needs to fund and authorize a second FAA training academy. One facility in Oklahoma City cannot produce enough controllers for 900 million annual passengers. Members of Congress from Oklahoma have actively blocked this. That needs to end yesterday.
The FAA needs to cut certification time. Six years from application to fully certified controller is absurd. The agency’s own data shows tower simulators reduce certification time by 27%. They’ve installed them at 95 facilities. That should be every facility, and the simulated hours should count toward more of the certification requirement.
The FAA needs to stop plugging staffing gaps with mandatory overtime. Controllers at understaffed facilities are working six-day weeks rotating between morning, mid, and night shifts. The NTSB has flagged fatigue repeatedly. The controller last night was managing overlapping emergencies during a nighttime operation. Overtime is not a staffing plan. It’s a countdown to the next runway collision.
The controller said “I messed up” to a Frontier pilot who watched the whole thing. The pilot responded “No man, you did the best you could.”
One of them is right. The answer determines whether this happens again.
Thread, continued, 02:49 UTC · 20 likes
Shockingly nobody wants to work mostly nameless and faceless behind the scenes, working hard for decent but not extraordinary pay (maxes out below $200k I think), while being responsible for the lives of thousands of people who stand to die if you fuck up even once
Thread, continued, 02:49 UTC · 8 likes
Pilots at least get to fly airplanes (and are relatively high status in society)
Reply to @realBoKane, 03:06 UTC · 2 likes
Bo Kane (@realBoKane): what are you using upworkers for?
Outsourcing a pretty monotonous and lower-skilled task (basically, google, click around to identify where a pdf might live, download, and confirm for ~400 specific academic open access papers) that would probably take me like maybe 10 hours, as part of a larger project that I myself am contracting for
tbc these 400 were selected directly for being the ones that any sort of programmatic or LLM-based or API-based process failed to capture
The idea was like: I don’t want to do this, it’s not really a good use of my time for advancing the project, the organization I’m contracting for doesn’t want to pay me to do this because I’m expensive, a lot of other people do want to do this for $200 (and tbc I’m not collecting US wage hours for the work done by someone in India)
Like probably I could figure out a system to do this but it would take longer than manual collection
Tweet, 03:36 UTC · 287 likes
Lol I might create an “S-corp” and I’m just now learning the details and is truly truly stupid that it exists
Thread, continued, 03:36 UTC · 92 likes
Basically: If you’re an independent contractor, you
1) set up an LLC (legal entity) with “s-corp tax treatment”
2) have *your* LLC pay *you* a salary less than your actual net earnings
3) as owner of the s-corp, you take the profit as well as the salary
Thread, continued, 03:36 UTC · 81 likes
4) but you don’t pay payroll taxes on the profit part, only the salary, even though the distinction is purely an on-paper formality: doesn’t correspond to any distinction irl
It gets better:
Thread, continued, 03:36 UTC · 82 likes
5) why not just pay yourself $1 and then take all of it as profit and avoid all the payroll tax?
Because the salary has to be “reasonable” according to the IRS, but the IRS *won’t tell you* what that means
Thread, continued, 03:36 UTC · 73 likes
Tbf apparently they give you some broad guidelines and considerations but there’s no formula, and you can’t ask ahead of time - you’d just learn ex post that you picked an unreasonably salary and that’s a tax violation.
Thread, continued, 03:36 UTC · 86 likes
And throughout all of this, most people pay an accountant to deal with the Byzantine paperwork and certification, so there’s a whole sub-industry that simply should not exist
Tweet, 04:56 UTC · 146 likes
“Timelines” are getting much less abstract
I think I personally have ~2 years of relatively in-demand skills. Could be 4, won’t be 8
Thread, continued, 04:56 UTC · 42 likes
The economic game we think of as a geological feature of earth (broadly: substantial nonfarm employment; most people sell labor) is only 300ish years old really looks like it’s gonna come to an end soon
Thread, continued, 04:56 UTC · 27 likes
Maybe it will take 3 years instead of 1 to proliferate throughout the economy, maybe something crazy will happen and we’ll make it to 2032
But things are not going to be the same, I hope for better but all things considered I think got worse
Reply to @AndyMasley, 04:56 UTC · 18 likes
Andy Masley (@AndyMasley): Andy Masley Research LLC is also an s-corp. All the employees here are big fans
Wait is this real
Reply to @RealRiddimHours, 04:57 UTC · 14 likes
(The tweet being replied to by @RealRiddimHours is no longer available.)
