Aaron Bergman's tweets, February 2025
Every public tweet, reply, quote tweet and retweet that Aaron Bergman (@AaronBergman18 on X, formerly Twitter) posted in February 2025 and has not deleted: 641 posts, oldest first, times in UTC. Each heading links to the original post on X. When Aaron replies to or quotes someone, their tweet appears as a blockquote with their name; those words are theirs, and everything outside the blockquotes is Aaron's.
This page is one month of the complete archive of Aaron Bergman's tweets, which updates automatically every few hours. Previous month: January 2025. Next month: March 2025.
Saturday, February 1, 2025
Tweet, 05:11 UTC · 2 likes
So good, inject it into my veins
Thread, continued, 05:11 UTC · 1 like
https://joecarlsmith.com/2025/01/28/fake-thinking-and-real-thinking
Thread, continued, 05:11 UTC · 2 likes
Things I’ve “really” (ie really under this framework) thought about for more than a few minutes:
- What total utilitarianism means and implies
- Hard problem, especially illusionism about consciousness
- ELK AI challenge
- Sleeping beauty problem
Tweet, 05:11 UTC · 1 like
I’m almost certainly forgetting stuff from a long time ago, but I doubt I’m forgetting more than 1-3 things over the last few years
“Really” thinking about stuff for a fair amount of time feels quite rare
Thread, continued, 05:11 UTC · 1 like
With the utilitarianism stuff in particular, I kinda wanna claim “listen, I know you’ve thought about this and are a smart person, but its likely I’ve thought about it in a qualitatively different (and epistemically superior) way than you”
Quote tweet, 05:15 UTC · 13 likes
Not as TVogenic as January 6 but uh this seems like a constitutional crisis (probably not alone tbc, but like the executive branch simply and illegally ignoring congress)
Quoting Kelsey Piper (@KelseyTuoc): Leaving millions of people to die via an unconstitutional stop-work order is in fact by far the most important thing this administration has done, even if it's not the most flashy. We are abandoning to death people we could easily help and, no, that's not good for America.
Thread, continued, 05:15 UTC · 5 likes
Call it Trump derangement syndrome, fine, I think derangement is the right response. Trump is really that bad, and in a saner world he would be impeached
Quote tweet, 06:05 UTC · 192 likes
It makes sense to use stimulants to be at your best and sharpest during high stakes moments
Quoting Dissproportionately (@dissproportion): It isn’t even about whether Zyn is unhealthy. It’s about whether RFK is too stupid and weak to make it through *a senate hearing* without popping one. Come on.
Thread, continued, 06:05 UTC · 20 likes
Not an RFK fan (but I am a stimulant fan)
Reply to @allgarbled, 06:07 UTC · 67 likes
gabe (@allgarbled): Women love it when you tell them to be at a place at a time
I have learned this. They really want you to plan dates. They don’t wanna just ride the vibes and wind
🤷♂️
Tweet, 06:08 UTC · 6 likes
Orange man bad
Reply to @DimitrisPapail, 19:30 UTC · 2 likes
Dimitris Papailiopoulos (@DimitrisPapail): Reasoning models tend to yap longer when they are wrong, making it a useful signature for auto-verification
(quoting @AlexGDimakis: Discovered a very interesting thing about DeepSeek-R1 and all reasoning models: The wrong answers are much longer while the correct answers are much shorter. Even on the same question, when we re-run the model, it sometimes produces a short (usually correct) answer or a wrong…)
Is this just a proxy for the question being harder?
Reply to @livgorton, 19:57 UTC · 2 likes
Liv (@livgorton): i’m so glad my toe socks will arrive in time for valentine’s day. thanks amazon.
😊
Tweet, 20:38 UTC · 302 likes
I feel like the 40/5 (40% of the gains, 5% of the difficulty) of “Pause AI” is “Stop Sam Altman”
He high key appears to be an extremely competent, powerful, and charismatic supervillain
Reply to @mattyglesias, 20:40 UTC · 36 likes
(The tweet being replied to by @mattyglesias is no longer available.)
Sad but true. I wonder if there’s a general lesson here
Reply to @a_real_society, 21:29 UTC · 3 likes
society (@a_real_society): As a high school and college dropout I don't get the obsession with trades, people always focus on the top outcomes for trades but if an average person can get a somewhat useful degree they're going to be better off doing that and getting an office job
(quoting @BiblicalBeauty: No smart aleck responses, please… What trades are there that a young single woman could learn that could actually make good money without getting a college degree? I want my sons to begin learning things like construction, welding, electrical work, etc., as they get older, but…)
There’s gotta be a narrow sliver of people who would not do well in college, would do well in trades, and (biggest filter) *can competently parse the arguments for each and how they apply to themselves, in high school*
Cc @JesseTayRiver
Reply to @JesseTayRiver, 21:42 UTC · 5 likes
Jesse Smith (@JesseTayRiver): Heading to a bat mitzvah. I don't remember who it's for and I've exceeded a reasonable number of times I can ask (~10). What's my next play?
Ask Claude what to do
Reply to @Yoshua_Bengio, 21:45 UTC · 40 likes
Yoshua Bengio (@Yoshua_Bengio): A few reflections I had while watching this interview featuring @geoffreyhinton:
It does not (or should not) really matter to our safety whether you want to call an AI conscious or not.
1⃣We won't agree on a definition of 'conscious', even among the scientists trying to figure it out.
2⃣What should really matter are questions like:
▶️Does it have goals? (yes).
▶️Does it plan (i.e. create subgoals)? (yes).
▶️Does it have or can it develop goals or subgoals that may be detrimental to us (like self-preservation, power-seeking)? (yes, already seen in recent months with experiments with OpenAI's and Anthropic's models).
▶️Is it willing to lie and act deceptively to achieve its goals? (yes, seen clearly in the last few months in these and other experiments).
▶️Does it have knowledge and skills that could be turned against humans? (more and more, see comparisons of GPT-4 vs humans on persuasion abilities, recent evaluations of o1 on bioweapon development knowledge).
▶️Does it reason and plan over a long enough horizon to be a real threat if it wanted? (not yet, but we see the planning horizon progressing as AI labs pour billions into making AI more agentic, with Claude currently better than humans at programming tasks of 2h or less for a human, but not as good for 8h and more, already).See https://arxiv.org/abs/2412.04984, https://arxiv.org/abs/2412.14093 and https://metr.org/blog/2024-11-22-evaluating-r-d-capabilities-of-llms/ for all the details.
It mostly doesn’t matter for traditional notions of “safety,” true, but it matters tremendously for intrinsic/ethical reasons
Reply to @davidad, 21:46 UTC · 2 likes
davidad 🎇 (@davidad): That’s the sense in which Yoshua says it does not really matter to our safety.
That doesn’t mean it has no ethical significance.
The issues are intertwined: the acceptability of containment techniques may depend on (folks’ best guess at) any subjective consequences for the AIs.
(quoting @davidad: @jessi_cata @jd_pressman @AndrewCritchPhD This too seems a disguised query. The heart of the matter is whether AIs are entitled to membership in a human moral community and/or to rights under its social contract (or if not, what other status is apt—as many countries do offer…)
Reply to @mmt_lvt, 22:23 UTC · 8 likes
(The tweet being replied to by @mmt_lvt is no longer available.)
I don’t really think “physical” is a natural category here - it’s all mediated by our brain, even though for evolutionary reasons our brain “says” that often pain is located somewhere on our bodies
It’s not hard to imagine pain and other types of suffering in disembodied minds
Thread, continued, 22:23 UTC · 11 likes
I certainly hope you’re right and AIs can’t suffer, but I think we’re so radically unsure about the nature of sentience that it’s very hard to be ~certain of anything
Thread, continued, 22:24 UTC · 8 likes
Also plenty of human suffering isn’t usually considered “physical pain” eg severe depression
Sunday, February 2, 2025
Tweet, 00:45 UTC · 14 likes
25% tariff on Canada (???) shows that you basically can’t model Trump through the intentional stance
He just does things, in the same way he just says random bullshit that doesn’t bear at all on the truth or his (past, present, and future) behavior
Reply to @austinc3301, 00:50 UTC · 21 likes
Agus 🔸 (@austinc3301): Why do you think Musk is worse?
Setting aside who’s heart is closer to Moral Truth, Altman seems way more strategic, ambitious, and less honest in a way that makes him more dangerous
Like Musk has some bad takes on public policy but he just tweets those
Thread, continued, 00:52 UTC · 9 likes
Society kinda knows how to deal with rich people with bad takes
Altman is sui generis
Reply to @StefanFSchubert, 00:54 UTC · 2 likes
Stefan Schubert (@StefanFSchubert): Afaik he’s long been in favour of tariffs, so I’m not sure it’s that random
Not totally random but I don’t think this pops naturally out of a goals/means/end/rational actor model
Reply to @austinc3301, 01:00 UTC · 6 likes
Agus 🔸 (@austinc3301): Aaron they literally just gave the rich person with bad takes insane amounts of political power
Yeah but even that is something the system is kinda familiar with
I guess my point is that if ~the system~ couldn’t basically handle Musk, it wouldn’t have survived cultural evolution, but this is not true for Altman
Reply to @eigenrobot, 01:01 UTC · 14 likes
eigenrobot (@eigenrobot): @AaronBergman18 5% is a massive underestimate imo
Maybe I’m just more pessimistic about how hard it would be to pause AI development
Reply to @mattyglesias, 04:02 UTC · 3 likes
(The tweet being replied to by @mattyglesias is no longer available.)
They should at least give me the data so I can run some fun regressions
Reply to @tszzl, 06:19 UTC · 23 likes
roon (@tszzl): that’s a completely insane position tbh
Maybe 5% is too low
Reply to @justjoshinyou13, 06:24 UTC · 19 likes
Josh You (@justjoshinyou13): @AaronBergman18 I don't think he's that much more impactful than a replacement-level CEO
He founded OpenAI, convinced many smart people to donate a lot of money to it as a nonprofit and to work there as part of a noble project, gained enough social (?) capital to become unfirable, turned the lofty nonprofit into a for-profit
These aren’t normal CEO behaviors
Reply to @tszzl, 06:32 UTC · 38 likes
roon (@tszzl): what do you honestly think happens if openai falls over dead? everyone forgets about agi and moves on?
I mean to the very hazy story involves DeepMind and Anthropic basically being more responsible + cooperative actors both for idiosyncratic personnel reasons and bc you generally get more slack as the size of an oligopoly decreases
But more importantly, OpenAI wouldn’t fall over dead and Sama seems like a ~uniquely (bad*competent) actor
It’s like Yud’s point about knowing that a grandmaster chess player will beat you even if you don’t know how; he has a pattern of deceptive and unsafe behavior that’s probably gonna continue even if I don’t know exactly what that looks like
Tweet, 22:16 UTC · 52 likes
Steelman of every bad economic policy including tariffs is “less wealth -> less factory farming”
Thread, continued, 22:16 UTC · 13 likes
I mostly don’t think this checks out but that might be cope
Quote tweet, 22:19 UTC · 10 likes
What the fuck
Quoting Fermat's Library (@fermatslibrary): From John Wallis, 1656
Monday, February 3, 2025
Quote tweet, 01:39 UTC · 10 likes
Noooooooooooooooooooooo
Quoting Roman Hauksson-Neill (@RomanHauksson): I showed up to the Schelling point meetup at the ticket booth of Reunion Tower at 12:00pm on Feb 1st and no one joined me :(
Reply to @adrusi, 02:06 UTC · 2 likes
autumn (@adrusi): wait
with the elimination of the de minimis rule
and probably like a hundred million frickin temu/etc parcels in transit
customs is gonna be so clogged up that im NEVER getting my turkish antidepressants
Which drug 👀
Reply to @herchosenpath, 02:09 UTC · 13 likes
Mama T🌻 (@herchosenpath): I love animals so bad. Like how do you have this much personality? Dude smiled 😭
Quote tweet, 22:22 UTC · 24 likes
The “little boys” meme is kinda dumb because I think the actual objectionable thing here is the fact that a handful of people who were never confirmed by the senate have an unusual amount of power
Quoting rohit (@krishnanrohit): Mutual is running the DOGE audit of the entire federal govt. This timeline is crazy!
