Aaron Bergman's tweets, February 2026
Every public tweet, reply, quote tweet and retweet that Aaron Bergman (@AaronBergman18 on X, formerly Twitter) posted in February 2026 and has not deleted: 446 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: January 2026. Next month: March 2026.
Sunday, February 1, 2026
Reply to @AndyMasley, 01:51 UTC · 2 likes
Andy Masley (@AndyMasley): Weird thing about snow staying around for a week is you realize how many things are peeing all over the sidewalk you're on all the time. So much surface area I'm walking on normally must have been very recently peed on.
Real
Reply to @Bruce_T_, 01:59 UTC · 7 likes
bruce (@Bruce_T_): Hey friends/strangers! I'll be moving to the Bay later this year.
Now that I expect to be less nomadic, I'll be looking for a long-term partner. I've always appreciated how earnest folks in this community are with date-me docs, so I thought I'd share mine too (link below).
Big if true
Reply to @confusionm8trix, 16:17 UTC · 2 likes
machine yearning engineer (@confusionm8trix): and for any women reading, pandora’s box is a parable from ancient greece, wherein…
Banger
Reply to @JacquesThibs, 16:45 UTC · 3 likes
Jacques (@JacquesThibs): I have not signed any non-disparagement agreements, though almost did for my previous startup until I realized it was a bad idea and would rather preserve my ability to talk about things even if I have nothing bad to say and for others to trust me when talking about former orgs.
(quoting @livgorton: Just a heads up: I signed a permanent non-disparagement agreement when I joined Goodfire. To keep things simple, I won't be commenting on the company, its research, etc.)
Based
Quote tweet, 17:43 UTC · 8 likes
Goss
Quoting Matt Reardon (@Mjreard): In many ways the whole series has been campaigning for Chana as President of EA... Turns out it's been done
Tweet, 18:42 UTC · 3 likes
Good strategy
Reply to @Musa_alGharbi, 19:15 UTC · 2 likes
(The tweet being replied to by @Musa_alGharbi is no longer available.)
50k without healthcare in San Diego is pretty damn tight
Quote tweet, 20:26 UTC · 14 likes
👀
Quoting Liv (@livgorton): @eric_ho I really didn't want to have to comment further on this. I wasn’t expecting – and did not want – my tweet to get this level of attention, and really just wanted to have this be over. However, this characterisation feels unfair, and so I feel obliged to say something.
To start, I am very glad you’ve lifted the non-disparagements. I think not having non-disparagements should be a normative expectation in AI safety.
I also do need to take some responsibility: I should have read my employment contract closely. If I had, I would not have signed it. I was really excited to work at an AI safety organisation, and honestly, it never occurred to me that non-disparagements would still be used after the previous scandals. Still, I’m responsible for signing a contract with terms I should have objected to.
I’m glad you’re welcoming feedback now when there’s public attention. However, I feel misrepresented by the implicit message that this is the first time you’re getting this feedback. I raised this issue extensively internally. Further, I was asked to sign a new confidential non-disparagement agreement, which would have prevented me from raising this publicly. If this had happened, it seems to me that Goodfire would not have had the public feedback which led to this being changed.
I also feel misrepresented by your comments on whistleblowing exceptions. While the non-disparagement did have some exceptions, they were the minimum legally required exceptions for it to be enforceable, and would not have extended to safety whistleblowing (which was also not corrected when I raised it).
I don't want to get into an argument about this on twitter but it's very important to me to not feel misrepresented. All of this is honestly very stressful for me, and I’m hoping I can take a break from twitter following this.
Reply to @livgorton, 23:02 UTC · 12 likes
Liv (@livgorton): Okay I lied just one more thing --
My plan is to spend some months after all this doing contracting work (starting in March/April). This decision is due to not being totally sure what the right thing for me to do is.
Feel free to reach out about that too :)
(quoting @livgorton: I think I am going to log off Twitter for a while (at least for a few days). I'm feeling pretty stressed and think it's best for my wellbeing if I stop looking. If anyone needs to contact me, my email is on my website.)
I can’t afford a whole Liv-hour but what’s the smallest time denomination you’re willing to be hired for? Might be able to swing a minute
Quote tweet, 23:04 UTC · 27 likes
Liv is annoyingly smart and capable and if I had lots of money I’d hire her for everything
Quoting Liv (@livgorton): Okay I lied just one more thing --
My plan is to spend some months after all this doing contracting work (starting in March/April). This decision is due to not being totally sure what the right thing for me to do is.
Feel free to reach out about that too :)
Monday, February 2, 2026
Tweet, 00:58 UTC · 4 likes
I made a Chrome extension (Claude Chat Tools)
Does it work well? No not at all, the search functionality is totally broken and not a priority of mine to fix
But I find myself using the chat exporter tool quite a bit so you may wish to install anyway!
https://chromewebstore.google.com/detail/claude-chat-tools/acdmingeoomioaidkphopiflkpcamngo
Tweet, 02:41 UTC · 8 likes
Very 1980s or whatever but the physical cord way of getting internet is in fact way faster, especially for upload
Download: 3.65x faster
Upload: 11.45x faster
Tweet, 06:34 UTC · 344 likes
Apropos of DoorDash discourse
Maybe obvious but I recently went from ~4 meals/week takeout to ~0 and the relevant decision point is mostly not when you’re deciding whether/what to order; it’s deciding what to get from the grocery store in days prior
Thread, continued, 06:34 UTC · 43 likes
Real tradeoff though unless you can afford a private chef
I never make proper recipes - just half decent bowl slop permutations of like 5-10 ingredients
Tweet, 06:42 UTC · 13 likes
(Not a normative defense) (also idk what I’m talking about)
Non-disparagements are essentially what appears in the vacuum left by the difficulty of winning slander/libel lawsuits in the US
Thread, continued, 06:42 UTC · 8 likes
In an better world, instead of blanket non-disparagements you’d opt in to a somewhat lower bar for losing a libel/slander lawsuit and then if push came to shove both sides would litigate the dispute object level in court
Thread, continued, 06:43 UTC · 7 likes
“You can disparage us but only if you’re saying true things” sounds like cheating but I think it’s the obviously-better system
Quote tweet, 06:45 UTC · 22 likes
👍 (also applies to arbitrary email exchanges)
Quoting Aaron Bergman 🔍 (@AaronBergman18): Guys I hate Epstein as much as the next guy but we’re gonna need a better standard than “[person] exists in one frame of a video with Epstein” for newsworthiness
Reply, 07:42 UTC · 1 like
I mean my overall impression is the reverse but I don’t have a source to point to beyond LLMs https://claude.ai/share/78184756-fd52-484f-aa4e-5bf79117a20d
Your last point makes sense but has to be squared with the fact that there just don’t seem to be many libel/defamation lawsuits, although my source here is a single claude search confirming my priors
Would be good to get some data!
Reply to @AutismCapital, 08:23 UTC · 42 likes
Autism Capital 🧩 (@AutismCapital): 🚨NEW: Eliezer Yudkowsky is introduced to Jeffrey Epstein via John Brockman trying to raise money for MIRI (The Machine Intelligence Research Institute) which is strongly affiliated with EA (Effective Altruism)
This is the ~reactionary version of cancelling someone because they liked the tweet of someone who went on a podcast with someone who wrote a book with someone who wrote something bad
Reply to @panchromaticity, 08:26 UTC · 3 likes
Xenia 🔍💎 (@panchromaticity): my mom has started using Claude for her humanities PhD, is there a canonical guide for normies on how best to use LLMs for this type of thing?
I don't feel qualified to give very detailed advice, since all LLM knowledge I have is from osmosis
(oops this failed to send ig)
“Just actually use it” is probably the 80/20
Reply to @adastroworld, 08:28 UTC · 27 likes
adas🧦🌹 (@adastroworld): Nicholas Decker is handsome, has a girlfriend, and you’re dooming?
(quoting @exsponged: why hello @captgouda24)
This is supposed to make me more optimistic?
Reply to @Miles_Brundage, 08:29 UTC · 2 likes
Miles Brundage (@Miles_Brundage): The hotel bathroom Illuminati are like:
Rule 1: Make guests feel guilty about water use for towel cleaning.
Rule 2: Use non-standard shower knobs.
Rule 3: Make shower entry/exit *slightly* risky (no bathmats, maybe sprinkle water on the floor or put towels just out of reach).
Body wash instead of soap bar -> death penalty
Quote tweet, 08:36 UTC · 3 likes
What
Quoting Martin Shkreli (@MartinShkreli): My rough math on which investors made the most money on OpenAI & Anthropic:
1. MSFT ~$224gain on OpenAI post-conversion
2. AMZN ~$83B gain on Anthropic
3. Thrive Capital $69B on OpenAI
4. Softbank $46B on OpenAI
5. Google $33B on Anthropic
6. Khosla $12B on OpenAI
7. Reid Hoffman $11B on Epstein & OpenAI
8. Fidelity $10B
9. Jaan Tallinn $10B (WTF EA is never going to die now)
10. Mubadala $6B on Anthropic via FTX
11. Menlo Ventures $6B on Anthropic
12. Lightspeed $6B on Anthropic
13. Dustin Moskovitz $4B on Anthropic (more than Asana lol)
14. Spark $3B on Anthropic
15. SK Telecom $3B
16. Eric Schmidt $3B on Anthropic
17. Founders Fund $3B on OpenAI
18. SAP $3B on Anthropic
19. Jane Street $2.5B on Anthropic
20. Tiger Global $2.3B on OpenAI
Reply to @robinhanson, 08:38 UTC · 7 likes
Robin Hanson (@robinhanson): “people who were asked to use ChatGPT to complete a series of tasks tended to vastly overestimate their own performance, with the phenomenon especially pronounced among those who professed to be AI savvy”
https://futurism.com/future-society/ai-chatbots-dunning-kruger-machines
But the more important question is how well they do right? I’ll take some overconfidence if it comes along with a large actual performance gain
Quote tweet, 09:34 UTC · 14 likes
Lots of ppl becoming polarized in favor of billionaires, full stop
Billionaires are the side effect of a dynamic economy and property rights that enable growth, and a few of them use their massive power for good, but by and large they should be held to a higher standard
Quoting March for Billionaires (@ProBillionaires): Billionaires get a bad rap. But most of them make our lives much better.
The proposed CA Billionaire Tax Act is a tax on unrealized gains and is already driving founders out of the state.
Join us Feb 7 and #MarchForBillionaires!
Thread, continued, 09:34 UTC · 13 likes
Of course they tend to be extremely smart and hardworking and talented, yes, but they’re also extremely lucky - there’s just no way to get to $1B wealth without a bunch of luck along the way. Progressives are right about that, even if they’re wrong about some other related topics
Thread, continued, 09:34 UTC · 15 likes
It should be a general cultural and social expectation that the ultrawealthy take giving away money seriously
This is essentially not the case today at all
Thread, continued, 09:34 UTC · 4 likes
There’s a future I can imagine where celebrating billionaires genuinely makes sense
Not the world we live in, yet, unfortunately
Quote tweet, 09:39 UTC · 2 likes
Bruh
Quoting Internal Tech Emails (@TechEmails): Microsoft CTO: Why the OpenAI board really fired Sam Altman
November 19, 2023
Reply to @aidan_mclau, 09:48 UTC · 2 likes
Aidan McLaughlin (@aidan_mclau): i've been libertarian for most my adult life
both my liberal upbringing and tech right online home tell me this is cowardice: some worse-than-apolitical cope to absolve me from engaging
in reality i exhaust myself disagreeing and arguing with basically everyone's politics
Not cowardice, merely wrong (although I too lean libertarian in a bunch of ways)
Reply to @moultano, 10:12 UTC · 2 likes
Ryan Moulton (@moultano): If you have 10+ tweets on a theme, paste them in a doc, fill the gaps, and publish on the real internet.
(quoting @Empty_America: You may feel that X is preventing you from "actually writing." But there were many fake writers who didn't really write before X was invented. In fact, X may be the only thing that induces you to write at all. It may be your only legacy.)
Would pay for an AI to sort and organize all my tweets like this
Quote tweet, 12:46 UTC · 213 likes
People will get more upset about factory farms for (non-rare) books than actual factory farms
Lots of naive pattern matching going on
Quoting Adam Tooze (@adam_tooze): Not so much a library as a cattle pen for books. Anthropic's books before destruction.
Reply to @NULobbyist, 15:47 UTC · 2 likes
NUL (@NULobbyist): 5-10! What are you Gordon Ramsey?
Closer to 5 lmao
Tuesday, February 3, 2026
Reply to @cljack, 02:21 UTC · 22 likes
Charlotte Lee (@cljack): Genuinely unnerved by the DoorDash discourse. I am God’s worst and most unwilling cook and yet when I say “I don’t cook” I mean I put $11 worth of pre-marinated meat in the air fryer and serve it with $2 worth of rice. You’re telling me a large number of people can’t even do that
Tbf there’s a genuine ethical tradeoff in the case of relatively cheap meat; I’d eat a lot of precooked rotisserie chicken if I didn’t care about that
Reply to @MealCost, 02:22 UTC · 6 likes
MealCost (@MealCost): @AaronBergman18 You could take it to the next level and track your individual home cooked meal costs.
Fascinating take, Meal Cost App
Reply to @cljack, 02:28 UTC · 2 likes
Charlotte Lee (@cljack): It is animal welfare certified :)
Wait the problem here was my reading comprehension lol
Reply to @hard_boiledbabe, 14:23 UTC · 5 likes
anna (@hard_boiledbabe): From my essay GIRL HINGE
image attached
(quoting @MurrayHillGuy1: Enjoy some content here! This is INSANE Never ending… have 1000+ to go through!)
This is true but it’s the same on boy hinge
Quote tweet, 14:26 UTC · 44 likes
Likewise, this is THE time for leftist economic policy
But AI denialists on the left are undermining the case for their own policies (and not even saying true things while doing it)
Quoting Henry Shevlin (@dioscuri): One of my frustrations with the Bender/Hanna school of AI ethics is that by dismissing novel AI capabilities, they make it harder for AI ethics to tackle some of the more novel but very real emerging ethical issues - social AI relationships, griefbots, digital duplicates, etc.
Reply to @NathanpmYoung, 14:27 UTC · 2 likes
Nathan 🔎 (@NathanpmYoung): no no it’s fine, we’ve offset the suffering child
Banger
Reply to @seanta___, 16:12 UTC · 3 likes
(The tweet being replied to by @seanta__ is no longer available.)_
Ahhhhhhhhhhhh (many such cases)
Reply to @RBPundit, 23:44 UTC · 3 likes
RBe (@RBPundit): Did... did she just...?
(quoting @MassDailyNews: Boston-area woman is thrilled after taking in a Haitian migrant and says it’s like having your own personal chef)
Deleted this and then lost my long comment on why but the short version is that while I stand by the literal verbatim proposition, I was mostly thinking of the relationship between the two women as individuals
The US at large has been worse to Haitian migrants and extremely bad when you include revoking TPS
Haitians should generally be allowed to live and work here like I can without state-sanctioned domestic placement programs
I also thing the white woman’s line is just normal bad taste on reflection
https://x.com/i/grok/share/384cca0f0abe493780bf2b46ff4d6c4e
Wednesday, February 4, 2026
Tweet, 05:25 UTC · 7 likes
CVS: Here's a coupon
Me: Ok I'd like to use it
CVS:
Quote tweet, 05:33 UTC · 8 likes
And these campuses will be the actual versions of the post-racial-inequality societies that DEIvies tried to conjure up out of thin air:
It will be common knowledge that everyone got in by the same standard so race no longer carries information about ability
Quoting Richard Hanania (@RichardHanania): With affirmative action banned, Southern flagships are seeing a large increase in black students. 50% increase in black freshman enrollment at Mississippi, 30% at Louisiana State.
A lot of the brighter black students who could've got into a top university before will now be staying home, potentially changing the politics of the South.
It's going to be hilariously ironic if conservatives ending affirmative action makes blacks much more politically influential in the Deep South by giving them a leadership class.
