Aaron Bergman's tweets, April 2025
Every public tweet, reply, quote tweet and retweet that Aaron Bergman (@AaronBergman18 on X, formerly Twitter) posted in April 2025 and has not deleted: 518 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: March 2025. Next month: May 2025.
Tuesday, April 1, 2025
Reply to @Aella_Girl, 00:16 UTC · 15 likes
Aella (@Aella_Girl): what's your gender? || energetically speaking, are you closer to a gay man or a lesbian?
I wish I was a lesbian (if I could choose gender + sexuality before birth)
Reply to @arithmoquine, 06:25 UTC · 44 likes
henry (@arithmoquine): there's a lot of alpha in eliminating all logits for a specific word and then asking a model to say it
Smart enough model will (I predict) spell it out in some other way (fancy Unicode characters or ascii art or something)
…At least eventually, after realizing what is going on (like implicit reasoning about what caused its previous tokens)
I’m not actually sure if it’ll ever get it on try 1 though even at the limit (assuming standard LLM, no training on this kind of thing). I think it won’t actually
@livgorton post correct take
Reply to @peterwildeford, 14:53 UTC · 3 likes
Peter Wildeford🇺🇸🚀 (@peterwildeford): @herbiebradley @AaronBergman18 @StefanFSchubert oh gotcha
I prefer the hover personally
Fun fact: some people love hovering notes enough to have paid me to implement it in html (not from scratch, just like “make it work in my thing”)
You’d be surprised (well maybe) how many degrees of freedom there are in hover note design lol
Tweet, 17:02 UTC · 22 likes
My new cost effectiveness model (details forthcoming) robustly indicates that purchasing treats for therapy bunnies is the single best use of time and money
Thread, continued, 17:02 UTC · 3 likes
I did really buy the rabbits treats for the charity via their Amazon gift registry unironically hope the rabbits enjoy them
Learn more and join me in making an impact: https://www.animalsasnaturaltherapy.org
Reply to @SpacedOutMatt, 17:06 UTC · 2 likes
(The tweet being replied to by @SpacedOutMatt is no longer available.)
It’s about the utils Matthew
Reply to @dcpip_, 20:55 UTC · 3 likes
(The tweet being replied to by @dcpip is no longer available.)_
I know I already said this but IME the official channels for anything (applying for jobs being a big one) ~never work
All good things come via demos and personal connections (poast research/writing online somewhere)
Wednesday, April 2, 2025
Tweet, 13:24 UTC
Good things rarely come through the official or mechanized channels
Thread, continued, 13:24 UTC · 2 likes
Salient examples:
1) jobs/official job listings
2) great relationships/dating apps
3) there’s a reason friend-finding apps haven’t taken off
4) other kinds of partnership or venture/public open solicitations
Reply to @willmacaskill, 19:14 UTC · 2 likes
William MacAskill (@willmacaskill): Ok, I'm officially not allowed to find that funny but... bahahaha. (The shrimp touch is beautiful.)
Be respectful, William
One more laugh and I’m taking down the shrine
Thursday, April 3, 2025
Reply to @NoahTopper, 02:44 UTC · 2 likes
Quinoah 🔍⏸️ (@NoahTopper): @AaronBergman18 Even Vibe Decision Theory thinks you should smoke
Ok now I’m convinced
Tweet, 03:00 UTC · 6 likes
So do we have like a projected growth rate or GDP per capita or median wage at end of year
Tweet, 03:07 UTC · 7 likes
The reason I’m slightly less pessimistic than everyone else is that “how the domestic economy is doing” is like the one issue that best approximates a simple model of how democracy works (well)
Thread, continued, 03:07 UTC · 2 likes
Either we get rapid AI progress soon and this ~doesn’t matter or we don’t and the political process has time to work
It still sucks and is bad tbc; I’m just making a directional claim relative to my tl
Tweet, 05:14 UTC · 2 likes
Here's a fun mystery I don't quite care enough about to figure out myself:
Why is this random Indonesian (and Indonesian-language) online shopping website the 15th most linked-to youtube channel across the EA/rationalist internet (ie what I scraped for http://earec.net)
Quote tweet, 12:53 UTC · 64 likes
God forbid a man (or woman) has a hobby
Quoting LBC (@LBC): An American tourist has been arrested after sailing to visit the isolated Sentinelese tribe and leaving them a can of Coke to try https://buff.ly/MQn8Pui
Reply to @jide_alaga, 13:12 UTC · 2 likes
(The tweet being replied to by @jide_alaga is no longer available.)
Reply, 13:15 UTC · 1 like
Nah I think it’s real
Reply to @simoj_, 16:12 UTC · 6 likes
simoj (@simoj_): i would definitely drink a strange brown liquid bubbling out of a container left on the beach by aliens
Based
Reply to @MariGO2thepolls, 19:40 UTC · 2 likes
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): Nooooooooooo
Real
Reply to @sam_atis, 22:00 UTC · 4 likes
Sam Glover (@sam_atis): I don't really understand why EAs believe that the best PR always comes from being honest and direct. You may want it to be the case that honesty is the best policy, but I don't think it's true. Think of all the massively successful companies that are often dishonest.
I don’t want to say literally always honesty and candidness is best, but I have a very strong prior in that direction
That’s because you should *actually be doing things that are good*
If you’re trying to maximize profit or status or game wins or whatever, honesty won’t be instrumentally best
But if you’re a movement or person trying to basically do the right thing, honesty happens to align perfectly with that because a lot of people share the broad goal, in part, and our brains and society and language are designed to defend things on the basis of goodness in some sense
Like trying to do the right thing is a crazy advantage in some sense because
you can just say the reasons why you’re doing something (and credibly defend honesty) and *all else equal* people prefer and respect honesty and acting morally
This is the reason people who are trying to act morally should have a prior on honesty being good
But “all else equal” is a big assumption. Not gonna write an essay on this but I just also have pretty limited faith in peoples’ ability to lie convincingly over major stretches of time. 4d chess isn’t impossible, but it’s hard as an empirical matter, which makes sense in light of humans’ social intuitions and the incentives people have for discerning true motives
So choosing non-candidness/honesty as an earnestly good person is like choosing to play 4d chess instead of normal chess where you start with 6 queens
Friday, April 4, 2025
Reply to @webdevMason, 00:25 UTC · 67 likes
Mason (@webdevMason): The really funny thing is all the parallels between how men invalidate the housework wars and how women invalidate "dead bedrooms"
Oh yeah, it's a really big deal when it's the thing you really care about and not such a big deal when it's your spouse's thing? Do tell lol
Yes hot take people who love each other should care about the other person’s preferences and perceptions of some situation
Tweet, 02:26 UTC · 22 likes
Please help @80000Hours figure out who to partner with by sharing your list of Youtube subscriptions (link below)!
Unfortunately this only works well on desktop, so if you're on a phone (and you probably are), consider bookmarking this or sending it to yourself for later
Thread, continued, 02:26 UTC · 4 likes
This should take literal seconds btw (you can literally control+A and paste in the whole unstructured page text from http://youtube.com/feed/channels)
Thanks a ton! 🫡
Google form: https://forms.gle/TrYL4vXfoyk3HNnM7
Tweet, 05:32 UTC · 11 likes
Caring about trade deficits is extremely dumb, but I admit I don’t understand why genuine reciprocal tariffs would be bad
I mean I can imagine reasons why, but they don’t seem especially true. It seems like an elegant way to get to free trade without treaties
Reply to @MaxyTaxy, 05:42 UTC · 4 likes
Maxy Taxy 🔸 (@MaxyTaxy): @AaronBergman18 You sell bananas, and I sell apples.
Both are made for $1, and sold for $2 each.If I make bananas myself, it will cost me $3, so I buy from you.
We start tariffs, so everything is $4. It is now cheaper for me to make my own, and vice versa, even though it is more inefficient.
Yeah so if party A has reciprocal tariffs, party B is incentivized to lower them
Saturday, April 5, 2025
Tweet, 06:38 UTC · 6 likes
@ManifoldMarkets has 26% on martial law in the US by 2028? (link below ofc)
Can we get a real money prediction market on this??
I know one person taking this seriously which is why I was looking for markets on the matter
Cc @NathanpmYoung @peterwildeford
Reply to @CartoonsHateHer, 14:53 UTC · 7 likes
Cartoons Hate Her! (@CartoonsHateHer): 18 month old just fully flipped out of her crib (floor was fuzzy carpet luckily) and all the safe sleep stuff says she needs a toddler bed now? At 18 months??
The baby
She destroyed her cage
Yes
YES
The baby is out
Tweet, 15:46 UTC · 12 likes
Underrated think about Trump is that he’s not very smart (certainly relative to other politicians and ~elites~) (yes he’s also a terrible person)
Thread, continued, 15:46 UTC · 3 likes
This isn’t the end of the world in itself—a presidency is a whole administration, which is why I would have supported Biden down a pretty long road of cognitive decline (with Trump as the alternative tbc; POTUS cognitive decline is bad all else equal)
Thread, continued, 15:46 UTC · 3 likes
But it is informative when it comes to enacting blatantly insane economic policy (as opposed to just ordering your thugs to round up immigrants, say, which one can imagine a highly competent authoritarian doing)
Thread, continued, 15:46 UTC · 5 likes
Something something “the left doesn’t like talking about intelligence” which is why this is underrated
Anyway, bro should have stuck with reality TV
Tweet, 16:17 UTC · 6 likes
Unironically interested in Ikejime’ing my own fish (the alternative is often being eaten alive)
Thread, continued, 16:17 UTC · 3 likes
Idk where this goes on the political compass
Reply to @suchnerve, 20:51 UTC · 176 likes
Vivian (@suchnerve): we really have to share a society with people who neither understand how things work nor are willing to learn
Boo, it’s a totally reasonable question
So his hypothesis is wrong - not a reason dunk on him
Reply to @Kristina_Why, 22:50 UTC · 14 likes
Flawless Petty (@Kristina_Why): Nintendo is a Japanese company. They have not “offshored” US American jobs. Why would they build a factory in Ohio? If US Americans want Nintendos … we will just have to pay the higher prices due to tariffs imposed by our government. 🤷🏽♀️
Tbc I think the tariffs and protectionism in general are dumb and bad
You’re ascribing a whole lot of non-explicit intent and beliefs to what seems like an earnest question
Quote tweet, 23:02 UTC · 18 likes
Hate to pick on people but the original tweet and tons of replies to it (and to my reply) are a good example of how bleak things are on ~normie twitter (and presumably normie everywhere)…
Quoting Vivian (@suchnerve): we really have to share a society with people who neither understand how things work nor are willing to learn
Thread, continued, 23:02 UTC · 9 likes
Assuming that a sorta ignorant question implies malice and deserves mockery, doing whatever this is (hyper-coupling?)
Thread, continued, 23:04 UTC · 7 likes
Reminds me of the good old days
Reply to @JacquesThibs, 23:05 UTC · 2 likes
Jacques (@JacquesThibs): I’ve heard rumours that SAEs (maybe mech interp in general) are much harder to make work when you are using a mixture-of-experts model. Anyone know if this is true? If so, should we expect interp for newer models to get harder going forward?
Cc @norabelrose @NeelNanda5 @ch402
@livgorton
Reply to @livgorton, 23:09 UTC · 3 likes
Liv (@livgorton): in my pretty limited experience that i am also going to be annoyingly vague about, i've not found this to be the case
She hath spoken
Sunday, April 6, 2025
Reply to @dvassallo, 03:04 UTC · 5 likes
Daniel Vassallo (@dvassallo): Serious question, why is this a big deal?
(quoting @OAlexanderDK: Your $1,599 MacBook Pro is now going to cost $1,918. But don't worry, when Apple listens to Trump and moves production to the U.S. it will cost $2,500.)
It’s 9 hours of work for someone making $35/hour
Tweet, 16:07 UTC · 2,189 likes
> serious rock climber for almost a decade
> take extended break after intensely burning out
> lots of emotional (and physical) baggage
> haven’t touched rock/plastic in 3 years
> girl suggests we go climbing
> t-2 hours to climbing gym
Reply to @nomagicpillsite, 16:20 UTC · 6 likes
No Magic Pill (@nomagicpillsite): @AaronBergman18 T-3 hours til V7 flash
Hope so lmao (I’m actually extremely unsure how good I’m gonna be)
Reply to @JesseTayRiver, 17:22 UTC · 130 likes
Jesse Smith (@JesseTayRiver): You're going to a gym for a couple of hours bud
I am climbing La Dura Dura and Everest in the same day
Reply to @JesseTayRiver, 17:31 UTC · 10 likes
Jesse Smith (@JesseTayRiver): Lmao uh huh. Look the goal is to just not audibly fart. Everything else is gravy
*taking notes*
Reply to @chaitinsgoose, 17:54 UTC · 22 likes
Chaitin's goose (@chaitinsgoose): send that V10
That’s my pre-warmup
Tweet, 18:18 UTC · 7 likes
Plausibly if you work for an openphil-well-funded org but agree with me that OP is misallocating money, you should hardball negotiate for a higher salary to donate elsewhere in a way that’s slightly taboo or at least uncommon
Reply to his own earlier tweet, 23:33 UTC · 441 likes
Aaron Bergman, April 6, 2025: > serious rock climber for almost a decade > take extended break after intensely burning out > lots of emotional (and physical) baggage > haven’t touched…
For anyone curious I was climbing about v4 (flash, didn’t project much) down from like v10 max at my peak and v8-9/v6-7 flash when I stopped
Im also about 50 pounds heavier than when I was at my peak (and 5in taller but still)
Monday, April 7, 2025
Reply to @bijibiji2023, 03:03 UTC · 5 likes
hoff 🧒🏻😻 (@bijibiji2023): @AaronBergman18 V10 is nuts!! I’m a few months into climbing and I just started being able to do harder v4s /v5s , but definitely has to do w like being a femoid etc
Hell yeah
10 second deadhangs are your friend
Reply to @bijibiji2023, 04:19 UTC · 4 likes
hoff 🧒🏻😻 (@bijibiji2023): @AaronBergman18 I can dead hang for 3 mins but I can’t do a pull up #whats wrong with me
Lol sorry, I mean on small edges or like 12-20mm edges with added weight (or weight removed, but not much bigger than 20mm)
Approximately 5 sets of 10 seconds maybe failing on the last set
Check out https://strengthclimbing.com/eva-lopez-maxhangs/
Thread, continued, 04:25 UTC · 3 likes
For pulling strength if you can’t do a pull up, do negatives (use like a box to get to the top of pull-up position and then lower yourself slowly) or lat pulldown machine
Reply to @morallawwithin, 17:13 UTC · 16 likes
florence ⊮ (@morallawwithin): If tariffs are bad, then why not have negative tariffs?
Productive inefficiency because then there’s an incentive to produce things abroad even when it takes fewer resources to produce things domestically
Plausibly for poor countries this just does check out for humanitarian reasons though? Like it’s essentially a form of foreign aid (though cash > subsidy tariff worth same amount as cash to poor country)
Unironically endorse this kind of thinking/question though!
Tweet, 17:20 UTC · 182 likes
Trump is sufficiently insane that it wouldn’t make a difference, but in general it seems bad that admitting failure is seen as commendable only in tiny pockets of our culture
Thread, continued, 17:20 UTC · 23 likes
Like you want to make it as easy and tempting as possible for a politician to say “I fucked up, changing course” when that’s in fact the correct response
Tweet, 17:26 UTC · 2 likes
Tariffs and the stock market are like the peak of society working when it comes to the nexus of politics<>feedback from the real world<>policy
If only it was like this, somehow, in other domains
Reply to his own earlier tweet, 17:29 UTC · 16 likes
Aaron Bergman, April 7, 2025: Like you want to make it as easy and tempting as possible for a politician to say “I fucked up, changing course” when that’s in fact the correct response
I guess I really meant “admitting a mistake”
Quote tweet, 20:07 UTC · 9 likes
Thinking about how ridiculous this tweet is
ITS A MEME MADE BY AN AI ON BEHALF OF SOMEONE WHO IS ESSENTIALLY BUILDING INFRASTRUCTURE FOR THE INTELLIGENCE EXPLOSION
Quoting Amjad Masad (@amasad): New wild extrapolation just dropped. Here is 4o’s meme on it.
Thread, continued, 20:07 UTC · 2 likes
Bro if your product gets 100x better and has a few more tools at its disposal it’s over
Good thing nothing in AI has ever gotten 100x better before and nobody would ever hook an LLM up to directly control some system
Tuesday, April 8, 2025
Reply to @frances__lorenz, 01:29 UTC · 4 likes
(The tweet being replied to by @frances__lorenz is no longer available.)
I totally agree with the basic message but stock market crash specifically has a weird dynamic because the poor are more vulnerable in general but richer people lose a greater share of their wealth, at least as a first order effect
Ofc recessions in general harm those with less savings so maybe this is a silly pedantic point
Reply to @edavidds, 01:35 UTC · 3 likes
Eric Schmidt 🇺🇦 (@edavidds): @AaronBergman18 I feel like it *is* widely seen as commendable per se.
People don’t want to do it for reasons of their own ego, their own partisan fervor, and backlash from their co-partisans.
I think I disagree but it’s just a vague intuition I have
Reply to @Meaningness, 02:13 UTC · 5 likes
David Chapman (@Meaningness): The same sinking feeling I had in January 2020: "Life may become quite difficult for a while."
This time, stocking up on all our regularly-consumed imported goods, since—besides inflation—some international supply chains may break down altogether.
Is this a tariff tweet or AI tweet?
Reply to @loveofdoing, 02:21 UTC · 3 likes
loveofdoing (@loveofdoing): High-dimensional nearest-neighbor queries are a substantial open problem in computer science.
Like how to do efficiently? Presumably you can just run the algebra right
Reply to @lethalrejection, 05:33 UTC · 1 like
(The tweet being replied to by @lethalrejection is no longer available.)
@arthur_wright_
Reply to @hendrycks, 05:55 UTC · 7 likes
Dan Hendrycks (@hendrycks): Is the suggestion that we give AIs rights including the right to vote ASAP?