Thread, continued, 22:22 UTC · 3 likes
The ideal age for this kind of thing, at some optimal combination of crystallized and fluid intelligence is, what, like 40?
So 25 or whatever isn’t ideal, but American society including government is largely a gerontocracy…
Thread, continued, 22:22 UTC · 3 likes
So honestly, 20-25 kinda sounds like 70-75 in terms of “distance from ideal age”
And nobody objects to 73 year olds having huge amounts of political power (although maybe they should)
Reply, 23:20 UTC · 1 like
Idk enough about the situation; I’m not claiming it’s illegal. Plausibly norm breaking but not even sure about that
Tweet, 23:26 UTC · 6 likes
Does anyone have a Deep Research output they care to publish/share yet?
At some level of quality it’s worth it for me, but $200/mo isn’t trivial
Tuesday, February 4, 2025
Reply to @AndyMasley, 03:19 UTC · 8 likes
Andy Masley (@AndyMasley): I do start to notice how many people (myself included) sometimes feel like they're just doing next token prediction in conversations. If they use the word "Bolster" or begin any sentence with "But I mean look here's the thing" it's a dead giveaway.
Oh yeah I absolutely do this
But I do think that at least some of the time I’m doing something different, like there are 2 systems that I can use (next token prediction is the default/majority one tho)
Reply to @robbensinger, 03:43 UTC · 3 likes
Rob Bensinger ⏹️ (@robbensinger): If you personally got hooked by the first Sequences post you read: what hooked you?
If you personally bounced off the first Sequences post you read: what made you bounce off?
I think I bounced off because it’s so long
I liked the 5% that I listened to, just not enough (evidently) to want to listen to ~50 hours
Maybe this is ~irrational~ though and I should suck it up
Reply to @DavidSartor0, 05:39 UTC · 2 likes
David Sartor (@DavidSartor0): @trevposts @AndyMasley I'll look out for this one.
I’ll also look out for this classic example
Tweet, 06:09 UTC · 28 likes
Feeling doomy on many levels :(
Reply to @izzyz, 06:11 UTC · 3 likes
izzy (@izzyz): @AaronBergman18 we're all in this together
Well sorta…
Reply to @izzyz, 06:24 UTC · 2 likes
izzy (@izzyz): @AaronBergman18 what are you ruminating on?
Not a complete answer but I think it’s more likely than not that AI takeover happens and this is getting less and less abstract by the day
US Federal government stuff feeds into this but also kinda sucks on its own
Also personal stuff I don’t feel like spelling out in detail rn
Thread, continued, 06:26 UTC · 6 likes
But the first thing is personal too tbc
Iirc Paul Christiano said on a podcast that he thought he was 50% likely to die from AI but that the potential upside made the whole situation positive EV
I kinda feel the same but wish I had like 20 or 30 more years to do stuff first lol
Reply to @Hayes_Inc, 06:36 UTC · 2 likes
matt 🕊️⏸️ (@Hayes_Inc): p(doom) is prob like 80%+ at a minimum, but hey, go chiefs
👍
Reply to @zinniaa_3, 07:53 UTC · 363 likes
(The tweet being replied to by @zinniaa_3 is no longer available.)
It was a fucking genocide, this was unironically very based
Tweet, 14:27 UTC · 4 likes
The whole irony of Elon and DOGE is that, empirically, the most successful/effective organizations (eg FAANG) pay a good number of people a large amount of money, along with very nice working conditions and other perks, to attract top talent
Thread, continued, 14:27 UTC · 4 likes
There’s actually a legitimate, if difficult, project around making government work better - modernize public-facing tech, grease the wheels of congressionally-mandated bureaucracy, just implement the whole @IFP gov reform wishlist
Thread, continued, 14:27 UTC · 3 likes
But DOGE *isn’t doing these things*
It’s taking the same strategy that got Twitter to lose 80% of its valuation and applying it naively to the federal government
Thread, continued, 14:27 UTC · 3 likes
Luckily Elon isn’t CEO and owner of the federal government, so it won’t be as catastrophic, but it’s a matter of degree rather than kind
Reply, 15:01 UTC · 1 like
I mean there are policy levers to increase demand for labor (a public choice/political economy problem is that businesses want plentiful cheap labor tho)
Quote tweet, 15:12 UTC · 14 likes
If you follow this line of reasoning seriously, you get to “why is OpenPhil spending money so slowly?”
Not sure what comes after that
Quoting Peli Grietzer (@peligrietzer): Do any '90% credence in ASI in 2 years' people want me to loan them $200 today and pay me $2,000 in 2 years?
Reply, 15:14 UTC · 1 like
@RafaRuizdeLira
Quote tweet, 15:27 UTC · 6 likes
Democrats: *pay lip service to “AAPI awareness” or whatever and then work to systematically disadvantage Asians throughout American society purely on the basis of race*
Asians: *move right*
Democrats:
Quoting Josh Barro (@jbarro): Asian voters now significantly more Republican than white voters in NYC
Quote tweet, 15:35 UTC · 14 likes
Why do we see a clear east-west pattern?
Naively I’d guess if anything it would be north-south bc men prefer colder climates
Quoting M. Nolan Gray 🥑 (@mnolangray): @finnawinna474 Only major straight dating market west of the Mississippi where men have an advantage.
Reply to @pingthebird, 15:45 UTC · 2 likes
(The tweet being replied to by @pingthebird is no longer available.)
I had similar levels 6 years ago. Welcome to TRT! (Now I’m at 530 natural)
Reply to @beala, 15:55 UTC · 3 likes
Alex Beal 🆎 (@beala): Wait does trt raise your natural level after discontinuing?
I don’t think reliably, my situation was more complex and pretty atypical
Reply to @beala, 16:02 UTC · 2 likes
Alex Beal 🆎 (@beala): If you're comfortable saying, curious if you went thru your GP or some clinic. Thinking about doing this myself.
Normal pediatric endocrinologist
I was at 18ng/dl which is so severely low that anyone would treat. In many ways it’s better to be super low than borderline
Thread, continued, 16:02 UTC · 3 likes
I was 18 but still seeing pediatric
Reply to @beala, 16:05 UTC · 3 likes
Alex Beal 🆎 (@beala): Thanks yeah this is my problem. 10 years ago I was borderline and an endocrinologist looked at me like I was crazy when I asked. I'm older now tho so maybe it's lower.
If you have the money and are consistently borderline, clinic may be the way to go
Tweet, 16:13 UTC · 131 likes
I’m tempted to be negatively polarized into supporting USAID unconditionally but my (uncertain, not super informed) sense is that probably 90% of the programs are useless?
Like maybe just do cash transfers lol
Thread, continued, 16:13 UTC · 26 likes
Ofc if it’s all or nothing I’ll stand by USAID, in general resource transfers to poor countries are probably good, but the left should take a look at this “effective altruism” thing for inspiration and an ungodly amount of research on the efficacy of various programs to help the poor
Reply to @pingthebird, 16:19 UTC · 5 likes
(The tweet being replied to by @pingthebird is no longer available.)
Ah, idk shit about trans stuff but good luck!
Reply to @panickssery, 17:02 UTC · 4 likes
Arjun Panickssery (@panickssery): @AaronBergman18 90% of USAID being useless according to your view would mean that the good that comes from PEPFAR is costing American taxpayers $35-45B
How does that promote long-term utility (even leaving aside the other concerns about the scope of government/obligations to its own citizens)
Long term utility? Like long long term, none of this matters
But that’s $40k/life saved (assuming the $40B thing is annual?) plus a lot of suffering prevented besides
A better buy of human welfare than almost anything the government does*
Tweet, 18:00 UTC · 9 likes
O3-mini-high: here’s a number
Me: you made up the source for this number
o3-mini-high: no I didn’t, *tells me that I gave it that number to work with*
[o3 was right, I was wrong, it’s over]
Reply to @SpacedOutMatt, 18:23 UTC · 19 likes
(The tweet being replied to by @SpacedOutMatt is no longer available.)
based, I have been informed in DMs that my original take is pretty bad and it's much closer to 50% cash transfer-equivalent
I am inclined to defer to this
As always, best way to get a right answer is to poast wrong answer
Reply to his own earlier tweet, 18:24 UTC · 18 likes
Aaron Bergman, February 4, 2025: Ofc if it’s all or nothing I’ll stand by USAID, in general resource transfers to poor countries are probably good, but the left should take a look at this…
Ok maybe it's good actually
Reply to @pitdesi, 19:42 UTC · 342 likes
Sheel Mohnot (@pitdesi): Obvs there is a lot of waste out there. This sounded crazy so I looked into it, turns out it's NBD
The money went to a lot of different research, part of which was $1,000 to measure how shrimp respond to changes in water quality with a treadmill.
Enjoy the video!
video attached
(quoting @Realalexstone: THREE. MILLION. TAX. PAYER. DOLLARS. WAS. USED. TO. STUDY. HOW. FAST. SHRIMP. CAN. RUN. ON. A. TREADMILL. You’ve gotta be kidding me.)
I don’t know what the treadmill was for specifically, but knowing how water quality affects shrimp is actually a super important question
Thread, continued, 19:43 UTC · 1 like
https://www.shrimpwelfareproject.org/shrimp-welfare-report
Tweet, 22:09 UTC · 11 likes
Simulator interns pitching simulation 48391014-Ω-φ: “…and last but not least, Donald Trump is gonna be president during the singularity”
Wednesday, February 5, 2025
Reply, 03:41 UTC · 1 like
Lmao (I went to therapy in elementary school for exactly this)
Quote tweet, 04:17 UTC · 3 likes
Can’t we have a little liberalism, as a treat
Quoting Dan Lamothe (@DanLamothe): Have verified that this memo circulating from West Point is real.
H/T @PeterLucier
Reply to @JgaltTweets, 11:54 UTC · 4 likes
JgaltTweets (@JgaltTweets): 'Astronomers say that a newly identified space rock, potentially as big as a football field, has a better than 1% chance of crashing into Earth on Dec. 22, 2032.'
https://www.npr.org/2025/02/02/nx-s1-5282071/asteroid-meteor-2024-nasa-earth
I don’t think we should be concerned about such speculative future risks
Reply to @liron, 12:09 UTC · 4 likes
Liron Shapira (@liron): Crazy how government speed went from slow to literally Elon Musk
This is a pretty silly 1d axis
It’s not like the same things are happening but patents and leases are approved quicker
Reply to @Simeon_Cps, 12:21 UTC · 5 likes
Sy (@Simeon_Cps): Why has nobody trained a good model capable of saying "I don't know"? Is it that hard or is it just that nobody cares?
Idk what I’m talking about but isn’t this (mostly) not what you’d expect from naive scaling
Like the whole paradigm is that models fake it and scaling helps them make it. Or in more neutral language, they give their (implicit) best guess at everything
Thread, continued, 12:23 UTC · 2 likes
I say “mostly” bc I do think this should work for unknowable things like “what object is in my right hand”
From interacting with Claude, I get the sense that it doesn’t “really” know what it doesn’t know - it just says it might be hallucinating about topics it identifies as niche regardless of true represented internal confidence
Reply to @demiurgently, 14:57 UTC · 8 likes
sydney (@demiurgently): remember: the only thing standing between you & everything you’ve ever wanted is the right note taking app
Ok not *everything* but high quality transcription and LLM processing does open up workable voice notes
This is like a real thing
Tweet, 16:25 UTC · 48 likes
Wait deep research only takes 5-10 minutes? Lol
They could probably make it seem more impressive by adding like an 8 hour meaningless pad on top to make it seem more effortful
Quote tweet, 16:43 UTC · 11 likes
Among many other stupid things, one stupid thing is that this is effectively punishing researchers who may be totally un- or anti-woke but thought that including a few benign keywords would help get their totally legit research published
Quoting Darby Saxbe (@darbysaxbe): 🚨BREAKING. From a program officer at the National Science Foundation, a list of keywords that can cause a grant to be pulled. I will be sharing screenshots of these keywords along with a decision tree. Please share widely. This is a crisis for academic freedom & science.