Lol but also 💀
Reply to @AndyMasley, 04:58 UTC · 10 likes
Andy Masley (@AndyMasley): Yes
Based, I will start brainstorming names (suggestions welcome)
Reply to @chrislakin, 05:30 UTC · 31 likes
Chris Lakin (@chrislakin): oh, you don't have an s-corp? (lol)
Guy who doesn’t have an s-corp by 26 (pathetic)
Reply to @polisisti, 05:32 UTC · 2 likes
Alessandro (@polisisti): @AaronBergman18 I think it's going to take quite some time for even very capable AI to be woven into organizational structures. Look at how long self-driving cars are taking to roll out. I think big changes will happen in our professional lives, but <10 more years of work seems quite unlikely
Do you want to bet on % of people in US (or US and Canada) in the workforce and earning most of their money from labor in 2035?
Reply to @polisisti, 05:54 UTC · 2 likes
Alessandro (@polisisti): @AaronBergman18 Sure 🙂 I can come back to this later in the week with a specific proposal after I see what data are easily available
Based
Reply to @helenlewis, 14:56 UTC · 3 likes
Helen Lewis (@helenlewis): I’ve written about the backlash to Adult Braces: “I do feel great sympathy for Lindy West. How was she to know that the great omertà of Millennial Feminism—that we had to take whatever people said about their life stories at face value—had broken?”
Interesting but I object to this tiny bit
As an epistemic matter it’s true that one cannot just declare that they have some condition
Otoh there’s a totally separate case to be made that people should control their own medical care for normal human freedom reasons; if she wants stimulants she should get stimulants for the same reason I can buy drinks at a bar for myself without proving to anyone that I have Mild to Moderate Social Anxiety Disorder that Ethostat (ethanol) has been approved by the FDA to alleviate
Reply to @TallRomanStatue, 15:02 UTC · 7 likes
Tall Roman Statue (@TallRomanStatue): Claude is too woke to add value beyond routine work which is already a race to the bottom.
While it can innovate non politically on command (ex: personalize vaccines), it cannot do any serious thinking or problem solving while it is programmed for equality
It's just a spreadsheet brained slave
Have you ever tried Claude? It’s not actually woke
Tweet, 20:47 UTC · 23 likes
Who up watching the Anthropic vs DoW hearing?
https://cand.uscourts.gov/judges/rfl/lin-rita-f
Tweet, 20:47 UTC
Direct zoom link: https://cand-uscourts.zoomgov.com/j/1615482105?pwd=cTZiSVdXelllSXpjQ1p0Nm5HaXRYUT09
Thread, continued, 20:47 UTC · 9 likes
Bruh DoW is getting (appropriately) cooked
Thread, continued, 20:47 UTC · 2 likes
Why does it matter if the tweet is *announcing* the supply chain risk order or *is* the supply chain risk order? This seems totally inconsequential but everyone is treating it like it matters
Thread, continued, 20:51 UTC · 7 likes
Bruh this poor DoW lawyer
DoW should argue that the judge asking their own lawyer questions is cruel and unusual punishment (it's painful fr)
Thread, continued, 20:58 UTC · 2 likes
The Trump administration is so plainly incompetent whenever there is like some sort of legitimate or competent process/institution mediating between them and the world (eg a court hearing)
Thread, continued, 21:04 UTC · 3 likes
Lmao DoW is arguing that Anthropic doesn't have standing to complain about Congress not being notified, which is written directly into law, because that harmed Congress rather than Anthropic
Look their legal team isn't stupid, they just don't have any substance to work with
Tweet, 21:13 UTC · 1 like
Lmao they're saying that the law that orders the agency to report various things to Congress first doesn't have a built-in mechanism for congress to veto the supply chain risk designation...
Thread, continued, 21:13 UTC · 2 likes
Indeed there is no *built in* mechanism but there is a mechanism: which is that Congress can pass or change a law
Reply to @ben_j_todd, 22:07 UTC · 3 likes
Benjamin Todd (@ben_j_todd): I wrote 80,000 Hours ten years ago because I was frustrated at how terrible career advice can be.
Today it’s even worse: still focused on traditional paths like law & medicine when we face AGI.
To fix that, Penguin are publishing a fully updated edition.