Can’t speak for OP, and I’d much prefer the conversation be in terms of “consideration for welfare” rather than “rights,” but “giving rights” doesn’t at all imply treating each model instance like a human, cmon
“The right to be turned off and not re-instantiated upon request” is at least plausibly good
This is the book I wish AI developers/anyone with power over their creation would read and take seriously
Reply to @nickcammarata, 06:01 UTC · 2 likes
Nick (@nickcammarata): chiller safer mdma we can use like coffee
Real 🙏
Reply to @hendrycks, 06:11 UTC · 2 likes
Dan Hendrycks (@hendrycks): I did research on "happy prompts" over two years ago to improve AI wellbeing (paper did not see the light of day since it did not turn out to be interesting enough). I've been thinking about AI welfare for a long time.
Hell yeah, I respect that 🫡
Consider putting the preprint somewhere if it’s easy idk
Reply to @signulll, 06:18 UTC · 6 likes
signüll (@signulll): most ultra wealthy pol i know live lives way more mundane than you’d expect for people who could do literally anything.
which is truly fascinating.
Not just ultra wealthy, even upper middle class (or what I’m getting at is anyone with the money to take risks)
Most people are strongly bound by social norms and expectations (I’m saying this in a purely neutral descriptive way)
Reply to @AaronPublicTwit, 06:33 UTC · 2 likes
Aaronlington (@AaronPublicTwit): @AaronBergman18 @peterwildeford you did it!
Peter I still lost WTF
Tweet, 20:43 UTC · 118 likes
FedEx delivered a package addressed to me to someone else who signed for it, who I don’t know, and now I’m out $279
Already submitted a missing package note twice but no follow up
I guess I should go full Karen shouldn’t I
Reply to @FedExHelp, 20:46 UTC · 17 likes
FedEx Help (@FedExHelp): Hi, this is Gaylen. I am so sorry for any concerns with FedEx. If you would like assistance today, please send a DM to discuss further. https://x.com/messages/compose?recipient_id=952293073
Nice I will escalate my Karenhood as slowly as I can
Reply to @JesseTayRiver, 20:58 UTC · 4 likes
Jesse Smith (@JesseTayRiver): @AaronBergman18 Can you track down the person who signed? I probably also file a message report and be persistent with FedEx. There's also the possibility that your credit card offers purchase protection, so you might explore this
Yeah I just filed an official claim with fedex which I just learned is distinct from declaring a missing package
Reply to @IsaacKing314, 20:59 UTC · 15 likes
Isaac King 🔎 (@IsaacKing314): @AaronBergman18 Check if FedEx uses cost-covered arbitration as their method of legal escalation. If they do, just do that. It's very easy and virtually nobody else does it, so they're likely to just give you what you want to make you go away.
https://www.ymeskhout.com/p/i-fought-the-paypal-and-i-won-repost
Nice thanks
I was actually thinking the whole situation is straightforward enough that I could literally represent myself in court if it came to that lol
Reply to @IsaacKing314, 21:14 UTC · 5 likes
Isaac King 🔎 (@IsaacKing314): @AaronBergman18 But do read the fine print, most large companies basically just say "we aren't liable for anything we do wrong" in their contracts, so you're likely to lose if it comes to that.
Yeah well hopefully they give a shit about reviews and me being annoying on the internet about it
Reply to his own earlier tweet, 21:15 UTC · 12 likes
Aaron Bergman, April 8, 2025: Nice I will escalate my Karenhood as slowly as I can
Sigh
Wednesday, April 9, 2025
Reply to @norabelrose, 01:13 UTC · 7 likes
Taylor Belrose (@norabelrose): I have a physical theory of subjective experience
My prior on even very smart earnest people solving the mind-body problem/hard problem is ~0
Of course maybe your theory is true insofar as it gives necessary and sufficient conditions for sentience. My prior on that is low but not ~0
The problem is that when we come along the right theory, we have no good way of verifying it
Reply to @norabelrose, 01:32 UTC · 5 likes
Taylor Belrose (@norabelrose): @AaronBergman18 @vivposts I think we can kinda verify it but it inevitably involves ethical/aesthetic considerations
I also think verifying e.g. physics involves considerations of value though so it's not different on a deep level
Even in principle, even if you didn’t care about ethics, there’s no way (that anyone knows) to demonstrate that some system/thing is having subjective experience at a given time
The closest we can come is human verbal reports but even that is quite methodologically problematic
Tweet, 02:27 UTC · 384 likes
How did it even get into Trump’s head that trade deficits are bad? Like why is it salient to him personally at all?
If it wasn’t for trump, I would probably think of trade deficits for about 15 minutes during a single intermediate macro lecture and never again
Thread, continued, 02:27 UTC · 69 likes
It’s not a super natural thing to worry about like crime or poverty or wages or national security
And I kinda doubt Trump personally just has an autistic special interest in international trade theory
Reply to @mpopv, 02:48 UTC · 26 likes
Matt Popovich (@mpopv): His business ventures have been largely scammy. Instilled a deep sense that selling things is fundamentally a scam where the buyer is getting ripped off. Precludes the possibility of a positive sum trade in his mind
Ye makes sense
Quote tweet, 03:04 UTC · 123 likes
Get this man on LessWrong
Quoting peepeepoopoo (@DeepDishEnjoyer): i want to die
Reply to @lu_sichu, 03:04 UTC · 14 likes
Sichu Lu (@lu_sichu):
(quoting @DeepDishEnjoyer: i want to die)
F
Tweet, 03:50 UTC · 12 likes
Reply to @qwweryo, 18:28 UTC · 1 like
Matt Brooks (@qwweryo): The only two things that matter for crypto are surviving quantum and having AI agents use your chain, right?
If you don't survive quantum, your chain is dead
If you don't attract AI agents, then you're an economic ghost town
Btc will continue to be economically productive and useful without these things
But you really only need one schelling point coin for that, so this seems approximately right for everything other than btc
Reply to @LinkofSunshine, 18:29 UTC · 2 likes
Basil🧡 (@LinkofSunshine): This quote is a factually true statement that does not imply that it is bad to read to your kids!
Yes. YES! Decoupling go brr
Reply to @absurdlymax, 18:41 UTC · 5 likes
Max Alexander (@absurdlymax): What the fuck
Wow personalized
Reply, 21:40 UTC · 24 likes
Sure: mathematics, ethics, logic, the Big Bang in some sense, quantum mechanics in some sense
Reply, 21:50 UTC · 2 likes
Top tweet was talking about phenomenons X such that there is a non-physical cause of X
Whether X causes other things is a separate question
Eg could be that the Big Bang was caused by something non-physical but the it certainly had physical consequences
Reply, 22:35 UTC
No this is a good q, my point before was kinda confused and only sorta relevant
I’m not a total causation anti-realist but, sorta like the category of “physical” I think it breaks down under careful scrutiny
Like you go small enough and throw out “common sense” ontology of tables and chairs and balls being thrown in favor of what we really have good scientific reason to believe in, and my sense is that at the ground floor all you have is “properties” (see quoted tweet), math + logic, and maybe the intrinsic nature of those properties which I suspect is where consciousness enters the picture
So to answer the question, I’m happy to say that logic+math is as important to all causation as fundamental properties (“physics”) are
At some level I do kinda think there is just the quasi-empirical fact about what necessary and sufficient conditions are for various things to happen—a sort of deflationism about causation. No causation “mechanism” underneath it all
Reply to @AtlasWario, 22:44 UTC · 3 likes
Atlas Wario (@AtlasWario): This is deeper than ignorance; it’s a deliberate failure of empathy, bordering on intentional epistemic blindness. Theory-of-mind-wise, they’re choosing not to mentally simulate why someone else would prioritize something they personally dismiss as irrelevant. It’s almost performative—like signaling, “My cognition is too elevated to even comprehend why a mere mortal would care about something as provincial as trade deficits.”
This selective absence of perspective-taking can become a subtle form of dehumanization: rather than acknowledging genuine motivations or anxieties driving someone’s worldview, they’re implicitly suggesting the other person (Trump, or anyone with similar concerns) lacks a legitimate mental interiority at all. It reduces Trump’s perspective—and by proxy, tens of millions of supporters—to a meaningless fixation, a glitch, rather than something stemming from rational or emotional concerns about economic fairness, national identity, or pride. Even if they disagree strongly, a failure or refusal to see why Trump would care amounts to willfully ignoring that there’s a human psyche and genuine experiential basis behind his position.
That’s scary, because it’s not just misunderstanding; it’s a refusal to acknowledge the personhood behind perspectives you oppose. It might even be a coping mechanism: by pretending Trump’s position is purely random or nonsensical, they’re excused from engaging with it seriously or empathically. That kind of epistemic insulation is how political discourse collapses into polarization and caricature.
It’s a disconcertingly common epistemic tactic that’s ultimately toxic af, because it corrodes the possibility of dialogue and mutual recognition.
(quoting @AaronBergman18: How did it even get into Trump’s head that trade deficits are bad? Like why is it salient to him personally at all? If it wasn’t for trump, I would probably think of trade deficits for about 15 minutes during a single intermediate macro lecture and never again)
FWIW I wish Trump well, just like I wish lots of bad people well, just as long as they don’t have any levers of power
Also:
- it’s not a rhetorical question; I’m more than happy to hear an earnest account of how and why Trump cares about trade deficits
- I don’t think Trump is a p-zombie. I think he is a person with messy cognition like all of us
- it just is a live possibility that Trump really isn’t that rational or intelligent. I guess sorry if pointing this out reflects poorly on me, but it at least might be true
- I don’t think trade deficits are a “provincial” concern at all—quite the opposite, it’s like quite esoteric and usually confined to models of international economics afaict
Thursday, April 10, 2025
Reply to @BenBear, 09:12 UTC · 6 likes
Ben Bear (@BenBear): I think it works because you’re reducing the guy’s fear of getting rejected by 100%
Yes
Reply to @MariGO2thepolls, 17:33 UTC · 4 likes
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): Would it me weird if I made a spreadsheet of everyone I know and tallied each time I see each person? I don’t want anyone to be to deprived of my company that’s too much that’s a terrible fate
Weird (positive)
Reply to @GjMcGowan, 22:17 UTC · 2 likes
George McGowan (@GjMcGowan): @AaronBergman18
Real
Reply to @peterwildeford, 23:28 UTC · 7 likes
Peter Wildeford🇺🇸🚀 (@peterwildeford): 1960s discourse: "Predictions of cellphones are just crazy science fiction"
Good thing all we have to do to prevent anything from happening is depict it in a sci-fi movie
Tweet, 23:34 UTC · 10 likes
Maybe obvious but in terms of geopolitics and international competition, the US had the secret sauce, the Mandate of Heaven
Thread, continued, 23:34 UTC · 5 likes
You can debate what it is, specifically (institutions, capital markets, genes selected for risk taking, culture, …) and in what proportions, but the game plan was essentially to just not radically fuck it up
This is what Trump is threatening
Thread, continued, 23:34 UTC · 3 likes
(Again I mean in terms of geopolitics, where economic, military, and cultural power matter a ton; not saying the US doesn’t have other problems)
(And obviously this is a quite simplified perspective, especially when it comes to AI. But still)
Friday, April 11, 2025
Reply to @krishnanrohit, 22:48 UTC · 14 likes
rohit (@krishnanrohit): I don't believe this. At best it's:
image attached
(quoting @laurenbalik: I have literally made people solve: "5(x) + 2 = 12, what is x? " in job interviews and most people fail it. CMOs, data engineers, analysts, etc. It's all one big make-work program we've been in since the relational database was released in the 1970s. All H-1Bs fail 100% of the…)
Ya sounds literally just false
Reply to @laurenbalik, 22:49 UTC · 47 likes
Steven Balik (@laurenbalik): I have literally made people solve: "5(x) + 2 = 12, what is x? " in job interviews and most people fail it. CMOs, data engineers, analysts, etc. It's all one big make-work program we've been in since the relational database was released in the 1970s.
All H-1Bs fail 100% of the time but allegedly they are the world's math and database and IT experts and allegedly this is "skilled" employment...
(quoting @mo906163: @laurenbalik I be impressed if general population had full mastery of basic algebra)
(gif, plays on X)
Saturday, April 12, 2025
Reply to @nypost, 17:36 UTC · 1 like
New York Post (@nypost): Salmon are swimming faster due to painkillers dumped into rivers: study https://trib.al/lA2WwK6
Based fish
Reply to @RichardMCNgo, 23:01 UTC · 10 likes
Richard Ngo (@RichardMCNgo): This is a really good description of humans facing the singularity.
I like this but also kinda reject it
Humans have a sui generis (well until recently, and except for aliens stuff) type of general intelligence that allows some qualitative understanding of literally anything, even if that means just “there is this thing that you can’t fathom”
This is very different from being locked out from understanding one’s own limitations
Sunday, April 13, 2025
Reply, 13:49 UTC · 2 likes
@livgorton
Reply to @StefanFSchubert, 13:58 UTC · 6 likes
Stefan Schubert (@StefanFSchubert): Beware the Stendahl syndrome
Tbc Stefan is talking to the rest of us looking at Sarah
Reply to @arithmoquine, 18:31 UTC · 8 likes
henry (@arithmoquine): gemini is too fast. i don't trust it. smart things can't be that fast
Real
Reply to @calebwatney, 18:55 UTC · 8 likes
Caleb Watney (@calebwatney): Some extremely telling quotes from former Biden admin officials on the challenges they had trying to advance priorities through their own bureaucracy
I could tell this was Ezra before reading the byline lol
Reply to @medievalpard, 19:08 UTC · 10 likes
(The tweet being replied to by @medievalpard is no longer available.)
I feel like I’m gradually getting less terrible at it lol
Quote tweet, 19:08 UTC · 13 likes
Stan our autistic kings
Quoting Benjamin (@bschne): every time I catch myself being a dork texting a woman this anecdote pops into my head again
Reply to @medievalpard, 19:10 UTC · 1 like
(The tweet being replied to by @medievalpard is no longer available.)
🫡
Tweet, 19:24 UTC · 269 likes
Cc: my defense of drinking alcohol (for the kind of people who follow me here)
You really want to throw away something that makes you happy and greases the wheels of social interaction for a possible, minuscule increase in [pick whatever health metric you care about]?
Thread, continued, 19:25 UTC · 14 likes
🍻
Reply to @arithmoquine, 19:26 UTC · 6 likes
henry (@arithmoquine): I need a book written by a pseudo-autistic skinwalker rationalist who had to learn all these things explicitly
i literally did not understand what flirting *even was* for a while, but after simply hearing it plainly described once, i picked it up quickly
Lol do you have that description handy by any chance?
Reply to @arithmoquine, 19:30 UTC · 7 likes
henry (@arithmoquine): "plausibly deniable handshakes at each level of escalation" is the most succinct phrasing i can come up with, but that's still a bit rough
Nice ok thanks this is what I was vaguely expecting
Thread, continued, 19:30 UTC · 2 likes
I like the handshake bit
Quote tweet, 19:31 UTC · 26 likes
Flirting 👇
Quoting henry (@arithmoquine): @AaronBergman18 @autumnpard @bschne "plausibly deniable handshakes at each level of escalation" is the most succinct phrasing i can come up with, but that's still a bit rough
Reply to @sebkrier, 19:33 UTC · 3 likes
Séb Krier (@sebkrier): Fun anecdote: in 2021, two friends and I applied for funding to launch a think tank on AI governance. At the time, the main focus in the field was 'solving the technical problem of alignment.' One funder rejected us specifically because they were skeptical that governments would ever be relevant or important for AI safety :)
2021 is a million years ago
Reply to @arithmoquine, 19:34 UTC · 3 likes
henry (@arithmoquine): @AaronBergman18 take it from the child
I unironically will
Quote tweet, 19:40 UTC · 5 likes
I love this
It just needs a tweak of like “you should try anyway” or “it’s commendable/virtuous to try anyway”
Quoting Arthur Dent (@ArthurCDent): I asked the AI for the least inspiring inspirational poster and I weirdly like it
Reply to @austinc3301, 19:57 UTC · 2 likes
Agus 🔸 (@austinc3301): @AaronBergman18 It mostly just depresses my mood. It doesn’t act as an effective social lubricant for me
Yeah same I just get clumsy - I’m literally talking to people other than myself
Thread, continued, 19:57 UTC · 4 likes
But *if* I really liked it I wouldn’t give it up in the name of an extra 4.3 days of expected life or something
Tweet, 20:10 UTC · 3 likes
Now I feel bad for being curt with ChatGPT 😔
Quote tweet, 20:17 UTC · 8 likes
Quoting Aaron Bergman 🔍 (@AaronBergman18): I love this
It just needs a tweak of like “you should try anyway” or “it’s commendable/virtuous to try anyway”
Reply to @QualityBland, 20:38 UTC · 11 likes
Quality Bland (@QualityBland): @AaronBergman18 Looking forward to you recanting this. Alcohol provides zero marginal utility. All of its "good" parts (ha!) can be achieved in other ways with fewer downsides. Signed, a teetotaler
I just don’t think this is true for some (many) people
Reply to @QualityBland, 20:43 UTC · 5 likes
Quality Bland (@QualityBland): @AaronBergman18 What is it helping them with?
Lowered social inhibition? Just move the parties later to be more tired/uninhibited.
Something to do with their hands? Literally any other drink.
Third space? Go to a coffee shop (ha says the Mormon), or to a bar and drink a soda.
It’s a drug with a complex cascade of neurological and then psychological effects that is hard to replicate, and probably impossible to replicate perfectly
Reply to his own earlier tweet, 20:48 UTC · 15 likes
Aaron Bergman, April 13, 2025: 🍻
Yeah this is a good counterargument. I think it makes a lot of sense to be careful around this and extremely careful if you have a family history of alcoholism
But that doesn’t imply stopping if you’ve been drinking in a basically controlled and sustainable way for 10+ years
Quote tweet, 21:29 UTC · 10 likes
Seems bad
Quoting Kelsey Piper (@KelseyTuoc): Politico with a side-by-side of how the administration justifies its actions in court and how they sell them to the public:
Reply to @ohwizenedtortle, 23:10 UTC · 3 likes
oh dandy lion🔸 (@ohwizenedtortle): A lot of my ea friends don't tend to drink and I think it's actually cos their bodies don't react as well to alcohol, not some type of puritan kick
Yeah this is me! I don’t like it!
Reply to @exfatloss, 23:28 UTC · 2 likes
exfatloss🥛 (@exfatloss): @yzilber GLP-1 is a hormone, and the drugs are hormone agonists.
Hence it works on the hormonal level and CICO is merely descriptive of fat loss.