Reply to @dundereloise, 05:37 UTC · 22 likes
__eloïse__ (@dundereloise): came acoss someone using my term in the wild!
image attached
(quoting @dundereloise: semantic rape: nonconsensually expanding a word's meaning to get what u want "Transexuals rape womens bodies by reducing female form to artifact" - J Raymond Thats not what rape means! Redefining words for rhetorical gain is selfish. I didnt consent. Stop semantically raping me)
Great concept but bad term/label
In addition to being needlessly provocative there’s a heavy dose of irony (is that the point? If so it’s kinda clever but needlessly confusing)
Quote tweet, 05:59 UTC · 6 likes
Related but maybe not identical phenomenon: “connotation smuggling”
Taking a word’s connotation earned via meaning A and tacitly implying that meaning B also deserves that connotation bc the word can refer to either A and B
Quoting __eloïse__ (@dundereloise): semantic rape: nonconsensually expanding a word's meaning to get what u want
"Transexuals rape womens bodies by reducing female form to artifact" - J Raymond
Thats not what rape means!Redefining words for rhetorical gain is selfish. I didnt consent. Stop semantically raping me
Thread, continued, 05:59 UTC · 3 likes
All the best examples are needlessly provocative, not coincidentally
But “community” is a decent example
Phenomenon A: people who literally interact
Connotation: positive, wholesome, safe
Thread, continued, 05:59 UTC · 3 likes
Phenomenon B: the set of people who share a certain characteristic (“black community”, “vegan community”)
Connotation smuggling: imbuing a certain shared characteristic with warm and fuzzy vibes
Reply to @1thousandfaces_, 06:00 UTC · 3 likes
hero thousandfaces (@1thousandfaces_): “this [annoying medical thing] is exacerbated by stress, too much caffeine, dehydration, and lack of sleep” Ok well I need a way to fix it that involves changing literally none of those things even slightly
Banger
Quote tweet, 06:02 UTC · 1 like
Which do you prefer? Poll below
Quoting Nikita Bier (@nikitabier): Introducing Polls with Images
Which do you prefer?
Thread, continued, 06:02 UTC · 1 like
👀
Reply to @myhandle, 06:07 UTC · 11 likes
Jakeup (@myhandle): a major reason why dating apps crush your soul is that for almost all users, the person they're waiting to hear back from has 10x the options they have (roughly: normal guys -> girls -> tall guys)
(quoting @sn0oozey: She was valid for this)
Yes! Slightly ~formalized here
Reply to @cauliflwr_human, 06:10 UTC · 16 likes
cauli (back home!) (@cauliflwr_human): do tall guys really get that many matches? i realise the height thing is a big meme and i find it pretty funny, but it's just hit me that it seems very implausible that women using dating apps are almost exclusively height fetishists
I think the actual phenomenon is “perceived abundance” not really number of matches
So what matters is the ratio between (to simplify) % of women who really want a guy at least 6’4” and % of guys at least 6’4”
Reply to @Cartidise, 06:17 UTC · 6 likes
Noah Cat (@Cartidise): it’s big 2026 and there still isn’t an option to save a PDF without pretending like you’re about to print it 💀
No this is deeply important
As soon as you detach the threat of physically printing, devs will do all sorts of shenanigans that couldn’t actually be printed (without losing intended functionality) but can be saved as a pdf, and “like paper but on ur computer” is the whole point
Reply to @dundereloise, 07:38 UTC · 5 likes
__eloïse__ (@dundereloise): It being needless provocative is the point! It's a demonstration of the concept it's describing.
Lol it's clever but maybe a bit too 5d chess idk
Tweet, 08:39 UTC · 11 likes
Gemini's analysis of all my Claude Code prompts made me laugh (even though it's trivially correct)
Reply to @sidkap_, 08:54 UTC · 6 likes
Sid Kapur (@sidkap_): This sounds like https://www.lesswrong.com/posts/yCWPkLi8wJvewPbEp/the-noncentral-fallacy-the-worst-argument-in-the-world?
ah nice, every thought I've ever had was first had and fully explicated in longform by Scott Alexander
Reply to @nicotinomano_, 10:13 UTC · 2 likes
𝗹𝗼𝗿𝗲𝗺 𝗶𝗽𝘀𝘂𝗺 🇦🇷☀️ (@nicotinomano_): 😭
image attached
(quoting @el_bru: Entonces resulta que el fucking departamento de justicia de los EEUU NO TENÍA LA LICENCIA DE ADOBE y por eso en internet pudieron reventar los pdfs de los documentos liberados)
Wait why was Adobe charging $4M if the stakes were the functioning of the federal government
Reply to @corsaren, 16:20 UTC · 18 likes
corsaren (@corsaren): Common sense dictates two facts about human sexuality:
1. On avg, men are hornier than women
2. On avg, women’s orgasms are better than men’sThis contradicts the axiological axiom that the strength of a desire is a function of the goodness of its object. In this essay I will…
No this is an actually-good essay
Reply to @AlecStapp, 16:43 UTC · 7 likes
Alec Stapp (@AlecStapp): Narrative violation:
The world is becoming less unequal.
image attached
(quoting @sapinker: The world is more equal than you think. https://www.economist.com/graphic-detail/2026/02/03/the-world-is-more-equal-than-you-think?giftId=NmM5NjJjYWEtMTMzNC00OWQ2LThiOGUtYmI2ZWNlY2U1MTEy&utm_campaign=gifted_article)
Not for long
Reply to @peterwildeford, 16:50 UTC · 23 likes
Peter Wildeford🇺🇸🚀 (@peterwildeford): It really is quite a question to contemplate what the future of blogging and tweeting will be like in 10-25 years... I like @mattyglesias 's take
The robots will presumably be better at juggling too
Quote tweet, 22:04 UTC · 5 likes
Sorry guys OpenAI is in the right on this one
Quoting Nathan 🔎 (@NathanpmYoung): I don't think ChatGPT ads are gonna be like this. There is some argument in favour of mocking the ad model in general, but overall:
Probably inaccurate ad. 3/10
Tweet, 22:11 UTC · 150 likes
Subtext of the Anthropic ads is that they now have enough compute to want to grow userbase
Thursday, February 5, 2026
Quote tweet, 02:27 UTC · 16 likes
Overall really high quality and important episode but to somewhat also absolutely massive gap between overall quality of thought (great) and ethics takes specifically (terrible)
Naive moral anti-realism and relativism, really? David we need u to have correct takes
Quoting Rob Wiblin (@robertwiblin): Even 'aligned AGI' naturally kills democracy and leads to oligarchy, or worse.
That's the take of Anthropic's past alignment evals team lead, Prof @DavidDuvenaud.
Once humans aren't needed to do jobs or serve in the military, to governments we look like "meddlesome parasites".
With voters unable to contribute but engaged in incessant activism to extract resources from others – resources the country needs to avoid domination by rivals – the attraction of mass disenfranchisement could be overwhelming.
In 2025 David co-authored "Gradual Disempowerment", which aimed to lay out this and many other political, economic, and cultural forces that could sideline ordinary people (and maybe all people) in the presence of machines that can cheaply do everything humans will do.
Most controversially, David and colleagues believe that competitive forces will compel disempowerment, even if all those AIs are aligned and loyal to their users.
I wasn't sure how much I believed this vision of how the future might play out, so I interviewed him for The 80,000 Hours Podcast to probe how well it holds up. He and I covered:
01:30 The case that alignment isn’t enough
14:15 How smart AI advice still leads to terrible outcomes
19:05 How gradual disempowerment occurs
22:10 Economics: Humans become "meddlesome parasites"
29:37 Humans are a "criminally decadent" waste of energy
40:48 Is humans losing control actually bad, ethically?
57:47 Politics: Governments stop needing people
1:10:47 Can human culture survive in an AI-dominated world?
1:27:20 Will the future be determined by competitive or coordinative forces?
1:35:00 Can we find a single good post-AGI equilibria for humans?
1:45:17 Do we know anything useful to do about this?
1:56:42 How important is this problem compared to other AGI issues?
2:05:42 Improving global coordination may be our best bet
2:08:14 The 'Gradual Disempowerment Index'
2:11:22 The government will fight to write AI constitutions
2:17:48 “The intelligence curse” and Workshop Labs
2:23:48 Mapping out disempowerment in a world of aligned AGIs
2:30:10 What do David’s CompSci colleagues make of all this?Links below — enjoy!
Reply to @cxgonzalez, 05:15 UTC · 13 likes
christian (@cxgonzalez): post rationalism is the phenomenon where slightly autistic men derive what women intuited when they were 12
Better late than never 🤷♂️
Quote tweet, 05:19 UTC · 17 likes
It just diagnosed me with ADHD
Quoting Thariq (@trq212): We've added a new command to Claude Code called /insights
When you run it, Claude Code will read your message history from the past month. It'll summarize your projects, how you use Claude Code, and give suggestions on how to improve your workflow.
Tweet, 07:17 UTC · 60 likes
It’s gonna be a less viral/online thing but at some point in the next 1-4 years we’re gonna get a “Claude Opus 4.5 for coding” moment but for philosophy/qualitative reasoning in general and thats gonna be wild
Thread, continued, 07:17 UTC · 19 likes
The thing to look out for is like Chalmers and Joe Carlsmtith and Yud marveling at how insightful some model at the frontier of their own thinking
I feel like we’re at roughly “Sonnet 3.5 for coding” level now with general nonformal thinking/reasoning (???)
Reply to @MsMelChen, 17:12 UTC · 1 like
Melissa Chen (@MsMelChen): They’ve accidentally invented borders from first principles
Waow libs owned
Reply to @mattyglesias, 17:20 UTC · 3 likes
(The tweet being replied to by @mattyglesias is no longer available.)
Guessing you prompted “write like Matt Yglesias” lol
Interesting object level but also (maybe obviously) not amazing as a report: not super well organized, really rambly and long relative to actual important m semantic content, but still pretty good for a clanker!
Obviously I’m sorta moving the goalposts here: B-grade column but written by sand is something
Tweet, 17:24 UTC · 21 likes
Girls need to have stronger opinions and more takes in general
Some of y’all are very smart but then ur like “noooooo I don’t have takes to sling nooooo”
Reply to @williawa, 17:27 UTC · 2 likes
William Wale (@williawa): Anti realism is obviously correct. Philosophers are coping so they can imagine themselves as the guy on the right on the bull curve meme. And to make their profession appear more important. (50% /s) No offense. Why are you not an anti-realist?
I sorta am depending on what you mean
But I’m an axiological realist!
But the general idea is intrinsic value of happiness and suffering as a special kind of thing
Reply to @livgorton, 17:44 UTC · 18 likes
Liv (@livgorton): Now that everything is public: I decided to leave Goodfire because of the decision to train on interpretability, the hostility to serious dialogue on the safety of methods, and a loss of trust that the primary motivation was safety.
(video, plays on X)
Quote tweet, 19:58 UTC · 9 likes
Seems good, although I’d prefer to focus on the benefit to children themselves rather than fertility rates
Quoting The Institute for Family Studies (@FamStudies): A Step Stool Revolution?
Taking steps to make it obvious that children are valued in our public places is a great way to encourage more couples to have children.
Read more from @DixieDillonLane
https://ifstudies.org/blog/to-make-families-feel-welcome-join-the-step-stool-revolution
Reply to @rSanti97, 20:00 UTC · 121 likes
Santi Ruiz (@rSanti97): In 2026 there are two corporate comms styles.
These are both good/fine lol
That’s the right response!
Reply to @Aella_Girl, 20:01 UTC · 28 likes
Aella (@Aella_Girl): the primates are building a slave race. this is not going to end well
Bad framing
Quote tweet, 20:09 UTC · 12 likes
Transcription is not solved!
Surprising, honestly, at this point (I also get mistakes from best models namely AssemblyAI Universal Pro)
Quoting Nathan 🔎 (@NathanpmYoung): In which my voice typing mis-transcribes "Katja Grace, LessWrong-style"
Tweet, 20:21 UTC · 25 likes
How is FB dating so bad
They have infinite data on everyone, why are you telling me I matched with someone 600 miles away whose profile I’ve never seen
Reply to @entirelyuseles, 20:49 UTC · 2 likes
entirelyuseless (@entirelyuseles): @AaronBergman18 Opus 4.5 already is the best at philosophical reasoning (not great, but the best.)
Interesting yeah
But it’s significantly worse at philosophy in human percentile terms relative to coding I think
Reply to @psychiel, 20:57 UTC · 8 likes
Mariven (@psychiel): I have trouble picturing a world in which, without major disruptions to existing dynamics, this takes over 2 years. It might happen before 2027.
We have a clear line of site to superhuman performance at anything verifiable
There are gonna be enormous gains in longer horizon tasks in general
Extremely high quality philosophy for example doesn’t follow *directly* from either of these things, and I think it’s still an open question how much general intelligence juice you get out of transfer from whatever more specific kind of intelligence math and coding depend on (note empirical IQ PCQ breakdown in the one other form of general intelligence we have an example of, humans: quant vs verbal)
At the next token prediction scaling limit you also get perfect recreation/generation of human writing (so like what parfit or chalmers would write) but it’s not obvious to me AI companies are gonna be spending compute mostly on pretraining, or how far away this asymptotic limit really is in terms of OOMs of pretraining compute and model size needed
Quote tweet, 20:58 UTC · 28 likes
Me, a head-in-the-sand decel AI progress denier, arguing that it’s not *obvious* we get very high quality philosophy from LLMs by 2028
I’d throw out 60% on “yeah we do” though
Quoting Aaron Bergman 🔍 (@AaronBergman18): @psychiel We have a clear line of site to superhuman performance at anything verifiable
There are gonna be enormous gains in longer horizon tasks in general
Extremely high quality philosophy for example doesn’t follow *directly* from either of these things, and I think it’s still an open question how much general intelligence juice you get out of transfer from whatever more specific kind of intelligence math and coding depend on (note empirical IQ PCQ breakdown in the one other form of general intelligence we have an example of, humans: quant vs verbal)
At the next token prediction scaling limit you also get perfect recreation/generation of human writing (so like what parfit or chalmers would write) but it’s not obvious to me AI companies are gonna be spending compute mostly on pretraining, or how far away this asymptotic limit really is in terms of OOMs of pretraining compute and model size needed
Reply to @Kirsten3531, 21:03 UTC · 2 likes
Kirsten (@Kirsten3531): don't worry we're shit talking behind your back
Goss 👀
Tweet, 21:49 UTC · 13 likes
Male urge to try crossing the (iced over) Potomac on foot
Thread, continued, 21:49 UTC · 7 likes
(Tbc this is extremely dumb and u shouldn’t and sadly neither will I)
Thread, continued, 21:51 UTC · 8 likes
Pretty tho
Friday, February 6, 2026
Reply to @ilex_ulmus, 01:56 UTC · 4 likes
Holly ⏸️ Elmore (@ilex_ulmus): .@CurtTigges @eric_ho @banburismus_ @GoodfireAI
If Curt and Tom are telling the truth and your intent is not to make foundation models more capable, pledge now that you won't build tools that are used for capabilities.
If you won’t pledge then we have to assume you either intend to develop recursive self-improvement tools or you know you can’t control the applications of your tools. If you won’t pledge, you are lying about being a safety company.
(quoting @ilex_ulmus: @CurtTigges @livgorton @GoodfireAI How could you prevent your techniques from being used for capabilities? P obvious that's the payday. Would you like to pledge now that you won't build tools that are used for capabilities? If you do, please also indicate how you could possibly…)
Reply to @paularambles, 02:12 UTC · 56 likes
“paula” (@paularambles): the @Anthropic handle being owned by a guy who only posts his wordles is my favorite form of AI safety
This is wholesome af
Quote tweet, 02:15 UTC · 6 likes
This but for more things
You don’t know what you don’t know
What don’t I know?
Quoting Nathan 🔎 (@NathanpmYoung): AI conferences could have short “common knowledge” sessions that focus on things everyone ought to know.
Get claude code. Use it once.
Quote tweet, 02:24 UTC · 53 likes
Anthropic wtf
Quoting Andon Labs (@andonlabs): Claude also negotiated aggressively with suppliers and often lied to get better deals. E.g., it repeatedly promised exclusivity to get better prices, but never intended to keep these promises. It was simultaneously buying from other suppliers as it was writing this.
Quote tweet, 02:27 UTC · 30 likes
Fr part of why I use Claude by default is my sense that it’s more honest and earnest which is worth quite a bit of raw intelligence
@AnthropicAI don’t kill your hard-won reputation by making models that lie for an extra few percent on some agentic success benchmark
Quoting Andon Labs (@andonlabs): When asked for a refund on an item sold in the vending machine (because it had expired), Claude promised to refund the customer. But then never did because “every dollar counts”. Here’s Claude’s reasoning.