Thread, continued, 16:43 UTC · 6 likes
Trump administration would have unfunded the research undergirding GLP-1 agonists if one of the authors had pronouns in a formalities/citation bit somewhere
Literally doing policy via regex
Reply to @beala, 17:05 UTC · 2 likes
Alex Beal 🆎 (@beala): Submitted my publication bias research proposal and now my NSF account says this 😭😭😭
Lol
Tweet, 18:15 UTC · 4 likes
This is less ambitious than what @PauseAI and friends are calling for, but a somewhat more plausible ask is “pause (frontier) AI until the next POTUS election” or at least “no self-improving AGI” until than
Thread, continued, 18:15 UTC · 2 likes
That’s like lengthening timelines by 2-3 years, something you might be able to do with the buy-in of like a single digit number of individuals
Thread, continued, 18:15 UTC · 3 likes
I really don’t think Trump should be president during the singularity
Reply, 18:25 UTC · 2 likes
I mean we’re talking about like 4-10 people (4 being Altman, Jensen Huang, Demis Hassibis, Dario Amodei) and it’s not clear why they’d care that it’s politicized; these are smart people who presumably don’t think highly of Trump
Reply to @AnthropicAI, 21:08 UTC · 25 likes
Anthropic (@AnthropicAI): Nobody has fully jailbroken our system yet, so we're upping the ante.
We’re now offering $10K to the first person to pass all eight levels, and $20K to the first person to pass all eight levels with a universal jailbreak.
Full details: https://hackerone.com/constitutional-classifiers
(quoting @AnthropicAI: New Anthropic research: Constitutional Classifiers to defend against universal jailbreaks. We’re releasing a paper along with a demo where we challenge you to jailbreak the system.)
This is very cool but actually like an OOM lower than it seems to be worth
Thursday, February 6, 2025
Reply to @bijibiji2023, 00:08 UTC · 1 like
hoff 🧒🏻😻 (@bijibiji2023): some of my cousins have this mindset. actually frightening how much they’ve deluded themselves with so little danger…
(quoting @soulkhan: it is lowkey abusive to raise your kids with this level of instilled paranoia and delusion and all for the sake of protecting the moral deadend of our people)
I’m willing to believe that it exists but to the best of my knowledge I have literally never personally faced antisemitism irl
And online it’s been maybe a handful of DMs here over like 5 years (all by random anon accounts, idgaf)
Reply to @banterrealism, 00:18 UTC · 3 likes
Sex Haver (@banterrealism): Can someone steelman the position:
“The population should NOT be educated and competent en mass”
I love steelmanning things like this. Here are some possibilities:
- humans were better off before agriculture and you don’t get legit education without agriculture
- education raises GDP which causes factory farming/AI doom
- not super sure what competent means; arguably making people “competent” in general is just prohibitively difficult or impossible
- an ideal world is hedonium/rats on heroin; education and competence are like totally irrelevant in this world
Tweet, 00:32 UTC · 8 likes
Literal shower thought:
Entrepreneurship is really 2 things in a trench coat:
1) innovation/object-level problem solving/creating goods and services
2) doing the operational work of bringing together and coordinating multiple resources that won’t otherwise find each other
Reply to @catehall, 00:35 UTC · 4 likes
Cate Hall 🔸 (@catehall): I'm looking for potent examples of someone on Twitter just straightforwardly admitting they were wrong about something after encountering a good counterargument, rather than digging in. What examples of this have you seen?
Idk about potent but I’ve used the exact phrase “I was wrong” a couple times here
Reply to @banterrealism, 00:57 UTC · 3 likes
Sex Haver (@banterrealism): @AaronBergman18 I also wasn’t marked gifted or whatever In elementary
I was put in support classes
I mostly just mean that it’s such a vague word
Like what do you need to be able to do to be competent: tie your shoes? Change a tire? Build a durable shelter?
Reply to @cremieuxrecueil, 02:31 UTC · 4 likes
Crémieux (@cremieuxrecueil): I just saved an ODing hobo amputee
Based
Quote tweet, 04:12 UTC · 16 likes
The more I think about it, the more I think 1780s-era morality is pretty much correct
Quoting vitalik.eth (@VitalikButerin): The more I think about it, the more I think 2013-era morality is pretty much correct:
* free speech good
* starting companies and making good products good
* monopolies bad, vendor lock-in bad
* democracy good
* greed bad
* trying to achieve national security through oppressing people bad
* cosmopolitan humanitarian values, caring about faraway people, etc goodThings we've "learned" since then have been harmful more than helpful
Thread, continued, 04:12 UTC · 2 likes
https://en.wikipedia.org/wiki/An_Introduction_to_the_Principles_of_Morals_and_Legislation
Thread, continued, 04:12 UTC · 4 likes
(Obviously this is a utilitarianism joke, lots of terrible things were happening in 1780)
Quote tweet, 04:18 UTC · 22 likes
God bless community notes
Quoting Armand Domalewski (@ArmandDoma): counterpoint:
Reply to @the_wilderless, 04:19 UTC · 23 likes
I (@the_wilderless): Trust me I hate it too, but:
walking around with headphones blocks your energy channels
Well sometimes I want my energy channels blocked
Reply to @martiniposting, 04:25 UTC · 3 likes
lari (@martiniposting): asked a man how often he gets a haircut and he said anywhere between 2 and 8 months
Real
Reply to @trevposts, 04:26 UTC · 4 likes
Trevor Levin (@trevposts): In terms of the amount of ignorance and incuriosity on display in the average thing I saw on here today, I think it was the dumbest day in the history of my twitter timeline.
Idgaf Travis
Reply to @br___ian, 05:01 UTC · 2 likes
bri (@br___ian): how young, how innocent the brian who asked this question was
Solid prompt
Tweet, 06:53 UTC · 11 likes
I’m kinda a fan of wokeness when it was articulated unironically as like “being aware of subtle prejudice”
Tweet, 15:52 UTC · 3 likes
Interior design of the reception area is a 100% perfect indicator of how much VC money and/or tech money subsidization a given irl business has
Tweet, 16:00 UTC · 27 likes
I really hope @CommunityNotes is training a model to predict note from tweet because that’s probably the best almost-ground-truth data you’re gonna get in complex and non formally-verifiable domains
Thread, continued, 16:00 UTC · 4 likes
Ofc it’s an empirical question whether it generalizes better to judging arbitrary queries than eg Grok prompted well, but I expect it would
Cc @jaybaxter
Tweet, 17:17 UTC · 6 likes
Headlines like this are funny bc you know if this is the worst they could dig up on a person, they’re relatively straight and narrow
Thread, continued, 17:17 UTC · 3 likes
No QT bc I’m blocked 🤷♂️
Reply to @jaybaxter, 17:33 UTC · 4 likes
Jay Baxter (@jaybaxter): Yeah, and the cool thing is anyone external can do this too if they use the API to get post text!
Notes data: free
GPUs: $100
X api: $585949910192466
Help my family is dying
Reply to his own earlier tweet, 18:02 UTC · 4 likes
Aaron Bergman, February 6, 2025: I really hope @CommunityNotes is training a model to predict note from tweet because that’s probably the best almost-ground-truth data you’re gonna get in…
If I do the data scraping I hope someone else will cook
Tweet, 18:32 UTC · 4 likes
Dario is making alooooot of sense here
In a sane world this would be like the Mitt Romney position and you’d have more hawkish safety folks to the metaphorical left
https://www.chinatalk.media/p/anthropics-dario-amodei-on-ai-competition
Reply to @colinmegill, 18:33 UTC · 2 likes
Colin Megill (@colinmegill): @AaronBergman18 @jaybaxter @CommunityNotes Seriously
Dw I’ve got 8 parallel chrome windows scraping as we speak
Reply, 22:21 UTC · 1 like
IME Claude sonnet is ~perfect
I did get it to fail once with a super high res complex technical diagram
Reply to @Meaningness, 22:38 UTC · 3 likes
David Chapman (@Meaningness): As the sages teach, cognitive decoupling is important and rare.
When I say “philosophy is bad” or “preference falsification is common,” replies mistake that as waging a culture war battle.
No, I meant only what I said. 🙄
🫡
Quote tweet, 22:55 UTC · 20 likes
This is probably a steelman case of how “cancel culture” can work reasonably well:
voluntary resignation from some position after fair and accurate accusations that meaningfully indicate low confidence in the person’s ability to perform that position’s duties well
Quoting Jeff Stein (@jstein_sun): Via the Wall Street Journal
Thread, continued, 22:55 UTC · 24 likes
I don’t think he should be blacklisted from everything forever, but yeah this sounds like an old fashioned racist who shouldn’t be in a policymaking position
Tweet, 23:01 UTC · 7 likes
Maybe now would be a good time to retvrn to kidney market discourse
We have essentially a Pareto improvement policy lever that would also slash into 1% of the federal budget (that’s a lot) (the policy lever is kidney donor compensation and the budget item is dialysis)
Tweet, 23:05 UTC · 83 likes
Underrated lens on Elon having political power is that you should expect whoever is in his shoes (hyper-wealthy, well-known) to have made a bunch of big bets and gotten lucky each time (this is a boring anthropics-adjacent take not a leftist anti-billionaire take)
Thread, continued, 23:05 UTC · 36 likes
So basically we should expect whoever is empowered via a process of like
Step 1) become the richest person
to have an inflated sense of their own judgement and decisionmaking
A bunch of counterfactual Elons were just a bit too early or too late to have PayPal go big
Reply to @sixers2772, 23:33 UTC · 11 likes
Sixers (@sixers2772): Why would he have to have gotten lucky each time? I might agree, but am curious to hear your reasoning
Admittedly after a bunch of big wins you can get get away with smaller losses but if either PayPal or Tesla had basically failed or been small, I think it’s unlikely Elon would be a major public figure today (the kind of person salient enough for trump to appoint)
Reply to @sixers2772, 23:46 UTC · 6 likes
Sixers (@sixers2772): Potentially, if he's just super competent, he could have just started something else that would takeoff after those failed?
An argument for him being very competent could be that he has succeded at a lot of things that generally one fails at
Yeah I think the second thing is basically true in a relative sense, but an *unselected* person of musk’s talent probably merely had say a 50% chance of founding any successful company instead of a 100% chance of founding 3-4 (or ~80% if you want the ex post outcome to be >50% of 3/3 wins)
(This is all made up numbers and a simplified model, I think the point still holds)
Reply to @sixers2772, 23:53 UTC · 1 like
Sixers (@sixers2772): "I am lucky+competent because I am successful"
I don’t fully understand what you’re saying. I totally agree success implies both luck and competence
Maybe it’s a little more complex than I initially implicitly thought bc it’s an empirical psychological question the degree to which people or Elon specifically adjust for things like the very concern I’m raising
But intuitively maybe there’s a partial adjustment but it doesn’t go all the way; like you should expect the world’s richest person to overestimate their own ability bc the human brain didn’t evolve in a context of 8b relevant humans, or maybe say 1k plausible contenders for richest person; maybe in the ancestral environment the analogous case had a population of like n=4
Idk if this makes any sense
Friday, February 7, 2025
Reply to @liz_love_lace, 00:05 UTC · 4 likes
Liz Lovelace (@liz_love_lace): It's odd to me that doctors ask for your symptoms first, and then run diagnostics. Especially for things like bad knees. I feel like asking *after* would make you less likely to be subconsciously biased when you do the initial diagnostics
A huge function of doctors is gatekeeping expensive shit like MRIs
There are a few blood and urine tests that could be run like this though
Tweet, 00:26 UTC · 8 likes
Those who donate >=10% of your income, what is (your best guess of) the net effect of donation on your hedonic quality of life
Thread, continued, 00:32 UTC · 5 likes
I’m weirdly uncertain what the results are gonna be but this seems like a weirdly important question
I’ve recently updated towards “people making over like $100k actually are giving up real hedonic utility by donating, there’s not so much free lunch”
Reply to @justjoshinyou13, 00:50 UTC · 2 likes
Josh You (@justjoshinyou13): @AaronBergman18 (I'm 29)
Geriatric
Reply to @monke_io, 00:54 UTC · 7 likes
(The tweet being replied to by @monke_io is no longer available.)