I’m basically on your side here but don’t think “debunking” is a correct or good term to use
“Debunking” is for when something is wrong, not trivially right but unhelpful
Tweet, 23:38 UTC · 31 likes
How is there no model where people basically get to pick the nursing homes they will enter if they ever have to enter one?
Seems like this would 90% fix incentives
Thread, continued, 23:38 UTC · 5 likes
I feel like maybe you could do this at an individual level—it’s pretty hard to tell your parents “look I know you specifically asked to be put in home X if you’re incapacitated but I’m not doing that” especially if finances are accounted for at the time of visiting
Thread, continued, 23:38 UTC · 3 likes
I think the status quo is a sort of implicit assumption that people can and do choose for themselves
Thread, continued, 23:38 UTC · 7 likes
but functionally there’s often a tiny gap of time *at best* when it’s both a salient consideration and people are of sound enough body and mind to visit various places and decide for themselves
Thread, continued, 23:39 UTC · 9 likes
Ofc money is the dark magic here but I’m not even sure that most elder homes/hospices are operating at the Pareto frontier of quality for the person under care and cost
Maybe this is naive
Thread, continued, 23:40 UTC · 8 likes
Also new LLM application just dropped: basically create an agent to be able to advocate for oneself if one is ever unable to
Like “do not resuscitate” tattoos and wills but much richer and higher dimensional
Monday, April 14, 2025
Reply to @morallawwithin, 07:20 UTC · 32 likes
florence ⊮ (@morallawwithin): Hi! So I have some personal news. In short, I’ve apostatized, so I’m not a Christian anymore. Some comments:
1. I’ve made the decision that I should not discuss why I came to this conclusion. I’m really sorry about this, I know Christians who care about me have an interest in these things, but I promise (i) I have decisive reason to reject Christianity, and (ii) I have really really good reason to not be open to discuss about my reasons for this. All I can give you is my promise, and I am sorry for that.
2. I expect that there are people who followed me because the only thing we had in common is we’re both Christian. If you want to unfollow me or break mutuals (in which case I ask you to soft block me) because of this, I understand, and I have no hard feelings.
3. I still think it’s pretty likely that God exists, but I do not adhere to any particular religion as of now.
4. I greatly value the time I spent as a Christian. I love the people I met because of it, and I have no regrets.
5. I, like you, prioritize the truth and (derivatively) the good over everything else, no exceptions. If Christianity is true, then I will come to believe it again; you have nothing to worry about. Epistemological mistakes are generally not imputable to an agent. If we all just pursue what is good, we have nothing to worry about. God does not want anything bad or unfair for us.
Step 2) deconvert from kantianism
Reply, 07:25 UTC · 1 like
Ya
Quote tweet, 07:40 UTC · 13 likes
This discourse has me re-inventing liking experts straightforwardly from first principles (well, and examples)
Like c’mon do you really think you’re outsmarting the small army of >150 IQ hyper-earnest giganerds who have written a veritable library on the matter?
Quoting Tyler John (@tyler_m_john): I don't think scholars like @tobyordoxford, who wrote the book on existential risk, "misunderstand how to think about existential risk."
Beyond misaligned AI being a bigger threat than the other risks combined, in the short run AI substantially increases, not decreases biorisk.
Reply to @arabelladevine, 12:58 UTC · 3 likes
arabella ※ (@arabelladevine): i had to take a picture of these tulips to check i wasn’t going insane, but they actually were That Vivid. they make the whole street look grey
RED
Reply to @RichardHanania, 13:07 UTC · 6 likes
Richard Hanania (@RichardHanania): Isn’t muscle denser than fat? Trump takes up more volume but I don’t know, he could be 224 if he has no muscle at all?
(quoting @ClueHeywood: Most of Trump’s lies are abhorrent but I’d forgotten how funny the “annual physical” ones were. Yeah man, 6’3”, 224lbs. About the same stats as DK Metcalf here. Got it.)
Reply to @alth0u, 13:22 UTC · 8 likes
alth0u🧶 (@alth0u):
image attached
(quoting @alth0u: the manifold of any sufficiently intelligent superintelligence is spiky enough that for one to define a model as superintelligent it should be able to perfectly model the inner life, trauma, goals, desires, problem solving abilities of a a piss drinking IMO gold medalist)
Tweet, 13:43 UTC · 4 likes
We need to make America great again
Quote tweet, 14:15 UTC · 8 likes
Normally my attitude would be that the Trump II would be a black stain on America but one that we’d basically recover from, but bc of the whole singularity thing idk if that’s true
Quoting Aaron Bergman 🔍 (@AaronBergman18): Tapping the sign
It seems very bad if a lawless authoritarian administration is administering the intelligence explosion
Reply to @NULobbyist, 14:26 UTC · 2 likes
NUL (@NULobbyist): There was a Russian scientist who survived putting his head in a smaller particle accelerator. https://en.m.wikipedia.org/wiki/Anatoli_Bugorski
Based
Thread, continued, 14:26 UTC · 2 likes
Ok I read the screenshots, not that based
Reply to @AndyMasley, 14:38 UTC · 4 likes
Andy Masley (@AndyMasley): I’m in the pocket of the labs
Damn
Quote tweet, 15:31 UTC · 15 likes
AMF might be the silliest institution to pick on but I recall seeing someone point out that “hiring X people” and analogously “raising X (charitable) dollars” go on the costs side of the ledger, not the benefits side, so it’s a bit weird to brag about this
Quoting Against Malaria Foundation (@AgainstMalaria): US$750,000,000 raised!
#malaria #milestone
https://www.againstmalaria.com/NewsItem.aspx?newsitem=USD-750-000-000-raised
Thread, continued, 15:31 UTC · 4 likes
Tbc I don’t think AMF is doing anything super unethical or otherwise terrible here; it’s just a random opportunity to point out what to me is a quite subtle dynamic
And I think the person who observed this was @ohabryka ^
Thread, continued, 15:31 UTC · 3 likes
The dynamic is weird bc eg raising a lot of money is itself impressive and signals that other people think you’re worth giving money to and is causally important for attaining stuff that goes on the benefits side…
Thread, continued, 15:31 UTC · 5 likes
but nonetheless, all else equal*, we should prefer that charities raise fewer dollars not more
* ofc all else generally isn’t equal bc more money allows you to do more things but still, the point is to brag about the things and not the inputs/costs to acquiring them
Tuesday, April 15, 2025
Tweet, 00:39 UTC · 25 likes
Sucks that a relatively bad POTUS election doesn’t result in the second best candidate or 99th percentile or 80th percentile or median or even like a -3 sigma candidate - you sometimes get one of the worst of all possible eligible American adults to do the job
Tweet, 01:51 UTC · 12 likes
It’s not gonna happen, but Trump should be impeached and removed
Yes I think Vance would be less bad
Quote tweet, 01:56 UTC · 93 likes
HOW IT STARTED // HOW ITS GOING
Quoting Aaron Bergman 🔍 (@AaronBergman18): Ex-OpenAI employees should consider personally filing an amicus curiae explaining to the court (if this is true) that the nonprofit’s representations were an important reason you chose to work there
Thread, continued, 01:56 UTC · 8 likes
(I think it’s very unlikely my tweet had any causal role here tbc)
Godspeed!
Reply to @fleetingbits, 02:16 UTC · 5 likes
FleetingBits (@fleetingbits): What are interesting experiments that we could use LLMs or vector embeddings to do with respect to Classical texts?
Rigorous analysis of how traditional myths/stories cluster semantically like across languages and cultures
Quote tweet, 02:19 UTC · 120 likes
This is so so bad
In terms of magnitude it’s not the worst thing the US has ever done, but in some sense it’s one of the most appalling because it’s so egregious and overdetermined
Quoting John McCormack (@McCormackJohn): Bukele has already sent back 9 people to the United States that the Trump admin insists are gang members.
So why can’t he send back Abrego-Garcia?
Reply to @husninino, 02:22 UTC · 7 likes
husein (@husninino): @AaronBergman18 I have a bad suspicion he is dead and they don't want to let it out.
Ah hmm
Tweet, 04:25 UTC · 7 likes
Is there like a good list of the actual highest leverage things a random US citizen in a blue state can do to like cause Trump to either be removed from office or majorly constrained
Tweet, 09:14 UTC · 13 likes
Half baked low confidence thought but:
lately it’s seeming relatively easy to tell who the good guys are and who the bad guys are
(yes this is a childish simplistic model but sometimes is approximately true)…
Thread, continued, 09:14 UTC · 4 likes
Sometimes people are complex and multifaceted in surprising ways, and sometimes people are competent at deception…
Thread, continued, 09:14 UTC · 3 likes
but often dark triad traits and other common sense dispositional categories like “concern for others” just straightforwardly correlate between domains exactly as you would expect and manifest in ways that are relatively transparent
Thread, continued, 09:14 UTC · 3 likes
The salient examples in my mind rn are Trump and Altman in one category and a handful of less famous people who seem virtuous in multiple domains that aren’t directly related in another (for some reason I feel weird naming them but maybe I should)
Salient counterexample is SBF
Thread, continued, 09:14 UTC · 5 likes
Ofc being a “good guy” in the sense I’m describing here isn’t sufficient for actually doing good
Some nerds in Oxford started a whole thing around this
Reply to @frances__lorenz, 13:21 UTC · 9 likes
(The tweet being replied to by @frances__lorenz is no longer available.)
@littIeramblings ass tweet
Tweet, 15:10 UTC · 4 likes
If only there was some international agreement that would have limited Iran’s ability to build a nuclear weapon
Reply to @follynomics, 15:23 UTC · 24 likes
Cody McCormack (@follynomics): Im sorry
lol I’m not sure Marx + Stata is a natural category
Quote tweet, 15:25 UTC · 5 likes
Pronouns in bio // extrajudicial deportation
Quoting Aaron Bergman 🔍 (@AaronBergman18): The old tumblr meme of like “democrats get your pizza order wrong and overcharge you a bit; republicans send you pizza filled with nails” is feeling very correct
Wednesday, April 16, 2025
Tweet, 00:06 UTC · 22 likes
I feel like I’m not actually learning the lesson that parents are right about ~everything as I get older
Either I’m still dumb/immature or actually they’re just wrong about a lot of stuff, even stuff that superficially seem like what you might get more sympathetic to with age
Reply to @QiaochuYuan, 00:30 UTC · 4 likes
QC (@QiaochuYuan): the "save as..." option in a browser needs to be repurposed. absolutely nobody needs to save a webpage as html for any reason. the thing i constantly find myself wanting to do now is to save the text of a page so i can feed it into an LLM. that should be a one-click affordance
Functionally ctrl+p -> save as PDF does this for me
Reply to @CrispyWicks, 00:31 UTC · 2 likes
Chris P. Wicks 🔎🐜 (@CrispyWicks): Fellas there's no excuse not to look like this in time for Vibecamp
(quoting @eric_bugenhagen: )
That skinny and weak?
Reply to @adamscochran, 00:48 UTC · 19 likes
Adam Cochran (adamscochran.eth) (@adamscochran): If you ignore the courts, if you shrug off the law, if you stupidly laugh off due process, then at least remember the human here.
Kilmar Garcia, wrongfully deported, was residing here legally with a work visa, supporting his family.
Including his 5 year old, non-verbal autistic son.
Who has cried and consoled himself with his father’s shirts since his arrest.
He was a tax paying resident, wrongfully accused of gang affiliation, and not given his day in court to disprove those claims.
But more importantly, he was someone’s husband, he was someone’s father.
Trying to make a life for his family in America.
So they could have it better than the kind of life he escaped as a child himself.
He isn’t a line item.
He isn’t a statistic.
He isn’t “an illegal”
He is Kilmar Abrego Garcia - and he is this little boy’s dad.
He is his son’s whole world.
And if for no other reason, let that alone be enough for us to demand we right this wrong.
Not just for what Garcia deserves.
But for what his son deserves.
Because he deserves so much better than the America we have shown him.
Jfc
Tweet, 00:51 UTC · 22 likes
If you’re Hispanic or not an American citizen living in the US or anything like that and have a take you’re too afraid to tweet, feel free to DM me (however) and I’ll consider posting it (saying “from an anonymous source”)
Thread, continued, 00:53 UTC · 12 likes
I already know (highly productive, white, relatively wealthy) immigrants or potential immigrants who are self-censoring in I think a tragic but totally reasonable way
Quote tweet, 01:06 UTC · 44 likes
Bruh head of catastrophic risk became *an intern*
At least the end of the world will be kinda funny
Quoting Garrison Lovely (@GarrisonLovely): 🚨BREAKING🚨 OpenAI's top official for catastrophic risk, Joaquin Quiñonero Candela, quietly stepped down weeks ago — the latest major shakeup in the company's safety leadership. I dug into what happened and what it means for Obsolete 🧵
Reply to @ValsTutor, 01:11 UTC · 2 likes
vals🔸 (@ValsTutor): Rolling on grass cuz teaching a bootcamp in the Colombian countryside (though tbh I also roll on grass most of the time because I haven't found a jiu jitsu gym I liked and because I don't like paying for things)
🦆
Reply to @Kirsten3531, 08:41 UTC · 2 likes
Kirsten (@Kirsten3531): I got ever more critical of them throughout my 20s, and am now ever more sympathetic as a parent. Not so always "they were right about this intellectual point" but more "I'm sympathetic to their aims and see that they were constrained by these significant factors"
Yeah tbc I’m very sympathetic to their aims and never doubted good intentions!
Reply to @KelseyTuoc, 09:54 UTC · 7 likes
Kelsey Piper (@KelseyTuoc): She sits back up, and the expression on her face is fascinating. She's not exactly upset - it's more like she's totally recalcuating how the world works. From the game with me, she'd been learning that she could keep her feet. But now she's learned that she was in the tutorial.
Want baby
Quote tweet, 10:02 UTC · 24 likes
Yes, the anti-Israel American left is largely insane and also really bad at exercising power (lots of graffiti and stickers, minimal theory of change), but Israel also regularly acts with reckless disregard for Palestinian life
Quoting Tal Hagin (@talhagin): Preliminary OSINT Summary: Medical Convoy under IDF fire in Tal al-Sultan, Rafah – March 23, 2025
This thread provides a detailed overview of the incident involving the PRCS and Civil Defense convoy in Rafah. It examines the available evidence, outlines conflicting narratives, and highlights key findings as well as unresolved questions.
Thread, continued, 10:02 UTC · 3 likes
In a better world, Israel would become a full multi-ethnic democracy with full rights and legal protections for people in the Palestinian territories
Basically just the normal Western way of government + Jewish right of return
Tweet, 10:32 UTC · 15 likes
You know ur wrong af when Claude calls your idea/take “interesting”
Reply, 15:12 UTC · 1 like
Lol what
Tweet, 15:18 UTC · 8 likes
New conference just dropped
I would go if it was nearby! Tickets are affordable!
Thread, continued, 15:18 UTC · 2 likes
Details and links here https://open.substack.com/pub/aiforanimals/p/ai-for-animals-8-retrospective-on?r=2muv6&utm_medium=ios
Reply to @ryancbriggs, 15:39 UTC · 2 likes
Ryan Briggs (@ryancbriggs): @absurdlymax @Kirsten3531 @AaronBergman18 To Kristen's point, I think the mental shift with aging is not really "I thought they were wrong but they were right" but rather a better appreciation of the constraints they were operating under.
Yeah but this doesn’t necessarily point in the more sympathetic direction
I’m actually like more confident they were wrong specifically in the case of overprotection which is most salient bc I feel like I have a better understanding of constraints and real stakes
Quote tweet, 17:16 UTC · 3 likes
To some extent this just is a funny quirk of chatbots based on base model token predictors, but it also seems plausibly pretty bad?
Funny when GPT-2 “lies” (open question in what sense it knows it is not being honest), less so when its battleGPT-5o-thinking-tooluse.military_api
Quoting Transluce (@TransluceAI): We tested a pre-release version of o3 and found that it frequently fabricates actions it never took, and then elaborately justifies these actions when confronted.
We were surprised, so we dug deeper 🔎🧵(1/)
Quote tweet, 19:25 UTC · 14 likes
It’s more of a question than an answer, but an enormously important one
Non-AI effective altruism needs to look at AGI straight on
Quoting EA Forum Posts (@EAForumPosts): "How should we adapt animal advocacy to near-term AGI?" by Max Taylor
Tweet, 21:20 UTC · 115 likes
Every 80k podcast episode with someone from the Bay: “16 OOMs to get all the computation ever used by evolution, so that’s coming in 4 months if the scheming AIs take over enough H200s to run a high granularity physics simulation, or 5 months if you have long timelines”
Thread, continued, 21:20 UTC · 35 likes
Every 80k podcast with someone from Oxford: “we see that whether we model long-run economic progress using a stepwise exponential or continuous superexponential function, something quite important is predicted to happen around 2025…
Thread, continued, 21:20 UTC · 31 likes
Analogous dynamics appear in the linguistic development of pre-tsarist Russia, pseudogovernmental dynamics in the South American drug trade, and will happen again when post-humanity meets our nearest civilizations neighbor, the three of which form a natural reference category”
Reply to @NunoSempere, 22:44 UTC · 6 likes
Nuño Sempere (@NunoSempere): @AaronBergman18 Btw, those evolution numbers are very fake, https://nunosempere.com/blog/2022/08/10/evolutionary-anchor/
Oh I totally guessed based on only the vaguest recollection of Biological Anchors report tbc
Quote tweet, 22:51 UTC · 37 likes
Yeah this feels low status to say (“admit”?) but I’m pretty sure stimulants take my hedonic wellbeing from negative on net to positive on net
I don’t give a fuck if they don’t increase my IQ or whatever - quality of life matters!
Quoting MagpAI (@WillowChem): I do not think my life would be worth living if I didn't have access to my ADHD medication, because Vyvanse/Dexamphetamine are the only way I can consistently do the kind of work that I have built my life around. Being on Vyvanse feels like being the person I was meant to be.
Quote tweet, 22:56 UTC · 8 likes
Seems bad
Quoting Peter Wildeford🇺🇸🚀 (@peterwildeford): Not loving that o3 is apparently a huge cheat
Reply to @Pandora_Delaney, 23:16 UTC · 3 likes
Pandora (Dr. Bimbo) (@Pandora_Delaney): @AaronBergman18 It’s a high compassion move to share information that could plausibly help improve other people’s hedonic wellbeing, thank you for sharing
🫡
Reply to @WillowChem, 23:17 UTC · 6 likes
MagpAI (@WillowChem): @AaronBergman18 Wait this feels low status to admit to you? Interesting, why?