Goss
Reply to @MariGO2thepolls, 22:13 UTC · 2 likes
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): Party ideas. Each guest is tasked with bringing the most controversial person they know as a plus one
Mar make it happen
Tweet, 23:42 UTC · 12 likes
I’m anti-anti-Sabrina Carpenter
Thread, continued, 23:42 UTC · 11 likes
Some of her physical traits are too exaggerated to be attractive to you? Congrats bro, I bet she’s devastated
Thread, continued, 23:43 UTC · 2 likes
Wednesday, March 25, 2026
Tweet, 00:03 UTC · 12 likes
Losing a well-written lengthy prompt bc the page refreshed or you hit back by accident or something is like losing a child
Thursday, March 26, 2026
Reply to @AmandaAskell, 22:08 UTC · 6 likes
Amanda Askell (@AmandaAskell): Tech companies pay millions of dollars for their employees and then stick them in open-plan offices that make it nearly impossible to get work done. Best strategy for poaching employees is probably to just offer them an office with a door.
Shh ur gonna give away anthropics alpha
Reply to @moldbugchaser, 22:14 UTC · 3 likes
xenoduck (@moldbugchaser): monkeys, typewriters, and selection pressure. we have trouble intuitively grasping just how much stuff there is and just how long it's been evolving
Ya fr im always like “I get evolution in theory (mostly) but no way it’s been going on long enough for all this right”
Reply to @ohlennart, 22:20 UTC · 1 like
Lennart Heim (@ohlennart): p(bloom) peaked today
Lennart Heim has gone full Hinge girl
Reply to @litteralyme0, 23:14 UTC · 18 likes
Miilato (@litteralyme0):
image attached
(quoting @CultureCrave: First look at Paapa Essiedu as Snape in the 'Harry Potter' series)
Not to be too woke but snape isn’t a real guy
Friday, March 27, 2026
Reply to @deanwball, 00:27 UTC · 26 likes
Dean W. Ball (@deanwball): This is a devastating ruling for the government, finding Anthropic likely to prevail on essentially all of its theories for why the government’s actions were unlawful and unconstitutional. One of the things she mentions is the huge range of amici briefs supporting Anthropic (by the way, 0 supported USG)—so thanks to everyone here who signed on to FAI’s brief, or to one of the many many others. These things do matter. More importantly, you were on the right side of history.
On a personal note: some friends and allies of mine on the right have been angry at me for my own words and actions in all this. Anyone who thinks I spoke out for personal gain or trivial reasons against an administration I served in is crazy. It was a hugely costly decision for me. But Judge Lin’s ruling shows why I did it: this is a staggeringly illegal act by the government. That is why I am particularly honored to have been (implicitly) quoted in the ruling for calling this what it was when Secretary Hegseth initially made his announcement: an attempted act of corporate murder.
The case continues, but Anthropic has scored a very large win here. The real victors, however, are all red-blooded Americans who are, as the founders would have said, “jealous of their liberties.”
(quoting @Hadas_Gold: BREAKING: Anthropic has been GRANTED a preliminary injunction re: Pentagon 'supply chain risk' designation by Judge Rita Lin in California but is allowing a stay for one week https://storage.courtlistener.com/recap/gov.uscourts.cand.465515/gov.uscourts.cand.465515.134.0.pdf)
🫡🇺🇸
Saturday, March 28, 2026
Reply to @JeffLadish, 21:30 UTC · 13 likes
Jeffrey Ladish (@JeffLadish): This was already an embarrassing take in 2023. But it’s 2026.
It’s not even true that pretraining produces models that “can only regurgitate what other people have written”. But setting that aside, have these people even heard of reinforcement learning?
(quoting @Quillette: Most of today’s “artificial intelligence” is better described as artificial autocomplete than artificial mind, writes @pllevin. https://quillette.com/2026/03/27/ai-is-not-about-to-become-sentient-moltbook-openclaw/)
Also what does any of that have to do with sentience
Sunday, March 29, 2026
Reply to @ShadowyZephyr, 17:10 UTC · 5 likes
zephyr (@ShadowyZephyr): I actually love this because GPT-3 era LLMs would always come up with a reason to justify their initial claim, usually by messing up the math. SOTA LLMs today are capable of reflecting and realizing their initial response was wrong!
Yeah this is actually good cognition
Quote tweet, 17:21 UTC · 9 likes
What
Quoting euan ashley (@euanashley): Models performed well without, and a little better with, the images. In one case, our no-image model outperformed ALL of the current models on the chest x-ray benchmark—including the private dataset—ranking at the top of the leaderboard. Without looking at a single image.