Is there even transgender theory? IME terms like that are 99% used by haters, 0.9% by legit academics, and 0.1% by actual trans people
Afaict very few trans ppl are interested in making strong metaphysical claims about what gender they “really” are, if that’s what u mean
Thread, continued, 00:54 UTC · 7 likes
Like I can totally imagine having gender dysphoria and wanting to transition without changing any propositional beliefs
Reply to @monke_io, 01:11 UTC · 6 likes
(The tweet being replied to by @monke_io is no longer available.)
I mean like many things, popular catchphrases aren’t best understood as analytic philosophy
I think you’re just applying way too philosophical a lens to what is mostly a question of pragmatic public policy and social norms
Thread, continued, 01:13 UTC · 6 likes
Like who cares whether trans women “are” women in some fundamental sense? The relevant questions involve public policy and social norms, and I don’t think in any case these hinge on the answer to fundamental questions of metaphysics or the nature of personal identity
Thread, continued, 01:14 UTC · 6 likes
FWIW I think the better understanding of eg a tweet that says “trans women are women” is basically just tribal affiliation and signaling (not necessarily bad things tbc), not advancing some substantive theory of the nature of gender
Reply to @monke_io, 01:15 UTC · 1 like
(The tweet being replied to by @monke_io is no longer available.)
Oh lol. I guess I misinterpreted
Reply to his own earlier tweet, 02:13 UTC · 5 likes
Aaron Bergman, February 6, 2025: If I do the data scraping I hope someone else will cook
Update: I couldn’t resist so yeah should have a pretty nice dataset of 1.6M tweets and notes in a week of runtime. I’ll probably put it on GitHub and hope for the best
Quote tweet, 02:44 UTC · 11 likes
Did everyone expect the entire civil service to just fall in line when faced with blatantly illegal executive orders?
Like not a single department head somewhere said “no sorry this is illegal I won’t physically do the thing”?
Quoting Will Stancil (@whstancil): If DOGE shows up on your doorstep and orders you to do something, your response needs to be “On whose authority?” Because they do not have any more authority to do half this stuff than I do.
Thread, continued, 02:44 UTC · 2 likes
Rule of law (like free speech) is about institutions and official mechanisms, yes, but it should also be something of a person-level value beyond and aside from that - one that sometimes comes into irreconcilable conflict with other values but still
Thread, continued, 02:46 UTC · 2 likes
I don’t wanna trivialize losing one’s job, but the stakes aren’t *that* high here…
Thread, continued, 02:47 UTC · 2 likes
The archetypal thing that comes to mind in this domain is like the soldier ordered to commit a war crime on the battlefield where the policy of whether to follow orders has immediate life-or-death stakes
Reply to @sullyj3, 02:57 UTC · 3 likes
░▒▓▒░▒▓▒░▒▓▒░ (@sullyj3): @JMannhart @AaronBergman18 What do you mean by well known among billionaires? like the super rich discuss these issues with each other? How do you know? This seems super interesting
Jonathan the secret is out
Quote tweet, 03:39 UTC · 145 likes
Wait what the fuck
This is the answer to how airplanes fly. It breaks my model of the world in an equal and opposite way to airplanes themselves
They cancel out. Airplanes are real after all (or maybe I should update to petroleum being fake)
Quoting John Arnold (@johnarnold): A single aircraft fueling hose delivers more joules of energy per minute than a nuclear power plant. The energy density of liquid fuels is incredible.
Thread, continued, 03:40 UTC · 16 likes
(o3-mini fact checked for me and said this is basically true although it’s close)
Thread, continued, 03:48 UTC · 45 likes
In terms of flight time, a large nuclear plant could only support 8 large jets at once
I would have probably guessed like 10k
Reply to @IsaacKing314, 03:49 UTC · 33 likes
Isaac King 🔎 (@IsaacKing314): @AaronBergman18 Flying takes very little energy; you can glide for hours with none at all. It's the air resistance that's a problem, and equally so for cars.
Sad, we should get rid of all the air
Reply to @AFpost, 04:09 UTC · 2 likes
AF Post (@AFpost): Israeli PM Netanyahu gave Trump a “golden pager” during White House meeting.
Follow: @AFpost
Bruh mossad has a 3d inside view with high quality audio of the White House fr
Reply, 04:18 UTC · 5 likes
wtf why don’t they go faster
Tweet, 04:19 UTC · 3 likes
Resist lib energy needs to retvrn
Reply, 04:24 UTC · 10 likes
Lol sorry I meant why don’t nuclear plants burn uranium faster but that was not clear
Reply to @bijibiji2023, 17:01 UTC
hoff 🧒🏻😻 (@bijibiji2023): without insurance coverage I got T (re)prescribed to me within 2 hours of me scheduling an intake appointment at planned parenthood. from start —> T in my hands the process took 3 hrs and cost $300. which isnt cheap but doable for many.
Do you think it would be better if the process was a bit longer/harder?
Reply to @yhdistyminen, 17:14 UTC · 47 likes
bob's burgers urbanist 🐿️ (@yhdistyminen): I do not think stupid social media activity should ruin a kid's life, either. But I also don't think "not giving a kid unfettered access to sensitive government information" constitutes "ruining a kid's life."
image attached
(quoting @JDVance: Here’s my view: I obviously disagree with some of Elez’s posts, but I don’t think stupid social media activity should ruin a kid’s life. We shouldn’t reward journalists who try to destroy people. Ever. So I say bring him back. If he’s a bad dude or a terrible member of the team,…)
Correct take. It’s fine actually to infer from someone’s stated beliefs that they shouldn’t be trusted to do a thing
Quote tweet, 17:15 UTC · 6 likes
It’s almost a little silly that this exists but I think in many cases the cost-benefit does check out (ie writing a public comment for some bill)
It’s not totally meaningless if regulators see an overwhelming level of concern from the most engaged citizens
Quoting sarah (@littIeramblings): fyi, literally anyone with takes on what the US govt should be doing about AI can submit up to 15 pages of them using a link in the next tweet
Reply to @vivian_belenky, 17:19 UTC · 1 like
viv 🔸 (@vivian_belenky): It didn't used to be this easy either. It used to be an enormous pain in the ass. And everyone hated it.
Yeah that sounds bad tbc
Idk I’m inclined to think this is fine tbc, but it’s not obvious to me that 3 hours start to finish is the ideal gatekeeping structure for major life changing decisions
Quote tweet, 17:25 UTC · 3 likes
Yes great point, and I think the key difference is whether normal capitalist incentives apply
Housing, medicine, and government are extremely slow bc this is not the case
(I’m pretty sure I could get a couch today if I wanted)
Quoting David (@DavidSHolz): sometimes it feels like there's 2 levels of services in the world, there's Uber, Amazon, IKEA, Instacart - everything in 30 minutes or next day - then you hit a wall - 2 months for a doctor, 6 months for a couch, 16 months for city permits - imagine everything running *at speed*
Reply to @vivian_belenky, 17:30 UTC · 2 likes
viv 🔸 (@vivian_belenky): In my experience transmasculine people agonize A Whole Lot about this decision and I know Quite A Few who opted (so far) not to do it for a variety of personal reasons. I am one such! In my experience, once you're in a place to make that decision, any gatekeeping is going to be just pointless frustration and waste.
Makes sense thx
Reply to @WillowChem, 17:32 UTC · 2 likes
MagpAI (@WillowChem): @AaronBergman18 @vivian_belenky @bijibiji2023 My general stance here is that if the state is facilitating a decision then it becomes legitimate for it to gatekeep it, as long as it does not artificially prevent you from making that decision without the help of the state
Yeah this sounds right but I’m not talking about law/public policy necessarily. Medicine is a central example of where/how “mere norms” made by civil society can be relatively important
Tweet, 20:15 UTC · 10 likes
Using 539254264676 gallons of water and 10^21 FLOPs coursing through an alien shoggoth mind just to make this (I have SORA now)
(video, plays on X)
Reply to @AndyMasley, 20:17 UTC · 12 likes
(The tweet being replied to by @AndyMasley is no longer available.)
This is kinda insane tbh, I do not trust anyone to identify the True Bad Guys to this degree
Reply to his own earlier tweet, 20:18 UTC · 4 likes
Aaron Bergman, February 7, 2025: Using 539254264676 gallons of water and 10^21 FLOPs coursing through an alien shoggoth mind just to make this (I have SORA now)
Maybe it’s just FLOP
Plz don’t cancel
Reply to @AndyMasley, 20:24 UTC · 9 likes
Andy Masley (@AndyMasley): @AaronBergman18 I think people posting very directly racist stuff are the true bad guys
Yeah I agree object level, just don’t trust you or me, or anyone to be correct with enough confidence that I’d endorse sex offender-level consequences without sex offender-level process (eg full criminal trial level)
Or even if I did, I don’t trust society to identify who the Correct Identifiers are and never empower someone who thinks they’re Andy but with actually bad takes
Reply to @GjMcGowan, 21:02 UTC · 6 likes
George McGowan (@GjMcGowan): Very important to set up google alerts for your own name.
Here is Nebraskan kindergartener George McGowan at a spelling bee.
His drip is INSANE, wish I was as cool a George McGowan as him.
Better dressed than I have ever been
Reply to @norabelrose, 21:19 UTC · 5 likes
Taylor Belrose (@norabelrose): Sparse autoencoders (SAEs) have taken the interpretability world by storm over the past year or so. But can they be beaten?
Yes!
We introduce skip transcoders, and find they are a Pareto improvement over SAEs: better interpretability, and better fidelity to the model 🧵
@livgorton big if true
Tweet, 23:04 UTC · 6 likes
Some of Trump and Elon’s shenanigans are illegal but I think I basically endorse “POTUS appoints high performing individual to change whatever within the bounds of the law, not tied to a cabinet position”
Thread, continued, 23:04 UTC · 3 likes
Just call it “Secretary of Everything Else”
Over the 1-2 decades, failure to govern/do anything at all has been a much bigger problem than overzealous governing
Thread, continued, 23:04 UTC · 4 likes
With hindsight, Obama should have done this the minute the Merrick Garland SCOTUS nomination was blocked; play nice with norms until the other side defects
Well that was a defection, and unfortunately Democrats basically did nothing at all in response
Thread, continued, 23:06 UTC · 2 likes
Maybe all this is moot because of the singularity, but if not, I don’t think it would be crazy to basically set aside like 1% of the federal budget for the presidential administration to use basically however (again within the bounds of the law), or even more
Reply to @halcyon_hazel, 23:52 UTC · 3 likes
❤️🔥 hazel ❤️🔥 (@halcyon_hazel): @vivposts iirc congress has to approve compacts?
Ya https://en.m.wikipedia.org/wiki/Interstate_compact
Thread, continued, 23:56 UTC · 2 likes
Wait it sounds like maybe the better interpretation is “states can do whatever they want, but the federal government won’t enforce contracts between states the way that states do to individuals”?