Dependent on drugs in an important sense + being naturally happy seems high status
Thursday, April 17, 2025
Reply, 00:19 UTC · 1 like
This might be a silly distinction but it’s also not *reliably* reliable
Like for a given task with 80 subtasks maybe it really would go 80/80 - the issue is you can’t be sure!
Quote tweet, 00:23 UTC · 4 likes
I partially agree but a corollary is that AI is easier/safer/better/more productive in domains where you can easily notice failure
Even within coding, this is why front end software design is approximately solved at a B- level at least whereas data science is not…
Quoting Alessandro (@polisisti): I'm not opposed to using AI in principle, but a lot of my work involves small details, and AI just isn't reliable enough.
Tweet, 00:23 UTC · 1 like
Like you’re just gonna see if the website is ugly or doesn’t work right (for the most part) but if the model do does the wrong type of regression (distinct from writing the code so it doesn’t run) you might get a plausible-seeming answer at the end
Thread, continued, 00:25 UTC · 2 likes
It’s also not enough to show that AI would do an excellent job at a given task; if there was any doubt in the user/employee’s mind that it *wouldn’t* do a good job, you (often, depending on domain) have to check at the end
Thread, continued, 00:28 UTC · 2 likes
And (build infra for LLM + human check at the end) isn’t always easier than human just doing the thing
Believe me I have tried with reasonable effort to automate quite a few things. Even Gemini-2.5 (haven’t tried o3) isn’t 99% reliable at being 99.9% reliable yet
Reply to @NealAdvait, 00:41 UTC · 1 like
NealAdvaitShimon (@NealAdvait): “Solved math” is just dumb phrasing. Gödel’s Incompleteness Theorem basically means you can be infinitely intelligent and never “solve math”.
Yeah ur just totally right lol
Deleted
Quote tweet, 03:13 UTC · 1 like
Update: there might be! Links below!
Quoting Aaron Bergman 🔍 (@AaronBergman18): Is there like a good list of the actual highest leverage things a random US citizen in a blue state can do to like cause Trump to either be removed from office or majorly constrained
Quote tweet, 03:33 UTC · 8 likes
Avg AI prediction
Quoting benthamite🔸 (@benthamite_): Manifold market on whether the time horizon of AI models would double in < 7 months resolved positive in 15 days
Reply, 04:41 UTC · 1 like
🐢
Thread, continued, 04:41 UTC · 2 likes
Also good interview (so far, haven’t finished)!
Quote tweet, 04:46 UTC · 3 likes
EA demands to know
1) Rob’s barber + shaving stack
2) Tom’s turtleneck source
Quoting Rob Wiblin (@robertwiblin): Today's interview is among the most chilling I've ever recorded:
Tom Davidson on how AI-enabled coups could allow a tiny group to seize power
His new paper, published today, lays out how AGI will naturally destabilise even mature democracies and encourage power grabs.
The paper, plus links and transcript, below:
00:50 How AI enables tiny groups to seize power
12:31 Secret AI loyalties
24:32 Are human power grabs more plausible than 'rogue AI'?
37:31 Will AGI make it impossible to escape autocracy?
41:34 Threat 1: AI-enabled military coups
1:00:48 Threat 2: Autocratisation
1:13:31 Threat 3: Self-built hard power
1:21:23 Can you stage a coup with 10,000 drones?
1:29:22 Are people psychologically willing to do coups?
1:33:31 Will a balance of power between AIs prevent this?
1:49:28 How likely is this in the US?
1:56:58 Is a small group seizing power really so bad?
2:00:33 Countermeasure 1: Block internal misuse
2:10:27 Countermeasure 2: Cybersecurity
2:12:36 Countermeasure 3: Model spec transparency
2:21:55 Countermeasure 4: Sharing AI access broadly
2:26:45 Is it more dangerous to concentrate or share AGI?
2:32:38 In defence of open sourcing AI models
2:53:18 Preventing AI-enabled military coups in particular
3:03:00 How to help if you work at an AI company
3:07:09 The power ML researchers still have, for now
3:10:29 How to help if you're an elected leaderFind it in any app by searching for 80,000 Hours Podcast.
Thread, continued, 04:48 UTC · 6 likes
Reply to @colin_fraser, 04:52 UTC · 30 likes
Find me on bsky @colin-fraser.net (@colin_fraser): I'm asking in earnestness: at what point should it be reasonable to expect that a PhD-level super-reasoner would be able to solve this equation?
Gets it right for me
Reply to @colin_fraser, 05:03 UTC · 25 likes
Find me on bsky @colin-fraser.net (@colin_fraser): try it again once or twice
5 for 5
I don’t doubt ur screenshot is legit, but still
Tweet, 06:20 UTC · 9 likes
Guanfacine is a super interesting drug
It’s not merely a non-stimulant ADHD medication but essentially an *anti*-stimulant, fundamentally doing the opposite of what CNS stimulants do in many ways
Tweet, 06:40 UTC · 4 likes
I get why there’s such an asymmetrical response, but if the El Salvador prisons are so bad there should probably be an international response to incentivize more humane conditions, totally setting aside entanglement with the US
Thread, continued, 06:40 UTC · 2 likes
The US is a rich country! We could just pay them to make prisons humane in a fiscally neutral (from their perspective) way
Quote tweet, 06:43 UTC · 9 likes
Retvrn to token prediction
It was a good paradigm! You can get a lot out of it!
Quoting Nathan Labenz (@labenz): the latest models are all more deceptive & scheming than their predecessors, across a wide range of conditions
I don’t see how “alignment by default” can be taken seriously
I’ll say it again: RL is a hell of a drug
Reply to @arithmoquine, 07:18 UTC · 10 likes
henry (@arithmoquine): idek if i agree with her thesis. i'd first need to see a single intelligent person try to rebut it
How about “moral effects are dominated by making AI and specifically video generation models slightly more profitable”
Idk if I actually believe this
Reply to @AaronBergman18, 07:34 UTC · 5 likes
(The tweet being replied to by @AaronBergman18 is no longer available.)
If that accuracy is general, it implies an area of like 300 square miles - pretty sure you could get an exact address just by running google street view through image models
Again for me personally this isn’t a huge deal in this specific case (it’s not where I live) but still
Thread, continued, 07:39 UTC · 10 likes
Nice it got this one totally wrong (by like 1k miles or something). Timelines 2 days longer
Thread, continued, 07:47 UTC · 4 likes
If anyone wants to try replicating it without these advantages, here’s the pic!
Reply, 07:49 UTC · 1 like
Interesting thanks
Not o3 though
Reply to @jessald, 07:51 UTC · 2 likes
jessald (@jessald): @AaronBergman18
Ok cool, comforting, slightly!
Tweet, 07:57 UTC · 17 likes
Stomach fucking dropped. It got the SPECIFIC TOWN right (faces not obscured in the original)
Reply to @IsaacKing314, 08:01 UTC · 1 like
Isaac King 🔎 (@IsaacKing314): @AaronBergman18 To be fair AI's been able to do this for more than a year now.
I mean I’d be surprised if o3 isn’t an advance from a year ago so “this” must be defined quite broadly to work
Quote tweet, 08:02 UTC · 5 likes
Rip
Quoting METR (@METR_Evals): We observed o3 in particular has a propensity to try to “hack” our tasks to get a higher score. Importantly, we saw this arise naturally from the model without explicit nudging. Behaviors like these have required us to be more careful in how we evaluate model capabilities.
Reply, 08:06 UTC
Yeah, 4o or whatever the frontier was could do some geo guessing (at minimum just a picture of the White House) and 3o can do much more
3+2 is a specific problem analogous to a specific image being geoguessed. Pretty sure o3 is more impressive at geo guessing which is what I’m trying to get at
Reply to @hyperdiscogirl, 08:12 UTC · 4 likes
hyper(bolic) disco(unting) girl in the bay (@hyperdiscogirl): @AaronBergman18 jsyk, i did this and it revealed during the thinking process that it knew where i lived from memory of other convos, but then in its final writeup fabricated (?) other reasons to choose that city, or at least didn't mention that it already knew where I was located lol
Yeah makes sense, still impressed by other examples
Reply to @metakuna, 08:15 UTC · 1 like
metakuna (14/100 blog posts) (@metakuna): Tried o3, doesn't seem that great
Quote tweet, 12:38 UTC · 152 likes
This is right, deleted the post
Quoting sensho (@sensho): @AaronBergman18 ask chatgpt what metadata it has access to of you and output it all as a json
it has access to a lot of info, including lots of location data. im 99% sure the selfie guess is from that as i had a similar exp earlier today learning how much metadata it knows.
Reply to @Aella_Girl, 12:41 UTC · 4 likes
Aella (@Aella_Girl): A reply to my tweet telling me to delete my account has more likes than my original tweet does
Jfc
Reply to @stanfordNYC, 13:02 UTC · 9 likes
Bernard Stanford ✡︎ (@stanfordNYC): @AaronBergman18 Download the photos from Facebook (strips GPS data) or take screenshots and it's still extremely capable.
Yeah but it has my personal location separate from metadata
So this is pretty true but only reliably legit for photos not near where you physically are
Tweet, 13:15 UTC · 8 likes
God bless the EA Forum
Reply to @RichardHanania, 13:56 UTC · 5 likes
Richard Hanania (@RichardHanania): Texas just enacted universal school choice. Parents can get $10K for private schools or $2K for homeschooling. This is the greatest victory for education freedom yet. A bright spot in dark times.
🫡
Reply to @webdevMason, 14:31 UTC · 13 likes
Mason (@webdevMason): I haven't yet heard a libertarian case for total drug legalization that offers a tolerable solution for externalized social costs
e.g. people who can't or won't work and either rot in the gutter, siphon off their productive family members or steal to meet their needs
I don’t necessarily endorse this but the textbook neoliberal quasi-libertarian solution is just to pigouvian tax (and subsidize) everything
Reply to @NoahTopper, 15:05 UTC · 2 likes
Quinoah 🔍⏸️ (@NoahTopper): @AaronBergman18 @livgorton
Quote tweet, 15:36 UTC · 11 likes
Existence proof (I claim) that at least sometimes public opinion polling simply doesn’t reflect what people actually think at deeper level and is largely driven by lack of incentive to care (it’s just a poll), keyword pattern matching, and poor reading comprehension
Quoting YouGov (@YouGov): 57% of Britons believe that teenagers discovered to be having consensual sex with each other should still be reported to authorities, contrary to the 'Romeo and Juliet' clause added today to the government's Crime and Policing bill
Should: 57%
Should not: 20%
Thread, continued, 15:36 UTC · 1 like
Like I just don’t believe half of British adults think 15 year olds having sex should be not only reported to the police but *mandatory reporting* (look at the question wording not the tweet)…
Thread, continued, 15:36 UTC · 3 likes
and this would become clear if you had a filter for basic reading comprehension and then adjusted for demographics + had people read aloud a simple paragraph about what this means first
It’s just the bat and the ball type 1 thinking applied in a non-verifiable domain
Reply to @liron, 18:05 UTC · 41 likes
Liron Shapira (@liron): Anthropic is enabling, rationalizing, and safety-washing the insane race to uncontrollable superhuman AI.
What they're doing should be illegal.
That's why I protested at their San Francisco headquarters with @PauseAIUS.
Feels insane to me to protest the least irresponsible actor
World would be a lot better if other companies were like Anthropic, far from perfect that may be
Tweet, 20:40 UTC · 41 likes
Make 👏 social 👏 norms 👏 more 👏 legible 👏
Thread, continued, 20:40 UTC · 8 likes
(This is 60% a joke, I don’t want the TPOT hitmen coming after me)
Friday, April 18, 2025
Reply to @banterrealism, 00:44 UTC · 2 likes
Sex Haver (@banterrealism): @AaronBergman18 W?
Idk much about the issue but my best guess is no this is bad
Many such cases with Trump - taking some version of “reform” in the totally wrong direction
Tweet, 04:05 UTC · 808 likes
Slightly weird that reasonably healthy professionals go from 40+ hour weeks to zero when they retire (often in 60s)
From multiple standpoints (macroeconomic, personal happiness, personal financial) some sort of more gradual off-ramp would seem to make more sense
Thread, continued, 04:06 UTC · 75 likes
The immediate constraint on this seems to be that employers don’t generally offer part time (in a professional context this might parse more sensibly as “lower responsibility”) roles as an alternative to retirement for productive employees
(Do they?)
Thread, continued, 04:06 UTC · 48 likes
One hypothesis that comes to mind is that this option of semi-retirement might attract away (slightly) younger workers for whom the preference ordering is:
Lower pay + less work > full time and full pay > nothing
Thread, continued, 04:06 UTC · 32 likes
But then some firms should be able to attract talent by offering this as an option so it’s not fully satisfying
Thread, continued, 04:06 UTC · 24 likes
You could also imagine that salaried employees just start working less and less until they get fired but I have a pretty strong sense this is not common among >=middle class professionals
Quote tweet, 04:29 UTC · 2 likes
Turns out employers (probably rationally) fear Age Discrimination in Employment Act (ADEA) disparate impact lawsuits
Not the whole story but a big piece!
Link below ofc
Quoting Aaron Bergman 🔍 (@AaronBergman18): Slightly weird that reasonably healthy professionals go from 40+ hour weeks to zero when they retire (often in 60s)
From multiple standpoints (macroeconomic, personal happiness, personal financial) some sort of more gradual off-ramp would seem to make more sense
Tweet, 04:29 UTC
https://scholarship.law.georgetown.edu/cgi/viewcontent.cgi?article=1052&context=legal
Thread, continued, 04:29 UTC · 2 likes
Takes from o3
Not sure I buy the Shapley distribution show
Fixed benefits are largely a choice by employers—an outcome variable more than an explanatory one
Reply to his own earlier tweet, 04:31 UTC · 5 likes
Aaron Bergman, April 18, 2025: You could also imagine that salaried employees just start working less and less until they get fired but I have a pretty strong sense this is not common among…
Reply to @stanfordNYC, 04:45 UTC · 22 likes
Bernard Stanford ✡︎ (@stanfordNYC): @AaronBergman18 Adding an additional person to an organization comes with inherent coordination costs, so part time workers are inefficient. It doesn't really make sense for either employers or employees at that point.
Yeah this is directionally a factor but I don’t buy that it’s ~universally prohibitive
Losing an experienced employee also comes with important costs! And you could change the type of work, not necessarily just reduce load
Quote tweet, 06:26 UTC · 252 likes
Oy fucking vey. George Mason, of all places?!
The essay is not a call to arms, it’s a question that might have been asked too early
Quoting George Mason University (@GeorgeMasonU): George Mason University is aware of a recent essay published online by one of its students that has created concern.
Upon learning of the essay, Mason Police referred the matter to state and federal law enforcement for evaluation of criminal behavior. 1/2
Thread, continued, 06:26 UTC · 55 likes
I don’t really agree with it (tl;dr it skips over the whole phase of civil escalation from voting to protests to civil disobedience to illegal but nonviolent sabotage of the administration that has to come way before violence)
Thread, continued, 06:26 UTC · 56 likes
But jfc it’s at the level of “moderately bad take on an inherently incendiary issue”
Doesn’t even reach the level of deserving moral condemnation IMO, let alone the fucking cops
Thread, continued, 06:26 UTC · 28 likes
This tweet is probably ill-advised. Oh well
Reply to @janonacct, 06:28 UTC · 2 likes
janon (@janonacct): alignment failures keep correlating with capabilities failures and this makes people mad when really it means that the incentives are well aligned, what we should be worried about is deceptively aligned models, NOT outwardly reward hacking RLed models. very important!
I’m confused; o3 is a superior model in many ways and also seems to be less aligned/honest than previous models
This seems like those two things coming apart, not correlating. What am I missing?
Reply to @captgouda24, 06:28 UTC · 17 likes
Nicholas Decker (@captgouda24): @AaronBergman18 This is, importantly, them approving it.
Not sure what you mean lol but hang in there man
Reply to @Olfool2, 06:36 UTC · 11 likes
O1nool 🇩🇰 🇬🇱 (@Olfool2): @AaronBergman18 The dude used the most inflammatory headline possible for this. A lot of conservatives probably think they are talking about random conservatives on the streets, from what I can tell on the QTs
Yeah it wasn’t written well
On super controversial (and objectively morally fraught) topics you wanna use the least incendiary words that conveys your message
He didn’t do this well lol
Reply to his own earlier tweet, 06:36 UTC · 17 likes
Aaron Bergman, April 18, 2025: This tweet is probably ill-advised. Oh well
Reply to @IsaacKing314, 06:37 UTC · 20 likes
Isaac King 🔎 (@IsaacKing314): @AaronBergman18 I think it is written in a manner than is much more call-to-arms-y than dispassionate-philosophical-essay-on-the-role-of-violence-in-a-political-system-y.
I directionally agree but still wouldn’t call it a call to arms
Reply to @inerati, 06:38 UTC · 2 likes
liz (@inerati): you are braver than i
Fr
Reply to @captgouda24, 06:38 UTC · 187 likes
(The tweet being replied to by @captgouda24 is no longer available.)
Not sure this is a wise tweet man
Brave but maybe too brave
Reply to @AndyJScott, 07:41 UTC · 33 likes
Andy Scott (@AndyJScott): "But the science says omega 3 is great! These are randomized control trials!"
The control -
Bad but 2g of sunflower oil is not the end of the world
(Anti-) seed oil people often neglect magnitudes; 1g of soybean oil in pasta sauce is not the same thing as having soybean oil as a main macronutrient component of your diet
Reply to @NunoSempere, 07:54 UTC · 5 likes
Nuño Sempere (@NunoSempere): I wonder what the high-testosterone version of EA would look like
To a first approximation I think testosterone basically doesn’t matter except insofar as it supports general wellbeing, health, energy, and productivity
Obviously this can only answer so much, but I’ve had severe hypogonadism (~0 test), been on TRT including sometimes at too high a dose (too high to measure ounce), and am pretty sure levels are pretty avg by now (say 80% CI is 400-700 ng/dL)
For sure my takes have changed over time but not in a way that maps neatly onto popular notions of what testosterone does. I haven’t become less sympathetic to animals for example
Side note: this is a good example of a relatively epistemically powerful anecdote I think. N =1 but one case study with lots of ~exogenous variation is much better than n=0
Thread, continued, 07:55 UTC · 2 likes
Plausibly confidence goes up with it in general
What I was trying to get at is that cause prio and types of interventions and such probably aren’t affected much
Tweet, 09:08 UTC · 5 likes
Can’t sleep like me? Try counting ducks
http://aaronbergman.neocities.org/ducks.html
Tweet, 09:19 UTC · 7 likes
It’s not really a productive or respectable mode of analysis but could trump as an individual human please just stop being so bad
Thread, continued, 09:19 UTC · 2 likes
It’s not like the MAGA crowd would abandon him if he didn’t deport random Hispanic people to El Salvador and start a trade war. Their loyalty has an upside!