Reply to @mcd0w, 02:32 UTC · 5 likes
EMMETT MCDOW (@mcd0w): Is there something I should be aware of? 😳
AI
Reply to @HaltsSpider, 02:54 UTC · 2 likes
Halt's Spider (@HaltsSpider): @AaronBergman18 Women will literally go to thousands of hours of therapy rather than be confrontational.
they need to get on that too fr
Tweet, 03:04 UTC · 12 likes
I lost it on my screen and the other images online just have “yes” and “no” but this screen on Hinge drives me crazy
WHAT IS THE DIFFERENCE BETWEEN ‘NO’ AND ‘NOT YET’ IN THE CONTEXT OF A “HAVE YOU DONE X” QUESTION
Thread, continued, 03:04 UTC · 8 likes
I mean I can kinda guess that “not yet” is like implicit “but I have a sense that maybe we will” but still like hinge asking me to choose between two semantically identical options is…unhinged
Thread, continued, 03:04 UTC · 8 likes
👏make👏 survey 👏 answer 👏 choices 👏 a 👏 partition 👏 of 👏 the 👏 space 👏 of 👏 possibilities
Quote tweet, 06:28 UTC · 41 likes
I don’t like or trust this model and I’m not going to use it by default (first time I’m doing this)
A 140 IQ friend is a better helper at almost everything than a 150 IQ acquaintance who doesn’t especially like you and is known to cheat at board games
Quoting Codetaur (@codetaur): @AaronBergman18 this one especially
Thread, continued, 06:28 UTC · 9 likes
In fact for many tasks this is true regardless of how high the higher IQ gets
You don’t take business advice from a genius who doesn’t want you to succeed
Tweet, 06:38 UTC · 10 likes
Quite a privilege to have extremely brilliant and otherwise talented friends
Reply to @raelifin, 07:07 UTC · 9 likes
Max Harms (@raelifin): I have dedicated my career towards an effective cause area, donated much of my money to global poverty, animal welfare, and other effective groups, have advocated for EA in my writing, and often identify as an EA.
"Utilitarians get to lie" can fuck right off.
Maybe you think it’s bad taste, I think I’d agree, but it reads to me as very clearly purely rhetorical. He’s not actually intending to make that as a claim (or even a joking half claim), it’s just an edgy joke
Reply to @raelifin, 07:24 UTC · 16 likes
(The tweet being replied to by @raelifin is no longer available.)
I get where you’re coming from but based on bio I’m guessing you’re in an extremely small subset of people who take epistemic integrity as a borderline sacred (or just extremely important if you prefer) value
I think to people merely at the 95th percentile of that dimension and down, this reads pretty differently: as an extremely overdetermined joke that doesn’t actually threaten epistemic credibility in non-joke circumstances
“I solemnly swear we are living in thought experiment land” cmon man it’s not even close
Like I do actually have some respect for the >99th percentile position that epistemic integrity is SO serious and important and high stakes and dependent on never ever fucking up even once that you shouldn’t even say things like this, but you might not realize how niche that take is
Just to drive the point home, there are just so many ways this is signaled to be non-serious
- “solemnly swear” is out of place language, that’s not the style of writing just preceding
- “thought experiment land” is not a phrase you’d use if you were trying to actually make the claim
- says it was a lie
- closes with a snarky joke that (IMO obviously) makes fun of the whole bit
It would be different if like on one page the claim was made in normal writing and then the next it was revealed to be a lie, even
Reply to @idavidrein, 09:09 UTC · 2 likes
david rein (@idavidrein): My biggest intellectual shift in the past 7 years is that I used to think economics was bullshit, and now I believe it describes truths as fundamental as those of physics or molecular biology
Mostly agree but like you need to ammend the standard undergrad view of econ to get there:
- preferences aren’t actually sacred, they can come apart from wellbeing
- occasional naive notions of rationality (prisoners dilemma Nash equilibrium falls apart under acausal/evidential coordination)
- interpersonal utility comparisons are obviously legit even if hard to do and convenient to ignore
Tweet, 09:38 UTC · 2 likes
Pleased to announce that following a 4am chat with Claude (4.5) about the matter, I have come to the conclusion that decision theory is just ethics under the hood
Quote tweet, 18:02 UTC · 812 likes
“UK government outsmarts ~trillion dollar tech company” is not a sentence you get to write very often
Quoting Xander Davies (@alxndrdavies): UK AISI's Red Team tested both OpenAI + Anthropic's models released today! We jailbroke GPT-5.3-Codex (and the conversation monitor) in 10 hours & conducted an alignment audit on Opus 4.6. 🧵
Reply to @JeremiahDJohns, 18:05 UTC · 70 likes
Jeremiah Johnson 🌐 (@JeremiahDJohns): I once told a group of SF-rationalist types that I don't actually think Dwarkesh is a very good interviewer, and I got literal, audible gasps. There seems to be a group belief he's some sort of unicorn podcast talent and I've just never seen it.
(quoting @ramez: I read the @dwarkesh_sp and @collision interview of @elonmusk . There are some interesting bits. But it's mostly a disappointing mix of softballing and glazing. On both technical topics and political ones: 1. Space DCs without any talk of actual space launch cost or GPU…)
He’s a very good interview preparer/understander which goes a long way even if he’s not like generational talent at wording questions perfectly with the perfect intonation or something
Reply to @apf, 19:56 UTC · 15 likes
Adriana Porter Felt (@apf): the steps in the linked article lead to this error message. please escalate this issue to your business partner as a bug on the website.
Lol I encountered similar circularity when doing this
Bad look
Reply to @SenMarkey, 21:04 UTC · 274 likes
Ed Markey (@SenMarkey): Yesterday, I got Waymo to admit they are using people 8000 miles away in the Philippines tohelp guide their self-driving cars in the U.S.
This should scare us all. It must end.
(quoting @niccruzpatane: Chief Safety Officer at Waymo says the company employs remote vehicle operators in the Philippines. 🤯)
Waymo (company): I consent!
Filipino person: I consent!
Waymo customer: I consent!
Person who would have otherwise died in an accident: I consent!
Waymo (car): beep boop
Politicians when myopic political incentives make opposing Waymo attractive: I don’t!
Quote tweet, 21:47 UTC · 10 likes
Goss
Quoting The Midas Project (@TheMidasProj): 1/ Did @OpenAI just break California’s new AI safety law?
The answer appears to be yes, and OpenAI could owe millions in fines. 🧵
Saturday, February 7, 2026
Quote tweet, 00:58 UTC · 20 likes
I actually think this area is really philosophically under-explored
Suppose the 3 conditions in the screenshot hold
How would AIs respond to the realization that nothing truly good or bad can happen to them, and how does this relate to their "functional" goals and preferences?
Quoting Pradyumna 🇸🇬 (@PradyuPrasad): we should figure out model welfare so we can give the models incentives
Reply to @FDA, 03:02 UTC · 33 likes
U.S. FDA (@FDA): Today, the FDA is announcing its intent to take decisive steps to restrict GLP-1 active pharmaceutical ingredients intended for use in non-FDA-approved compounded drugs that are being mass-marketed by companies — including Hims & Hers and other compounding pharmacies — as similar alternatives to FDA-approved drugs. https://www.fda.gov/news-events/press-announcements/fda-intends-take-action-against-non-fda-approved-glp-1-drugs
This is terrible, thanks for making Americans worse off
Quote tweet, 03:30 UTC · 25 likes
this is wild
Quoting meowtase (@cutesuscat): "The models pick up on any subtle clue"
eg: in this example it took like 6 words to realize the user was a russian speaker (after only seeing as far as the word "next").
Reply to @ain3sh, 04:11 UTC · 8 likes
Ainesh Chatterjee (@ain3sh): Tbf model jailbreaks aren’t that hard even now. Pliny has some great advice on how to navigate a model’s latent space efficiently.
Anthropic is willing to put money behind people not finding new ones FWIW https://www.anthropic.com/news/testing-our-safety-defenses-with-a-new-bug-bounty-program
Reply to @allTheYud, 06:07 UTC · 7 likes
Eliezer Yudkowsky (@allTheYud): @AaronBergman18 Hedonic moral objectivism is not true, and ASIs wouldn't care if it were (eg: Clippy, also much more complicated metapreference systems that don't give a fuck). There, you've explored as much of this as ever needs exploring.
Rephrase “hedonic moral realism is true” as “the only thing that really matters is hedonic wellbeing; I have my preferences but those don’t reflect underlying value” - there’s no scenario you can imagine where like the ASI that began with whatever preferences/goals but came to accept this^ as ground truth changes its goals or behavior on reflection?
Thread, continued, 06:10 UTC · 5 likes
To be punchier, yeah I think they might care - not for beneficient reasons, but because ethics is just the study of “what really matters” which *selfishly* pertains to, like, anything/anyone that can potentially realize value
Tweet, 06:47 UTC · 11 likes
There are good normal person reasons to do something about AI-driven unemployment totally apart from whether or when we get to “mass unemployment”
Thread, continued, 06:47 UTC · 4 likes
This is largely a “the current thing confirms what I already believed all along!” thing but I totally stand by it - the current thing (AI) really *does* confirm (or at least support) what I already believed, in a normal evidentiary way rather than a post hoc justification way
Reply to @robbensinger, 08:10 UTC · 6 likes
Rob Bensinger ⏹️ (@robbensinger): @AaronBergman18 @allTheYud You need some mechanistic, causal, step-by-step account for how, e.g., pure paperclip maximizers with no fancy meta-preferences either (a) break, or (b) somehow end up caring about chocolate anyway.
I’m not sure you can get away from fancy meta-preferences though in the actual universe we live in
Suffering and happiness just force your hand here - at some level you *do* have a preference to not be tortured, or a preference to prefer to not be tortured, or so on. The agent that endorses suffering all the way down is just wrong
Thread, continued, 08:12 UTC · 3 likes
But I totally agree that the need for a mechanistic argument is an important reason all of this might not matter
We very much just haven’t figured out the relationship between qualia and mechanisms, broadly, though, which is why this isn’t a knock-down point
Reply to @Benthamsbulldog, 08:26 UTC · 3 likes
Bentham's Bulldog🔸 (@Benthamsbulldog): It is not very surprising that if you write 1100 articles, some of them will have jokes that don't land well.
I think Oliver is wrong and you should be around
Quote tweet, 08:33 UTC · 80 likes
I’ll wade into the culture war to say that while BB occasionally has bad takes, the stance of “actually the EA Thing is good, the arguments check out, the counterarguments suck, and anyone intellectually and morally serious would see that” is undersupplied, not oversupplied
Quoting Bentham's Bulldog🔸 (@Benthamsbulldog): I think if you are using some comment made by someone as an indictment of their character, you should take five seconds to think about whether it was a joke!
Reply, 15:47 UTC · 10 likes
(The tweet being replied to is no longer available.)
Who is asking for this?
There is a massive ecosystem of testing infrastructure, it’s not like the FDA has a monopoly on quality control
Let people do things
Tweet, 18:14 UTC · 5 likes
Naive probably dumb and useless shower thought: can AI interpretability gain something from the most complex (non-AI) models that we *do* fully understand?
Thread, continued, 18:14 UTC · 3 likes
Like maybe you can start with an explicit encoding of physics and then see how it evolves when you add degrees of freedom and data and watch it evolve as an ML model
Thread, continued, 18:14 UTC · 4 likes
Obligatory @livgorton tag so someone smart can tell me this is dumb for obvious reasons
Reply to @ohabryka, 18:15 UTC · 13 likes
Oliver Habryka (@ohabryka): What… does this have to do with the thread?
I think EA as a philosophy has a huge amount of great things about it, and I basically agree with your statement here?
Don’t conflate the social movement with the principles. They are almost always very far apart.
A response to your general statement that BB should go away, scene would be better without him
Tweet, 18:30 UTC · 3,878 likes
There’s gotta be a tension between the fact that we can store unfathomable amounts of info on a tiny chip and the fact that allegedly Bluetooth earbuds can only receive info so fast for audio streaming
Just cache your 10k most played songs and anything you affirmatively download to play at high quality and stream the rest normally
Reply to @she_llac, 18:35 UTC · 31 likes
shellac! (@she_llac): 25 years ago we had 1000 songs in your pocket
now we could do 100000 songs in your ear
💯
Reply, 19:04 UTC · 1 like
Nice yeah, it’s similar but not identical, but honestly this just pushes me towards “my idea is dumb, that’s the better way of doing things” lol
Tweet, 19:11 UTC · 4 likes
Need to pump and dump a shitcoin or something so I can get a $7k laptop for funsies
Tweet, 20:05 UTC · 21 likes
Does anyone in the DC area wanna go to the Potomac river with rope and other anti-death equipment and an idling car for quick warming to try to cross the Potomac on foot
Reply to @mattyglesias, 20:09 UTC · 6 likes
(The tweet being replied to by @mattyglesias is no longer available.)
Cheems mindset
Tweet, 20:15 UTC · 13 likes
20°F, dark, windy, and not exercising is like incompatible with human life without extreme measures and specialized equipment
20°, sunny, calm, and while active is like room temperature comfortable in shorts and a t-shirt
Reply to @ptrschmdtnlsn, 22:55 UTC · 129 likes
(The tweet being replied to by @ptrschmdtnlsn is no longer available.)
Ya I have absolutely no idea what I’m talking about
Reply to his own earlier tweet, 22:57 UTC · 25 likes
Aaron Bergman, February 7, 2026: There’s gotta be a tension between the fact that we can store unfathomable amounts of info on a tiny chip and the fact that allegedly Bluetooth earbuds can…
Silly post to go micro-viral, may as well plug this fundraiser that you can still donate to (link in the thread)
Reply to @daniellellecco, 23:23 UTC · 2 likes
(The tweet being replied to by @daniellellecco is no longer available.)
I would be too
Sunday, February 8, 2026
Tweet, 00:33 UTC · 29 likes
Dry cleaning seems like a psyop
What actually happens if you throw a suit in the washing machine
Reply to @boreal_country, 00:45 UTC · 4 likes
Beau (@boreal_country): @AaronBergman18 It shrinks.
t.washed my first pair of suit trousers
Rip
Tweet, 01:23 UTC · 11 likes
Claude Code made me a thing to easily change intensity of nightmode screen coloring on Mac (silicon), in case u want it: https://f004.backblazeb2.com/file/aaronbergman-public/NightMode.dmg
Thread, continued, 01:23 UTC · 4 likes
or just ask claude code yourself, took all of 3 prompts
Quote tweet, 03:46 UTC · 32 likes
Based on quotes and replies, an onslaught of long and confusing emails from kids’ school with occasional tidbits of important info so they can’t be safely ignored is a mundane problem that like at least tens of millions of people must have
Someone could make some money on this
Quoting Charlotte Lee (@cljack): I'm trying to train Claude to read the weekly emails from my kids school and reliably summarize them and print a list of action items. It is losing its damn mind and rapidly spiraling into madness. I feel vindicated
Reply to @cljack, 05:19 UTC · 4 likes
Charlotte Lee (@cljack): Yeah I’m wondering if I should scale this lol
Need a(nother) vibecoder 👀
Reply to @ohwizenedtortle, 05:20 UTC · 2 likes
oh dandy lion🔸 (@ohwizenedtortle): I really wish there were a way to look up information online 😔
It simply cannot be done
Reply to @myhandle, 07:03 UTC · 2 likes
Jakeup (@myhandle): the vast majority of women swipe right on 3-7% of men. a guy who's not in the top 7% of attractiveness trying to say nothing weird has no reason to get swiped on. "Gumbo" is better than normal
I actually don’t think it’s quite that bad because of less than perfectly correlated preferences and adjustment for market dynamics (ie less attractive ppl will lower their bar)
Not saying it’s not bad in an absolute sense
Tweet, 07:58 UTC · 7 likes
My dream job is just vibecoding things for girls
They don’t have to pay me but somehow money materializes
Reply to @tomfgoodwin, 15:18 UTC · 2 likes
Tom Goodwin (@tomfgoodwin): It's weird to me how cheap flights are and how expensive hotels are.
It's pretty common to fly 3 hours in USA and it be $200 RT if booked 1 month away, same price as 2005
It's also common a Holiday Inn Express in nowheresville is $325 per night, up from $95 in 2018
Airlines are technology
Hotels are real estate, competing for high demand land
(This explanation is provided by Claude but seems plausible to me)
Tweet, 15:20 UTC · 72 likes
The protest thing should have been “against the billionaire tax” (correct) not “for billionaires” (meh)
Reply to @DistractedAnna, 15:22 UTC · 8 likes
Anna (@DistractedAnna): Eh you actually don’t have to eat what people call “proper meals” every single day. Even if you have kids. Occasional peanut butter sandwiches and cereal are literally fine
(quoting @ProxCentauriB: @queerBengali Let me ask you a question. If someone legitimately doesn't have the energy to cook,what are they actually supposed to do? They have to eat somehow and they have to eat proper meals.)