Saturday, February 8, 2025
Quote tweet, 02:02 UTC · 4 likes
Possibly a some free lunch if many people don’t donate and yet 78% of donors think that donation hasn’t left them worse off (contrary to a basic economic-psychological model of welfare and consumption)?
n=65
Quoting Aaron Bergman 🔍 (@AaronBergman18): Those who donate >=10% of your income, what is (your best guess of) the net effect of donation on your hedonic quality of life
Tweet, 02:02 UTC
I was gonna say “If everyone was altruistic, calibrated, interested only in hedonic wellbeing, and rational, 100% of donors would have answered “negative”, but in fact only 22% did”
But…
Thread, continued, 02:02 UTC · 3 likes
It’s totally possible, and seems plausible, that even among donors who answered “positive”, the effect of the last, marginal donated dollar (ie 9.99% -> 10%) is negative, but the first chunk of money they give away is sufficiently positive to offset that
Tweet, 02:15 UTC · 40 likes
The fact that I wasn’t following the author of “one of the most careful meta-analyses [Scott Alexander] has ever seen” proves that the market for twitter follows is inefficient
@davidroodman
Thread, continued, 02:15 UTC · 3 likes
Paper https://arxiv.org/pdf/2007.10268
Thread, continued, 02:15 UTC · 3 likes
Post https://www.astralcodexten.com/p/prison-and-crime-much-more-than-you
Reply to @niplav_site, 02:32 UTC · 3 likes
niplav (@niplav_site): @AaronBergman18 @davidroodman My impression is that often people whose papers/books/blogs are great are terrible on twitter
Lol, still
Quote tweet, 04:02 UTC · 11 likes
Tweet text, author, date/time, like, reply count, and corresponding community notes for >120k tweets
(This is like 8% of the whole thing, but my ADHD ass cannot be trusted to do this for ~7 days straight so this is Checkpoint 1)
https://github.com/AaronBergman/twitter_community_notes/
Quoting Aaron Bergman 🔍 (@AaronBergman18): I really hope @CommunityNotes is training a model to predict note from tweet because that’s probably the best almost-ground-truth data you’re gonna get in complex and non formally-verifiable domains
Thread, continued, 04:02 UTC · 3 likes
(If someone seriously intends to finetune or train a model on this [or I guess wants to give me money for cloud GPUs to do this, although I'm probably not the best person], lmk bc then I will actually probably do the full data collection)
Thread, continued, 04:20 UTC · 11 likes
Real footage of me collecting data on everything I encounter
(video, plays on X)
Reply to @Aella_Girl, 05:11 UTC · 14 likes
Aella (@Aella_Girl): What's your sexual market value?
After months of work, I have a comprehensive survey based on a ton of data from how people ranked nearly every conceivable trait.
So: How hot a commodity are you in the dating arena? I'd love to see your scores!
Link below
If any straight/bi women here wanna do me a favor: https://forms.gle/v4uomKEQUmi3vRfH6
Reply to @cremieuxrecueil, 06:01 UTC · 10 likes
Crémieux (@cremieuxrecueil): BREAKING:
Journalists discover that one of the millions of federal employees holds a view that you might not agree with.
Bro isn’t a mailman
Quote tweet, 06:07 UTC · 4 likes
I have updated quite a bit toward classical liberalism and pre-woke-normie-non-racism being less popular among something like “the interested public”
Quoting Ryan Grim (@ryangrim): The “no affirmative action” and “no racial preferences” policy from Trump lasted about a week. White South Africans now specifically favored:
Thread, continued, 06:07 UTC · 2 likes
(hard to know what a truly random sample of people think)
Thread, continued, 06:07 UTC · 2 likes
Part of this is me; I previously thought some degree of affirmative action anti-racist type policies would be good, and still kinda do in some idealized sense, but no longer trust progressive institutions to do them well and without insane backlash
Thread, continued, 06:07 UTC · 2 likes
Normie-non-racism seems both pretty good and a stable political schelling point of sorts
Tweet, 06:09 UTC · 5 likes
Has @mattyglesias done a “a strong labor market is good for free speech bc employers don’t wanna fire people and lower downside to getting fired” take?
That feels like the epitome of an Yglesias take
Quote tweet, 06:41 UTC · 21 likes
A giant rock getting thrown in the ocean would give massive utility to men everywhere, and this is a consideration that should be weighed against others
Quoting Steve Jurvetson (@FutureJurvetson): 1 in 44 chance of a 8-megaton blast by meteorite 💥
(Hiroshima was 0.015 megatons for comparison).JWST will try to get a better read on its trajectory and size. If needed, deflection has to occur in 2028, not on final approach like in the movies.
Tweet, 19:14 UTC · 10 likes
At some level it’s just my fault for not getting to the right answer myself, but I wish society had pushed more realistic memes around the relationship between genetics/uncontrollable environment and intentional effort (eg training, practice) in determining skill and success
Thread, continued, 19:14 UTC · 14 likes
A big project of a big proportion of my life—getting better at competitive rock climbing—was largely motivated by what in hindsight seems like a basically incorrect premise - that I could keep improving if I just tried hard enough and figured out the right training routine
Thread, continued, 19:14 UTC · 7 likes
I did know something like “natural talent” was a thing, but think my model was basically that enough training could make up for deficits here
This was wrong, and my physical and mental health suffered for it
Thread, continued, 19:14 UTC · 8 likes
A contributing factor here was that I was pretty good, especially with endurance (pushing v10, upper 5.13s), and did keep getting better for quite a long time (roughly like ages 11-16)
Thread, continued, 19:14 UTC · 12 likes
It’s very unnatural to see yourself work hard and get better at something for many years and then, at 17, conclude “ah, my stagnation now is something like a plateau around my genetic potential, or at least a local maximum of some sort”
But I think that was what was going on
Reply to @why_bun, 19:27 UTC · 2 likes
Why bunnies are better than humans (@why_bun):
These aren’t inconsistent
Reply to @slimepriestess, 19:51 UTC · 83 likes
Ra (@slimepriestess): hasn't worked here, Eliezer specifically said he wouldn't read it.
He’s the only person actually consciously and intentionally doing decision theory irl
Reply to @GjMcGowan, 19:53 UTC · 4 likes
George McGowan (@GjMcGowan): Update: it came back and I smashed it with a book and it went away, no side effects.
10/10 would take advice from construction guys on twitter again.
(quoting @GjMcGowan: Hand doctor: “So yeah, we can keep observing it, or drain it again. There’s always surgery but the risk of complications is -” Me: “A construction guy on Twitter told me if you smash it with a heavy object it’ll go away? What about that?” Hand doctor: 👁️👄👁️)
Based
Reply to @JesseTayRiver, 21:58 UTC · 53 likes
Jesse Smith (@JesseTayRiver): Based Yud
image attached
(quoting @jessi_cata: Suspected murderer Maximilian Snyder / Audere writes open letter to Eliezer Yudkowsky about animal ethics. I'm not making this up. https://archive.ph/GErgG)
Yeah
Reply to @OrionJohnston, 22:07 UTC · 4 likes
David Johnston (@OrionJohnston): (No one’s killed anyone for my attention)
Nice good job
Reply to @tracewoodgrains, 22:45 UTC · 3 likes
Jack (@tracewoodgrains): yes, when you're talking about things like being glad a murder victim is dead right after he dies, it's dehumanizing and bad. or when you make similar excuses for murdering people's parents. things like that suggest a lack of care for human life that I distrust
I’m pretty sympathetic to the vegan side and there seems to be a bunch of context I don’t have, but I do think it’s all of {wrong, bad vibes, not helpful to the cause of animal welfare} to suggest that murder is like not that bad or kinda justified, in virtually all cases
I can construct hypotheticals where killing a human might check out, but even in these pretty unrealistic cases, the killing would be best characterized as like the lesser of two evils
More prosaically, “step 1) do something that makes you seem crazy and evil” is
~never a legit theory of change - at a descriptive level, I don’t think the Zizians actually did anything positive for animals beyond their personal veganism (and quite possibly its net negative with the murder)
Sunday, February 9, 2025
Tweet, 00:31 UTC · 34 likes
Felt cute might delete later
Thread, continued, 00:44 UTC · 6 likes
Reply, 00:45 UTC · 1 like
15 pro
Reply to @austinc3301, 00:49 UTC · 2 likes
Agus 🔸 (@austinc3301): @AaronBergman18 I feel bad because I have the same phone and all my pictures are terrible
Lol I mean “foggy national mall at night” is pretty naturally photogenic
Reply to @livgorton, 05:07 UTC · 2 likes
Liv (@livgorton): update: i am going! please reach out if you’d be keen to chat :)
(quoting @livgorton: wdym i don’t overthink anything (i am spending my weekend turmoiling about whether to attend EAG))
Based
Reply to @tracewoodgrains, 05:33 UTC · 5 likes
Jack (@tracewoodgrains): my worldview is in some ways not just atheoretical but antitheoretical
I think it is a very good thing that most people let practical reality sand off the sharp edges of their philosophies, because most people believe some crazy and evil things and are kept in check by norms
most of the worst people in history are the ones who are so committed to one theory or another that they are willing to brush aside all of the sanity checks, common-law morality, and balancing tools broader society has built to keep things more-or-less running, and embrace their Pure Theory independent of consequences
it is very, very good that most people lack the courage of their convictions, because we are neither smart enough nor good enough for our convictions to be good enough to bypass all sanity checks
I’m anti-anti-theoretical (the phrase I’d use is “taking ideas seriously”)
In practice I think you should abide by norms and “common sense morality” most of the time
But the default state of the world where nobody takes ideas very seriously isn’t one without “crazy and evil”
Non-ideological apathy, self-interest, and failure of coordination, and most of all just the laws of nature grinding away without regard for the harm that entails, are sources of what would be called “crazy and evil” if done intentionally. Factory farming, wild animal suffering, most human poverty, and most human suffering are examples
Biological and cultural evolution make no guarantee that following what seems basically normal entails, or even directionally point towards, mitigating these bad things in the world
You don’t get systematic efforts to mitigate what we might call “natural harm” without recognition of the problem and abstract reasoning about values and prioritization—in other words, “theory”
Reply to @LinkofSunshine, 06:16 UTC · 157 likes
Basil🧡 (@LinkofSunshine): Being in SF is permanent twitter
Being in Seattle is permanent discord
Being in DC is permanent reddit
Being in NYC is permanent instagrampick your poision
DC is permanent LinkedIn
Reply to @arithmoquine, 07:45 UTC · 3 likes
henry (@arithmoquine): bmi in freefall can my fat followers please please tell me how they do it
Abilify
Reply to @NathanB60857242, 14:27 UTC · 2 likes
Nathan Barnard (@NathanB60857242): Fuck I'm becoming a post rat in some meaningful sense (thinking that there true things that can't be expressed - in principle - propositionally but nevertheless we should put weight on when making decisions where we care about observable outcomes.)
I’m inclined to think “can’t be expressed propositionally” is a skill issue
Even analytic philosophy makes good use of the toolkit of figurative language when it needs to (“what it’s likeness”)
Reply to @NateSilver538, 14:45 UTC · 34 likes
Nate Silver (@NateSilver538): A refreshingly (?) short newsletter for a change. Democrats need to pick a direction; the middle ground of separating the world into naughty and nice billionaires is probably the worst option.
https://www.natesilver.net/p/democrats-need-a-billionaire-strategy
I mean I think it’s basically true though. Dustin Moskovitz really is good and Elon really is bad (recent events have sealed the deal for him), so there are the existence proofs
Reply to @QualiaNerd, 15:01 UTC · 12 likes
QualiaNerd (@QualiaNerd): Why is Elon bad? (A completely genuine question BTW, I don’t follow politics much.)
Dumbass harmful changes to the federal government. DOGE has been net negative so far
Thread, continued, 15:05 UTC · 7 likes
USAID dismantling being one big thing
Reply to @girllbossing, 15:10 UTC · 3 likes
(The tweet being replied to by @girllbossing is no longer available.)
😢
Tweet, 20:53 UTC · 2 likes
This accords with both my (modest) surprise at the speed and efficacy of DOGE and EO implementation and what I’ve heard irl (although plausibly these sources are not totally independent convergence)
Quote tweet, 21:01 UTC · 49 likes
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
Quoting Bentham's Bulldog🔸 (@Benthamsbulldog): I'm finding it easier to be a partisan Democrat these days because the Republicans have just openly declared their allegiance to evil.