🤯🤯
Quote tweet, 17:24 UTC · 9 likes
There really do be patterns
Quoting euan ashley (@euanashley): New AI paper from us this week. When my student first showed me his initial findings, I really didn’t know what to make of them. I felt that this was an interesting but curious loophole phenomenon that would shortly be closed. I was very wrong.
https://arxiv.org/abs/2603.21687
Reply to @MatthewGNagler, 17:30 UTC · 2 likes
Matthew G Nagler (@MatthewGNagler): Nematodes’ brains induce them not to move in the direction of food smells, but to move in the direction in which food smells *increase*
We are neurologically wired for reference dependence. It’s amazing it took economists so long to figure out that utility works in this way.
Wait am I crazy or aren’t these just the same thing when the reference frame is “the level of food smell right here”
Increasing ~ level
Thread, continued, 17:33 UTC · 1 like
Or like “moving in the direction of the food smell” was never a possibility without a model of where the smell was originating, which is exactly the problem that the organism has to try to solve
Reply to @vikhyatk, 17:43 UTC · 5 likes
vik (@vikhyatk): if you decide not to start a family because of the bullshit ai companies and doomers are peddling about AI, this will be you in 50 years
Which doomers are telling ppl not to have kids?
Reply to @TetraspaceWest, 17:53 UTC · 6 likes
Tetra 💎⏹️🌌 (@TetraspaceWest): airline attendant: final boarding, the plane is literally taking off, if you are hearing this message it's too late
me (boarding group 1 premium gold priority): guess it's time for me to go to the gate
(quoting @troller_Adi18: )
What is this “Megan” situation
Quote tweet, 17:57 UTC · 33 likes
I’ve started using a service that uses AI to make phone calls and this has the potential to be good but should really incur a tax of some sort
Quoting nxthompson (@nxthompson): OK, this is a pretty amazing use of AI.
Thread, continued, 17:57 UTC · 7 likes
I’d be happy to pay something like $0.25/min of answered call and $0.05/unanswered call in order to dissuade mass enshittification
Reply to @MichaelAlleged, 18:05 UTC · 2 likes
Michael (Allegedly) (@MichaelAlleged): I like when a good work buddy becomes a real friend beyond and outside of work
Yea and people should stop compartmentalizing social spheres in general
Tweet, 20:34 UTC · 7 likes
(From my drafts, a couple weeks old)
Doctor’s waiting room - woman in her late 70s-80s probably. But looks and sounds as healthy/vital as one can be at that age
Thread, continued, 20:34 UTC · 3 likes
Telling a story about how she found herself in something like a nursing home (I didn’t catch the beginning) and basically nobody glanced at her for 5 days
Thread, continued, 20:34 UTC · 3 likes
They also wouldn’t let her out, so she threatened to kill herself unless they let her leave
This led to her being sent to the psych department (perhaps not shockingly)
Thread, continued, 20:34 UTC · 3 likes
I only caught a few bits and pieces but she must have convinced someone she was the real deal (mentally and physically able to leave) because I couldn’t hear what else transpired but here she was, in a normal outpatient setting, as a free woman
Thread, continued, 20:34 UTC · 5 likes
You probably already know my schtick on this - seems really bad, there should be an exceptionally high bar for involuntary commitment, etc
Thread, continued, 20:34 UTC · 3 likes
But one additional point is that insofar as you’re worried about a crazy person harming others if they’re not committed, there are probably Pareto gains to be had by recognizing that the elderly (and especially elderly women)
Thread, continued, 20:34 UTC · 3 likes
Are just at much lower risk of harming anyone else (probably due to both disposition and ability)
Thread, continued, 20:34 UTC · 2 likes
Complementing this is the fact that the elderly sometimes literally cannot leave/escape without permission
Thread, continued, 20:34 UTC · 2 likes
Like if I’m inside a hospital, lucid, and not physically restrained, it’s because in *some sense* I want to be there (or else I’d just leave)
Thread, continued, 20:34 UTC · 2 likes
But even a relatively healthy and independent 82 year old woman, even one who perhaps can walk with a walker or cane, needs de facto consent to get anywhere: instructions on how to leave, someone to hand her the walker, whatever
Thread, continued, 20:34 UTC · 6 likes
So the elderly are both at higher risk of being de facto imprisoned in a medical institution and lower risk of hurting anyone else and so we should probably give them even stronger protections
Reply to @ronSzab9, 20:34 UTC · 6 likes
Áron Szabó (@ronSzab9): Someone (maybe Tetra) asked Claude recently about the origin of the Shoggoth ai meme, and the reply said the creators name was "Megan Tetraspace". So Tetra did the only logical thing in response.