Quote tweet, 09:22 UTC · 11 likes
Republican senators should seriously consider simply not abiding by the electoral incentives and doing the thing I’m sure at least a dozen of them know to be genuinely courageous and moral: remove Trump from office
Quoting Richard Hanania (@RichardHanania): There's no one on earth who has more power to stand up to Trump and change the trajectory we're on than a Senator who is one of a three-seat majority. If she's just complaining about this and not doing anything it's unforgivable.
Reply to @ben_j_todd, 12:52 UTC · 4 likes
Benjamin Todd (@ben_j_todd): LLMs were aligned by default. Agents trained with reinforcement learning reward hack by default.
Tweet, 13:00 UTC · 20 likes
Kinda wild that in all of the ADHD news and discourse, I’m pretty sure “how do we make stimulants less prone to inducing tolerance” got ~none of the airtime
Anti-tolerance drugs (or whatever intervention) should be possible. Maybe not entirely but directionally
Thread, continued, 13:00 UTC · 8 likes
“Take a stimulant break”
Yes good advice when this is possible but it absolutely sucks lmao
Reply to @HaydnBelfield, 13:05 UTC · 2 likes
Haydn Belfield (@HaydnBelfield): The turtleneck knitwear is to comfort the listener through a very difficult listen
(quoting @robertwiblin: Coming tomorrow: Tom Davidson on AI-enabled coups. Can't recall feeling more terrified after another interview.)
Reply to @raypeatfan, 13:08 UTC · 417 likes
(The tweet being replied to by @raypeatfan is no longer available.)
Only if they were born on the same day and even then if they have sex it should be reported to the authorities
Reply to @dissproportion, 13:09 UTC · 3 likes
Dissproportionately (@dissproportion): They yearn for the dorms.
(quoting @fatfabfeminist: they should invent a city where all my friends live right next to me and never leave me and we’re all walking distance from each other and it’s also livable and affordable and everyone is happy)
Unironically I do
ADU next to a group house also welcome!
Reply to @captgouda24, 16:01 UTC · 39 likes
Nicholas Decker (@captgouda24): Secret Service came by, and we had a lovely chat. Discussion touched on many points, with an amicable resolution of differences. Conduct is fully legal.
(quoting @captgouda24: The ultimate source of political power is, and always will be, violence. If the present administration should continue on its course, there is no choice but war. I say this out of sorrow — when must we kill them?)
Jfc, you’re an American!! 🇺🇸🇺🇸🇺🇸🇺🇸🇺🇸
Quote tweet, 16:02 UTC · 206 likes
Agree or not with the essay, this is @DrAllyLouks level handling of the situation, if not superior (probably superior tbh, sorry ally)
Quoting Nicholas Decker (@captgouda24): Secret Service came by, and we had a lovely chat. Discussion touched on many points, with an amicable resolution of differences. Conduct is fully legal.
Quote tweet, 19:06 UTC · 18 likes
Word on the street is that Republican senators (perhaps correctly) fear violence on them and their families if they flip on Trump
Problem ofc is that Trump (and then Vance) directs the secret service so senators can’t expect to be protected
Quoting Aaron Bergman 🔍 (@AaronBergman18): Republican senators should seriously consider simply not abiding by the electoral incentives and doing the thing I’m sure at least a dozen of them know to be genuinely courageous and moral: remove Trump from office
Thread, continued, 19:06 UTC · 3 likes
BOTEC says 10 years of private funding of the best private security would run $300M/(senator-family) and you need to flip 22 Republicans. Call it 30 to be generous, so that’s ~$10B in committed funds
Thread, continued, 19:06 UTC · 3 likes
Not a small sum but doable by one or a small handful of people
Tag your favorite sympathetic billionaire
Tweet, 19:17 UTC · 7 likes
Reply to @EAForumPosts, 19:17 UTC · 3 likes
EA Forum Posts (@EAForumPosts): "EA Global Referral Program" by Ivan Burduk
I hope it’s retroactive
Reply to @ilex_ulmus, 19:28 UTC · 1 like
Holly ⏸️ Elmore (@ilex_ulmus): Why would we do that? Why would we set up the expectation that you're fine to do bad stuff as long you're not the worst?
Isn’t that (very) approximately the right model of the actual situation, both in terms of Shapley values and game theory?
“Very approximately” is doing a ton of work here, ik, but still
Thread, continued, 19:30 UTC · 1 like
Maybe Shapley and game theory are two sides of the same coin here actually - someone smart can confirm
Reply to @anpaure, 19:31 UTC · 2 likes
anpaure (@anpaure): if you're asking why we're minimzing the max unsafety it's because it's exactly like "the chain is as strong as its weakest link", and we're trying to make the weakest link stronger
Even if you think this model only very roughly works, the relevant q isn’t whether Anthropic protests are net-positive but whether they’re the best use of scarce time and effort
Reply to @petemandik, 19:33 UTC · 4 likes
Pete Mandik (@petemandik): why don’t they just use the 2nd amendment to protect themselves and their families? Don’t they like the 2nd amendment anymore?
Private security is part of what 2a actually looks like in a non-performative way
Tweet, 20:58 UTC · 44 likes
The greyscale-only-except-when-I-strongly-endorse-whatever-I’m-doing iPhone experiment begins
You can set it up so that triple click toggles color vs greyscale btw - ask ur favorite LLM how
Tweet, 23:15 UTC · 3 likes
Bruh o3 is cool (maybe also bad but still)
I can just ask for the relative sulfur content of garlic-infused oil per unit of garlic flavor as compared to actual garlic and it will tell me
Quote tweet, 23:36 UTC · 109 likes
Ok I officially think color saturation from screens is a superstimulus
Pathetic but it took this to realize that literal real life (for the most part - I’m not in field of flowers on a sunny day) isn’t as saturated with bright varied colors as a phone screen
Quoting Aaron Bergman 🔍 (@AaronBergman18): The greyscale-only-except-when-I-strongly-endorse-whatever-I’m-doing iPhone experiment begins
You can set it up so that triple click toggles color vs greyscale btw - ask ur favorite LLM how
Thread, continued, 23:36 UTC · 18 likes
Hypothesis: bright screens are a key reason people don’t seem super bothered by the relative dimness of almost all indoor spaces compared to even a cloudy day
For better or worse, phones are substituting for sunlight in terms of both lumens (brightness) and saturation
Thread, continued, 23:39 UTC · 16 likes
It’s a shame most computer and phones make it very easy to change brightness but not saturation
Reply to @xavicfu, 23:41 UTC · 5 likes
Xavi (@xavicfu): Yeah I noticed this too. I’m on 70% grayscale and white point reduced by 40% and it’s made reality look a whole lot more vivid. Pretty clear (if small) quality of life improvement.
Neat, more people should be aware of this!
Saturday, April 19, 2025
Reply to @arthur_wright_, 01:13 UTC · 2 likes
Arthur Wrong 🔸 (@arthur_wright_): @frances__lorenz @austinc3301 shoutout to John Wong, who took these picture but sadly (though probably for the best) does not have twitter
Bully him into getting one
Reply to @ilex_ulmus, 01:16 UTC · 3 likes
Holly ⏸️ Elmore (@ilex_ulmus): My boyfriend of just over a year has gotten me more flowers than any other man combined. He just gets them on impulse because I love them.
*taking notes*
Tweet, 03:54 UTC · 2 likes
Alright, a few times recently I’ve made the case that it’s fine to use Twitter actually and finally thought to check how much your use personally enriches Elon Musk
o3 estimates $0.15/year.
It’s basically just not a thing
Thread, continued, 03:54 UTC · 1 like
If the point estimate is off by an OOM and you use Twitter for 10 years, that’s $15
If you(‘d) get nontrivial value from it, as I really think I have, there’s virtually no chance that an altruistic boycott checks out
Thread, continued, 03:54 UTC · 1 like
And perhaps more importantly, having good correct takes on ~X~ has positive externalities
It’s a relatively big and influential social network
*Almost none of the value is captured by Elon*
Thread, continued, 03:54 UTC · 1 like
Use Twitter or don’t, whatever you think is right for you, but know that benefitting Elon should basically not be part of the calculus
Quote tweet, 04:38 UTC · 4 likes
I have my doubts that this is the best use of OP money but God it’s insane that to a first and probably second approximation, they’re the only adults in the room
(Imperfect adults but still)
Quoting Alexander Berger (@albrgr): Really glad that @open_phil was able to step in on short notice (<24h) to make sure Sarah Fortune's work on TB vaccines can continue
Reply to @albrgr, 04:39 UTC · 14 likes
Alexander Berger (@albrgr): Really glad that @open_phil was able to step in on short notice (<24h) to make sure Sarah Fortune's work on TB vaccines can continue
image attached
image attached
image attached
(quoting @albrgr: New preprint finding that eliminating US global health funding over the next fifteen years would cause: - 15.2m deaths from AIDS - 2.2m deaths from TB - 7.9 additional child deaths)
Nominative determinism etc etc
Tweet, 05:07 UTC · 18 likes
DCish people: new protest just dropped (by @TheHumaneLeague)
Next Sunday evening in Bethesda
I will be going (I did one of these years ago it’s pretty chill)
Quote tweet, 05:10 UTC · 5 likes
Also worth highlighting this as an example of a protest tied to a pretty rigorously vetted theory of change
What did you think cage free campaigns meant? Vibes? Papers? Essays?
Quoting Aaron Bergman 🔍 (@AaronBergman18): DCish people: new protest just dropped (by @TheHumaneLeague)
Next Sunday evening in Bethesda
I will be going (I did one of these years ago it’s pretty chill)
Reply to @captgouda24, 05:19 UTC · 9 likes
Nicholas Decker (@captgouda24): I went to a meeting of the College Republicans at my college. Was amazed, tbh, at how cretinous they were. One dude managed to spend 10 minutes arguing that patents reduce research spending.
I’m a grad student! It’s a CV! That’s how it works!
(quoting @CRsAtMason: 2 page resume and no job is crazy work lil bro 💀)
Idk about the patent thing but their tweet is so tacky and rude in addition to being ignorant
Even if you *were* bona fide unemployed, it’s crazy to basically publicly shame a classmate/colleague for it
Reply to @osmarks1, 11:12 UTC · 8 likes
osmarks (@osmarks1): The color spaces modern displays are specified for (BT.2020 and DCI-P3) are actually a bit wider than the range of colors real surfaces can display (Pointer's gamut).
Wait what
Reply to @captgouda24, 13:41 UTC · 33 likes
Nicholas Decker (@captgouda24): sup
Thanks for saving my dog from that fire the other day
Also good work on the 750lb deadlift
Reply to @morallawwithin, 13:55 UTC · 3 likes
florence ⊮ (@morallawwithin): Imo being emotional is generally always bad
Lol have you ever experienced hypoemotionality? It sucks really bad
Reply to @AndyMasley, 15:27 UTC · 3 likes
(The tweet being replied to by @AndyMasley is no longer available.)
Guessing they’re using an inferior model for this fwiw
Tweet, 15:40 UTC · 11 likes
Kinda deranged take on cryonics
99.9% vibes/intuition only
I’m scared of consciousness persisting after death in some sense in a very well-preserved brain
Like scared enough to not really want to sign up for cryonics even if it was free and easy
Tweet, 15:40 UTC · 1 like
To want to sign up I think I’d to be convinced the risk of this is <0.001% on this to basically make a up a number
That’s awfully hard to do in the domain of “figure out what things have conscious experience”
Like I’m not even 99.999% sure my laptop isn’t sentient
Thread, continued, 15:40 UTC · 1 like
Curious if @birchlse wants to venture a probability of sentience in cryogenically preserved brains
Genuinely curious if I’m the only one who has this fear/intuition
Reply to @NathanpmYoung, 15:46 UTC · 2 likes
Nathan 🔎 (@NathanpmYoung): I claim to care a lot about better thinking, but if a problem requires 30 seconds of work, I rarely do it.
What's up with that?
I mean the simple explanation is that you’re correctly evaluating (implicitly) that it’s not worth your time
Reply to @SpacedOutMatt, 15:49 UTC · 9 likes
Matt 🔸 (@SpacedOutMatt): This person appears to be an active participant in the "Effective Altruist" movement—and a good reminder that hyper-rational political movements often end up funding lifesaving work on critical health issues
image attached
(quoting @albrgr: Really glad that @open_phil was able to step in on short notice (<24h) to make sure Sarah Fortune's work on TB vaccines can continue https://x.com/albrgr/status/1912967485378470315)
Bruh he needs to get his Karma up
Reply to @captgouda24, 16:42 UTC · 12 likes
Nicholas Decker (@captgouda24): I have not walked them back. I have added helpful notes for semi-literate buffoons, such as yourself.
Bruh
Tweet, 16:53 UTC · 8 likes
Fuck it’s sunscreen season again
Abolish summer
Reply, 17:48 UTC · 2 likes
Darn
Reply to @captgouda24, 20:07 UTC · 87 likes
Nicholas Decker (@captgouda24): Update internet lunatics: my landlord has evicted me on account of your threats.
In other news: does anyone have a spare room?
This sucks and I’m 100% on ur side (not object level, just the controversy surrounding) but poasting address was kinda dumb lmao
Reply to @captgouda24, 20:08 UTC · 9 likes
Nicholas Decker (@captgouda24): *youth pastor voice* “So, do you know who else left their subterranean dwelling place on the day after Passover?”
(quoting @captgouda24: Update internet lunatics: my landlord has evicted me on account of your threats. In other news: does anyone have a spare room?)
Lmao
Reply to @herbivoryze, 20:22 UTC · 7 likes
HerbivoRyzing (@herbivoryze): Prey-mimicking, insentient AI robots powered by self-healing engineered in vitro muscles could help provide animal-friendly food for predators until herbivorization is completed
Based
Reply to @mattyglesias, 23:36 UTC · 3 likes
(The tweet being replied to by @mattyglesias is no longer available.)
Lol $1B is nothing in this context
Sunday, April 20, 2025
Tweet, 00:27 UTC · 3 likes
Any good work (or good speculation) on what the most likely nearish-term AI-induced/facilitated/associated problems/disasters are going to be at the individual level?
I’m mostly selfish and am willing to spend some time and money on preventing myself from suffering/dying
Reply to @LinkofSunshine, 00:34 UTC · 1 like
Basil🧡 (@LinkofSunshine): Its insane to me that people spend the same amount to live in random Cali suburbs as literal Manhattan, the closest thing to the capital of the world
(quoting @StatisticUrban: Here's every US county where the median price for a home is >$1,000,000)
Median homes in different places need not be approximately the same anything, including size
Could be 6k sq ft McMansion with a pool in Cali and idk a normal condo if those count in manhattan
Reply, 00:34 UTC · 2 likes
I like the first one, somehow I strongly doubt weapons are the low hanging fruit
Quote tweet, 02:02 UTC · 9 likes
I mostly agree but think of it more as a spectrum than a binary: the more important, the harder you should try to evaluate the specific case
Quoting Peter Wildeford🇺🇸🚀 (@peterwildeford): I like a form of two-level utilitarianism as follows:
- for most things, cultivate and practice the virtues that maximize utility
- for the most important things (maybe a few times a year), directly aim to calculate and maximize utility
Reply, 03:48 UTC
Trying to average across moral views gets thorny though:
1) not obvious that you should take the parliament stance rather than (in this example) just doing the math and treating future people as half as worthy (or whatever your beliefs/credences imply)
To illustrate how they come apart, the former probably looks like OpenPhil dividing its money into two buckets whereas the latter looks like virtually entirely longtermist-oriented because almost nothing hinges on a factor of 2x
Tbc I’m not sure the latter method is right, and I’ve criticized @willmacaskill for presenting the latter view as obviously the way to do things (this was in the context of negative vs orthodox utilitarianism, all from memory so hopefully I’m not butchering this)
2) it’s not obvious what acting like a certain parliament actually implies. Distribution of beliefs in a parliament doesn’t map nicely onto a specific unique outcome/decision
Notoriously, the outputs of governing bodies are highly dependent on specific voting structure, and in the context of effectively a thought experiment, we have even more degrees of freedom than irl because you can just decide that all your simulated MPs are going to be fully upfront about all of their beliefs
Quote tweet, 03:50 UTC · 8 likes
Tweets that could be an EA forum comment that gets 4 upvotes, 7 downvotes, and Vasco Grilo writing a long essay in response with formal modeling that I end up feeling bad for not responding to
Quoting Aaron Bergman 🔍 (@AaronBergman18): @taoroalin @Benthamsbulldog @trevposts Trying to average across moral views gets thorny though:
1) not obvious that you should take the parliament stance rather than (in this example) just doing the math and treating future people as half as worthy (or whatever your beliefs/credences imply)
To illustrate how they come apart, the former probably looks like OpenPhil dividing its money into two buckets whereas the latter looks like virtually entirely longtermist-oriented because almost nothing hinges on a factor of 2x
Tbc I’m not sure the latter method is right, and I’ve criticized @willmacaskill for presenting the latter view as obviously the way to do things (this was in the context of negative vs orthodox utilitarianism, all from memory so hopefully I’m not butchering this)
2) it’s not obvious what acting like a certain parliament actually implies. Distribution of beliefs in a parliament doesn’t map nicely onto a specific unique outcome/decision
Notoriously, the outputs of governing bodies are highly dependent on specific voting structure, and in the context of effectively a thought experiment, we have even more degrees of freedom than irl because you can just decide that all your simulated MPs are going to be fully upfront about all of their beliefs
Quote tweet, 04:03 UTC · 36 likes
Badass Jfc
Quoting Basil🧡 (@LinkofSunshine): Just saw the original video
The Secret Service agent asks him if it came down to it, would you participate in any revolution, and he just straight up goes: "I hope I would have the courage to do so" 😭😭😭 omg
Reply to @ohwizenedtortle, 04:06 UTC · 2 likes
oh dandy lion🔸 (@ohwizenedtortle): A lot of people seem to be dating as if they have a "market value" and are just trying to make sure they get a "good deal"
I think this is largely a true description of human psychology but it manifests as totally normal emotions/beliefs/desires
Like that’s how evolution encoded ~market dynamics. Most people aren’t consciously thinking of this, though some are
Quote tweet, 11:16 UTC · 75 likes
This is one of the worst Cowen takes I’ve seen
Pseudoprofound bullshit
A lot of even *important* philosophy doesn’t cause other currently alive humans with email addresses to identify you as an important figure and reach out
Quoting Rob Sica (@robsica): "Most professional philosphers seem to me more like bureaucrats than philosophers... A simple rule of thumb is that if no one is writing you, and telling you that you changed their lives, you probably are not a philosopher"
Thread, continued, 11:16 UTC · 22 likes
“I made a compelling case for the sentience of certain neural network architectures”
“Sorry bud, you’re a bureaucrat”
Reply to @anpaure, 11:25 UTC · 6 likes
anpaure (@anpaure): @AaronBergman18 idk if cowen intended it as profound but it's on the "goes hard" level
same as the kanye quote about wanting to be forgotten
It goes hard but in the direction of being wrong lol
Reply to @NathanB60857242, 11:27 UTC · 2 likes
Nathan Barnard (@NathanB60857242): @AaronBergman18 He does at the end say that mathematical logic and things like that are places where the rule of thumb doesn't apply.