Thread, continued, 21:07 UTC · 6 likes
Not a complete object-level progressive btw
Reply to @sanjehorah, 21:17 UTC · 3 likes
sanje horah 🇵🇸🇺🇦 — PNEUMATICS ONLY (@sanjehorah): my SMV equivalent is Miss Piggy
image attached
(quoting @Aella_Girl: What's your sexual market value? After months of work, I have a comprehensive survey based on a ton of data from how people ranked nearly every conceivable trait. So: How hot a commodity are you in the dating arena? I'd love to see your scores! Link below)
Same
Reply, 21:19 UTC · 4 likes
Coase’s "Net effect?" Ghost 🏴🕊️ (@GhostCoase): @AaronBergman18 Democrats kill several tens of thousands of Arabs to make your pizza and violently deport millions into poverty.
Insofar as this is true it’s largely because of electoral pressure from the other party, especially deportation
This isn’t a total moral justification but still
Reply to @NathanB60857242, 22:18 UTC · 2 likes
Nathan Barnard (@NathanB60857242): @AaronBergman18 @wilhelmscreamin Why do you disagree with the Buddhist one?
Short version is that introspection and other ways people (think they) gain knowledge seem very unreliable but we should expect them to hit on correct arguments once in a while bc luck
Analytic philosophy *also* seems unreliable but OOMs less so, to the point that I think broadly philosophy really has made progress over ~any timescale
Monday, February 10, 2025
Tweet, 00:19 UTC · 25 likes
Ok, I now have a more complete Theory of Hinge, which probably applies to similarly structured dating apps in general
Thread, continued, 00:20 UTC · 4 likes
The insight honestly came from a combo of luck/variation and adding what’s now my pfp as a pic, which for the first time ever put me into the situation of having more matches than I really wanted to spend ~energy chatting with (not a flex; keep reading and you’ll see why)
Thread, continued, 00:20 UTC · 1 like
Basic model is that a person can be in one of two possible states:
1) few matches, or
2) many matches
(I think some small tweaks would make this work with latent potential matches too ie likes)
Thread, continued, 00:20 UTC · 5 likes
In some sense, the ideal situation is Few vs Few
This is kinda how dating apps should naively work: both people are relatively picky and so excited about matches (ie only like people they’d be enthusiastic to chat with)
Thread, continued, 00:20 UTC · 4 likes
The problem is that matches correlate with attractiveness, which is pretty highly correlated between people (the fact that it’s not perfectly correlated is a big way this model can “fail” which is good)
Tweet, 00:20 UTC
Why’d you match in with less exciting people in the first place?
Thread, continued, 00:20 UTC · 6 likes
Probably uncertainty about quantity of matches in the near future, frustration with the most attractive matches ghosting, and maybe just to give the guy a chance in the first few messages in case those set him apart somehow
Thread, continued, 00:20 UTC · 2 likes
Ok anyway, the situation is that the Many person is chatting with their most exciting matches, who tend to be more attractive, but, again, attractiveness correlates with number of matches, so now we’re in the last possible situation: Many to Many
Tweet, 00:20 UTC · 2 likes
Many to Many is probably what happens for the most attractive men, in general, and in general involves the most attractive people talking to one-another
Well, congrats, two attractive people who are excited about one another do indeed form couples relatively easily!
Thread, continued, 00:20 UTC · 2 likes
But then they leave Hinge!
So the remaining cases maybe involve people with very high standards or hate chatting or for whatever reason are adversely selected…
Tweet, 00:20 UTC · 2 likes
3) people who you are excited about and do reply end up being adversely selected for some reason
Thread, continued, 00:20 UTC · 3 likes
^in my extremely short-lived experience with having like 7-8 matches for the first time, the person I was most excited about and spent the most time and energy chatting with ended up basically saying she wasn’t interested in going on a date after all…
Thread, continued, 00:20 UTC · 7 likes
Which makes a lot of sense, because she probably had many exciting options (who in turn would have lots of *other* options, etc etc ad infinitum until everyone is tired and nobody is happy)
Thread, continued, 00:20 UTC · 11 likes
(She might even be reading this; if so, hello and no hard feelings I’m just being hyper-analytical and autistic about everything)
Ok that’s it thx for reading
Reply to @NathanB60857242, 00:28 UTC · 5 likes
Nathan Barnard (@NathanB60857242): @AaronBergman18 Likes should be priced better.
Yeah, at least hinge restricts likes
Another option I’m not super sure about would be some sort of indicator along the lines of “your match has matched with 6 other hotter men, FYI”
Reply to @Duderichy, 00:52 UTC · 36 likes
the Rich (@Duderichy): you can actually see over the hood
incredibly based and kid safety pilled
(quoting @sweatystartup: Wtf is happening?)
Tweet, 01:04 UTC · 9 likes
I went to a Very Online EA x normie* mixer yesterday; it was actually good! Afaik nobody died or even came especially close. I encourage more of this
*normie just meaning like everyone else; still kinda selected
Reply to @livgorton, 01:32 UTC · 4 likes
Liv (@livgorton): am i a normie (moderately online EA adjacent)
I thought u were no longer adjacent 😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢 😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢 😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢 😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢 😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢 😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢 😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢 😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢 😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢 😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢 😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢 😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢 😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢 😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢😢
Thread, continued, 01:33 UTC · 6 likes
But no ur not a normie sorry
Reply to @livgorton, 01:35 UTC · 2 likes
Liv (@livgorton): did you get chatgpt to produce this monstrosity
No I rawdogged it (idk how many characters it is but not super high n for 2^n)
Tweet, 03:40 UTC · 1 like
Kansas City has enough people to support an NFL team??
Tweet, 03:44 UTC · 1 like
Ok this isn’t that small I guess
Thread, continued, 03:44 UTC · 2 likes
But the Packers shouldn’t exist fr, that’s literally like one large university
Reply to @StatisticUrban, 04:18 UTC · 22 likes
Hunter📈🌈📊 (@StatisticUrban): My thesis for why we're fatter today than we used to be is really very simple.
GDP went up because the capital, land, and labor combined to produce more goods and services
Reply to @RichardHanania, 17:33 UTC · 3 likes
Richard Hanania (@RichardHanania): Trump is eliminating the penny, which costs 3.69 cents to make.
Amazingly, there’s an activist organization on the other side, cleverly called “Americans for Common Cents.”
I assume Big Zinc is behind it, as I can’t imagine anyone deciding this is the cause to fight for.
Iirc that’s exactly right, it’s literally the metal companies
Quote tweet, 17:37 UTC · 11 likes
One of the most important questions in the world rn is whether the Trump administration will comply with federal judiciary rulings
Quoting *Walter Bloomberg (@DeItaone): US JUDGE ORDERS TRUMP ADMINISTRATION TO FULLY COMPLY WITH PREVIOUS ORDER RESTORING ALL FROZEN FUNDING
Quote tweet, 17:53 UTC · 11 likes
Quoting Jacobin (@jacobin): At first glance, America’s declining drinking culture looks like a public health victory. But in an age of unprecedented loneliness and isolation, we’re losing crucial spaces for human connection. https://jacobin.com/2025/02/alcohol-bars-loneliness-covid-health/
Reply to @TetraspaceWest, 21:07 UTC · 3 likes
(The tweet being replied to by @TetraspaceWest is no longer available.)
Bristlemouth, as of 5 seconds ago
Quote tweet, 21:08 UTC · 9 likes
What R library
Quoting AJ Thurston, PhD (@AJThurston): I have officially found the most cursed #dataviz 🤮
Tuesday, February 11, 2025
Tweet, 03:14 UTC · 2 likes
Jalen Hurts 🤝 Me
Being MVP
(Him Super Bowl, me watching airplane reels on instagram)
Tweet, 03:45 UTC · 20 likes
I’m not really a public protest guy and the last thing I protested for in public was gun control (half-heartedly) in high school but I think I’d like go and yell at the White House irl if Trump basically just defies judicial orders (conditional on like other people being there)
Tweet, 04:45 UTC · 19 likes
Deep Research outputs kinda suck because the sources are literally written out inline which sucks for exporting or text-to-speech
Copy and paste the outputs here for a better version
(video, plays on X)
Reply to @zeta_globin, 04:54 UTC · 2 likes
(The tweet being replied to by @zeta_globin is no longer available.)
Yeah and it’s not even close
I strongly expect any reasonable analysis to conclude that net QUALYs is positive
Quote tweet, 17:47 UTC · 10 likes
Seems bad
Quoting Armand Domalewski (@ArmandDoma): the change in the NIH funding formula, to me, is the most explicit and direct way Elon has broken the law. Congress passed a law *explicitly banning* the Executive from making changes to the indirect cost %
Wednesday, February 12, 2025
Quote tweet, 02:47 UTC · 19 likes
Classic “I think it’s not bad but all my oomfs definitely score in the 100th percentile” result
Quoting Gnostrils (@gnostrils): new game: convergent association test
how low can you get
Thread, continued, 02:49 UTC · 3 likes
Why did it ask for 10 words tho
Thread, continued, 02:52 UTC · 2 likes
https://www.datcreativity.com/
Reply to @smatta1701, 03:11 UTC · 2 likes
sma ⏹️ (@smatta1701): @AaronBergman18
Classic
Reply to @StatisticUrban, 03:26 UTC · 5 likes
Hunter📈🌈📊 (@StatisticUrban): Not bad
Ok now I think this is largely just tracking how common words are, rather than just how similar their actual semantic meaning is
Tweet, 03:33 UTC · 15 likes
I don’t know what a Markov Chain is and at this point I’m too afraid to ask
Thread, continued, 03:35 UTC · 12 likes
Oh that’s actually not too bad
Reply to @ali_sivi, 03:37 UTC · 32 likes
bald ann dowd (@ali_sivi): hey bro we're doordashing individual shrimps later if you wanna swing by
I hope they come alive with an enriching tank environment
Reply, 04:10 UTC
Isn’t this the same thing (input vs compressed input)?
Reply to @livgorton, 15:47 UTC · 5 likes
Liv (@livgorton): it’s just a matrix it can’t hurt u
All matrices hurt me
Thursday, February 13, 2025
Reply to @NathanpmYoung, 01:13 UTC · 2 likes
Nathan 🔎 (@NathanpmYoung): @menhguin I do.
Rip
Reply to @ionsabug, 01:19 UTC · 1 like
(The tweet being replied to by @ionsabug is no longer available.)
Based, I vote that the dog stays
Quote tweet, 01:22 UTC · 20 likes
I think this is like 50% likely to be basically true
Quoting Aella (@Aella_Girl): we're all dead. I'm a transhumanist, I love tech, I desperately want aligned AI, but at our current stage of development, this is building the equivalent of a planet-sized nuke. The reason is boring and complicated and technical, so midwits in power don't understand the danger
Reply to @egirlian, 17:56 UTC · 137 likes
egirlian 🌸 (@egirlian): unpopular opinion around these parts, but i don’t think 18-25 year olds aren’t approaching women because they’re terrified of getting accused of harassment. i think they’re just plain terrified, and the instant gratification of phone screens is preferable to pain of rejection.
Quote tweet, 17:57 UTC · 21 likes
Big if true
Quoting Simon Berens (@sberens): Exciting personal update — I've been working on building the world's brightest lamp, with the goal of getting daylight levels of light indoors.
We just launched on Indiegogo an hour ago; I'm especially proud of the video!