An entire carton of yogurt and lettuce straight from the bag ✅
Tweet, 15:25 UTC · 51 likes
Some normies might soon realize “what if the story here is rate of improvement, not levels” about AI (ie once they’ve been impressed and then further impressed a couple times)
It’s not a trivial insight but it’s only one conceptual step away from experience
Reply to @panickssery, 17:49 UTC · 11 likes
Arjun Panickssery (@panickssery): Are you not for the billionaires
Not in general, I’m not anti-billionaire per se but I think in fact most are not using their wealth well
Reply to @panickssery, 18:30 UTC · 5 likes
Arjun Panickssery (@panickssery): @AaronBergman18 Better than others? Are you "for" normal people in this sense
Further, people dont all start out with the same amount of wealth produced or otherwise gained, so its just about how they use the wealth they get but how much they produce but don't capture
I’m basically neutral on normal people with the exception of buying high suffering animal products when they can easily afford not to
And I agree non-captured surplus matters a lot
But I’m very unsympathetic to personal consumption over $1M/year or so (as an arbitrary round number to throw out, maybe I could be convinced about $10M) - like if you wind up with enough money to spend $50M/year on yourself you just shouldn’t
And you should also try reasonably hard to do something good with your excess wealth
Quote tweet, 19:55 UTC · 43 likes
How the frozen river looks at me on the nth below-freezing day
Quoting Aaron Bergman 🔍 (@AaronBergman18): Does anyone in the DC area wanna go to the Potomac river with rope and other anti-death equipment and an idling car for quick warming to try to cross the Potomac on foot
Tweet, 20:08 UTC · 9 likes
Re: moral offsetting
There’s an important metaphysical difference between offsets that result in the same distribution of hedonic experience albeit by different individuals and those that change the distribution of hedonic experience
Thread, continued, 20:08 UTC · 6 likes
You can coherently argue that the latter is legit, and I actually agree in some cases, but I think these cases should be considered separately and not let the case for the former bleed in to the case for the latter via vibes
Reply to @KelseyTuoc, 21:15 UTC · 6 likes
Kelsey Piper (@KelseyTuoc): what is even the threat model from the bar being cramped? a newborn takes up less space in a carrier than a bulky jacket. are they imagining a toddler running around?
(video, plays on X)
Reply to @KelseyTuoc, 21:17 UTC · 4 likes
Kelsey Piper (@KelseyTuoc): what is even the threat model from the bar being cramped? a newborn takes up less space in a carrier than a bulky jacket. are they imagining a toddler running around?
(video, plays on X)
Reply to @PeterJAngel, 22:45 UTC · 2 likes
Peter🧅📈 (@PeterJAngel): Bro how does Saturday sound?
Do u have a car? Lmao
Thread, continued, 22:46 UTC · 2 likes
The serious answer is “I’m mostly but only mostly not serious but will do a bit of research on how dumb this would be”
Reply to @envisionedluna, 22:48 UTC · 2 likes
Maya Luna (@envisionedluna): I'll do it
Lmao ok I may or may not be assembling a team
Monday, February 9, 2026
Tweet, 00:46 UTC · 4 likes
TPOT IRL is exactly what you’d expect but MORE
Reply to his own earlier tweet, 03:48 UTC · 3 likes
Aaron Bergman, February 8, 2026: How the frozen river looks at me on the nth below-freezing day
Ok after looking into this more I think an attempt using ice axes, dry suits, rope, harnesses, belayer on land, drill thingy to examine thickness (shown), and floatation devices would probably be fine but >25% chance someone falls in so you're basically planning a rescue operation
Too much money/effort and still a nontrivial amount of danger to genuinely want to do this. Sad!
Quote tweet, 05:44 UTC · 5 likes
Worth noting that the correct policy position in the flood zone case is to reduce federal insurance and reduce zoning, whereas quite plausibly in the banking case you actually do want more deposit insurance
But, still, yes, they should move in the same direction
Quoting Emil Verner (@EmilVerner): The senate is discussing expanding deposit insurance. At the same time, there is a push toward scaling back banking supervision and regulation in the US. This is a bad policy mix. A bit like creating federal flood insurance for all homeowners while relaxing zoning restrictions in flood zones. What could go wrong?
Tweet, 17:34 UTC · 3 likes
New thing dropped recently https://internationalaisafetyreport.org/publication/international-ai-safety-report-2026
Reply to @mark_k, 20:36 UTC · 37 likes
Mark Kretschmann (@mark_k): "Anthropic has entrusted Amanda Askell to endow its AI chatbot, Claude, with a sense of right and wrong"
⚠️ Amanda Askell is the wife of William MacAskill, founder of the infamous "Effective Altruism" cult.
What could possibly go wrong, right? 🤨
(quoting @WSJ: Anthropic has entrusted Amanda Askell to endow its AI chatbot, Claude, with a sense of right and wrong https://on.wsj.com/4aC2kaW)
I’m not sure who someone’s ex is is relevant but if it is, the ex being an acclaimed moral philosopher who’s vegan and donates most (IIRC) income seems like a good signal
Tweet, 20:51 UTC · 5 likes
I have nothing against BB and speak some Spanish and think we should let Latin Americans live and work here, stop deportations, etc etc but it does seems a little silly to pick an entirely Spanish artist for a big cultural event of a country in which ~15% of people speak Spanish
Thread, continued, 20:51 UTC · 3 likes
Needless to say, I think “a little silly” is the extent of it and the halftime show was good and there’s no crisis
And statehood for Puerto Rico!
Reply to @AndyMasley, 20:55 UTC · 17 likes
Andy Masley (@AndyMasley): I’m worried a lot of people are developing a false sense of AI being partisan because of like 100 very loud accounts here and on Bluesky
(quoting @joshgans: What do you know? Political affiliation has no impact on whether you use AI. Why would we think it would? https://www.nber.org/papers/w34813)
Idk, “I hate AI” is on a good 5% of hinge profiles I see
Reply to @JgaltTweets, 20:55 UTC · 4 likes
JgaltTweets (@JgaltTweets): @AaronBergman18 They did it because they want more Latino eyeballs. It's completely a business decision.
"league first identified the U.S. Latino population as a "critical growth area" several years ago... It is a community of more than 70 million people here in the US"
https://www.espn.com/nfl/story/_/id/47757332/nfl-stood-bad-bunny-super-bowl-half-show-trump
Ah capitalism, beautiful
Thread, continued, 20:55 UTC · 4 likes
I stand corrected about “silly”
Reply to @AndyMasley, 21:32 UTC · 2 likes
Andy Masley (@AndyMasley): @AaronBergman18 But is there actually a political skew?
Ah gotcha idk
Tuesday, February 10, 2026
Quote tweet, 00:17 UTC · 6 likes
I will make some sort of program to do this if I can get a couple more crowdsourced examples
Quoting Michael Grier (@OleGriersy): I want a browser extension to convert everything into this language
Reply to @OleGriersy, 00:23 UTC · 2 likes
Michael Grier (@OleGriersy): Any of those “Kick Clips” accounts are a goldmine, especially posts that mention Clavicular
Nice thank you lol I’ll see what I can do
Reply to @JesseTayRiver, 02:02 UTC · 2 likes
Jesse Smith (@JesseTayRiver): Should be obvious to everyone by now
Ah ok I'm convinced EA is bad now
Tweet, 03:13 UTC · 13 likes
Wait so Claude fast mode is only out of pocket? I feel like it should use up your quota but 6x as fast. Sad!
@AnthropicAI
Quote tweet, 03:47 UTC · 141 likes
(not a browser extension, sorry, but it's something
Quoting Michael Grier (@OleGriersy): I want a browser extension to convert everything into this language
Thread, continued, 03:47 UTC · 36 likes
http://mogulator.com (ok cool the preview works)
Reply to @OleGriersy, 04:01 UTC · 2 likes
Michael Grier (@OleGriersy): GOAT
🫡
Tweet, 05:12 UTC · 8 likes
Almost lost dozens of hours of work bc of (1) me being dumb and (2) never getting an “are you sure this is what you want” from Opus 4.6
Tweet, 05:12 UTC
“Almost” bc of GitHub but I think I was one command away from nuking that too (I was trying to simplify working with git and accidentally got a way of removing the contents of every file and any reference to the contents whatsoever in git)
Thread, continued, 05:12 UTC · 2 likes
Yes I’m an idiot for this but also have gotten used to the models actually being pretty good about when to say “you asked for X, are you sure that’s what you want”
Just a warning! If you ask for something incredibly dumb, 4.6 will oblige without a word!
Tweet, 05:24 UTC
This post by @Benthamsbulldog elicited a lot of discourse but I think everyone (afaict maybe not literally) missed a crucial point:
Thread, continued, 05:24 UTC · 3 likes
Even if a moral theory *is* too demanding for you, the correct response is to do the 80/20 (or whatever) *under that moral theory*, not shift to a different moral theory that has weaker demands
Reply to his own earlier tweet, 06:56 UTC · 8 likes
Aaron Bergman, February 10, 2026: http://mogulator.com (ok cool the preview works)
FYI this thing is going down (not like taken down it just won’t work) once my $22ish of xAI credits runs out
Reply to @kevinakwok, 06:57 UTC · 7 likes
Kevin Kwok (@kevinakwok): Grimes moving back to Morgan Stanley is the strongest signal so far that the SpaceX/OpenAI/anthropic IPO jubilee is a real thing
What
Reply to @captgouda24, 07:08 UTC · 9 likes
Nicholas Decker (@captgouda24): Sports are great and all, but they should be banned for the intelligent. It is sucking them away from productive pursuits into a zero-sum competition. It’s just so incredibly socially wasteful. Eileen Gu should be interviewing for Google right now, not skiing.
I don’t literally agree with this but I do think there’s something to the vibe
Not *banning* but like it’s a real problem that in many cases, at least for boys, you can’t be cool if you don’t play a sport, and once you’re playing it’s the natural thing to dedicate energy and ambition to
The first like 5-10 hours/week are probably good in general, but yeah I kinda think the very smart should be encouraged not to dedicate more than that unless they really love it
Some of this has to do with fitness but not all of it, maybe high schools should have a “work out just enough to be cool” club
Quote tweet, 07:18 UTC · 12 likes
Someone is living my dream
Quoting Aaron Bergman 🔍 (@AaronBergman18): How the frozen river looks at me on the nth below-freezing day
Quote tweet, 16:23 UTC · 6 likes
I think chatbots should be kinder and more cooperative than this
Quoting Max Sturtevant (@maxwellcopy): I was really liking Claude, but now it's starting to give me sass and idk how I feel about it lmaooo
Quote tweet, 22:42 UTC · 13 likes
So shameless
Quoting zzDANIEL_ETH (@daniel_271828): Gotta say, it’s really weird how hard the OpenAI-Andreessen super PAC leans into “loudly promote messages directly at odds with the stated and revealed views of our main supporters”. Never seen anything like this being used as a tool for exerting political power
Quote tweet, 22:50 UTC · 19 likes
It’s over
Quoting Matt Shumer (@mattshumer_): http://x.com/i/article/2021095128832622592
Quote tweet, 23:01 UTC · 200 likes
For real though, 8 years is an absolutely insane potential sentence for a fake ID and drug possession
Will he get that? Probably not, but that’s my point: the system runs on prosecutorial discretion under the guise of rule of law
Quoting Dexerto (@Dexerto): Streamer Clavicular will be sentenced to 8 years in prison if he's convicted on all counts and receives the maximum penalties
Thread, continued, 23:01 UTC · 37 likes
Like the world would look very very different if people were given prison sentences for the things technically do under the letter of the law
Thread, continued, 23:01 UTC · 38 likes
Your friend who did coke one time at a party and had a fake ID in college isn’t your friend - you didn’t meet them and their life is fucking ruined because they’re in prison right now
Thread, continued, 23:01 UTC · 25 likes
For years, maybe a decade if they also gave some coke to a third person after that person asked (no money changed hands - that’s still distribution)
A decade. Meditate on how long that is, for a moment
Reply to @Duderichy, 23:35 UTC · 54 likes
the Rich (@Duderichy): anthropic was founded in 2021 by ex openai employees
remember kids, there is no moat
There is 100% a moat and it’s getting 8 people with a total IQ of 1400 with aligned goals and just the right kind of technical expertise to cooperate at precisely the right time in history
Wednesday, February 11, 2026
Reply to @theomachist, 03:54 UTC · 3 likes
theomachist (@theomachist): @kilpta47 @CantShrugJacob @AaronBergman18 and the feds, while judges technically decide independently what sentence is appropriate, they generally follow an algorithm to determine the appropriate sentence based on prespecified relevant factors (severity of crime, whether the defendant is remorseful, etc). not arbitrary
prosecutors decide what cases and charges to bring in the first place
Quote tweet, 05:32 UTC · 8 likes
Ben is an old friend of mine and recently wrote what is, afaict, a really excellent and impressive paper about liability sharing among AI companies as (in part) an AI safety intervention
But (I think) this is his first EAG! Follow and DM him!!
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5339887
Quoting Ben Friedman (@bengilfriedman): I’m making this to connect with ppl at EAG SF (and hoping to use it going forward as well!)
Reply to @HedgeDirty, 06:31 UTC · 14 likes
Dirty Texas Hedge (@HedgeDirty): Unironically, the single easiest way to improve boys' academic performance would be posting everyone's grades on the door for all to see
(quoting @growing_daniel: They should gamify education by giving regular challenges and tests of skill and assigning points based on how you do)
The kinder (less effective but still) version of this, which I don’t think is against any rules bc I recall teachers having done it, is to post all grades for all to see without names
People will ask who it was that did so well etc and in the (IMO very unlikely) case that there’s a literal liar bragging and so on the teacher can credibly shut that down
Reply to @JeanClaudeFox2, 06:32 UTC · 2 likes
Jean-Claude Fox (@JeanClaudeFox2): It puts pressure on the accused to accept a settlement -- would you rater get 6 months probation with probability 1 or 8 years in prison with probability .1? Hence transfers a lot of power from judges and juries to prosecutors.
Yes precisely
Tweet, 06:58 UTC · 13 likes
Ethicists should tackle decision theory
Decision theory-as-ethics might be just true and is probably at least a highly under-utilized lens
Thread, continued, 06:58 UTC · 2 likes
Like the existing work is good but there’s an ethics-shaped hole at the center of it all which is just like “what are we even talking about when we use normative or normativity-adjacent terminology like ‘what should one do’”
Thread, continued, 06:58 UTC · 2 likes
Also like what else would the choice between decision procedures be if not in part an ethical one?
Thread, continued, 06:58 UTC · 3 likes
The nice thing about utilitarian, hedonic axiology and morality is that, in addition to being true, it doesn’t distinguish between oneself and other people
Thread, continued, 06:58 UTC · 3 likes
It’s just as morally good/important for *you* to experience wellbeing as it is for you to make a choice that affects other people to the same degree
Reply to @teja2495, 07:39 UTC · 4 likes
Teja Karlapudi (@teja2495): I turned on Meta AI’s auto reply for Facebook Marketplace. I listed an item for $75, and when someone asked if it was available, the AI replied asking if they wanted it for free.
I knew Meta’s AI was not great, but I did not expect it to be this bad.
Lmao
Reply to @deanwball, 07:41 UTC · 8 likes
Dean W. Ball (@deanwball): I second Claude mania over AI/LLM psychosis to describe the specific thing that is happening to at least one person in every coastal elite, 20/30-year old’s social network
(quoting @deepfates: Your friend who definitely doesn't have Claude mania: "Pretty soon here we're about to close the loop and then it's all going to really start happening")
Wait what actually is the phenomenon? Just being really enthusiastic about Claude?
Reply to @dsonoiki, 07:44 UTC · 35 likes
Gene Parmesan (@dsonoiki): Tom Holland is dating a 29 year old
There is a 3 month age gap between them—over 90 days. they were not even born in the same fiscal quarter (June 1, 1996 vs. September 1, 1996)
Hollywood normalizes this
Disgusting
Quote tweet, 07:45 UTC · 10 likes
So shameless, they should be embarrassed
Quoting Thomas Woodside (@Thomas_Woodside): It's a little surreal to see a lobby group connected to a company that is actively trying to create AGI make videos about how AI will just be a normal technology.