Ah
Reply to @billyhumblebrag, 20:44 UTC · 2 likes
billy (@billyhumblebrag): Grabbing all my old photos onto a hard drive....was going to organise them a bit but I figure in like 6 months I'll be able to point an opus level model at it and say "organise this, pick all the best photos" and spend like 1% of my $20 subscription allowance.
Grok-4.1-fast-thinking is cheap enough and ok enough to enable some use cases that aren’t otherwise worth it (and batch gets you down another 50%)
Still have to set up the infrastructure/system tho
Reply to @StefanFSchubert, 21:09 UTC · 18 likes
Stefan Schubert (@StefanFSchubert): Is intelligence more like a tower (which you can keep making taller far beyond the human level) or like a ball (which you can only make so much rounder)?
(See next tweet for context.)
The tower vs ball thing is more a property of the domain in which intelligence is being applied than intelligence itself
Solving integrals -> tower
Tic tac toe -> ball
Also I’m not sure perfect roundness is in fact best considered a limit of some sort?
Reply to @StatisticUrban, 21:25 UTC · 23 likes
Hunter📈🌈📊 (@StatisticUrban): Only a few specific cities have been granted an exemption from DCA's "perimeter rule." The rule forbids any non-stop flight more than 1250mi away from DCA, instead directing them to IAD. The goal is to "reduce congestion."
IAD is 30 miles from the US Capitol, DCA is only 5.
It’s insane to have specific route monopolies written into law like this (nobody can compete on DCA - SFO???)
Ik this is already a thing for very narrowly defined routes but realistically a 1pm and a 2:30pm route between the same airports are largely competitors
Quote tweet, 21:27 UTC · 27 likes
There’s gotta be a paper to be written with the 1250 mile line break as a discontinuity “natural” experiment
Quoting Hunter📈🌈📊 (@StatisticUrban): Only a few specific cities have been granted an exemption from DCA's "perimeter rule." The rule forbids any non-stop flight more than 1250mi away from DCA, instead directing them to IAD. The goal is to "reduce congestion."
IAD is 30 miles from the US Capitol, DCA is only 5.
Monday, March 30, 2026
Quote tweet, 02:21 UTC · 64 likes
Some people really do just be smarter than you exhibit 58595929245
Quoting Aashish Reddy (@_AashishReddy):
Reply to @doubtpointv2, 02:27 UTC · 4 likes
🦫🇺🇾 (@doubtpointv2):
https://pmc.ncbi.nlm.nih.gov/articles/PMC5789215/pdf/DialoguesClinNeurosci-19-381.pdf
Reply to @NicholasD91704, 02:34 UTC · 3 likes
(The tweet being replied to by @NicholasD91704 is no longer available.)
Post on main
Quote tweet, 02:53 UTC · 63 likes
People really do not care about the suffering of children very much (except insofar as it affects them into adulthood)
Quoting Nina (@NinaPanickssery): why are people so nonchalant about the frequency of daycare illnesses? colds don’t “build immunity”! it’s just pure harm and suffering. people should be more sad about this.
Quote tweet, 03:00 UTC · 11 likes
Erdos may have been a mathematical genius but he was evidently innumerate in one sense
One month of genius productivity for <$2.5k in today’s dollars? Not worth it, shoot at least an OOM higher, Paul!
Quoting Crémieux (@cremieuxrecueil): Paul Erdős won a bet by proving he wasn't addicted to amphetamines.
But he got nothing done: "You've showed me I'm not an addict... You've set mathematics back a month."
Reply to @NinaPanickssery, 03:01 UTC · 8 likes
Nina (@NinaPanickssery): yeah it’s sad
that’s where you can help most as a parent as well
Reinventing EA from first principles
Tweet, 03:33 UTC · 24 likes
It sounds like a vacuous platform and it kinda is but “make it to February 2029 with neither an intelligence explosion/superintelligence yet underway nor a psychopathic presidential administration” seems like an appropriately-difficult but doable thing to aim and optimize for
Quote tweet, 09:40 UTC · 219 likes
Forgive my lack of imagination but what is DoE gonna do with more compute than God
Quoting Samuel Hammond 🦉 (@hamandcheese): I remember when this was announced but didn't fully appreciate the size. That's a hell of a cluster. The Department of Energy will basically be a frontier AI company.