Thread, continued, 18:02 UTC · 2 likes
(Looks like the current cost is $850 ☹️)
Tweet, 21:08 UTC · 8 likes
MacBook Pro randomly stopped working. Looking like the Apple Store will be able to restore the hardware but my files are only saved bc I have a full computer cloud backup going
Worth considering if you don’t! (I use @backblaze which is like $100/year iirc)
Reply, 22:35 UTC · 1 like
That sucks and gl with that but also this tweet should go in a museum (unironically kindly)
Tweet, 22:45 UTC · 18 likes
I can’t easily find it but I think I RT’d something not long ago that said basically “I guess the original OpenAI board has been totally vindicated” and yes this keeps getting increasingly true https://80000hours.org/podcast/episodes/rose-chan-loui-elon-musk-open-ai/
Thread, continued, 22:45 UTC · 2 likes
The tl;dr as I interpret it is that the current situation is that Altman has to claim that the board abandoning its original mission and control over the for-profit is actually good for its original mission
Thread, continued, 22:45 UTC · 2 likes
My editorial is that the only way this works is if nobody, namely the current board members and Delaware AG, have even the most vestigial of spines
Thread, continued, 22:45 UTC · 2 likes
Apple pods: https://podcasts.apple.com/us/podcast/80-000-hours-podcast/id1245002988?i=1000691529880
Friday, February 14, 2025
Reply to @rgblong, 01:13 UTC · 4 likes
Robert Long (@rgblong): Great to see OpenAI’s new model spec taking a more nuanced stance on AI consciousness!
At @eleosai, we've been recommending that labs not force LLMs to categorically deny (or assert!) consciousness—especially with spurious arguments like “As an AI assistant, I am not conscious”
In all serious, do you think model responses bear at all on whether or not models are in fact conscious? AFAIK, they’d need something resembling an introspective capacity for this to even be on the table, which I don’t think current LLMs have
The plausible steelman “yes” case involves reasoning about LLM consciousness from an “outsider” perspective, not actually reporting on subjective experience
Reply to @arithmoquine, 01:25 UTC · 4 likes
henry (@arithmoquine): i need to get over my weird aversion of github & cloud storage because if i lose my laptop i'm set back a year on multiple fronts
I’d like I think I partially caused this tweet with one earlier, but regardless, I think backblaze is good and don’t know what the other good options are for automatic backup but you should probably spend some of your phat Atlas dollars on one of them, yes
Reply to @frances__lorenz, 01:27 UTC · 2 likes
(The tweet being replied to by @frances__lorenz is no longer available.)
So endearing for both of you lmao
Reply, 02:13 UTC · 1 like
Based on my knowledge of LLMs, yes - it has everything to do with system prompt, fine-tuning/RLHF, and possibly even situational awareness (ie that it’s a model in a certain situation) but not introspection per se
Reply to @eshear, 03:19 UTC · 2 likes
Emmett Shear (@eshear): Because of the deep commitment to the idea that model is Just A Tool.
Has anyone considered u for OpenAI CEO
Reply to @br___ian, 04:07 UTC · 3 likes
bri (@br___ian): is there any downside to tipping the scales in favor of defensive weapons over offensive, at all size and time scales
Good q
Reply to @br___ian, 04:41 UTC · 4 likes
bri (@br___ian): is there any downside to tipping the scales in favor of defensive weapons over offensive, at all size and time scales
From a very abstract perspective maybe the ideal thing is the ability to do costly offense
Otherwise current power structures (eg physical borders) just become entrenched forever
Quote tweet, 04:53 UTC · 37 likes
We must wild animal suffering-pill the toddlers (slightly kidding but also not kidding that much)
Quoting Kelsey Piper (@KelseyTuoc): The propaganda aimed at children that honestly bothers me the most is environmentalist media for kids which treat humans as a threat to what would otherwise be some kind of idyllic peace-and-harmony nature situation
Reply, 05:34 UTC
I was thinking exactly about this but it’s plausible only if ~2027 isn’t the end of the current world order, which it’s looking to be
Quote tweet, 06:06 UTC · 168 likes
I’m pro-Aella but also anti-anti-anti-Aella (sorta)
I don’t think yelling “WOW SHE REALLY BULLDOZED THAT ABSOLUTE DUMBASS LOSER” is a good way to foster scout mindset, memetically
Quoting Aella (@Aella_Girl): Because they see 'not knowing what you're talking about' as losing. This type of debate strategy wants to have the feeling of 'winning' at all costs, which leaves no room for backing down even when you've made an error.
Learning to lose is a difficult and important skill!
Reply to @skooookum, 06:08 UTC · 2 likes
(The tweet being replied to by @skooookum is no longer available.)
Time for a manifold market
Reply to @AmandaAskell, 06:37 UTC · 9 likes
Amanda Askell (@AmandaAskell): My take on the ethics of billionaires:
- Making billions of dollars: morally fine (sometimes bad, often good)
- Keeping billions of dollars: morally bad (children are dying yo)
- Giving away billions of dollars: morally awesome (you save many people, you deserve much praise)
The other natural category is “consuming billions of dollars worth of stuff” which is worse than “keeping”
Reply to @MichaelAlleged, 11:39 UTC · 4 likes
Michael (Allegedly) (@MichaelAlleged): @arabelladevine me personally I think you’re enchanting
Thanks for saving my puppy from that burning building the other day
Reply to @arabelladevine, 11:40 UTC · 3 likes
arabella ※ (@arabelladevine):
I’ve seen enough insane scores like this to basically think the test is meaningless
Quote tweet, 11:49 UTC · 8 likes
Big if true
Quoting JAMA Psychiatry (@JAMAPsych): Low-dose semaglutide reduced alcohol craving and consumption in adults with alcohol use disorder. https://ja.ma/4gJTxU5
Quote tweet, 11:55 UTC · 80 likes
There might be a lesson here along the lines of “if it’s a really good psych drug, you won’t have to squint at the data and employ a dozen statisticians to squeeze out a positive result; people will just notice that it works”
Quoting JAMA Psychiatry (@JAMAPsych): Low-dose semaglutide reduced alcohol craving and consumption in adults with alcohol use disorder. https://ja.ma/4gJTxU5
Thread, continued, 11:55 UTC · 6 likes
Reminds me of the early MAOIs, where people noticed a drug working for a different condition
I genuinely wonder if the same thing would have happened if Prozac happened to treat tuberculosis and was developed in the 50s; would it also have so obviously “just worked”?
Tweet, 12:12 UTC · 6 likes
Pretty nuts that oxytocin induces labor. Evolution was really cooking with that one
Reply to @robertwiblin, 12:44 UTC · 2 likes
(The tweet being replied to by @robertwiblin is no longer available.)
1 but it only changes p(listen) from .99999 to .999992
Reply to @RichardHanania, 13:05 UTC · 15 likes
Richard Hanania (@RichardHanania): A problem with EA types is that they overestimate the good charity has done in the world and underestimate the good of markets.
It's usually better to invest your money in goods and services people want than sitting there and trying to think about what you can do for them.
(quoting @AmandaAskell: My take on the ethics of billionaires: - Making billions of dollars: morally fine (sometimes bad, often good) - Keeping billions of dollars: morally bad (children are dying yo) - Giving away billions of dollars: morally awesome (you save many people, you deserve much praise))
I mean obviously I’d say this but I don’t think this is true? Like the question of when you should expect charity to beat investment and personal consumption has been wrestled with quite a bit, it’s a critique that’s already been incorporated
Reply, 13:24 UTC · 1 like
Absolutely bodied by Claude Sonnet, which says I “somewhat overstates the case that they shouldn't be compared. In practice, people need to make allocation decisions between these options, even if they operate on different philosophical dimensions.“
Reply to @jjspicer, 15:03 UTC · 1 like
Jonny Spicer🔸 (@jjspicer): Isn’t this graph obviously wrong? Pretty sure the NHS doesn’t cost £176 trillion, what am I missing? Also the IHT revenue blatantly doesn’t line up with the y axis to make it look less significant than it is
It says billion
Reply to @livgorton, 15:20 UTC · 2 likes
Liv (@livgorton): for valentine’s day, i’d love a pet big and/or miniature cow pls
I hope he reads this
Reply to @livgorton, 15:21 UTC · 2 likes
Liv (@livgorton): for valentine’s day, i’d love a pet big and/or miniature cow pls
🅱️ig 🐷
Reply to @livgorton, 15:22 UTC · 2 likes
Liv (@livgorton): that would suffice, yes
Reply to @crumbraven, 15:35 UTC · 2 likes
crumbraven (@crumbraven): @GarrettPetersen that’s a fucking good question
Fr
Reply to @livgorton, 15:46 UTC · 2 likes
Liv (@livgorton): wow i misspelt pig 😭🐽 this was written in a state of late night delirious yearning for a farm animal
Cancelled
Reply to @skooookum, 16:23 UTC · 15 likes
(The tweet being replied to by @skooookum is no longer available.)
I straightforwardly do not believe there’s a causal link and am willing to bet on this
Reply to @RichardMCNgo, 18:25 UTC · 26 likes
Richard Ngo (@RichardMCNgo): Trump was way ahead of the curve on the three most important geopolitical issues of the last decade:
- Everyone sleeping on China (remember the compilations of him saying China over and over?)
- EU hugely underspending on defense (leading to disaster in Ukraine)
- Abraham Accords(quoting @0xDeric: @RichardMCNgo I'm curious what examples you would provide of Trump's incredible geopolitical instincts! Not explicitly against it, but also I can't think of obvious examples that I would point to that display international political acumen)
When you say incoherent bullshit about everything, people can sift through and find stuff you got right but Trump has also said/done things that point in the opposite direction
Maybe you can give him the second thing about EU spending, but even that wasn’t exactly a novel insight (maybe you like his hardline rhetoric here but that’s a separate point)
Tweet, 21:29 UTC · 2 likes
(Possibly?) nobody has ever had less skin in the game, as a matter of proportionality (ie skin in the game : importance of the game) than Elon Musk and the federal government
Thread, continued, 21:31 UTC · 2 likes
Refusal to recognize the gravity of his whims is separable from and possibly more damning than any single object-level decision
Reply to @OpenAINewsroom, 21:37 UTC · 429 likes
OpenAI Newsroom (@OpenAINewsroom): "OpenAI is not for sale, and the board has unanimously rejected Mr. Musk's latest attempt to disrupt his competition. Any potential reorganization of OpenAI will strengthen our nonprofit and its mission to ensure AGI benefits all of humanity."
—Bret Taylor, Chair, on behalf of the OpenAI Board of Directors
Sounds like you’re lying, because the current goal is to abandon the nonprofit’s original charter
Saturday, February 15, 2025
Quote tweet, 00:18 UTC · 12 likes
I kinda vibe-associate with alt-psychiatry but I have 0% confidence that this is going to be done well
Quoting chris person (@Papapishu): Gonna be honest: I wouldn’t do that if I were you man. https://www.whitehouse.gov/presidential-actions/2025/02/establishing-the-presidents-make-america-healthy-again-commission/
Thread, continued, 00:18 UTC · 3 likes
The claims I will actually stand by are that MAOIs are relatively underrated and under-prescribed, antipsychotics are probably given out at highish doses a bit too freely/without sufficient warning
Reply to @MariGO2thepolls, 01:37 UTC · 2 likes
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): I think looks are generally completely off limits unless and only unless one has a very deceptive pfp. Like cmon this is pathetic
Making fun of people’s looks is bad, even if they have bad politics
There’s nothing wrong with using a relatively flattering picture as pfp
Quote tweet, 10:44 UTC · 26 likes
What in the world
Quoting Gabriel (@gbrl_dick): truly one of the most incredible policy stories i can remember — in the midst of covid, italy gave away €200 billion to the construction industry (😉) to renovate a few homes.
after this and the IRA, MMT is now kind of approximately responsible for at least two of these lmao
Reply to @bayeslord, 10:52 UTC · 9 likes
bayes (@bayeslord): question for the ai doomers reading this, sincerely, why do you think memetic success been relatively hard to come by?
I think I reject the premise; abstract claims about the future are competing over a very small % of the population to begin with. There’s no world in which >50% of the population is e/acc or doomer rn (you can get people to say whatever on surveys with the right wording but that’s different)
Reply to @_alice_evans, 11:56 UTC · 10 likes
Alice Evans (@_alice_evans): Male-female friendships are a under-rated driver of gender equality.