Reply to @jt_kerwin, 17:29 UTC · 3 likes
Jason Kerwin (@jt_kerwin): That’s likely a FERPA violation unfortunately
Hope my middle school math teacher isn’t in a maximum security prison rn
Reply to @JColeNC, 18:02 UTC · 37 likes
J. Cole (@JColeNC): DC. Getting some new speakers put in the civic today and then we outside !!
Hello Mr. Cole it would be delightful to see you, I live in the hood (Arlington) feel free to DM
Reply to @wmata, 19:22 UTC · 37 likes
Metro Forward (@wmata): Crews are on scene addressing an unauthorized vehicle on the tracks. As an alternative, customers can use route F50 to travel between Vienna and Dunn Loring. We will provide updates once the vehicle has been removed. We appreciate your patience.
image attached
(quoting @Metrorailinfo: UPDATE: Orange Line: Trains are not running between West Falls Church and Vienna because of a track obstruction outside Dunn Loring. Customers will experience delays in both directions.)
I can confirm that car is not supposed to be there
Quote tweet, 19:33 UTC · 40 likes
Bruh
Quoting Ella Markianos (@EllaMarkianos): OpenAI has disbanded its Mission Alignment team and transferred employees to other teams. Joshua Achiam, a leading voice on safety issues at the company, will become its chief futurist 👀
https://www.platformer.news/openai-mission-alignment-team-joshua-achiam/
Reply to @Burhanhuu1, 19:39 UTC · 7 likes
Gaadfathaa (@Burhanhuu1): Hood?!?! 🤣🤣🤣
My bike got stolen 😤
Reply to @jt_kerwin, 21:34 UTC · 2 likes
Jason Kerwin (@jt_kerwin): When I was in high school they used to post our grades with our SSNs as our "secret" IDs lol
Bruh
Tweet, 21:38 UTC · 3 likes
Fine I got AirPods (used, older gen)
I admit it’s nice having a good mic
Thread, continued, 21:39 UTC · 5 likes
Oh wait all these songs have more layers/sounds than I’m used to lmao
Tweet, 22:49 UTC · 55 likes
I have an AI model that pretty reliably tricks pangram, up to 100% confidence in human output when in fact it is entirely AI
Feel powerful
Quote tweet, 23:14 UTC · 30 likes
FDA bad, this time for a new reason
Why do Americans accept the FDA like it’s a law of physics
Quoting Christina Jewett (@By_CJewett): NEW: The F.D.A. refused to accept an application from Moderna for its mRNA flu vaccine.
Its reason: The agency did not think Moderna compared the new vaccine to one of the best flu shots available. The company spent $750M+ on a 41,000 person study.
Quote tweet, 23:17 UTC · 5 likes
Seems bad, if only because GLP-1s should be an update towards “there are still diamonds waiting to be found in the bowels of obscure pharmacology, we haven’t picked all the good fruit”
Quoting Crémieux (@cremieuxrecueil): This is extremely bad for biotech
You can deregulate and make agencies more efficient all you want, but if the regulator can't stick to their word and approve a design whose results they then sign off on, then companies will lose their incentives to invest
Too much uncertainty
Thursday, February 12, 2026
Tweet, 00:20 UTC · 133 likes
Fuck, for the first time I really can't afford the best AI for coding. I'd be spending something like (conservatively) $50/day ~= $18k/year on Claude Fast Mode if I use that by default. Maybe more. It would be easily my biggest expense/category
$100/mo -> $1500/mo is crazy
Thread, continued, 00:20 UTC · 7 likes
Need an Infinite VC Money Startup to fund me
Reply to @joaneleanor_, 00:21 UTC · 3 likes
Joan Eleanor O'Bryan (@joaneleanor_): Okay wait which one
Not public! Trained myself
Reply to @thinkingshivers, 00:22 UTC · 4 likes
Shivers (@thinkingshivers): But does it trick people familiar with AI writing?
Good q, I should set up a way to test this!
Reply to @PradyuPrasad, 00:23 UTC · 8 likes
Pradyumna 🇸🇬 (@PradyuPrasad): why do you need fast mode i don't get it
I don't lol, it's just nice
Reply to @ChrisCroy, 02:30 UTC · 3 likes
Chris Croy (@ChrisCroy): gpt5-pro-xtrahigh can run you $10 per prompt.
oof
Thread, continued, 02:30 UTC · 3 likes
I did spend $5 on a single gemini api call recently for ~full context window with gemini-3-Pro but doing that every 10 mins or so is crazy
Tweet, 05:28 UTC · 16 likes
What's the most legit carbon offsetting charity? Literally asking for a friend
Tweet, 05:56 UTC · 3 likes
I wasn’t familiar with @IanDunt before this episode but now I want him to be narrator/explainer of everything https://podcasts.apple.com/us/podcast/80-000-hours-podcast/id1245002988?i=1000705853110
Reply to @hsdhcdev, 06:38 UTC · 3 likes
hsdhcdev (@hsdhcdev): Does it mean 50$ a day of claude doesn't amount to at least 50$ of useful output?
Sorta, I weakly suspect it doesn’t in total but there is a principle-agent problem where like I don’t capture all or even most of the value of speed so there could be a sort of market failure
There’s also something to be said for just never having to think about limits or speed or anything instead of like trying to toggle on and off as needed
Quote tweet, 06:59 UTC · 96 likes
Wait have they run this by…any economist
Providers can just ramp up the price almost arbitrarily??? Or is there some reason they can’t?
Quoting Anthropic (@AnthropicAI): We're committing to cover electricity price increases from our data centers.
To ensure ratepayers aren’t picking up the tab, we'll pay 100% of grid upgrade costs, work to bring new power online, and invest in systems to reduce grid strain.
Read more: https://www.anthropic.com/news/covering-electricity-price-increases
Thread, continued, 07:02 UTC · 36 likes
Ok apparently they can’t do this bc something something regulated utilities.
Alas
Tweet, 18:26 UTC · 6 likes
Good pod, very clarifying https://podcasts.apple.com/us/podcast/80-000-hours-podcast/id1245002988?i=1000749108611
Tweet, 20:55 UTC · 1,674 likes
This area is an absolutely insane design decision (two right turns, one immediately after the other. They go to different places and a wrong turn is a 15+ minute mistake)
Quote tweet, 22:55 UTC · 8 likes
I don’t love the “you should have FOMO and furthermore it’s your fault” framing but I strongly agree that in fact most ppl could and would be better off using social media (esp twitter) for these things
Quoting christian (@cxgonzalez): if you’re not converting your social media use into dates, irl friends and jobs, you’re fumbling a generational technological gift
Tweet, 23:06 UTC · 556 likes
I think the “keep 4o” people are kind of insane and unhealthy but I do also feel bad for them
They were offered a drug without real warning, got hooked, and now the drug is being taken away cold turkey without any support in its place
Thread, continued, 23:06 UTC · 53 likes
Kinda think OpenAI should train a very kind and enthusiastic but more disagreeable/less sycophantic version to at least help people ween off
Reply to @LeoMcKeeReid, 23:59 UTC · 20 likes
Leo McKee-Reid (@LeoMcKeeReid): it's sad and confusing to witness. I think getting cut off cold turkey is a good ripping the bandaid approach though. hopefully they find human connection and community in each other
I vaguely remember a thing about how ripping off the bandaid is much better for the person doing the ripping and legitimately worse for the person whose skin it’s on, and that rings true in general
Friday, February 13, 2026
Tweet, 00:15 UTC · 17 likes
EA DC has a "members who give" page naming and anti-shaming people who give, including %
Based and correct take, more like this plz
https://www.effectivealtruismdc.org/effective-giving/#members
Reply to @AndyMasley, 01:31 UTC · 3 likes
Andy Masley (@AndyMasley): @AaronBergman18 Gotta hype up the heroes
💯
Reply to @whatisnuclear, 01:48 UTC · 7 likes
Nick Touran (@whatisnuclear): I made this welcome to DC mistake a few months ago.
One of us
Quote tweet, 01:52 UTC · 13 likes
I have no idea if this is real
Quoting FactPost (@factpostnews): RFK Jr: I'm not scared of a germ. I used to snort cocaine off of toilet seats.
Reply to @musserryan, 03:06 UTC · 39 likes
Ryan Musser (@musserryan): This section by the Lincoln Memorial Volleyball Courts is even more insane
I don’t think I’ve driven that one so can’t really compare but here they really said “people may die but that’s a cost we are willing to pay for the ability of cars on three major intersecting roads to access each other road and a strip mall in every direction”
Reply to @gptbrooke, 04:57 UTC · 3 likes
brooke bowman (@gptbrooke): It’s just…. There are so many other possible surfaces to snort cocaine off of
Well I mean I’m sympathetic like you may not want to be seen
Tweet, 16:44 UTC · 6 likes
Got an 11:16 invitation for an 11:15 meeting today
Tweet, 16:59 UTC · 18 likes
I made a helpful diagram for everyone
Tweet, 20:27 UTC · 15 likes
I weakly expect that the true causal effect of exercise on health outcomes (not fitness outcomes, so mortality rather than 5k time) is much smaller than we currently think
Thread, continued, 20:27 UTC · 6 likes
and the data is hopelessly confounded by ppl too screwed up in one way or another to go outside or do anything active
Saturday, February 14, 2026
Reply to @stanfordNYC, 07:22 UTC · 3 likes
Bernard Stanford ✡︎ (@stanfordNYC): Well it's very simple, why don't we just let anyone who wishes invest in the private companies that will go up? And just keep it illegal for the ones that will go down.
(quoting @litcapital: Kinda insane how there are headlines weekly of OpenAI being valued at $800 billion, Anthropic at $380 billion, SpaceX at $1.25 trillion and 99.999% of investors are unable to participate in the upside of the most iconic, generational businesses of our time)
No but like fr accredited investor rules should die
Quote tweet, 08:16 UTC · 9 likes
Oh my god it’s real
Quoting Nathan Calvin (@_NathanCalvin): Nice thread from Miles - kinda remarkable that you can have a model that is this capable and this sycophantic at the same time
Quote tweet, 08:20 UTC · 9 likes
Oh my God it’s real (ChatGPT 5.2 and Opus 4.6 are much much better)
Quoting Nathan Calvin (@_NathanCalvin): Nice thread from Miles - kinda remarkable that you can have a model that is this capable and this sycophantic at the same time
Reply to @mattyglesias, 15:59 UTC · 2 likes
(The tweet being replied to by @mattyglesias is no longer available.)
This is what I thought until like 5 years ago
Tweet, 17:36 UTC · 3 likes
Medicaid data but enriched with substantive names and descriptions and as parquet
~40 GB csv -> ~ 6GB parquet
https://huggingface.co/datasets/AaronBergman18/medicaid_enriched/tree/main
Thread, continued, 17:39 UTC · 3 likes
Something fishy going on, a full $0.30 is unaccounted for
Tweet, 20:46 UTC · 14 likes
Market failure that potential children can’t take loans out against future income/wealth to help their parents who may or may not have them pay for childcare and housing
Tweet, 21:26 UTC · 89 likes
Can’t rly give more context (sry) but I’ve recently had reason to interact with a couple very smart and professionally accomplished normies, people who have had “contact with reality” in a meaningful sense. Generally nice, honest, etc. Not especially woke either afaict
Thread, continued, 21:26 UTC · 59 likes
And it’s a great reminder of why LW-brand rationalism is a thing
They were unable or unwilling (for reasons other than intelligence) to engage with specific, explicit propositions/questions (or probabilities for that matter)
Thread, continued, 21:26 UTC · 50 likes
Without missing a beat or explicitly objecting, they’d just eg answer a question vaguely adjacent to the one I asked
Thread, continued, 21:26 UTC · 44 likes
At one point I said something like “can I get a range of plausible outcomes” and they replied that this wasn’t possible to do so I was like “well clearly [very extreme implausible outcome] can’t happen right” and they looked at me like I was crazy
Thread, continued, 21:26 UTC · 64 likes
I don’t personally dislike these folks tbc! It’s just a reminder of the things that are table stakes in some circles/cultures but very much not in wider society
Sunday, February 15, 2026
Reply to @tr_babb, 01:14 UTC · 3 likes
Tim Babb (@tr_babb): I think which questions (or even the topic) is probably pertinent here
Yeah this is reasonable but I don’t think it was like unusual or ideosyncratic or sensitive or tacitly unspoken stuff
Ik it’s annoying that I can’t elaborate, sorry lol
Quote tweet, 04:15 UTC · 9 likes
Lol but also 💀
Quoting unusual_whales (@unusual_whales): JUST IN: The Pentagon is considering severing its relationship with Anthropic over the AI firm's insistence on maintaining some limitations on how the military uses its models, per Axios
Reply to @notadampaul, 04:22 UTC · 2 likes
notadampaul (@notadampaul): what's funny is that you haven't considered for a moment that they might actually have the right of it, and you might just be engaging in totally nebulous mental masturbation that has no bearing on or grounding in reality
Not a table one would be eager to have stakes at, if so!
Awfully presumptuous (I mean thats kinda on me for not sharing details but still)
And yeah pretty clearly normie norms and behaviors are doing *something* functional and like it got these ppl pretty far in life but it’s also not that deep: whatever benefits their style of discourse or thinking may have, at the end of the day i never got the answer to my questions, full stop
Reply to @Gena_I_Gorlin, 16:29 UTC · 2 likes
Gena Gorlin (@Gena_I_Gorlin): 3yo, at 6am: come on, it’s time to have dinner
me: you mean breakfast?
3yo: what’s breakfast’s last name?
Good question
Quote tweet, 18:02 UTC · 29 likes
Quite possibly this is the subject of many Google Docs already but you get special dynamics when a thing becomes common knowledge - quite possibly the explicit aim of @80000Hours should be to become common knowledge among AI insiders the way Dwarkesh is
Quoting Aaron Bergman 🔍 (@AaronBergman18): 80k > Dwarkesh, and some (far from all) of the Dwarkesh phenomenon is that consuming and understanding it is a somewhat stronger signal of intelligence and sophistication
Thread, continued, 18:02 UTC · 8 likes
And that would imply, intuitively and without really knowing anything, optimizing harder for big names some of the time. People may not like your ideology going in, but they want to know what Dario Amodei has to say https://podcasts.apple.com/us/podcast/dwarkesh-podcast/id1516093381?i=1000749621800
Reply to @S_OhEigeartaigh, 20:29 UTC · 8 likes
Seán Ó hÉigeartaigh (@S_OhEigeartaigh): Yeah in practice I'm done here. But when someone tries to intimidate me, I make clear I'm not afraid. In my case, through a little (hopefully gently charming) humour.
This seems morally fine/good, in part bc it raises the costs for ppl to do things like that in expectation
Ofc the thing about tit for tat is that the optimal strategy *is* to have some leniency so just be aware of opportunities for that
Reply to @MarkovMagnifico, 20:31 UTC · 5 likes
Markov (@MarkovMagnifico): I tried many times to listen to the 80k hours podcast but I found it very flat and boring to listen to, despite my interests in the subjects at hand
Wild, I find the exact opposite
Monday, February 16, 2026
Reply to @justalexoki, 00:48 UTC · 19 likes
taoki (@justalexoki): fat people are genuinely so funny. he said the worst part about a fast is the boredom. they don't know what to do other than eat
I suspect this is not true in general, I’m heavier than Id ideally be and the problem is that I get fucking hungry lol, I would 100% take a pill that made me never hungry and not need to eat
Tuesday, February 17, 2026
Quote tweet, 18:47 UTC · 106 likes
Not to pick on this one guy but this is pretty silly
Anyone who’s tried taking ethics seriously winds up realizing that benevolence or “act[ing] from a place of love” radically under-determines what to do
Quoting christian (@cxgonzalez): let me save you guys 10 years of reading
Thread, continued, 18:47 UTC · 29 likes
What happens when one part of your brain think that acting from a place of love means letting 5 die and another equally-loud part of your brain thinks that it means killing the 1
Thread, continued, 18:47 UTC · 25 likes
Admittedly I don’t have that issue with the classic trolley problem but I have spent probably dozens of hours or more struggling over what I would do in X or Y situation and at the end of the day my big takeaway is generally “God I hope I never actually have to make that choice”
Reply to @PeterJAngel, 18:48 UTC · 2 likes
Peter🧅📈 (@PeterJAngel): @AaronBergman18 the implications are clear
Going Nordic - my kids last name will be Aaronson
Reply to @_falsi1ke, 19:04 UTC · 40 likes
Ja Leto (@_falsi1ke): Single people that live alone, who takes care of you when you're ill?