Reply to @Somakethefriend, 16:28 UTC · 5 likes
Sara makes the friendship bracelets (@Somakethefriend): My best bet:
image attached
(quoting @10thDocWhooves: PLS I WANNA KNOW IF OLIVIA SECRETLY ATTENDED ERAS 😭)
🕶️
Quote tweet, 18:51 UTC · 9 likes
Hate to pick on this one post but I’m against this genre of reasoning
Like yeah no shit your big sign/t-shirt implies there’s a reason people are asking questions about the thing - the answer is still plausibly exactly and entirely what the t-shirt/ad/whatever reads
Quoting shreya (@shreyabasu003): Our “we operate under US law” ad is raising a lot of questions already answered by the ad
Reply to @TheStalwart, 21:54 UTC · 4 likes
Joe Weisenthal (@TheStalwart): Something extremely funny to me about this poll showing that 38% of Americans say the pace of AI development is happening roughly in line with their estimates. https://poll.qu.edu/poll-release?releaseid=3955
Still, +43 towards faster
Reply to @booritney, 22:04 UTC · 6 likes
brit (@booritney): the http://cuties.app dropdown menu for community affiliations is an excellent litmus test for how chronically online you are
People should pitch their microcommunity (or I guess don’t if you don’t want it to be legible/bigger)
I’m familiar with 6-8 of all these depending on how you count and I’m pretty online
Reply, 22:25 UTC · 1 like
Thousands and thousands of hours and at least tens of millions of dollars have been spent trying to identify compelling reasons substantial AI risk is wrong
We would know by now if the core doomer arguments were weak (say for the claim p(very important bad outcome) < 5%); they’re not
Reply to his own earlier tweet, 23:28 UTC · 4 likes
Aaron Bergman, March 30, 2026: Forgive my lack of imagination but what is DoE gonna do with more compute than God
Idk anything about anything but not everything that requires compute scales to *this level* of compute
We know of exactly one thing that scales and justifies the cost, LLM pretraining, right?
Reply to @LarryPanozzo, 23:30 UTC · 1 like
Larry Panozzo (@LarryPanozzo): @AaronBergman18 Yes they’re interested in fusion and the climate and whatnot, but to me it 100% feels like an excuse to buy a bunch of compute that they can “strategically repurpose” later
This makes the most sense so far lol
Seems fine/good
I guess we can have some really excellent natural gas extraction simulations or something while the folks at DoW x NSA dream up how to utilize the whole cluster
Tuesday, March 31, 2026
Quote tweet, 00:20 UTC · 50 likes
“Patient relies on a drug therefore we must make it harder for them to access” is another bit of FDA logic that would correctly parse as insane and backwards to someone without a lot of context hearing it for the first time
Quoting QualiaNerd (@QualiaNerd): The FDA pondering whether to approve my drugs: "The upside is that the patient is no longer depressed. The downside is that the patient is happy."
A tough decision…
Reply to @deanwball, 01:48 UTC · 16 likes
Dean W. Ball (@deanwball): oh man my answer to this would be such an info hazard that I am not sure I’d even speak of it in a scif
You’re a smart guy I feel like you can probably come up with something that satisfies both constraints (not an infohazard, actually works)
Maybe Yud will grant a version that’s like “only has to have a 90% chance of working for a thousand years conditional on the policy actually being implemented, not literally forever”
Reply to @deanwball, 01:53 UTC · 16 likes
Dean W. Ball (@deanwball): intelligence 👏 is 👏 not 👏 not 👏 the 👏 bottleneck 👏 to 👏 “solving” 👏 a 👏 great 👏 many 👏 matters 👏 of 👏 public 👏 interest
Ok fine maybe the “you’re a smart guy” clause is irrelevant
Reply to @mealreplacer, 01:57 UTC · 2 likes
Julian (@mealreplacer): I wrote a blog post on AI safety grantmaking. Link below!