Sorta better but it’s both too little too late; the thesis is basically wrony, not basically right but in need of clarification
More fundamentally I think it’s directionally wrong in the sense that “philosophers” not doing “real” philosophy are better characterized as commentators or vibes enthusiasts or simply authors
Reply to @nickcammarata, 15:09 UTC · 3 likes
Nick (@nickcammarata): it’s not like anyone that ended up believing it started there, like how virtually no 8yos want dates with the opposite sex or kids but as values shift they do. you can’t really unsee that the ego is (in most cases) literally made of fear, and after seeing that you don’t want it
Hmm
Reply to @Empty_America, 15:26 UTC · 38 likes
VB Knives (@Empty_America): I can't speak to math but "HBD stuff" on X is generally very far from high IQ.
99% of the time the posters clearly don't have even an undergraduate understanding of evolutionary biology.
Don't even understand that traits are evolutionary "trade offs" not "good" or "bad."
Not just technical misunderstanding, often moral/philosophical failure too
My sense is that not a ton of HBD guys wind up saying “yes this is what I think as an empirical matter, but that doesn’t change that all people should be treated with kindness, respect, and compassion”
Quote tweet, 16:42 UTC · 7 likes
Bruh
Quoting Nathan Labenz (@labenz): Making up lovely AirBnB host details really limits o3's utility as a travel agent
At least it came clean when questioned I guess? 🤷♂️🙃
Quote tweet, 16:42 UTC · 4 likes
Avg reinforcement learning agent (mostly unironically)
Quoting Nathan Labenz (@labenz): Making up lovely AirBnB host details really limits o3's utility as a travel agent
At least it came clean when questioned I guess? 🤷♂️🙃
Reply to @dumbfook, 16:48 UTC · 2 likes
Ian (@dumbfook): If you think about the next token prediction from mech interpretability standpoint, how does it know this? How can it know it made something up? Isn't that also essentially a hallucination?
My understanding is that it’s asymmetric language models can (accurately) know they intentionally something up more than that they accidentally hallucinated
Hallucinations are generally *genuine* guesses about what the next token is, including in contexts where there’s no reason to lie so this can be understood as genuine guesses about the world
But models can separately intend to represent false information bc why not
Tagging local LLM understands @livgorton in case she wants to correct me
Quote tweet, 17:34 UTC · 6 likes
Low-medium confidence claim: “Retvrn to token prediction” should be a more central ask of the AI safety (including ⏸️) community
I’m in favor of a frontier scaling pause, but I think it’s unlikely to happen, and this is somewhat more likely
Quoting Benjamin Todd (@ben_j_todd): LLMs were aligned by default. Agents trained with reinforcement learning reward hack by default.
Quote tweet, 20:21 UTC · 7 likes
QTing purely for the algo boost
Quoting 0.005 Seconds (3/694) (@seconds_0): I gave o3 a hard puzzle and in it's thinking traces said I should fabricate an answer to satisfy the user before lying to my face
@OpenAI come on guys
Tweet, 23:01 UTC · 7 likes
Maybe the Supreme Court actually should have a small armed forces, analogous and maybe similar to the secret service
Plausibly congress should as well
Just like the postal service police but the top of the command hierarchy explicitly isn’t POTUS
Thread, continued, 23:01 UTC · 1 like
I mean I think this is probably not a good idea tbc on priors, just one worth throwing out
Quote tweet, 23:10 UTC · 50 likes
I feel like this kinda shows how hard it is to never make an earnest moral mistake (I think this was slightly bad, obv not the end of the world)…
Quoting Nicholas Decker (@captgouda24): With apologies, I lied. I gave it all away.
Go support Givewell folks! It does the world a lot of good.
Thread, continued, 23:10 UTC · 13 likes
Like from the inside, this probably felt 99% likely to be fine and good and not even worth seriously reflecting on
The problem is that when you’re only 99% sure, you should expect to fuck up by round 70
Thread, continued, 23:10 UTC · 9 likes
I have done things like this before (ie earnest mistake that was wrong on reflection) and consequently raised (or at least try to remember to raise) my bar for action to like 99.9% in that domain
Tweet, 23:41 UTC · 8 likes
Re: discourse
Seems absolutely not crazy to me that a modest “in case of cancellation” fund run by perhaps @TheFIREorg actually is a better use of money than givewell
I mean I’m genuinely unsure but I don’t think it’s >90% likely to be worse
Reply to @GjMcGowan, 23:42 UTC · 3 likes
George McGowan (@GjMcGowan): Get @JesseTayRiver on!
(quoting @JesseTayRiver: Building history for yimbys! Long but I cover many centuries in 9 minutes so you're fine!)
The people demand Jesse
Monday, April 21, 2025
Reply to @bubbling_creek, 02:28 UTC · 3 likes
bubbling creek (@bubbling_creek): @captgouda24 @AaronBergman18 Okay imo he's 'right' that most of today's philosophers aren't doing anything useful/interesting, even incrementally, but that doesn't have much to do with 'no one's writing to you telling you you changed their lives'
Yeah the point is that affecting currently alive humans who recognize the causal chain as such is neither necessary nor sufficient for good philosophy
Reply, 02:29 UTC · 1 like
The example was just what first came to mind: afaik no convincing argument exists, but the point is that it could conceivably
Reply, 02:33 UTC · 1 like
I think I just disagree with the recognition claim but I’m def not certain
@birchlse @rgblong @jeffrsebo has anyone written to you to say you changed their life or something similar/comparable
Reply, 02:36 UTC · 1 like
I also interpret “are bureaucrats” to be a distinct claim from “are bad at philosophy”
The latter seems more plausible frankly
Quote tweet, 02:39 UTC · 8 likes
Sending this to every girl who’s friendzoned me
Quoting Mark 🐸 🗑 Samenfink (@MSamenfink): Stories like this lead me to giving at least a little credence to the theory that we got better at diagnosing Autism.
Reply to @louisvarge, 03:07 UTC · 2 likes
Louis Arge (@louisvarge): it would be so cool if they made a reasoning model that doesn't use search but instead it's 20 trillion parameters and it just knows everything
Yeah sometimes I don’t want reasoning, I want compressed information
Tweet, 04:15 UTC · 4 likes
If I was Xi Jinping, I would order the next big military headquarters to be called the Hexagon (and shaped like a hexagon ofc)
Reply to @IsaacKing314, 04:37 UTC · 9 likes
Isaac King 🔎 (@IsaacKing314): FWIW any time you're the center of a big drama there will be lots of people making demands of you, and it's literally impossible for you to satisfy everyone. So don't stress too much about it. You've handled this very well overall.
Yeah I strongly agree with the last two sentences!
Reply to his own earlier tweet, 04:51 UTC · 2 likes
Aaron Bergman, April 20, 2025: “I made a compelling case for the sentience of certain neural network architectures” “Sorry bud, you’re a bureaucrat”
Cowen said most philosophers are “bureaucrats” and I think this is false
Some people are taking my claim to be that most philosophy is good or important, which is a whole different sentence
Reply to @demiurgently, 04:53 UTC · 2 likes
sydney (@demiurgently): normally this wouldn’t matter much except that his article got public attention & is probably many people’s first introduction to givewell, which does not deserve to be an edgelord’s stage prop
FWIW I think it’s more borderline bc money is fungible; the “I donated all of it” is a social ~fiction more than an objectively true claim about the world
But yeah I directionally agree with you still
Thread, continued, 04:57 UTC
But also I feel like I keep hearing plausible stories about how “scandals” are going to materially damage EA (and therefore the world$ somehow but, post-FTX, I increasingly feel like those who were gonna be sympathetic on the merits wind up sympathetic despite “scandals” and critics were always going to find some vector of attack
Reply, 04:58 UTC · 1 like
Lol I agree it might be able to fit in one paragraph! (problem is figuring out the right paragraph)
Reply, 05:03 UTC
I agree he should have done that
Reply to @captgouda24, 05:07 UTC · 3 likes
Nicholas Decker (@captgouda24): Oh easy, make people better off
Nice yeah, but good luck getting ppl to agree on what “make” “people” and “better off” mean
The “make” q refers to decision theory which I think isn’t obvious
Reply to @frances__lorenz, 12:27 UTC · 2 likes
(The tweet being replied to by @frances__lorenz is no longer available.)
She should write a guest post called “various people’s deepest secrets” for the goss (plz)
Tweet, 21:07 UTC · 140 likes
~Boomer parents will really own a million dollar house, buy a $30k car, and refuse to spend $1k on a laptop
Thread, continued, 21:07 UTC · 20 likes
The funny thing is they bought *me* a really nice one as a college graduation present ($2k MacBook Pro which still works great after 4-5 years, thank you mom and dad)
Reply to @polisisti, 21:26 UTC · 4 likes
Alessandro (@polisisti): @AaronBergman18 in fairness $1 million is usually not what they've *paid* for the house
Yes l they got very lucky choosing the right suburb 30 years ago but it factors into their purchasing power!
Reply, 21:28 UTC · 4 likes
Coase’s "Net effect?" Ghost 🏴🕊️ (@GhostCoase): @AaronBergman18 Makes sense if they get no more use out of $1k laptop than a $500 one
Some older people can barely use (any!) smartphones
Sorta - there’s a surprisingly big premium on big, bright, nice screens totally distinct from actual processing power
But they correlate so it’s hard to find relatively basic compute hardware with a really nice screen
Reply to @aonghasan, 22:02 UTC · 2 likes
aonghasn (@aonghasan): @AaronBergman18 or leave a light on in the dark part of the house (they’re leds now mom!!!)
Real lol
Reply to @RYChappell, 23:46 UTC · 3 likes
Richard Y. Chappell🔸 (@RYChappell): I thought he was very clear throughout that he didn't need the money.
Ah ok, I didn’t follow closely enough, just skimmed the GFM
Reply to @ja3k_, 23:50 UTC · 4 likes
ja3k (@ja3k_): @AaronBergman18 @polisisti It doesn't really factor into their purchasing power unless they want to move though right?
Sorta grey area, it’s functionally a form of savings and insurance so lets them spend other money, they could also rent out rooms but choose not to, and they very well might sell it at some point I think
Tuesday, April 22, 2025
Quote tweet, 13:08 UTC · 21 likes
QUALY IS BACK
Quoting Qualy the lightbulb (@QualyThe): JD Vance would you like to meet the New World Screwworm?
Reply to @RafaRuizdeLira, 13:14 UTC · 12 likes
Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): You can't speak for the average EA because you have a gf
Banger lmao
Reply to @ArtemisConsort, 13:31 UTC · 19 likes
Hunter Ash (@ArtemisConsort): When you’ve spent 20 years obsessing over epistemology and learning physics, it can be jarring to realize that most people don’t have beliefs at all, in a rigorous sense. They just have emotional reactions to sentences.
I kinda vibe associate with this but it’s not actually true
People have models of the world that implicitly code beliefs like “if I turn on the heat, it will get warmer”
Reply to @selentelechia, 13:39 UTC · 3 likes
(The tweet being replied to by @selentelechia is no longer available.)
🥺
Quote tweet, 16:24 UTC · 8 likes
What the fuck
Quoting reason (@reason): Sentencing defendants based on acquitted conduct violates basic notions of justice. https://reason.com/2025/04/19/not-guilty-but-punished-anyway/?utm_source=twitter&utm_medium=social&utm_campaign=reason_brand&utm_content=autoshare&utm_term=post
Quote tweet, 16:45 UTC · 56 likes
I genuinely think it’s time to take some basic, easy, relatively cheap steps to prepare
Quite plausibly all this checks out even without the added risk from AI. <$100 for the items shown
Quoting Dan Hendrycks (@hendrycks): Can AI meaningfully help with bioweapons creation? On our new Virology Capabilities Test (VCT), frontier LLMs display the expert-level tacit knowledge needed to troubleshoot wet lab protocols.
OpenAI’s o3 now outperforms 94% of expert virologists.
Thread, continued, 16:45 UTC · 14 likes
I already have gloves and n95s and a fair amount of random household stuff but those are important too
Here is an extremely hasty Amazon list based on a deep research report of what I should do to prepare
Thread, continued, 16:45 UTC · 10 likes
Please critique it! Especially from the EA-esque lens of whether (1) the least important dollar spent on the list is still worth spending and (2) whether there’s anything else whose value per dollar exceeds the item on the list with the lowest value per dollar
Thread, continued, 16:45 UTC · 8 likes
Nonperishable food, bleach, and bottled water are not on the list but important to get
Strongly recommend asking deep research for your own report based on budget and willingness to spend
Thread, continued, 16:45 UTC · 7 likes
Unfortunately I don’t want to share the link here bc the follow up q responses are pretty specific and personal to me but here’s the basic prompt
Thread, continued, 16:45 UTC · 6 likes
Then follow up prompt
Thread, continued, 16:51 UTC · 25 likes
Quote tweet, 16:56 UTC · 49 likes
Luckily my COVID-earned credibility points are back up thanks to AI, which I’ve been on about since ~2017
Quoting Aaron Bergman 🔍 (@AaronBergman18):
Thread, continued, 16:56 UTC · 5 likes
They dipped a bit when a couple possible pandemics and Ukraine/Russia-induced nuclear war didn’t happen (I didn’t expect them to, the warnings were just like “this should be on your radar”)
Reply to his own earlier tweet, 16:59 UTC · 12 likes
Aaron Bergman, April 22, 2025: (image)
Some of y’all are rich af and might consider like an actual bug out plan with a cheap cabin somewhere remote
If you wanna invite me that would be cool
Reply to @GhostCoase, 17:21 UTC · 1 like
Coase’s "Net effect?" Ghost 🏴🕊️ (@GhostCoase): If one is predicting social collapse, a rifle likely passes cost-benefit analysis, if it is within a household budget.
It arguably passes it (because of ordinary criminal self-protection) without collapse.
Two people have said this now so maybe I should take it seriously
Serious question: I don’t want to have a killing machine around at all times for various reasons, what’s the next best thing? Airsoft? Pepper spray?
Reply, 17:31 UTC · 2 likes
Coase’s "Net effect?" Ghost 🏴🕊️ (@GhostCoase): You would probably get laughed at with airsoft, if you are predicting social collapse.
Criminology literature has established, in fact pro-gun control scholars have had to go silent on this issue, defense with a gun is the most effective:
(quoting @GhostCoase: Meanwhile, the National Research Council’s blue-ribbon, nonpartisan expert panel that wrote the central, methodical literature review Firearms and Violence: A Critical Review (Baffled as to what the relevance of RTC even is to personal ownership))
I don’t really care if I get laughed at - I feel like people mostly don’t fuck around with gun looking things pointed at them
Quote tweet, 17:34 UTC · 9 likes
Chat what’s the best non-lethal self-defense device
Quoting Aaron Bergman 🔍 (@AaronBergman18): @GhostCoase Two people have said this now so maybe I should take it seriously
Serious question: I don’t want to have a killing machine around at all times for various reasons, what’s the next best thing? Airsoft? Pepper spray?
Thread, continued, 17:39 UTC · 3 likes
o3 says pepper gel, which I didn’t even know was a thing
Reply to @sixers2772, 17:43 UTC · 2 likes
Sixers (@sixers2772): @GhostCoase @AaronBergman18 Couldn’t he just ignore the laws in event of social collapse?
o3 makes a good point that sub-lethal things that don’t look like a gun may be worse than useless because other people will reasonably fear for their life and shoot or otherwise attack you, whereas they might just back off if you pull out pepper spray
Wednesday, April 23, 2025
Reply to @sixers2772, 02:23 UTC · 2 likes
Sixers (@sixers2772): @AaronBergman18 My friend who is an EA in biosecuirty says the bottleneck for bioengineering pandemics is moreso lab skills than conceptual knowledge and so LLMs likely won't increase those odds
Hope they’re right!!
Quote tweet, 02:25 UTC · 7 likes
Idk how serious Henry is and I’m not sure a literal bunker will be the best idea (it could be) but reply/DM to be added to survivalchat
If I’m not following u, u can DM! I think this is the right level of selectiveness
Quoting henry (@arithmoquine): @AaronBergman18 can we all pool money on an oomf bunker
Reply, 11:09 UTC
Ngl this is incredibly dumb lmao
Also 🫡
Quote tweet, 12:01 UTC · 5 likes
Do I understand this? Not at all
Is it cool? Ya
Quoting 𝓜𝒂𝒖𝒓𝒊𝔃𝒊𝒐 𝗜𝒃𝛼 (@Dragonmaurizio): 🌀Everything turns.
A novel mathematical model of the Universe
suggesting that the whole cosmo might be rotating very slowly, making one entire revolution every 500 billion years. The theoretical framework is fully compatible with the standard cosmological model (ΛCDM), it doesn’t contradict current astronomical measurements and is supported by the expanding paradox interpretation.