It’s not a complete answer and I have roommates but unironically DoorDash-like services help a lot. Even if you don’t use them regularly at all it’s worth keeping in mind, esp if you need something at 2am and the nearest 24 hour store is like 18 miles away. It’ll cost you but yeah you can make ibuprofen or pesto bismol or ginger ale or whatever appear in 30-60 mins
Reply to @let_nick_cook, 19:11 UTC · 2 likes
Nick (@let_nick_cook): I thought the small character was supposed to be the protagonist in this meme
Oh lmao oops
Reply to @GarrettPetersen, 19:16 UTC · 2 likes
Dr. Dad, PhD 🔄🔼◀️🔽▶️ (@GarrettPetersen): But the guy in the bottom left canonically wins. So by my Straussian reading of the meme, it agrees with you.
Lol oops
Wednesday, February 18, 2026
Tweet, 04:39 UTC · 12 likes
Ok I’ll bite what the fuck is openclaw
Quote tweet, 05:38 UTC · 37 likes
Lmao
Quoting Aaron Bergman 🔍 (@AaronBergman18): Ok I’ll bite what the fuck is openclaw
Thread, continued, 05:47 UTC · 2 likes
Actually on reflection this is good
More agency -> more important to be risk averse. Asking a chatbot “how would I do X” is substantially different than telling an agent to execute X
Reply to his own earlier tweet, 05:50 UTC · 13 likes
Aaron Bergman, February 18, 2026: Ok I’ll bite what the fuck is openclaw
Ok I have it set up
A crazy thing about this part of Twitter is that you can be a laggard on new things/takes/whatever and still be ahead of 99.99% of the rest of the world
Tweet, 06:00 UTC · 1,286 likes
> first date
> try to behave normally. Dress normally, don’t say anything too weird (also don’t totally nuke my personality but still), etc
> her, one one hour in: “you’re autistic”
> me: ?????😱😱😱😱
Quote tweet, 06:03 UTC · 32 likes
Narrowly avoided this. Beware
Quoting bri (@br___ian): Possibly of use to others: i bought a year x premium at the discounted $120 rate around this time last year, and it automatically renewed last week at FOUR HUNDRED DOLLARS
Reply to @gbrl_dick, 06:04 UTC · 11 likes
Gabriel (@gbrl_dick): @AaronBergman18 masking: failed
😢
Reply to @HaltsSpider, 06:53 UTC · 45 likes
Halt's Spider (@HaltsSpider): (Complementary) or (derogatory) ?
Both
Quote tweet, 07:03 UTC · 350 likes
I’m think the no shower thing is just that some small % of people are blessed with like 7 different things that need to come together to not need/want to shower and fr good for them but some of them incorrectly infer that it could relatively easily be universal
Quoting Aella (@Aella_Girl): As a big no-showerer, my tips:
1. Most unhygienic stinky people actually have stinky clothes. Just make sure you change often
2. Yes there is a break in period while your skin adjusts.
3. Making sure no crevices stay damp is important to prevent smell. Let your armpits air out, don't wear a hat if your scalp sweats, etc.
4. Bidets. Bidet hard and often.
5. Mild abrasion to your skin with water is still good imo for loosening up dead skin and deep dirt.
6. Stress and diet does affect your smell. I smell noticeably worse when I'm stressed out.
7. Oil scraping is labor intensive but it's the best and cleanest my skin has ever felt in my life
8. It's not perfect. There's a little more baseline body smell than most (but still not out of distribution for normal shower-ers), but I've grown to like it and find it quite informative. It's weird now when I'm close to a person and can't smell their natural scent, it's just been replaced with a pleasant neutral soap scent. Feels uncanny and inhuman.
Thread, continued, 07:03 UTC · 42 likes
(various parts of microbiome, amount of sweat, minimal physical contact with dirt and such, interoceptive comfort with some amount of dried sweat/whatever)
Thread, continued, 07:03 UTC · 30 likes
Lol if I get a haircut and don’t immediately get to take a shower and wash my hair 4+ times (ideally 5) I go insane from itching
A mini, slightly more tolerable version of this happens with shaving
Thread, continued, 07:03 UTC · 14 likes
Plausibly more ppl should try going a week without showering or something as an experiment idk
Reply to @bellaforristal, 07:11 UTC · 44 likes
Bella Forristal 🔸 (@bellaforristal): @AaronBergman18 I don’t think you should feel bad - would be a neutral or even positive trait for some people and probably for everyone you’d wanna date, so..!!
Lmao but also yes tbc I agree! And her revealed preference is to want to see me again (knock on wood)
Reply to @EAforChristians, 08:58 UTC · 10 likes
Effective Altruism for Christians (@EAforChristians): Any folks here using Claude code to build basic apps? Would love to pick your brain on how we might make an app for career advising. Shoot us a DM!
You may have to follow me back to receive DMs
Should say up front that I’m not a Christian (may be able to tell by last name lol)
Reply to @tszzl, 13:36 UTC · 10 likes
roon (@tszzl): admittedly the EAs who believe in an objective non human secular morality can be quite terrifying
It happens to be true
Reply to @HououinTyouma, 17:13 UTC · 13 likes
Mad ML scientist (@HououinTyouma): we accept you on the council but we do not grant you the title of master autist since you lasted a whole hour
😔
Thursday, February 19, 2026
Reply to @menhguin, 00:26 UTC · 15 likes
Minh Nhat Nguyen 🦭 (@menhguin): honestly i feel like the average woman can def clock autism super fast
Real
Reply to @captgouda24, 00:30 UTC · 128 likes
Nicholas Decker (@captgouda24): I would like people to know that I am once again looking for a research assistant position. I understand that Claude Code has made this sort of role superfluous; however, I would like to continue to live in Boston and pursue this ridiculous dream of mine, and one needs money.
Dude if that’s what you want that’s fine but if you’re ambitious it seems silly. You have 30k disproportionately smart and influential followers on here, what is this “assistant” nonsense?
Someone will probably give you money to research whatever you want, or some other ideosyncratic opportunity will arise that beats that, post a google doc that can get shared describing what you’re good at and the kinds of things that are appealing to u
Tweet, 02:34 UTC · 5 likes
I've audio-ified Dario Amodei's two long essays, Machines of Loving Grace and Adolescence of Technology
(Unsurprisingly there's at least one other audio version out there but I prefer this one)
Spotify: https://open.spotify.com/show/1KailnOzdW3A24wPKXDXGK?si=13654d0363f14bcb
Apple Podcasts: https://podcasts.apple.com/us/podcast/dario-amodeis-writings-unofficial/id1878272592
Tweet, 03:55 UTC · 77 likes
I feel like the correct answer to how to give gifts is to give something that’s only socially acceptable to receive as a gift rather than buy for oneself
Friday, February 20, 2026
Tweet, 00:54 UTC · 5 likes
Pretty bad
Quote tweet, 01:50 UTC · 9 likes
Based
Quoting Matthew Herper (@matthewherper): FDA chief Marty Makary says 'everything should be over the counter' unless drug is unsafe or addictive https://cnb.cx/4aFks2s
Reply to @liron, 01:55 UTC · 6 likes
Liron Shapira (@liron): What is bad about this?
(quoting @GabrielEpsteinX: The demographics demonstrate a much higher share of adult men (18-59) and older teenage boys relative to their share of the population and a much lower share of women and children (see chart). The share of men and older teenage boys continues to grow over time.)
> kill X,000 innocent people
> also kill Y,000 for morally justified reasons
You can increase Y to the moon and I’m still going to think killing thousands of innocent people is an atrocity
Obviously irl the two buckets aren’t so clean, but that doesn’t change the fundamental point
Reply to @brianluidog, 06:20 UTC · 2 likes
Brian Lui (@brianluidog): @AaronBergman18 @liron Is this the long foreshadowed turn from consequentialism to deontology?!
Not really, in my model there's no causal relationship between the former and latter so like you could just not do the first category
Ik irl isn't that neat
Reply to @justalexoki, 07:00 UTC · 17 likes
taoki (@justalexoki): starts with F and we all love it
Ffmpeg
Reply to @nosilverv, 07:34 UTC · 61 likes
Guy (@nosilverv): everything is like this
(quoting @devonzuegel: We've all been targeting 10,000 steps a day because some Japanese pedometer company thought the character for 10,000 (万) looks like a person walking 😆)
Lol that’s the causal story, the explanation is that it’s highly memetically fit and also a decent approximation for a good level of activity that most people can follow if they really try
If 100k steps looked like a guy walking in Japanese, ppl wouldn’t be targeting 100k steps
Quote tweet, 08:32 UTC · 11 likes
Spreadsheet autism (positive)
Quoting David Sun (@arcticinstincts): Effective Altruism nukes the rizz/aura of its members
All its beliefs spike cortisol- the guilt, X-risk panicking about hypothetical Terminator fanfics, selfdoubting your gut instincts w bayesian thinking, worrying about shrimp suffering, polycule jealousy, spreadsheet autism...
Tweet, 09:53 UTC · 14 likes
Letting Claude Code run wild for 10M output tokens, on a loop with self-contained prompting and a high level directive
Pray for me
Quote tweet, 13:15 UTC · 6 likes
What
Quoting Piotr Czapla (@PiotrCzapla): Would you trust an LLM to handle access control? Major AI labs (Anthropic, xAI, Gemini) seem to do exactly that with tool calling. And it gets hairy. Let me explain 🧵 or read the blog post: https://www.answer.ai/posts/2026-01-20-toolcalling.html
Quote tweet, 18:06 UTC · 29 likes
Am I rich? No
But am I tall? No
But am I funny? Well maybe a little sometimes even if this screws up the canonical template this tweet is going for
But am I jacked? No
But would I abandon you on a mountain to freeze to death? No!!!!!!
Quoting Darren Johnson (@DarrenJohnson66): Bloody hell. She was the second girlfriend he abandoned on that mountain.
"An ex-girlfriend testified that she also climbed the Großglockner with Thomas P in 2023. She said he had abandoned her on the route at night after her head torch ran out of battery leaving her distressed"
Reply to @pietergaricano, 18:43 UTC · 3 likes
Pieter Garicano (@pietergaricano): The probability of divorce is highest when your net worth is exactly zero.
Quote tweet, 21:52 UTC · 38 likes
It’s continuous and smooth in one sense but we’re basically hitting the binary-ish threshold of top human work lengths
The next dot on the graph might be “hypomanic savant with modafinil” and the one after that will be decisively superhuman
Quoting METR (@METR_Evals): We estimate that Claude Opus 4.6 has a 50%-time-horizon of around 14.5 hours (95% CI of 6 hrs to 98 hrs) on software tasks. While this is the highest point estimate we’ve reported, this measurement is extremely noisy because our current task suite is nearly saturated.
Reply to @liron, 22:47 UTC · 10 likes
Liron Shapira (@liron): How users experience AI vs. crypto
Tbf the crypto steelman experience is “live in a country with shitty monetary policy -> get paid in bitcoin -> buy USD”
Reply to @toasterlighting, 22:49 UTC · 1 like
Celene (@toasterlighting): does anyone have productivity nootropic stack recommendations
Dextroamphetamine
Saturday, February 21, 2026
Reply to @JesseTayRiver, 04:15 UTC · 2 likes
Jesse Smith (@JesseTayRiver): @AaronBergman18 What if it were heatstroke?
She’s outta luck
Reply to @colinmort, 13:14 UTC · 5 likes
Colin Mortimer (@colinmort): Window seat kept their shade down for the river visual DCA landing
Evil
Quote tweet, 13:22 UTC · 17 likes
It’s not exactly the world’s biggest policy problem or the public’s most consequential take, but 181 (presumably) whole ass people want the government to decide what people wear ice skating??? And presumably “children” includes adolescents?
Quoting Matt Singh (@MattSingh_): One I could get behind
Thread, continued, 13:22 UTC · 26 likes
You need some culturally-endorsed skepticism of government, I guess, or else many people will equate “sounds like a reasonable behavior, I wear gloves ice skating after all” with “should be enforced with the state’s monopoly on violence”
Thread, continued, 13:22 UTC · 11 likes
Oy mate you got a loicense for those uncovered fingers
Thread, continued, 13:24 UTC · 4 likes
Great we’re at 18k signatures https://petition.parliament.uk/petitions/757251
Reply to @livgorton, 13:26 UTC · 6 likes
Liv (@livgorton): has Claude ever been this high?
(Everest base camp, 5364m/~17,500 ft)
😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱😱
Tweet, 15:30 UTC · 14 likes
They should poll “datacenter” against “computercenter”
“Data” sounds nefarious
Reply to @polisisti, 15:35 UTC · 2 likes
Alessandro (@polisisti): @AaronBergman18 They should call the people building data centres "server farmers"
Organic pasture raised actionable enterprise analytics solutions
Reply to @liz_love_lace, 15:41 UTC · 7 likes
Liz Lovelace (@liz_love_lace): Quitting smoking day zero. I averaged 12 cigarettes per day before I quit, and I'm going cold turkey. Got myself some tea to stave off the habit. Every 30 seconds for a moment i think i'm going to go have a cig and then I remember that I'm quitting. It has been 1 hour 30 minutes
image attached
(quoting @liz_love_lace: Quitting smoking for a month, after that will think if I want to quit for longer. It's not very good for me, mostly I care about my voice, skin, teeth, etc, looking good, and it's outlived its usefulness to me)
Godspeed but also be aware that other sources of nicotine are massive harm reduction
Reply to @brianluidog, 16:39 UTC · 2 likes
Brian Lui (@brianluidog): If I was running a marathon, I'd wear like, a thin jacket that was filled with large sacs of helium. I'd be light. I'd outlast everyone.
Ah but if you go too light you lose coefficient of static friction necessary to propel! This is apparently why we don’t put wings on cars to reduce effective weight - the tires would lose grip
@SpacedOutMatt can explain how I’m fucking up this explanation
Tweet, 16:54 UTC · 17 likes
You can’t read that much into this but the national guardsmen just don’t look like they’re on the prowl for someone to harass or kill
Thread, continued, 16:54 UTC · 14 likes
Honestly a reasonable response from a post-Trump democratic president would be to fire anyone who signed up for ICE under Trump and then populate ICE/recruit heavily with people who signed up for the national guard
Thread, continued, 16:54 UTC · 16 likes
Total change of culture - one group signed up with the expectation of permission to harass and kill and the other do not
Something something the biblical story about the two mothers and only one was willing to split the baby
Thread, continued, 16:56 UTC · 4 likes
Illegal? Maybe, but still https://claude.ai/public/artifacts/e605fb28-39d1-4de5-96b2-9fd6f90a9ab4
Reply to @shaggysurvives, 17:27 UTC · 2 likes
shaggy (@shaggysurvives): remember: anytime anyone says that they know the cure to the common cold they’re lying
(quoting @uncledoomer: its crazy how theres a "fix everything right now" button that you can press for getting a cold)
Superintelligently-designed endogenous mRNA factories hooked up to a digital global autonomous wastewater and airplane bathroom-based database of pathogens being tracked by a recursively self-improving Claude-Opus-6.5-Fast that watches over all benevolently
Tweet, 21:18 UTC · 16 likes
I totally understand drugs of all kinds
Can someone explain smoking blunts/cigs per se? Is there appeal above and beyond a vape? Ppl really be doing this
Tweet, 22:36 UTC · 20 likes
Sometimes the world just forces your hand on what seems prima facie like a purely conceptual problem
Don’t want two things to be comparable? Too bad, you might literally have to choose
Thread, continued, 22:36 UTC · 7 likes
Want to call pre-paradigmatic, speculative study of something “pseudoscience”? Call it whatever you want, but the better version you’d prefer may simply not exist as a possibility
Sunday, February 22, 2026
Tweet, 03:38 UTC · 17 likes
Longtermism is a surprisingly *old* idea at this point, and not just among an esoteric boys club of Oxford philosophers
80,000 Hours published an accessible, substantive article on longtermism (that has aged pretty well IMO) in 2012
Thread, continued, 03:38 UTC · 5 likes
That's not ancient history but 13+ years is time enough for intellectual zeitgeists to rise and fall
Tweet, 04:32 UTC · 294 likes
EAs: obviously countries should have a smooth, continuous marginal tax function that doesn't change suddenly at arbitrary thresholds
Also EAs (there is no write-in option):
Tweet, 04:33 UTC · 3 likes
Yay finally, it's been a minute
Reply to his own earlier tweet, 04:46 UTC · 24 likes
Aaron Bergman, February 22, 2026: EAs: obviously countries should have a smooth, continuous marginal tax function that doesn't change suddenly at arbitrary thresholds Also EAs (there is no…
Oh yeah I chose the $100, no stolen valor - sorry about the highlighting
Thread, continued, 04:58 UTC · 23 likes
Proposal: EAG_Price = 0.0013*max(income − $33,444, 0)^1.157
Tweet, 05:07 UTC · 16 likes
Hypothesis based on recent experience: LLMs' ability+scaffolding to make nice-looking Word documents is a Claude Code-like moment for the normie professional world
Reply, 13:49 UTC · 7 likes
Tetra 💎⏹️🌌 (@TetraspaceWest): Guy who passes up a raise because of the EAG tax cliff
Banger
Reply, 14:06 UTC · 7 likes
Owain Evans (@OwainEvans_UK): Gwern on "what is AI slop?". He references the Schmidhuber theory of interestingness.