Welcome back to Twitter
Reply to @business, 02:39 UTC · 11 likes
Bloomberg (@business): E-cigarettes are likely to cause cancer, including in the lungs and mouth, according to a sweeping review of scientific evidence that challenges their positioning as a safer alternative to smoking. https://www.bloomberg.com/news/articles/2026-03-30/review-challenges-claims-that-e-cigarettes-are-safer-than-smoking?taid=69ca83f7ce3e6b00011e2b35&utm_campaign=trueanthem&utm_content=business&utm_medium=social&utm_source=twitter
What’s the relative effect size
WHATS THE RELATIVE EFFECT SIZE
Reply to @AndyMasley, 02:40 UTC · 26 likes
Andy Masley (@AndyMasley): Okay folks, this study is being shared by prominent people already, and I am pretty confident I found a completely over the top goofy numeral issue that will destroy the whole argument pretty quickly involving... 2 orders of magnitude. Stay tuned.
He can’t keep getting away with this
Reply to @SpencrGreenberg, 05:15 UTC · 2 likes
Spencer Greenberg 🔍 (@SpencrGreenberg): A question for you: what's the most interesting topic in the world for you (to talk about or discuss)?
Consciousness, I think, followed by a tie between meta-ethics and ethics
Quote tweet, 05:20 UTC · 14 likes
I hope future archeologists find this and assume it was scripture
(video, plays on X)
Quoting Aaron Bergman 🔍 (@AaronBergman18): I gave myself a birthday present
It's the most egregiously self-aggrandizing thing I've ever done but I turned my most liked tweets into a 470 page book
Who wants a copy (mostly a joke but if u give me $31 I will in fact make sure it gets to you)
Wraparound cover:
Reply to @zagrebbi, 11:07 UTC · 5 likes
Werner Zagrebbi (@zagrebbi): Even at a 1600 SAT, Columbia only admits ~1 in 5 applicants. Apply early, kids!
I have a blog post about this! Yes, perfect academics does not get you in to the Ivies https://www.aaronbergman.net/p/most-ivy-smart-students-arent-at
Reply to @AndyMasley, 11:14 UTC · 9 likes
Andy Masley (@AndyMasley): 1.5k likes on this completely 100% fake story CNN's reporting on. The paper hasn't even been peer reviewed yet.
(quoting @CNN: The vast data centers that power artificial intelligence guzzle huge amounts of energy but they also have another alarming impact, according to new research. They are creating “heat islands,” warming the land around them by up to 16 degrees Fahrenheit, and making life hotter for…)
Nooooo Andy don’t fall for the peer review madness
That part is fine actually
Reply to @GreatKingCnut, 11:15 UTC · 2 likes
ɢʀɛǟȶK̶i̶n̶g̶ӄռʊȶ🤣```𝚙𝚢 (@GreatKingCnut): it's the Paul Christiano cluster, this is why he's gone underground in the gov for 2 years
I hope so
Reply to @mattparlmer, 11:28 UTC · 7 likes
mattparlmer 🪐 🌷 (@mattparlmer): This is absolutely revolting
(quoting @Telegraph: 🇮🇱 Israel has approved the death penalty for Palestinians but not for Israelis who commit the same crime. Sixty-two lawmakers, including Prime Minister Benjamin Netanyahu, voted in favour of the controversial bill which has been condemned by Britain ⬇️…)
Also it’s like 90 days to execution IIRC
Famously nobody has ever been falsely convicted of a capital crime before so that shouldn’t be an issue
Reply to @NathanpmYoung, 11:41 UTC · 5 likes
Nathan 🔎 (@NathanpmYoung): If I were gonna make the optimist case:
- AI companies want to make profit, not die
- things that follow commands will probably make more profit
- current and near AI seems very unlikely to kill us, we can work on aligning future AI
- problems are solvable, humans are apocalyptic(quoting @gnostrils: I watched The AI Doc. Could be partly an editorial thing on the part of the filmmakers, but what really stands out to me is how the optimists *never* have [valid] arguments or counterarguments. It’s just “don’t listen to the doomers”)
There’s no general rule that problems are solvable
Some problems are not solvable
Reply to @Babygravy9, 11:59 UTC · 4 likes
RAW EGG NATIONALIST (@Babygravy9): Gut bacteria also drive metabolic changes that result in obesity. I firmly believe obesity is also “infectious,” the implication being—stay away from fat people.
(quoting @NTFabiano: Gut bacteria was transferred from depressed humans to rats. The rats developed features of depression.)