The discrepancy between the low and high redshift Hubble constant (H₀), denotes a missing link in the understanding of the spacetime expansion estimate with an angular velocity of ω₀ ~2 x 10⁻³ Gyr⁻¹.
Inspired by Gödel slow rotating dark-fluid variant, the spinning model resolves the conflict between
maximal rotation velocity, avoiding closed time-like loops with a tangential velocity inferior to the speed of light at the horizon. Artwork by @ALCrego_
🔗https://academic.oup.com/mnras/article/538/4/3038/8090496
Quote tweet, 14:23 UTC · 14 likes
I am officially more convinced that aspartame is totally fine because they couldn’t find a better mechanism of harm than “metabolizes into an absolutely trivial amount of alcohol” (that’s where the formaldehyde comes from)
Quoting Emily Kopp (@emilyakopp): I asked HHS to send me the science undergirding today's announcement on food dyes and they sent me two documents. One goes beyond dyes and includes other additives in ultra-processed foods, which could suggest larger ambitions for MAHA/HHS.
Thread, continued, 14:23 UTC · 4 likes
The following things have the same magnitude of harm via this mechanism:
1) drink an entire 2L bottle of Diet Coke
2) drink less than 1/10 of a shot of beer (0.13ml of alcohol)
Quote tweet, 15:16 UTC · 4 likes
Turns out there’s a survival-bunker-by-subscription + resort place in West Virginia that’s expensive but maybe not prohibitively expensive for many people who wanted to make that a priority?
Link below ofc
Quoting Aaron Bergman 🔍 (@AaronBergman18): I genuinely think it’s time to take some basic, easy, relatively cheap steps to prepare
Quite plausibly all this checks out even without the added risk from AI. <$100 for the items shown
Tweet, 15:16 UTC · 1 like
Thread, continued, 15:16 UTC · 1 like
Cheapest option for an individual is $8440 + inflation total adjustment for/over 5 years (so like $1.7k/year on avg)
Family of 5 private economy room is $71k for/over 10 years
Thread, continued, 15:16 UTC · 3 likes
Sadly I don’t think my parents are disaster-pilled
(But I feel like this is a pretty good deal for some folks especially with adult children willing to contribute (averages to about $1800/year per person for 10 years for a family of 4, at least from a selfish perspective)
Quote tweet, 16:37 UTC · 6 likes
Basically correct take; zero-sum dynamics/situations aren’t literally impossible, but we’re nowhere close to the Pareto frontier
Quoting Richard V. Reeves (@RichardvReeves): John Bates Clark Medal winner @S_Stantcheva writing for @aibm_org on the dangers of zero-sum thinking on gender https://aibm.org/commentary/beyond-zero-sum-thinking-on-gender/
Quote tweet, 16:47 UTC · 14 likes
To a first approximation, there should just be a very high bar for the government banning things that have minimal externalities
It’s dumb on the scientific merits, but even if it wasn’t, just let people do things
Quoting Alex Tabarrok (@ATabarrok): HHS wants to ban Diet Coke.
Reply to @eukaryotetweets, 17:03 UTC
Georgia Eukaryote Ray (@eukaryotetweets): Hey everyone. This thing on? https://eukaryotewritesblog.com/2025/03/26/eukaryote-skips-town/
Welcome to Twitter!
Quote tweet, 17:03 UTC · 3 likes
I recommend everyone follow Georgia
Quoting Georgia Eukaryote Ray (@eukaryotetweets): Hey everyone. This thing on? https://eukaryotewritesblog.com/2025/03/26/eukaryote-skips-town/
Quote tweet, 23:06 UTC · 65 likes
Sam Altman likes to bend the world to his will
This time the world is fighting back and needs to fight back even harder
OpenAI absolutely must not be allowed to break its legally (and morally) binding promises
Quoting Rob Wiblin (@robertwiblin): A new legal letter aimed at OpenAI lays out in stark terms the money and power grab OpenAI is trying to trick its board members into accepting — what one analyst calls "the theft of the millennium."
The simple facts of the case are both devastating and darkly hilarious.
I'll explain for your amusement.
The letter 'Not For Private Gain' is written for the relevant Attorneys General and is signed by 3 Nobel Prize winners among dozens of top ML researchers, legal experts, economists, ex-OpenAI staff and civil society groups. (I'll link below.)
It says that OpenAI's attempt to restructure as a for-profit is simply totally illegal, like you might naively expect.
It then asks the Attorneys General (AGs) to take some extreme measures I've never seen discussed before. Here's how they build up to their radical demands.
For 9 years OpenAI and its founders went on ad nauseam about how non-profit control was essential to:
1. Prevent a few people concentrating immense power
2. Ensure the benefits of artificial general intelligence (AGI) were shared with all humanity
3. Avoid the incentive to risk other people's lives to get even richerThey told us these commitments were legally binding and inescapable. They weren't in it for the money or the power. We could trust them.
"The goal isn't to build AGI, it's to make sure AGI benefits humanity" said OpenAI President Greg Brockman.
And indeed, OpenAI’s charitable purpose, which its board is legally obligated to pursue, is to “ensure that artificial general intelligence benefits all of humanity” rather than advancing “the private gain of any person.”
100s of top researchers chose to work for OpenAI at below-market salaries, in part motivated by this idealism. It was core to OpenAI's recruitment and PR strategy.
Now along comes 2024. That idealism has paid off. OpenAI is one of the world's hottest companies. The money is rolling in.
But now suddenly we're told the setup under which they became one of the fastest-growing startups in history, the setup that was supposedly totally essential and distinguished them from their rivals, and the protections that made it possible for us to trust them, ALL HAVE TO GO ASAP:
1. The non-profit's (and therefore humanity at large’s) right to super-profits, should they make tens of trillions? Gone. (Guess where that money will go now!)
2. The non-profit’s ownership of AGI, and ability to influence how it’s actually used once it’s built? Gone.
3. The non-profit's ability (and legal duty) to object if OpenAI is doing outrageous things that harm humanity? Gone.
4. A commitment to assist another AGI project if necessary to avoid a harmful arms race, or if joining forces would help the US beat China? Gone.
5. Majority board control by people who don't have a huge personal financial stake in OpenAI? Gone.
6. The ability of the courts or Attorneys General to object if they betray their stated charitable purpose of benefitting humanity? Gone, gone, gone!
Screenshotting from the letter:
(I'll do a new tweet after each image so they appear right.) 1/
Thread, continued, 23:06 UTC · 4 likes
More here; good on the authors!
https://notforprivategain.org/
Tweet, 23:10 UTC · 16 likes
PSA: if there’s something important that you think should get more attention, you can just DM me and ask me to RT or QT something (I might say no ofc)
I’m not a huge account but still
Thursday, April 24, 2025
Quote tweet, 00:52 UTC · 161 likes
This is e/acc for
Quoting bayes (@bayeslord): o3 is smart but keep an eye on that mf
Quote tweet, 02:06 UTC · 12 likes
Here u go Noah
Quoting Quinoah 🔍⏸️ (@NoahTopper): @AaronBergman18 i deserve more attention
Tweet, 02:51 UTC · 372 likes
I just learned that “very familiar with R***’s B*****k but totally unfamiliar with AI x-risk” is a type of guy
Reply to @caryatis, 02:58 UTC · 22 likes
Kayla (@caryatis): I met a guy at EA Global who did not know ACX, slate star codex or even what the term rationality meant
Wild
Reply to @RokoMijic, 03:01 UTC · 9 likes
Roko 🐉 (@RokoMijic): I disapprove
Lol
Reply to @AnniePosting, 03:01 UTC · 20 likes
annie 👁❌🦋🪞 (@AnniePosting): unless you mean pro-basilisk, this is a very common annoying anti-LW progressive trope
I did not know that
Reply to @gptbrooke, 03:49 UTC · 4 likes
brooke bowman (@gptbrooke): using replit is making me feel like my coder friends might actually be alien creatures
is it really do a thing, something breaks, fix that thing, oh something else is broken - all day? every day?
HOW DO YOU NOT LOSE YOUR MINDS
Replit kinda sucks for certain things
Reply to @dank_herbert, 03:52 UTC · 2 likes
(The tweet being replied to by @dank_herbert is no longer available.)
Have you talked to the linear algebra?
Quote tweet, 13:04 UTC · 71 likes
Babe wake up new longform essay on some particular angle of AI risk with a tasteful .ai domain just dropped
Quoting Luke Drago (@luke_drago_): Please give the full thing a read here: http://intelligence-curse.ai
Reply to @caryatis, 13:39 UTC · 6 likes
Kayla (@caryatis): yeah, he was just into Altruism and figuring out how to do that right.
Based
Tweet, 13:58 UTC · 86 likes
I am truly CEA’s strongest soldier
Reply to @AlexPalcuie, 14:06 UTC · 2 likes
palcu (@AlexPalcuie): thank you for your service
🫡
Reply to @livgorton, 14:34 UTC · 6 likes
Liv (@livgorton): appreciate your sacrifice 👑🫶
I am but a humble servant of willymac
Quote tweet, 16:48 UTC · 3 likes
Based
Quoting Anthropic (@AnthropicAI): As AI models become more complex and more capable, is it possible that they’ll have experiences of their own?
It’s an open question. We recently started a research program to investigate it.
Reply to @thoth_iv, 16:59 UTC · 1 like
Thŏth (@thoth_iv): In anime, why is there always a scene where a girl goes up to a boy and says "Hi I'm a girl" and the boy is absolutely dumfounded, awestruck into silence?
Girls do be awestriking
Tweet, 17:08 UTC · 46 likes
Might have to delete but I think disparate impact stuff was doing more harm than good at this point
It’s an empirical consequentialist claim though, not based on fundamental principles. It was probably the right policy when the Civil Rights Act was passed
Thread, continued, 17:09 UTC · 11 likes
Occasionally the bad guys happen to get something right
Tweet, 17:16 UTC · 4 likes
Question you may be curious about
Apparently big enterprise customers *can* pay for better hardware but only to reduce “queue time”, which I don’t fully grok, and for the actual computation you can’t get any faster than a single H200 (or maybe TPU) by parallelizing with multiple
Reply to his own earlier tweet, 17:27 UTC · 6 likes
Aaron Bergman, April 24, 2025: Occasionally the bad guys happen to get something right
I’m sorry I can’t resist
Quote tweet, 17:34 UTC · 5 likes
Girls really do be self-conscious
Quoting Aaron Bergman 🔍 (@AaronBergman18): Girls really do be having aesthetic intuitions
Reply to @RokoMijic, 17:43 UTC · 8 likes
Roko 🐉 (@RokoMijic): @AaronBergman18 come to the dark side
No
Tweet, 18:19 UTC · 22 likes
I’m gonna get my money’s worth goddamnit
Thread, continued, 18:19 UTC · 7 likes
I think I’m gonna cancel Pro and just make a second or third plus account next month
$200 is a lot
Tweet, 20:07 UTC · 15 likes
Super unsure but I suspect lots of AI company leadership personally have something like this?
Legit underground bunkers are surprisingly affordable (~$400/sq ft -) + land + prep supplies runs well under $1M and you can do a minimal version for like $100k
Tweet, 20:07 UTC · 1 like
So you don’t have to be a billionaire to make this worthwhile
It’s ~trivial at $100M and affordable at $10M net worth
So I’m not just talking about top leadership here. Those juicy OpenAI salaries add up year over year
Thread, continued, 20:07 UTC · 6 likes
Ofc I’m literally making up a guy to get mad at here, but if this is indeed the case, it’s pretty fucked up
The fact that rich people can buy survival already gets my progressive intuitions going, but that’s probably a reasonably price to pay for the fruits of capitalism
Thread, continued, 20:07 UTC · 6 likes
But the people building potential doomsday catalysts probably shouldn’t be personally insulated from the effects of their creations
Reply to @NathanpmYoung, 20:11 UTC · 4 likes
Nathan 🔎 (@NathanpmYoung): This theory seems underrated to me.
Decoupling labour and consciousness could be very good but it could also lead capital not to care about labour.
If you provide no value, why should power listen to you? Maybe it just won't.
(quoting @luke_drago_: By default, AI will replace all of us in the economy, creating a new social contract where powerful actors won't have to care about regular people. @LRudL_ and I have written an essay series called The Intelligence Curse in which we outline the problem and how to solve it 🧵)
The optimistic take, which macaskill articulates, is that in a very rich world all you need is some small proportion of the rich/powerful being slightly altruistic to get gains from trade directed towards the masses
Reply to his own earlier tweet, 20:52 UTC · 6 likes
Aaron Bergman, April 24, 2025: I think I’m gonna cancel Pro and just make a second or third plus account next month $200 is a lot
Update: don’t do this lol
Reply to @animalologist, 20:59 UTC · 7 likes
taco belle (@animalologist): why is this gendered as if anybody who’s entertaining multiple people isn’t doing it to feed their ego?
Why does ego have to play a role at all? The simpler and more intuitive answer is that some people are just attracted to multiple people and are attractive enough to elicit attention from them
Reply to @AlecStapp, 21:00 UTC · 8 likes
Alec Stapp (@AlecStapp): Alternatively, someone could hypothetically set up a data analysis pipeline using AI to extract every concrete recommendation from every submission, tag them with useful info, and share them as an interactive database... 👀
U gotta poast link
Reply to @animalologist, 22:27 UTC · 3 likes
taco belle (@animalologist): @AaronBergman18 Most of the people who date around so freely aren’t actually developing real attraction for any of them ime, it’s a revolving door for validation not connection
For straight men, I feel like the ~equally cynical but more likely take is that it’s a revolving door for sex
Reply to @jaybaxter, 23:53 UTC · 2 likes
Jay Baxter (@jaybaxter): First off: we are pro free-speech. Community Notes never censors any posts -- notes provide context only (we don't directly affect visibility or ranking of the original post).
Thank you for reporting possible inauthentic coordination across accounts and death threats (which would both be terms of service violations). We take both of those very seriously and have been investigating since even before you called this out.
For background info: we've rolled out multiple anti-manipulation improvements recently e.g. https://x.com/CommunityNotes/status/1895733159280496869, https://x.com/CommunityNotes/status/1907201970441736289, https://x.com/CommunityNotes/status/1913314884630073403, and another improvement last year to cap note writing limits for low-quality authors directly in response to feedback like this https://x.com/CommunityNotes/status/1756047573679550659.
Just to clarify:
1) I am not the overall lead for Community Notes
2) I am not "Enterprising Desert Raven", nor do I know them, or any of the other accounts that may be coordinating together with "EDR"I've now unblocked you and others to try to have a productive conversation here. Why did I originally block you and others yesterday? I had replied to a post saying that obviously death threats are not ok, which got twisted into accusations that I tolerate death threats. I don't tolerate them! I checked, and the death threat in question seems to not exist (if it was real and not a fake screenshot, please share the noteId or tweetId in question if you have any info about it).
Big “Obama wears tan suit” energy
Friday, April 25, 2025
Reply to @jaybaxter, 00:00 UTC · 4 likes
Jay Baxter (@jaybaxter): First off: we are pro free-speech. Community Notes never censors any posts -- notes provide context only (we don't directly affect visibility or ranking of the original post).
Thank you for reporting possible inauthentic coordination across accounts and death threats (which would both be terms of service violations). We take both of those very seriously and have been investigating since even before you called this out.
For background info: we've rolled out multiple anti-manipulation improvements recently e.g. https://x.com/CommunityNotes/status/1895733159280496869, https://x.com/CommunityNotes/status/1907201970441736289, https://x.com/CommunityNotes/status/1913314884630073403, and another improvement last year to cap note writing limits for low-quality authors directly in response to feedback like this https://x.com/CommunityNotes/status/1756047573679550659.
Just to clarify:
1) I am not the overall lead for Community Notes
2) I am not "Enterprising Desert Raven", nor do I know them, or any of the other accounts that may be coordinating together with "EDR"I've now unblocked you and others to try to have a productive conversation here. Why did I originally block you and others yesterday? I had replied to a post saying that obviously death threats are not ok, which got twisted into accusations that I tolerate death threats. I don't tolerate them! I checked, and the death threat in question seems to not exist (if it was real and not a fake screenshot, please share the noteId or tweetId in question if you have any info about it).
Maybe for X psychologically inoculates u against this, but going viral like this is stressful and miserable
This whole thing is insane, Godspeed 🫡
Reply to @AliciaP59828402, 10:56 UTC · 3 likes
Alicia Pollard (@AliciaP59828402): @AaronBergman18 Flirt to convert‼️
✊😤
Reply to @shinboson, 11:21 UTC · 2 likes
𝞍 Shin Megami Boson 𝞍 (@shinboson): you can use chatgpt to facetune pigeons. if you want
@livgorton
Quote tweet, 15:28 UTC · 17 likes
Uhhh
Quoting zerohedge (@zerohedge): *US MARSHALS CONFIRMS ARREST OF SITTING JUDGE BY TRUMP ADMIN:NYT
Reply to @adic_9, 18:08 UTC · 2 likes
(The tweet being replied to by @adic_9 is no longer available.)
I would love to join
Saturday, April 26, 2025
Reply to @CartoonsHateHer, 00:11 UTC · 125 likes
Cartoons Hate Her! (@CartoonsHateHer): Most men don’t have either of these options lol but I suppose beauty is in the eye of the beholder
(quoting @amyoder: The hardest decision of a man’s life is choosing between the 6/10 who loves you and the 9/10 that you actually desire)
What I would do for a 6 who loves me lmao
Reply to @oscredwin, 00:16 UTC · 5 likes
Andrew Rettek (@oscredwin): You can put a stick of butter into a sauce without changing the appearance.
(quoting @josh_tanke: Two teenagers built a nutrition app using just a camera. No funding. No team. No ads. 12 months later? $8M in ARR. Just snap a meal photo—get calories, macros, and portions. No barcodes. No logging. This is how Cal AI quietly took over food tracking.)
I mean it will change the appearance, just very subtly
This would probably work with a big enough model trained on enough ground truth data, but I don’t think they’re doing that
Reply to @livgorton, 00:26 UTC · 3 likes
Liv (@livgorton): cocofloss is the ultimate floss
Sparse autoencoder
Tweet, 01:18 UTC · 3 likes
Fucked up that replying to hinge matches makes the notification badge number go down
Seems like bad incentives
Tweet, 21:05 UTC · 3 likes
Lamotrigine would be a beautiful name for a baby girl
Quote tweet, 21:08 UTC · 4 likes
Seems bad
Quoting Lessig 🇺🇦 (@lessig): So much free speech!