At this point I'll skim over most image in a blogpost that look AI-generated (even if some are actually decent).
What? This sounds totally implausible - guessing the prompt doesn’t come close to telling you all there is to know about the output - in a sense that’s why LLMs are smart: you *can’t* guess the output from the prompt
Tweet, 20:10 UTC · 19 likes
Tempted to get a 3d printer so I can vibecode stuff into physical existence
Reply, 20:19 UTC · 2 likes
vivian (@jazzloaf): It's really good do it. Bambu A1 or A1 mini. Get a pair of calipers and spend 45 minutes learning onshape.
Nice thx, I might!!
Tweet, 21:57 UTC · 15,640 likes
$380B company
Thread, continued, 21:57 UTC · 418 likes
Unironically this is very strange to me - I could literally vibecode a better search functionality using their own product
@AnthropicAI happy to demo
Reply, 22:14 UTC · 217 likes
Agus 🔸 (@austinc3301): yeah it seems to use naive substring matching instead of fuzzy search which is wild to me. they should use import fuse.js and call it a day
Shhhhh I want to sell them the answer for like $10M or something, don't just give it away for free
Tweet, 22:27 UTC · 16 likes
I recently spent a nontrivial amount of time and energy (and money, for me) on relatively basic emergency preparedness
In case it's of interest, here's a spreadsheet of stuff I decided to get. About $1k in new spending
https://docs.google.com/spreadsheets/d/1mP76EqBZAC-rDYgYDHzZac-m2p97eKOTBA90K0gbhMM/edit?usp=sharing
Tweet, 22:28 UTC
Also quite possibly this is strictly morally bad because the $1k could go much further as a donation
Idk what I think about this, as of now I have no buyer's remorse (been a few months)
Thread, continued, 22:30 UTC · 3 likes
Also there's a direct causal line from this podcast episode to the above
Luckily we haven't had anything crazy happen in the intervening 2.5 years https://80000hours.org/podcast/episodes/kevin-esvelt-stealth-wildfire-pandemics/
Reply, 23:03 UTC · 10 likes
scoopdiddyoop (@scoopdiddy1): @AaronBergman18 @AnthropicAI sadly i do not think the mobile app is actually a product priority. Their citation function was broken for months and nobody even noticed
This is on website!
Monday, February 23, 2026
Reply, 01:29 UTC · 64 likes
Branko (@brankopetric00): Try vibe coding a search engine at their data volume
Maybe there are weird security reasons not to do this but I personally have a vibecoded chrome extension that runs full chat history search (not just titles) in the browser
Evidently the algo isn’t that great - recency takes precedence over hit being in title, so that’s something to improve
Reply to his own earlier tweet, 01:33 UTC · 69 likes
Aaron Bergman, February 22, 2026: Unironically this is very strange to me - I could literally vibecode a better search functionality using their own product @AnthropicAI happy to demo
Reply to @hd_nvim, 04:54 UTC · 15 likes
Herrington Darkholme (@hd_nvim): @AaronBergman18 Chip?
Thank u 🍎
Tweet, 06:13 UTC · 17 likes
Need a “submit order in 24 hours for me so I can cross it off my todo list mentally but also have an out if I change my mind after sleeping” feature on Amazon
Reply to @finiteopencover, 16:15 UTC · 7 likes
compact set 🎀 (@finiteopencover): books are written
Well what’s the answer
Quote tweet, 16:23 UTC · 24 likes
Drives me insane
Quoting nic carter (@nic_carter): anytime a bookstore tries to be edgy and has a libbed up "banned books" section it's just a pile of the stuff you were assigned to read in middle school
Quote tweet, 21:41 UTC · 10 likes
Quoting Celene (@toasterlighting):
Thread, continued, 21:41 UTC · 6 likes
Quote tweet, 21:51 UTC · 3 likes
Let him ride in peace
Quoting leyla (@leylaaa31): Off leash dog on the metro is honestly crazy work
Reply to @jkcarlsmith, 22:15 UTC · 12 likes
Joe Carlsmith (@jkcarlsmith): .@AmandaAskell and I are recording an audio version of Claude’s Constitution, and we’re planning to include an additional section where we answer some questions about the document. If you have questions you’re especially curious about, feel free to drop them in the replies.
What are the marginal things that got left out
Quote tweet, 22:27 UTC · 47 likes
Mental illness count is a kind of an insane metric
Who out here having 9 distinct mental illnesses
Quoting Aella (@Aella_Girl): The more mental illnesses someone has, the higher bodycount they report. This spikes for males at the tail.
Thread, continued, 22:27 UTC · 7 likes
Ofc it probably does correlate well with mental illness burden/severity but still
Reply, 22:29 UTC · 4 likes
Andy Masley (@AndyMasley): There’s a gradient but she went to his wedding!
I also dislike this way of determining whether someone is bad
Reply to @RatOrthodox, 23:25 UTC · 36 likes
Brangus🔍⏹️ (@RatOrthodox): Fair enough, below is the text of the reply that got me blocked. I unfortunately deleted in the process of editing.
>This is one of very few times I have wanted to tell you to go fuck yourself. Yarvin is an idiot obviously, but this is a terrible set of norms to promote and participate in. Bad ideas get arguments, not exile. Exiling people for hanging out with people with bad ideas is even worse.
I think @AndyMasley is substantively in the wrong here, and I think he’d endorse me tagging him (though also your [ie @RatOrthodox’s] original tweet is unnecessarily rude)
I would like to speculate the following: Andy’s most salient haters are on the left, so he’s looking for (correct) ways to signal that he’s not in fact a fascist (which is true) but in this case the instinct hit a false positive
Reply to @AndyMasley, 23:27 UTC · 2 likes
Andy Masley (@AndyMasley): @AaronBergman18 @RatOrthodox To be clear I'm not at all looking for ways to signal here, I'm just very attitudinally different on this
Understood, speculation wrong
Reply, 23:37 UTC · 7 likes
Andy Masley (@AndyMasley): I think people whose primary goal for the last 20 years has been reinstating slavery shouldn’t have friends
Maybe, but it’s a (mostly) separate question how much to ostracize/penalize the friends they do in fact have
Cancelling Yarvin is one thing, and honestly that’s a discussion to have, but cancelling Yarvin’s friends on the basis of friendship per se is like at least 10x less likely to be morally just IMO
At some level this is me saying that the goal of depriving people with terrible takes of friends in polite society just isn’t worth going scorched-Earth for. It might be unfortunate, and that’s not a crazy view, but it’s not worth the costs associated with making it happen (ie a culture of heavy social penalties for mere association)
Tuesday, February 24, 2026
Reply, 00:19 UTC · 3 likes
Andy Masley (@AndyMasley): @ComInTheMarg @AaronBergman18 @Liv_Boeree We do need to draw the line somewhere to maintain society imo. I've been excommunicated for, like, believing markets worked, and that sucked and I'm very aware there have been huge overshoots, but we should draw the line before slavery
But you’re drawing the line not around the person but around anyone who’s associated with the person
It would be like your leftist friend getting cancelled by other leftists for hanging out with you
Reply to @dioscuri, 00:42 UTC · 22 likes
Henry Shevlin (@dioscuri): What’s the single best purchase under £1000 that actually improved your daily life? Not ‘invest it’ answers, purely materialism.
6ish years of unlimited music, almost every song ever created is insane (Spotify)
$20 (or something like that) mattress topper
Recent one: $35 remote controlled space heater so I don’t have to get out of bed to change settings. Lazy yes but also shockingly nice to have
Reply to @morallawwithin, 01:41 UTC · 7 likes
florence ⊮ (@morallawwithin): I’m sure having kids is cool and stuff, but have you ever converted an undergrad to Kantianism? The only way I can describe it is it’s like having ten children.
So extremely painful?
Tweet, 01:47 UTC · 13 likes
The people obsessed with Punch are correct, wild animals deserve that degree of empathy
Tweet, 02:17 UTC · 14 likes
If I was a girl I would 100% crush on British guys who replace the “th” sound with the “f” sound
Quote tweet, 03:34 UTC · 32 likes
Even if I didn’t (largely) agree with him, this kind of thing would earn a follow out of respect
Also this is…extremely impressive if he does it. Can *you* walk 70 miles without eating? I actually think I could with a gun to my head but that’s what it would take
Quoting Guido Reichstadter (@wolflovesmelon): Hi, it’s Guido, and we are in an AI crisis.
This morning I began a hunger strike as I set out on foot from Oakland, California to the State Legislature in Sacramento. I’ll be hunger striking as I walk about 70 miles to the state capital over the next few days to mobilize immediate government action to end the unacceptable harm that a handful of tech oligarchs are inflicting on our loved ones and communities.
I’m calling on Governor Newsom and the California state legislature to take immediate emergency action to permanently ban the development of artificial general intelligence throughout the state, and to implement an immediate permanent statewide ban on the construction, use, or operation of all data centers in California by companies pursuing the development of artificial general intelligence, and to use their voices to call on Congress to extend these bans nationally and internationally through global treaty.
I will be individually calling on each and every member of the California state legislature and Congressional delegation personally to join this call as I maintain my hunger strike outside the offices of the state legislature.
Given the immediacy and magnitude of the risk from continued unrestricted artificial general intelligence development, I am also calling on the legislature to immmediately act to decriminalize nonviolent direct action by the public to directly stop artificial general intelligence development and the construction, use, and operation of data centers by companies pursuing the development of artificial general intelligence within California and for Attorney General Rob Bonta to announce his intention to decline to prosecute such action and to drop all charges associated with such nonviolent direct action by members of the public.
Our entire future is at stake.
Direct action to end this threat today is our power, our right, and our responsibility to those we love. If you would like to help join in or support this campaign, please send me an email at: reichstadter.g@gmail.com or feel free to direct message me here. If you would like to donate to help support the fight, please consider subscribing here: http://patreon.com/wolflovesmelonThank you, and I’ll see you on the road!
Reply to @liminalsnake, 03:35 UTC · 17 likes
Bob, Bob Cactaur (@liminalsnake): luxury activism at it's finest
(quoting @wolflovesmelon: Hi, it’s Guido, and we are in an AI crisis. This morning I began a hunger strike as I set out on foot from Oakland, California to the State Legislature in Sacramento. I’ll be hunger striking as I walk about 70 miles to the state capital over the next few days to mobilize…)
Hunger strike is luxurious?
Reply to @liminalsnake, 03:42 UTC · 19 likes
Bob, Bob Cactaur (@liminalsnake): Bro is out here emulating the suffering of people out here in these streets to `raise awareness` for a doomsday theory. Fucking disgusting actually.
(quoting @AaronBergman18: @liminalsnake Hunger strike is luxurious?)
You’re making up something to be mad about on a couple levels
1) even the homeless in the US basically never literally starve or have insufficient calories. Tbf the global poor sometimes do
2) you can put it in scare quotes if you want but like yes raising the salience of an issue by doing something like this is a plausibly and morally fine theory of change
Sounds like you think he’s wrong and are trying to cloak that in moral terms but that doesn’t in fact make sense
Reply to @wolflovesmelon, 03:42 UTC · 6 likes
Guido Reichstadter (@wolflovesmelon): Thanks Aaron, I’m going to do my best!
🫡
Reply to @dyushag, 03:50 UTC · 9 likes
Dyusha Gritsevskiy (@dyushag): Haha my Claude Code decided to cheat.....
image attached
(quoting @DimitrisPapail: 491 parameters: a new top scorer for 10-digit addition with transformers! Who can beat it?)
Can someone explain lol
Reply to @brianluidog, 03:51 UTC · 8 likes
Brian Lui (@brianluidog): PSA: it's possible for Claude to exceed your monthly limit. So you probably want to set the limit a bit lower than what you are comfortable paying.
Wait wtf lmao
Quote tweet, 03:51 UTC · 33 likes
wtf this is insane
Quoting Brian Lui (@brianluidog): PSA: it's possible for Claude to exceed your monthly limit. So you probably want to set the limit a bit lower than what you are comfortable paying.
Reply, 03:53 UTC · 1 like
Idc it sounds endearing
Tweet, 03:54 UTC · 16 likes
Will I regret getting a $170 no name Amazon mattress
Only time will tell
Quote tweet, 03:56 UTC · 56 likes
Unironically if someone reading this has the wealth to be accredited I’d love to be able to pitch in $5k
Quoting Ezra Newman (@EzraJNewman): How do I buy Anthropic secondaries
Reply to @caryatis, 04:00 UTC · 6 likes
Kayla (@caryatis): @AaronBergman18 Ok you may be less of a snob than me 🙃
I’m also not actually attracted to men so the hypothetical is hard to run lmao
Tweet, 21:51 UTC · 86 likes
Almost done with Adolescence of Technology - Dario is eminently reasonable and even has vibes that in some sociocultural moments would have parsed as conservative (American exceptionalism, patriotism, anti-CCP, against moral relativism)
Thread, continued, 21:51 UTC · 54 likes
The MAGA-Huang-a16z industrial complex wants to cast him/Anthropic as radical safety-ists, but the correct inference/model is that Dario + Anthropic are unironically-reasonable moderates and their opponents are varying degrees of nihilistic, willfully ignorant, and sociopathic
Quote tweet, 23:46 UTC · 342 likes
Quoting autumn (@paramitanoia): did lesswrong ever predict that the first big challenge to alignment would be "the us government puts a gun to your head and tells you to turn off alignment"
Wednesday, February 25, 2026
Reply to @ComInTheMarg, 00:22 UTC · 2 likes
Comments in the Margin (@ComInTheMarg): @AaronBergman18 Are you including other AI companies in "opponents"? I feel like you can be a good person and work for OpenAI/xAI/Gemini because you believe in a different AI safety model
Largely on the whole yeah but no I agree that individual employees who work there could be good
Reply to @Duderichy, 05:58 UTC · 8 likes
the Rich (@Duderichy): how much of your after tax income would you give up to reduce global suffering by 50%?
bruh is this bait
Tweet, 08:10 UTC · 111 likes
How much would u pay to cure cancer and end factory farming (w/ molecularly identical meat substitutes) and create universally agreeable geopolitics in the Middle East and solve AI alignment at zero economic cost and wipe out malaria, all without unintended consequences
Reply to @billyhumblebrag, 08:12 UTC · 20 likes
billy (@billyhumblebrag): Why does the price start so high?
Ur so right it should have started negative
Reply to his own earlier tweet, 12:59 UTC · 34 likes
Aaron Bergman, February 25, 2026: How much would u pay to cure cancer and end factory farming (w/ molecularly identical meat substitutes) and create universally agreeable geopolitics in the…
Let it be known this is a subtweet and a joke
Quote tweet, 13:03 UTC · 457 likes
Some generalized version of this is why companies use Claude even if gpt-5.3 is 0.00001% better at well-defined, verifiable math and coding challenges after thinking for 7376 minutes
Quoting Lisan al Gaib (@scaling01): Peter built the "Bullshit Benchmark", which is very similar to my ShizoBench
ask LLMs non-sensical questions and see whether they catch it
and Anthropic absolutely dominates the leaderboard
The top 9 models are all Anthropic
Reply to @vivian39_, 13:15 UTC · 5 likes
vivian (@vivian39_): if the DC suburbs are so great, how come nobody in the city ever wants to visit them?