Come on dude lol there’s a whole industry around fecal transplants because of how *hard* it is to swap gut bacteria even with intention
Fat is not infectious
Reply to @sciencegirl, 13:33 UTC · 177 likes
Science girl (@sciencegirl): Capybaras have long legs
and they kept it a secret
This must be how it fits
Quote tweet, 13:36 UTC · 58 likes
2% of the US showing up to protest is more newsworthy than how it’s been treated
Quoting Politics & Poll Tracker 📡 (@PollTracker2024): Yahoo News: More than 8 million people turned out at over 3,300 “No Kings” protests across all 50 states on Saturday, organizers said, calling it the largest single-day demonstration in U.S. history.
Tweet, 14:07 UTC · 23 likes
Why doesn't Will buy shrimp welfare arguments?
Thread, continued, 14:07 UTC · 2 likes
Transcript
https://docs.google.com/document/d/1eUOlirxATtgzZ359jZd-tKoQw1eeLYgdNmrP0MZ-Z1E/edit?tab=t.0
Thread, continued, 14:07 UTC · 2 likes
Pod https://www.youtube.com/watch?v=5JF0gOQvskY
Reply to @wifiejwn38, 15:10 UTC · 4 likes
IntellectualDemon (@wifiejwn38): I'm not a utilitarian. I don't believe you can actually calculate suffering like that. I also don't think Graham's number of people getting dust in their eye is worse than torturing a person for 50 years.
Interestingly I consider myself a utilitarian but share your take on the torture/dust matter, and have I thought about it a lot lol https://forum.effectivealtruism.org/posts/je5TiYESSv53tWHC9/utilitarians-should-accept-that-some-suffering-cannot-be-1
Tweet, 15:25 UTC · 11 likes
If we make it to 2029 the US should probably spend at least hundreds of billions of dollars righting the wrongs (or trying) perpetrated on just so many people around the world
Thread, continued, 15:25 UTC · 5 likes
- afghan translators etc left to die
- immigrants (yes some illegal) sent to Salvadoran concentration camps
- the families of executed Irani schoolchildren
- anyone we abruptly and without warning cut off aid from
Thread, continued, 15:25 UTC · 4 likes
- random Venezuelan fisherman (maybe also carrying drugs but since when is moving cocaine around a capital and international offense)
- everyone we cut off from legitimate paths to legal status
- this is just stuff from memory, I’m definitely forgetting large categories
Thread, continued, 15:25 UTC · 3 likes
Clearly eventually there’s gonna be a decision about whether to spend another million dollars tracking down the marginal victim of Trump’s shitstorm but I’m pretty sure the ideal number of dollars (largely also as a proxy for real resources) Is >>0
Thread, continued, 15:25 UTC · 3 likes
Maybe this is actually cheaper than I intuit and it could be done for tens of billions of dollars or less, all the better
Purely from a strategic perspective it’s probably money well spent
Reply to @GWHayduke97, 15:34 UTC · 3 likes
Hayduke ⏹️ (@GWHayduke97): Correct, I'm not. EAs don't have a monopoly on this sort of discourse.
Lol I will attest to this, you’re allowed to say EA is wrong about stuff
Tweet, 16:47 UTC · 21 likes
Wait so I’m gathering that a set of incredibly influential papers (Chamley 1986 and Judd 1985) gave policy wonks the takeaway that the optimal capital gains tax rate is zero
but this relies crucially on an assumption that people consume their wealth before dying
Thread, continued, 16:47 UTC · 7 likes
Which just isn’t what we see irl (ie people who accumulate $100B gross wrt personal spending over a lifetime don’t in fact spend $100B or anything close to it)
Thread, continued, 16:47 UTC · 11 likes
So an interesting technical result has become neolib (positive, usually) gospel on extremely tenuous foundations
Unironically do I have this right?
Thread, continued, 16:47 UTC · 8 likes
Cc @NathanB60857242 @captgouda24 might know
Reply to @captgouda24, 16:59 UTC · 3 likes
Nicholas Decker (@captgouda24): It’s still approximately correct, because we don’t want to tax investment. Otherwise you tax a business process multiple times. Anyway, may be of interest.
I’m not sure I see why multiple taxation per se is bad
Like maybe you have a tax on paying your employees (i want to say this is fine for now but maybe you’d object) but then the employees go and make you a billion dollars; I don’t see why we shouldn’t tax the billion dollars (like we *could* try to transfer this onto the payroll tax but the whole idea is that you tax the winners specifically, in fact if anything we should take the payroll tax and try to move that revenue onto a capital gains tax)
Tweet, 17:00 UTC · 4 likes
Golden sentence
End of March 2026. Previous month: February 2026. Next month: April 2026. Back to the archive index.