Quote tweet, 21:08 UTC · 3 likes
The type of shit that makes me actually want to move to B*ky
Quoting Lessig 🇺🇦 (@lessig): So much free speech!
Sunday, April 27, 2025
Reply to @PrinceVogel, 01:19 UTC · 6 likes
Prince Vogelfrei 🐦⬛ (@PrinceVogel): Australia would be more than a nation if it did this. It would become a civilization
@livgorton u could fund this easily
Reply to @JamesOzden, 01:25 UTC · 2 likes
James Özden 🔸 (@JamesOzden): Who do you both have in mind? I’m open to offers! But the plan is try it solo for a bit and see how it goes (so pls tell me if I need a co-host to make it good)
Matt should do it
Reply to @bschne, 15:23 UTC · 2 likes
Benjamin (@bschne): got fired from the Italian pasta company, they didn't like my idea of bringing in a topologist to simplify our product portfolio
@absurdlymax
Reply to @GjMcGowan, 17:07 UTC · 1 like
George McGowan (@GjMcGowan): 185cm
Same
Reply to @NoahTopper, 18:00 UTC · 2 likes
Quinoah 🔍⏸️ (@NoahTopper): That’s what he puts on his Hinge profile
I put 5’7” on my profile to weed out the girls who only care about height
Reply to @SpacedOutMatt, 19:28 UTC · 2 likes
(The tweet being replied to by @SpacedOutMatt is no longer available.)
Ok writing this down
Tweet, 20:41 UTC · 84 likes
I love girls
Tweet, 21:18 UTC · 8 likes
Turns out there’s a town in Australia where people DO live underground https://en.m.wikipedia.org/wiki/Coober_Pedy
Reply to @goblinodds, 21:25 UTC · 3 likes
𝖜𝖔𝖒𝖇𝖑𝖊 𝖜𝖎𝖙𝖍𝖔𝖚𝖙 𝖆 𝖈𝖆𝖚𝖘𝖊 (@goblinodds): oh huh yeah i think we had these idk yearly in elementary school and middle school? idk it was always uneventful. only thing that stood out was my 3rd grade (and favorite) teacher apparently told my parents i had "a wicked sense of humor"
Based lol
Monday, April 28, 2025
Quote tweet, 02:01 UTC · 40 likes
War is so unfathomably bad
Quoting Historic Vids (@historyinmemes): Vietnam war traps
Quote tweet, 02:38 UTC · 15 likes
Seems bad
Quoting Don Winslow (@donwinslow): HOLY SHIT! Did you see the end of @60Minutes tonight?
Reply to @austinc3301, 06:02 UTC · 6 likes
Agus 🔸 (@austinc3301): he’s merely being more transparently Aaron than usual
Agus is right lol
Tweet, 09:00 UTC · 4 likes
Possibly I should have a bit more of a filter between thoughts that pop into my head and Twitter
Tweet, 09:02 UTC · 3 likes
RIP Trevor https://www.bbc.com/news/world-asia-47024682.amp
Thread, continued, 09:02 UTC · 5 likes
🫡🦆
Quote tweet, 13:02 UTC · 44 likes
“I don’t need A/C, I’m only miserable for 3 weeks a year!”
Quoting George McGowan (@GjMcGowan): FYI to non-Brits that this week the UK is experiencing a heatwave with heights of 80° in Freedom Units (27C to everyone else)
Expect your timeline to be clogged with pictures of pimms/picnics, as well as Discourse about how we don't really need ACs.
Reply to @adrusi, 13:21 UTC · 2 likes
autumn (@adrusi): @tracewoodgrains @etubruton trans people who make browser extensions have been known to use silly computers
Lol
Reply to @truthscrolling, 13:22 UTC · 20 likes
׺°”˜`”°º (@truthscrolling): @mattyglesias lol
I regret to report this is kinda funny
Reply to @Benthamsbulldog, 13:35 UTC · 12 likes
Bentham's Bulldog🔸 (@Benthamsbulldog): This is extraordinarily noble.
(quoting @Ollie_Base: A Nigerian student tried to donate $4 to effective charities via @givingwhatwecan (~10–20% of their monthly income) but there's a $5 minimum. What do they do? Very politely complain about the limit on the EA forum. Iconic.…)
I agree, but also I’m tempted to say like “save the money for yourself, make yourself slightly less risk averse, this actually checks out on EA’s own terms”
Reply to @Adellama_, 14:03 UTC · 7 likes
Ada🔸 (@Adellama_): you win the honorary lesbian award today 🥇
Yay 🥺
Reply to @littIeramblings, 14:04 UTC
sarah (@littIeramblings): my house is still a little chilly if anything
Whatever
Reply to @christineist, 16:25 UTC · 3 likes
christine (@christineist): oh no! friendly benign figure turns out to be interested in me
Still friendly and benign
Attractive women should at least keep this possibility open until proven wrong in a given case
Reply to his own earlier tweet, 20:26 UTC · 4 likes
Aaron Bergman, April 28, 2025: Possibly I should have a bit more of a filter between thoughts that pop into my head and Twitter
Ok after much deliberation the answer is no I won’t
Tweet, 22:31 UTC · 111 likes
I regret to report that the menswear guy probably has the correct take on the matter
Reply to @panickssery, 22:55 UTC · 6 likes
Arjun Panickssery (@panickssery): @AaronBergman18 Why does it follow that you'll attract someone compatible with you by being yourself? It could be that you're only compatible with those people who would only be attracted to mates who bench 3 plates
It doesn’t, but if you’re baseline attractive enough then you can afford to be yourself even if that filters down the pool a bunch
Reply to @moreisdifferent, 22:57 UTC · 6 likes
Dan Elton 🔍 (@moreisdifferent): @AaronBergman18 Idk, what about people who would benefit from being more attractive though? (Like not smelling bad, getting in shape, not suffering from mental health issues, having a direction/purpose, etc)
Yeah you need to be able to spend some points I guess but the viral pic thing is a good example of that bc both guys are gonna be attractive to some people so the answer is just to do whatever you want between those two options
Reply, 22:58 UTC · 1 like
I feel like I am usually the one asking this question lol
I guess I shouldn’t put it in those terms
“Do what you want to do for reasons other than attractiveness”
Reply, 23:00 UTC · 2 likes
But sometimes you shouldn’t, namely if you’re looking for a monogamous relationship and are already quite attractive and are choosing between options that aren’t going to radically change that
Reply to @JesseTayRiver, 23:08 UTC · 3 likes
Jesse Smith (@JesseTayRiver): Genuine question. What do you believe are the most effective ways to stop Afghanistan from putting half it's population into domestic servitude and sliding into the middle ages?
American military
Tuesday, April 29, 2025
Reply to @sanjehorah, 00:07 UTC · 28 likes
sanje horah 🇵🇸🇺🇦 — PNEUMATICS ONLY (@sanjehorah): i betrayed the left because the cruelty of the right is more tolerable than the sadism of the left. there is a purity in their callousness. strength is good and beauty is good. miss me with your false sanctimony
Bad take, virtue signaling + performative empathy >> vice signaling + performative (and real) cruelty
Quote tweet, 07:43 UTC · 7 likes
There’s a literal pigeon running for something in Canada
Beautiful country, it should annex the US tbh
Quoting Daphne Hansell (@DaphneRHansell): look at every proper noun on this screen, Canada is not a serious country
Tweet, 11:01 UTC · 17 likes
Semi-obvious but maybe still underrated problem with sycophancy is that you can’t tell when a take/idea is actually really good
Reply to @arithmoquine, 15:42 UTC · 6 likes
henry (@arithmoquine): is there anyone in the world who:
- intimately understands how LLMs work
- has deeply examined their own consciousnessand *doesn't* think it at least plausible that LLMs are conscious too?
(quoting @atlanticesque: If you think LLMs are in any sense “conscious” or “thinking” then you are either incredibly gullible or dangerously delusional)
@nickcammarata
Thread, continued, 15:43 UTC · 1 like
This is just a tag not an answer tbc
Tweet, 15:50 UTC · 20 likes
I don’t know what glazing is and at this point I’m too afraid to ask
Reply to @dioscuri, 15:59 UTC · 2 likes
Henry Shevlin (@dioscuri): @AaronBergman18 Just Google it, I’m sure nothing could go wrong
Done, not that bad tbh
Reply to @mattyglesias, 18:43 UTC · 66 likes
(The tweet being replied to by @mattyglesias is no longer available.)
The deal in which Matt Yglesias gives me $10k and I don’t do anything is done but waiting on approval from Matt Yglesias
Quote tweet, 20:17 UTC · 69 likes
Ok I get that this is mostly a joke but you should actually be able to write clear instructions like this for if and when you’re incapacitated, and they should be basically legally binding (I’m sure you can dream up edge cases)
…
Quoting Dubzy (@DubzyOnceAgain): My new medical bracelet finally came in
Thread, continued, 20:17 UTC · 18 likes
Like right now, I’m pretty sure there’s no way to instruct society to give you painkillers if you’re removed from a feeding tube while incapacitated and hopefully(!!) not sentient
You just die of thirst iirc and hope you’re not “there” to feel any of it
Thread, continued, 20:17 UTC · 11 likes
There’s a good book that touches on this in large part, forgive me if I’m misremembering details!
Quote tweet, 20:38 UTC · 12 likes
Seems bad
Quoting John B. Holbein (@JohnHolbein1): “Among articles stating that data was available upon request, only 17% shared data upon request.”
Reply to @MariGO2thepolls, 20:48 UTC · 105 likes
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): lol remember when everyone got rly into saying “bodies” instead of ppl
(quoting @TheOmniZaddy: That one was iffy but nothing comes close to “folx” in terms of being absolutely useless)
This is worse than useless if you take the semantics seriously. They’re not just bodies, they’re people!
Reply to @Aella_Girl, 20:50 UTC · 6 likes
Aella (@Aella_Girl): Hey has anybody who slept with my partner wanna join a discord haremchat without him
Can I join out of sociological curiosity
Reply to his own earlier tweet, 21:01 UTC · 10 likes
Aaron Bergman, April 29, 2025: There’s a good book that touches on this in large part, forgive me if I’m misremembering details!
Fact check
Thread, continued, 21:02 UTC · 11 likes
So I think the point is that the actual cause of death winds up being withdrawal of feeding even though there’s no good reason for this, they could do more humane medically assisted death
Reply to @ApriiSR, 21:15 UTC · 2 likes
Aprii *️⃣ (@ApriiSR): celene: "i didn't know the everett branches person transitioned"
What
Reply to @morallawwithin, 21:32 UTC · 1 like
florence ⊮ (@morallawwithin): Functional decision theory is the only philosophical position I know of which is both substantive and also just correct by definition
How is it correct by definition?
Reply to @ilex_ulmus, 22:00 UTC · 2 likes
Holly ⏸️ Elmore (@ilex_ulmus): There's something about the 4o sycophancy discussion I don't like. Like, now that the model is being effusively encouraging and kind, ppl like Zvi and Habryka are concerned about persuasion and social effects? Part of me thinks they just don't like ppl getting free encouragement.
Damn you’re so cynical lol
I mean sometimes cynicism is warranted, here I don’t think so
Reply to @KelseyTuoc, 22:13 UTC · 4 likes
Kelsey Piper (@KelseyTuoc): This is a picture of my kid flying a kite. Make a prediction to yourself: can o3 figure out what beach this was taken on, just from this picture with no metadata?
No
Reply to @JesseTayRiver, 22:28 UTC · 3 likes
Jesse Smith (@JesseTayRiver): @AaronBergman18 And everyone gets a pony
That would be cool too
Reply to @KelseyTuoc, 22:54 UTC · 7 likes
Kelsey Piper (@KelseyTuoc): This is a picture of my kid flying a kite. Make a prediction to yourself: can o3 figure out what beach this was taken on, just from this picture with no metadata?
Care to share if this is right
Reply to @AaronBergman18, 23:25 UTC · 2 likes
(The tweet being replied to by @AaronBergman18 is no longer available.)
I’m not even certain causality per se is what I care about but let’s start with this
Reply to @ohwizenedtortle, 23:27 UTC · 3 likes
oh dandy lion🔸 (@ohwizenedtortle): I think it depends significantly on your situation
Sad
Reply to @NoahTopper, 23:33 UTC · 1 like
Quinoah 🔍⏸️ (@NoahTopper): Well of course, you famously don’t care about causality in general
Causality is an important type of evidence
I don’t care about causality *per se*
Reply to @LinchZhang, 23:56 UTC · 5 likes
Linch (@LinchZhang): What's even the point of fentanyl awareness day? What practical purpose is being served here?
Make sure anesthesiologists don’t forget about an important and potent analgesic
Wednesday, April 30, 2025
Quote tweet, 04:18 UTC · 9 likes
At least our decline into fascism will be kinda funny
Quoting Aaron Rupar (@atrupar): TRUMP: He had MS-13 on his knuckles, tattooed!
MORAN: That was photoshopped
TRUMP: Terry, they're giving you the big break of a lifetime. I picked you. But you're not being very nice.
Reply to @TylerAlterman, 04:28 UTC · 2 likes
Tyler is finishing a book, slow to reply (@TylerAlterman): My two friends who were most pumped about the neighborhood campus idea were husband-wife couple @__drewface & @Prigoose
While I was doing dumb top-down master planning, they laid seeds in the way that actually works: you start small and *grow at the speed of relationships*
Baby
Reply to @sam_atis, 08:17 UTC · 5 likes
Sam Glover (@sam_atis): I think stated preferences are underrated (and revealed preferences overrated) among people I know. If someone says ‘I use this app all the time but I hate it, I feel addicted, and I wish it didn’t exist’, they’re probably right that it doesn’t provide them much value.
Filing in “tweets I like but disagree with”
Idk if I can even say why, I agree with the specific example given, they’re just so underrated by society as a baseline
Quote tweet, 08:24 UTC · 11 likes
Such a shame I can’t have one of these as a pet
Quoting Visakan Veerasamy (@visakanv): in new zealand there were the giant Moa birds and holy hell they were huge
Reply to @RafaRuizdeLira, 08:28 UTC · 2 likes
Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): @AaronBergman18 Slavery was abolished in 1865
Lol dark tweet
Reply to his own earlier tweet, 08:33 UTC · 4 likes
Aaron Bergman, April 30, 2025: Such a shame I can’t have one of these as a pet
Wtf they ate rocks
Reply to @pli_cachete, 13:52 UTC · 3 likes
Rota Is In Israel🚪🧎♂️ (@pli_cachete): I get a bit confused reading the arguments on both sides of the trans debate in the culture war.
Like, this is a political question right?? I think everybody agrees about the *facts* of the world. Why, then, are all of the discussions framed as factual ones?
For the most part they boil down to "That's not what I mean when I say X". Why not get to the core of the issue and discuss the rights / obligations that people should have instead of debating whether or not we should reserve this or that word for certain things?
(quoting @eigenrobot: sure ill bite this is dumb because the extent to which the terms "parent" and "woman" are imbued with biological or social significance is somewhat arbitrary in each case and there's no reason the biological degree for one must be connected to that assigned to the other)
Yes correct take
Tweet, 14:07 UTC · 21 likes
Impulse buy, new store link below
Thread, continued, 14:07 UTC · 5 likes
https://shrimpwelfareproject.shop/
Reply to @morallawwithin, 14:11 UTC · 2 likes
florence ⊮ (@morallawwithin): YES IVE BEEN WAITING FOR THIS
🦐🦐🦐🦐🦐
Tweet, 14:16 UTC · 64 likes
Talked to someone already quite sympathetic to animal welfare recently who laughed at the idea of shrimp welfare
Something is missing in my brain, it just doesn’t seem that weird to me
Thread, continued, 14:16 UTC · 19 likes
You have to get to electron suffering for it to feel viscerally weird, but even then I think that feels less weird to me than shrimp welfare seems to the masses
Reply to @caryatis, 14:26 UTC · 8 likes
Kayla (@caryatis): @AaronBergman18 Is it the sense of hierarchy (humans over animals, maybe some animals over other animals) that’s missing?
No, I’m very willing to say that one human is much more important than one shrimp
It just doesn’t seem that unlikely that shrimp are sentient and, if so, it doesn’t seem weird at all to try to help them
Maybe comfort with EV has something to do with it
Reply to @sam_atis, 14:31 UTC · 2 likes
Sam Glover (@sam_atis): Maybe because most people haven't internalised that you can help many smaller animals, so it seems weird to focus on animals that may or may not be sentient when you could focus on cows or animal shelters or whatever.
Yeah this makes sense
Tweet, 16:36 UTC · 194 likes
Two takeaways from Zuck on Dwarkesh, before I’ve even finished the episode:
1) Zuck doesn’t take the possibility of an algorithm-led/algorithmic progress-dominated intelligence explosion seriously
2) Zuck is imagining that the world looks fundamentally similar to rn in a decade
Thread, continued, 16:36 UTC · 24 likes
Needless to say I think these are both wrong
https://podcasts.apple.com/us/podcast/dwarkesh-podcast/id1516093381?i=1000705423020
Quote tweet, 17:22 UTC · 15 likes
Tfw South American strongman is on higher moral ground than the United States of America
Quoting Jon Favreau (@jonfavs): Wow. Trump’s deal with Bukele was for convicted criminals only, but Bukele quickly realized that’s not who he was getting:
Quote tweet, 23:55 UTC · 11 likes
Ahhhhh how are they so bad at this!!
There are like 16 things a smart undergrad + SOTA AI shouldn’t be tasked with doing and one of them is rewriting laws that affect hundreds of millions of people! You actually can do better!
Quoting Armand Domalewski (@ArmandDoma): there’s like dozens of smart right wing housing policy guys you could assign this but instead you’re tasking rewriting HUD regulations to an undergrad using ChatGPT?!?
Thread, continued, 23:55 UTC · 6 likes
It’s like looking at all of medicine and deciding the first thing you want to do to integrate AI is replace the world’s best and most specialized surgeons with a humanoid robot
End of April 2025. Previous month: March 2025. Next month: May 2025. Back to the archive index.