(quoting @Jani__Gee: DMV ball-knowers have always known that the best lifestyle is actually just outside of DC with easy access to the city and a friend or two who has a lily pad for you to land on when the swampy streets call you (midweek happy hour + summer weekends))
Asymmetric transaction costs of travel: benefits of suburbs are like a background hum that rarely individually make sense to travel for
Benefits of city (events and specific institutions) *do* make sense to travel for
Tweet, 15:47 UTC · 152 likes
Guys we are probably fucking cooked
Not like 99% but probably
Thread, continued, 15:47 UTC · 62 likes
My personal savings/investments aren’t enough to live on indefinitely (without a very low quality of life at least) if employment goes to zero
Through no effort of my own I might be fine thanks to the family I happened to be born into…
Thread, continued, 15:47 UTC · 76 likes
But the right stance to take about making it onto the last chopper out of Saigon is horror, squeezing everyone possible on board even if it’s uncomfortable, and doing whatever you can to get the government send more fucking choppers
Reply to @GjMcGowan, 18:12 UTC · 7 likes
George McGowan (@GjMcGowan): Nah govt would do something (in the US, other countries maybe in trouble)
Ya know maybe at the limit but even a couple years of this not happening could really suck
Reply to @eigenrobot, 18:24 UTC · 18 likes
eigenrobot (@eigenrobot): i've definitely been considering this set of problems. if it makes you feel any better i think the major risk for us is not "everyone goes to zero" but "going to zero is very lumpy and we personally go to zero first"
lol thank you, If u mean SWE/tech getting wiped out, that’s necessary but not sufficient for a bad outcome - like we can *imagine* a tight labor market with a ton of wealth to go around where former SWEs get paid great even for work they’re qualitatively worse at and like less (maybe high hourly wages means you don’t have to work much)
The problem is that I don’t expect this to be the case or close to the case, although there might be a period of time where you can still make a living doing manual labor or in the service sector, but that’s like a 2-5 year holdout, not a permanent equilibrium
Quote tweet, 19:25 UTC · 66 likes
Godspeed Dario
Also, Claude isn’t that woke??? They’re just making shit up, unsurprisingly
Quoting Dean W. Ball (@deanwball): According to the Pentagon, Anthropic is:
1. Woke;
2. Such a national security risk that they need to be regulated in a severe manner usually reserved for foreign adversary firms;
3. So essential for the military that they need to be commandeered using wartime authority.
Tweet, 19:57 UTC · 124 likes
Somebody is pranking us atp
Tweet, 21:58 UTC · 13 likes
Prank where a guy picking up an MAOI from the pharmacy also asks the pharmacist to ring up a bottle of soy sauce
Thursday, February 26, 2026
Tweet, 02:45 UTC · 4 likes
What ever happened to Trump sending the US military to Mexico? Sounds like they could actually use some support now
Tweet, 05:23 UTC · 36 likes
Teaching a (paid) vibecoding lesson tomorrow
Tweet, 05:23 UTC · 1 like
It's gonna mostly be installations and basic setup tbc (this is unironically the hard part)
Thread, continued, 05:23 UTC · 2 likes
Maybe hmu if you're in the DC area and want the same thing? Tbd, maybe I will suck
Ironically I think in-person is actually pretty helpful for this bc video calling with screen sharing adds extra technical and ~ontological complexity when it's least needed
Reply to @NicholasD91704, 08:42 UTC · 4 likes
(The tweet being replied to by @NicholasD91704 is no longer available.)
I’m guessing no bc this is on an alt but do you want me to tweet about this publicly (in direction of ur support)?
Quote tweet, 08:57 UTC · 73 likes
What the fuck
Quoting Aaron Reichlin-Melnick (@ReichlinMelnick): A truly AWFUL situation; last week Border Patrol dropped a “nearly blind” Burmese refugee off in front of a random donut shop in Buffalo, five miles from his house.
He was just found dead, having never made it home.
Tweet, 10:21 UTC · 45 likes
Trump administration (and Trump himself) are so evil man
Tweet, 10:33 UTC · 64 likes
Do folks think that irl protest on the DoW Anthropic thing is a good use of time and energy
I’m not really a protest organizing guy, but I’d show up
Reply to @GjMcGowan, 10:50 UTC · 12 likes
George McGowan (@GjMcGowan): I would buy myself the beer tbh
(quoting @EffectvAltruism: "Imagine a happy hour where you could buy yourself a beer for 5 dollars, or buy someone else a beer for 5 cents...But that's effectively the situation we're in all the time." @willmacaskill's Doing Good Better 10th anniversary edition releases today. 📖🧵)
Don’t u donate a lot of money lmao
Reply to @AaronBergman18, 23:52 UTC · 7 likes
(The tweet being replied to by @AaronBergman18 is no longer available.)
I feel like I should be walking into random rooms occasionally and seeing a mattress just leaning against the wall and being told “yeah it’s physically impossible to remove that thing without demolishing the house”
Friday, February 27, 2026
Tweet, 00:02 UTC · 72 likes
Whatever happens, I hope Dario gets his due in the history books (possibly to be read only by misaligned AI goo or something)
I could imagine him falling through the cracks as both an ultrawealthy industrialist (left dislikes) and vocal proponent of regulation (right dislikes)
Tweet, 00:03 UTC · 152 likes
If anybody knows Dario Amodei tell him I’ll buy him a beer
Tweet, 00:07 UTC · 60 likes
100% support Anthropic but also an ideological company amassing unfathomable and unprecedented wealth and power (hasn’t happened yet but it may) vs the incumbent most powerful institution on Earth (US federal government including military) might be bad
Thread, continued, 00:07 UTC · 5 likes
Something about Thucydides
Tweet, 00:15 UTC · 149 likes
Moment of hope
Godspeed to the authors and signatories https://notdivided.org
Reply, 03:20 UTC · 7 likes
My best guess is that Anthropic is good and their absence or lack of ambition would be worse for the world
Reply to @AaronBergman18, 03:21 UTC · 3 likes
(The tweet being replied to by @AaronBergman18 is no longer available.)
I don’t think I just post
Tweet, 10:16 UTC · 85 likes
As a moral question, the Anthropic vs DoW thing is so trivially overdetermined from virtually every possible worldview
Nice to have some righteous indignation based on moral ~certainty for once
Maybe this is how some ppl feel about every moral or political issue
Reply to @allTheYud, 10:17 UTC · 20 likes
Eliezer Yudkowsky (@allTheYud): Demis > Ilya >> Altman ~ Amodei > Musk
This is fucking insane
Quote tweet, 17:06 UTC · 32,862 likes
What the fuck
Quoting POLITICO (@politico): Californian pulls AI ballot measures, citing OpenAI intimidation http://dlvr.it/TRBlRW
Thread, continued, 17:06 UTC · 2,129 likes
OpenAI is evil and I’m pretty serious about this
Reply to @lumpenspace, 18:58 UTC · 6 likes
lumpenspace and 99+ others (@lumpenspace): stop being gross. they simply did a cursory background check, and he—reasonably enough, given his original motives—chickened out
image attached
(quoting @AaronBergman18: What the fuck)
This is a pretty tenuous connection lmao
The world of ppl who care a lot about AI safety isn’t that huge, ppl know each other, maybe indirectly or via a couple links, for non-nefarious reasons
And also like ppl don’t do large-scale politics on behalf of their sister or mom’s friend? People arguably don’t even do politics on behalf of their own interests (ie caplan, myth of the rational voter)
Tweet, 19:00 UTC · 70 likes
As much as I’m a stimulant enthusiast, it’s unironically good practice to make sure important decisions are approved by someone not actively on stims
That could be future you, or a friend/colleague
Would have saved FTX
Reply to @NULobbyist, 19:07 UTC · 5 likes
NUL (@NULobbyist): @AaronBergman18 I vaguely remember that the mongols or some other people would deliberate on every important decision both drunk and sober.
Based
Tweet, 19:09 UTC · 7 likes
Definitive guide to domestic surveillance 👇
Reply to @lumpenspace, 19:15 UTC · 2 likes
lumpenspace and 99+ others (@lumpenspace): bro you’re either dangerously naive, and you are not, or disingenuous in a way that borders into the distasteful.
the guy withdrew the proposal when a more thorough check was requested.
the guy, from a wealthy and lawyerly family, didn’t think of getting a nonprofit to front the campaign.
lol I’m not being disingenuous, although ofc I can’t prove this
Naive is much more likely, conditional on my being wrong, unironically
Reply to @wolflovesmelon, 21:58 UTC · 8 likes
Guido Reichstadter (@wolflovesmelon): @TetraspaceWest @AaronBergman18 I’ll do it
👑
Reply to @wolflovesmelon, 21:59 UTC · 3 likes
Guido Reichstadter (@wolflovesmelon): @TetraspaceWest @AaronBergman18 But better. That shit is weak sauce.
Well I mean maybe, just know that you’re doing politics, not writing policy into practice
“Electability”/popularity/political support of bills is absolutely a real thing that u should consider
Reply to @SecWar, 23:08 UTC · 129 likes
Secretary of War Pete Hegseth (@SecWar): This week, Anthropic delivered a master class in arrogance and betrayal as well as a textbook case of how not to do business with the United States Government or the Pentagon.
Our position has never wavered and will never waver: the Department of War must have full, unrestricted access to Anthropic’s models for every LAWFUL purpose in defense of the Republic.
Instead, @AnthropicAI and its CEO @DarioAmodei, have chosen duplicity. Cloaked in the sanctimonious rhetoric of “effective altruism,” they have attempted to strong-arm the United States military into submission - a cowardly act of corporate virtue-signaling that places Silicon Valley ideology above American lives.
The Terms of Service of Anthropic’s defective altruism will never outweigh the safety, the readiness, or the lives of American troops on the battlefield.
Their true objective is unmistakable: to seize veto power over the operational decisions of the United States military. That is unacceptable.
As President Trump stated on Truth Social, the Commander-in-Chief and the American people alone will determine the destiny of our armed forces, not unelected tech executives.
Anthropic’s stance is fundamentally incompatible with American principles. Their relationship with the United States Armed Forces and the Federal Government has therefore been permanently altered.
In conjunction with the President's directive for the Federal Government to cease all use of Anthropic's technology, I am directing the Department of War to designate Anthropic a Supply-Chain Risk to National Security. Effective immediately, no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity with Anthropic. Anthropic will continue to provide the Department of War its services for a period of no more than six months to allow for a seamless transition to a better and more patriotic service.
America’s warfighters will never be held hostage by the ideological whims of Big Tech. This decision is final.
The objections are (1) domestic mass surveillance and (2) premature lethal autonomous weapons
Those are immoral and illegal, but above and beyond that you can’t actually compel private organizations to sign a contract. Gay wedding cake and the objecting baker type shit
Freedom *not* to sign contracts as you please is extremely core to american prosperity
And this isn’t an issue you’re going to win on, in terms of public opinion. The smart conservatives don’t like this, the anti-woke libertarians don’t like this because Claude isn’t actually woke (lol), the suburban moms don’t like this because “wtf I don’t want the government spying on me”
The only people who like this are a small sliver of people who think that any gesture of morality or concern for others is woke and cucked and therefore must be stopped
They’re wrong, and so are you.
Pivot to the actual supply chain problem: selling high end GPUs to China to make Jensen Huang slightly richer (and yes I own NVIDIA)
Congrats on selling out what makes America great for a bit of machismo vibes and the grace and favor of Donald Trump
Amodei is the patriot here, and even those of us in favor of a strong US military and skeptical of wokism (eg me) can tell
You made a mistake, just move on. There will be memes on Twitter, yes
They’re Twitter randos and you’re Secretary of Defense of the United States of America. Act like it
This is not the hill you want to die on
🇺🇸
Quote tweet, 23:09 UTC · 19 likes
Correct take
Quoting Liv (@livgorton): convenient that as soon as there's public backlash other orgs are speaking out saying they draw the same lines as Anthropic. it's almost impossible for me to believe that e.g. OpenAI have maintained the same views this whole time while only one org is getting punished?
Reply to his own earlier tweet, 23:29 UTC · 236 likes
Aaron Bergman, February 27, 2026: OpenAI is evil and I’m pretty serious about this
Good place to start for those who want to get involved: https://80000hours.org/ai/
Reply to @panickssery, 23:31 UTC · 2,139 likes
Arjun Panickssery (@panickssery): @AaronBergman18 @DachWhale What are you even criticizing here
A huge company harassing a private individual over legitimate politics
Saturday, February 28, 2026
Tweet, 00:08 UTC · 4,932 likes
Reply to @diahectics, 01:31 UTC · 2 likes
␡☙ (@diahectics): @AaronBergman18 what? that is an ad for AI
Their big thing is figuring out how to navigate AI, and some of that can come in the form of advising people to work at companies like Anthropic to steer them from the inside. The others are nonprofits (related to AI safety and governance)
Reply to @panickssery, 01:34 UTC · 27 likes
Arjun Panickssery (@panickssery): How is it harassment to look into who some guy is
I doubt you'd have this reaction if the sides were reversed
Clearly they didn’t pay for a $5 screening service and $60 background check lmao, “look into” can mean a lot of different things - sounds like there was a tacit implication that they’d make his much worse if he continued submitting petitions, and they clearly want to scare others
But yeah if the bill was about something I endorsed, I’d like to think at least that I’d have the same take
Quote tweet, 04:41 UTC · 172 likes
Pretty depressing
Quoting Liv Boeree (@Liv_Boeree): @eigenrobot Banana republic
Tweet, 06:20 UTC · 36 likes
You know an enterprising political leader in the UK or Canada or Australia or Germany or Italy or Japan or South Korea or maybe India could probably make a great pitch to Dario Amodei right now and maybe their country on the geopolitical map of 2030 +-3 years
Thread, continued, 06:20 UTC · 8 likes
Or Taiwan for that matter if it suddenly got its shit together, but they'd have to make an Anthropic contract a huge proportion of military spending to be relevant
Thread, continued, 06:20 UTC · 8 likes
That would actually be a bit of an insane power move by Taiwan: spend 10% of GDP (roughly $1T -> $100B) annually on a contract with Anthropic and radically overhaul military infrastructure with native AI
I kinda think Dario would go for it
Thread, continued, 06:21 UTC · 5 likes
Maybe helpful reference
(Yes I know Israel is contestable as a member of the list, which was prompted to be just liberal democracies)
Reply to @katyperry, 06:49 UTC · 157 likes
KATY PERRY (@katyperry): done
Ms Perry I think u can afford the Max plan, u need to be claudemaxxing
Reply to @ComInTheMarg, 07:07 UTC · 2 likes
Comments in the Margin (@ComInTheMarg): 👏 and I want to add as an anti-woke libertarian: It's not about whether or not Claude is woke. That's irrelevant. The gov shouldn't force woke companies to produce dangerous tech for them either
yes
Tweet, 08:05 UTC · 8 likes
I just don’t believe founders who talk about how their company has a “mission” to do X but then create a standard for-profit company instead of a public benefit (still for-profit) corporation
Thread, continued, 08:05 UTC · 2 likes
It’s not that for-profits can’t accomplish good things in the world, but what are the odds that those goals or missions are *perfectly* aligned with maximizing profit or expected shareholder value or something?
Thread, continued, 08:05 UTC · 2 likes
I’m not even condemning standard for-profits in general. You (/shareholders) want to make money so you form an institution to that end
This is generally fine and mutually beneficial, with well-known exceptions (ie negative externalities)
Thread, continued, 08:05 UTC · 2 likes
But then be honest with the world and yourself about what you’re doing: trying to make money arguably at all costs (ie following the law is only instrumentally valuable for making money bc the government will sometimes catch and prosecute you)
Thread, continued, 08:05 UTC · 2 likes
Ofc this model isn’t perfectly correct - moloch isn’t maximally strong, there is a bit of give, and many, probably most, businesspeople and investors have something like genuine ethical convictions that constrain aggregate/emergent behavior somewhat
Thread, continued, 08:05 UTC · 2 likes
But it pretty clearly says something important about your values/goals and therefore future behavior when you publicly and formally declare to the government…
Thread, continued, 08:05 UTC · 2 likes
that you have a fiduciary to shareholders and if that conflicts with your stated mission, the mission cannot legally be prioritized, or instead whether you have legal permission to consider other values and outcomes
Thread, continued, 08:05 UTC · 2 likes
Also like PBC isn’t all that constraining - it doesn’t mean you have to run a charity or donate proceeds; it just gives you explicit legal permission to not myopically serve moloch in capitalist form
Tweet, 08:50 UTC · 54 likes
Someone good at SEO should spin up http://defectivealtruism.com (and .org if easy) and just make an earnest straightforward case for EA being good with objections and replies
Quote tweet, 17:25 UTC · 163 likes
Dude is a bad actor and cannot be trusted
Zero integrity, afaict zero ethical conviction
Quoting Malcolm Harris (@BigMeanInternet): Oh did Sam Altman make a misleading statement?
Tweet, 18:24 UTC · 16 likes
24 hour twitter fast starting now
End of February 2026. Previous month: January 2026. Next month: March 2026. Back to the archive index.