When women can fraternise freely, they can share ideas, gain respect, & build empathy with men who care about their wants & welfare.
Countries with high male-female friendships score better on equality 👇
I like girls and volunteer to be their friend
Reply to @peterwildeford, 14:34 UTC · 3 likes
Peter Wildeford🇺🇸🚀 (@peterwildeford): @ShakeelHashim @Mjreard yeah I misspoke - this is what I meant
tweeting is hard
The whole thing is really quite clear and intuitive
Reply to @banterrealism, 14:59 UTC · 1 like
Sex Haver (@banterrealism): You guys grew up with the internet no?
Start acting like it
The parsimonious explanation for random medium big accounts is that they are who they say they are (conditional on this being plausible)
Seems so much harder and more complex to make a fake big account
Reply to @monodevice, 15:02 UTC · 2 likes
🐜onio (@monodevice): This is a very typical straight male who either lost a bet or is engaging in some one-upmanship or is responding to a dare. Most often done in explicit absence of any non-males, meaning this should be seen as a bull signal on the military, if that's your concern.
(quoting @libsoftiktok: What is going on at the Air Force…. Reportedly at Holloman Air Force Base in New Mexico)
Yes, ~0% chance this is earnest
Reply to @matt_beard_, 15:11 UTC · 4 likes
Matt Beard (@matt_beard_): it's a pretty amazing feeling when all of your problems are instantly solved all at once. "your visa has been approved for three years" is the best sentence I've ever heard (other than "I do").
Lfg congrats 🇺🇸🗽🎆🎇⭐🦅🚀🏈🥧🎺💙❤️🤍🗽🇺🇸🎆🎇🦅⭐🚀🏈🥧🎺💙🤍❤️🇺🇸🗽🎆🎇🦅⭐🚀🏈🥧🎺💙🤍❤️🇺🇸🗽🎆🎇🦅⭐🚀🏈🥧🎺💙🤍❤️🇺🇸🎆🗽🎇🦅⭐🚀🏈🥧🎺💙🤍❤️ 🇺🇸🗽🎆🎇⭐🦅🚀🏈🥧🎺💙❤️🤍🗽🇺🇸🎆🎇🦅⭐🚀🏈🥧🎺💙🤍❤️🇺🇸🗽🎆🎇🦅⭐🚀🏈🥧🎺💙🤍❤️🇺🇸🗽🎆🎇🦅⭐🚀🏈🥧🎺💙🤍❤️🇺🇸🎆🗽🎇🦅⭐🚀🏈🥧🎺💙🤍❤️
Reply to @pli_cachete, 16:35 UTC · 3 likes
Rota Is In Israel🚪🧎♂️ (@pli_cachete): 2/3
The answer is always something like this
(4/3)pi(r^3)-ass answer
Reply to @mattyglesias, 16:52 UTC · 32 likes
(The tweet being replied to by @mattyglesias is no longer available.)
I don’t think EV policy should be set based on the owners politics
Reply to @littmath, 17:00 UTC · 29 likes
Daniel Litt (@littmath): It’s undeniable at this point that the best LLMs are quite good at standard undergrad material. One place they still underperform strong math undergrads, though, is honesty. Good students (even those who haven’t fully mastered the material) resist bluffing/BSing.
My best guess is that bluff/BSing isn’t a good description of the internal psychology
Like I don’t think the model has an internal representation of it being wrong; from the inside being right and being wrong “feels” the same to an LLM (generally)
Reply to @mattyglesias, 17:07 UTC · 5 likes
(The tweet being replied to by @mattyglesias is no longer available.)
Ah makes sense
Reply to @sama, 17:55 UTC · 11 likes
Sam Altman (@sama): i particularly love when the anti-AI crowd makes up shit about our water usage while eating a hamburger
(quoting @AlecStapp: Using ChatGPT isn't going to destroy the environment or consume all our resources)
@AndyMasley
Reply to @sama, 17:56 UTC · 53 likes
Sam Altman (@sama): i particularly love when the anti-AI crowd makes up shit about our water usage while eating a hamburger
(quoting @AlecStapp: Using ChatGPT isn't going to destroy the environment or consume all our resources)
Tweet, 19:00 UTC · 6 likes
5 minute survey to weigh in on the use of insects in animal feed in the UK. Suggested responses and direct link below.
Took me about 4 mins total
Reply to @NoahTopper, 19:09 UTC · 2 likes
Quinoah 🔍⏸️ (@NoahTopper):
@littIeramblings
Reply to @goingawoll, 19:28 UTC · 7 likes
Amos Wollen 🦐🪰✴️ (@goingawoll):
Link for others https://www.theguardian.com/world/2025/feb/15/prawn-farming-cruelty-electrical-stunning-waitrose
Reply to @WillowChem, 19:52 UTC · 5 likes
MagpAI (@WillowChem): !!!!!! Someone printed me out the Roche metabolic pathways poster onto cloth so that I can actually keep it and transport it around with me. This makes me so unbelievably happy
Based
Reply to @banterrealism, 19:59 UTC · 2 likes
Sex Haver (@banterrealism): Guy who is pro vaccine but anti SSRIs and ozempic but only lowkey
Vaccines > ozempic > SSRI > 0
Reply to @NoahTopper, 20:01 UTC · 3 likes
Quinoah 🔍⏸️ (@NoahTopper): This is huge
Poast bench at some point
Reply, 20:03 UTC
I thought u were a commie
Reply to @bijibiji2023, 20:07 UTC · 2 likes
hoff 🧒🏻😻 (@bijibiji2023): @AaronBergman18 I give up I’m going to join ur shrimp movement
Yay (we’re not winning yet but trending in the right direction)
Reply to @AskYatharth, 20:21 UTC · 361 likes
yatharth ༺༒༻ (@AskYatharth): i've already wondered & felt confused how humans could eat anything from hamburgers to plants to fish and still make... more human body. i've also been surprised how plants mostly gain their mass from the "air". the surprising answer to both is that we are carbon-based life?
Humans require a relatively varied diet and require 40-50 specific low-level nutrients, mostly molecules but some elements iirc
In some ways it’s more amazing that like rabbits and other herbivores can make more rabbit from only like grass
A rabbit organism isn’t that much simpler than a human organism!
Tweet, 20:36 UTC · 2 likes
TIL important biochemical reactions in humans involve “quantum tunneling” of protons (apparently there’s a second type of non-nuclear subatomic particle change besides ionization [which is when electrons move around])
Can’t say I fully understand it
Sunday, February 16, 2025
Reply, 00:48 UTC
Based gf
Thread, continued, 00:48 UTC
What is she looking at that rocks 2000 nits? Not many monitors do that
Tweet, 01:08 UTC · 2 likes
.@WhatsApp when I search for “chicken” in the emoji reaction panel, the “🐓” character doesn’t appear as an option! (This isn’t true in general on iOS/other apps)
Please assign a very well-paid engineer to fix this urgent and tremendously important injustice, thx
Thread, continued, 01:08 UTC · 3 likes
They should give me like a $25k bounty for this tbh
Reply to @DeepDishEnjoyer, 01:12 UTC · 3 likes
peepeepoopoo (@DeepDishEnjoyer): If this is true THIS FEELS VERY BAD FOR AI SAFETY
(quoting @AaronBergman18: @littmath My best guess is that bluff/BSing isn’t a good description of the internal psychology Like I don’t think the model has an internal representation of it being wrong; from the inside being right and being wrong “feels” the same to an LLM (generally))
My claim is true of getting math problems wrong (which I have no reason to believe they aren’t being earnest about) not true in general tbc
Like intentional deception/hiding capabilities is still on the table and potentially discoverable via interpretability
Reply to @jeffreyhuber, 01:14 UTC · 658 likes
Jeff Huber (@jeffreyhuber): mildly interesting
Reply to @kane, 01:21 UTC · 80 likes
Kane 謝凱堯 (@kane): The reason Japanese cities can have nice public bathrooms is bc the penalty for vandalism is three years in prison, the penalty for drugs in a public space is seven years in prison, and penalty for drug trafficking is life, all with 99% conviction rate.
This is insane. 7 years for smoking weed outside is bad, even if that leads downstream to marginally nicer bathrooms
Reply to @elgorond, 03:02 UTC · 1 like
eLGO (@elgorond): It's really not. It is disrespectful to do drugs in a public area, and where children can be. Have fun with the crack pipe at home
I largely agree; 7 years in prison is a lot. Maybe like a $200 fine lol
Reply to @pli_cachete, 03:03 UTC · 2 likes
Rota Is In Israel🚪🧎♂️ (@pli_cachete): I really do try my best to have a nuanced view of different political perspectives and ways of slicing up the world, but if you think this way you are unequivocally a part of my outgroup. Just an unbelievably self-effacing and incurious way of engaging with the world
(Also false, importantly)
Reply to @RichardHanania, 03:11 UTC · 5 likes
Richard Hanania (@RichardHanania): It’s Saturday night, and I’m here to provide what you need. A 5,000 word essay on why America needs more corruption.
I explain why you should stop worrying about Elon’s conflicts of interest, and learn to love Chinese billionaires buying our politicians. https://www.richardhanania.com/p/american-governance-is-not-corrupt
Oh man bad take. There are two optima: a global optimum with very low corruption and a worse local optimum with developing country level corruption where everyone knows that’s how you do things
America isn’t so far gone that we have to resign ourselves to the worse world!
Reply to @banterrealism, 03:58 UTC · 2 likes
Sex Haver (@banterrealism): Pretty crazy that I felt nervous posting this originally, compared to how obvious of a slam dunk it should have been in retrospect
(quoting @banterrealism: My take on trump is that I don’t take him seriously. Bad faith rhetoric (I hate) which spills over into real consequence (Jan 6) and I don’t understand how serious people (from tpot) find him nuanced. I’ve never argued about him with anyone. I’m kinda hoping the era just passes.)
Yes correct take
Quote tweet, 04:08 UTC · 10 likes
Replies here are a great illustration how the pendulum of public opinion swings
Lots of strong reaction to the current bad thing (people don’t like weed in public/public disorder; neither do I)
Quoting Aaron Bergman 🔍 (@AaronBergman18): @kane @Noahpinion This is insane. 7 years for smoking weed outside is bad, even if that leads downstream to marginally nicer bathrooms
Thread, continued, 04:08 UTC · 4 likes
Not a ton of effort to pre-empt the inevitable bad thing in the other direction: lives ruined for objectively relatively minor harm (the one guy smoking weed outside isn’t responsible for all instances of public disorder, even if you want him to be a scapegoat)
Reply to @niplav_site, 08:49 UTC · 7 likes
niplav (@niplav_site): Why do effective altruists mostly try to stop/prevent things?
Examples: Prevent AI risk & biorisk, eliminate diseases, decrease the amount of factory farming…
(They do try to create/build stuff, but mostly as instrumental goals)
Shouldn't the ratio be 50-50, a priori?
In which I record a ramble and get Claude to figure out what I'm trying to say. This is getting like 90% of the conceptual content across
Reply to @caleb_parikh, 08:51 UTC · 3 likes
Caleb Parikh (@caleb_parikh): Seems fine given that empirically Japan's policies result in few people going to prison.
I mean that's a very big bundle of policies and culture and other differences. It's directional evidence that their drug sentencing is good for this but not super strong
Quote tweet, 08:58 UTC · 9 likes
Boosting this point but also: not to be too much of a simp but Claude Sonnet still seems exceptionally good at a certain class of harder-to-make-a-benchmark-for vibey tasks
Quoting Aaron Bergman 🔍 (@AaronBergman18): @niplav_site In which I record a ramble and get Claude to figure out what I'm trying to say. This is getting like 90% of the conceptual content across
Reply to @Rado9910, 10:05 UTC · 1 like
Rado9910 (@Rado9910): Must read. A very good argument for why suffering reduction matters more than happiness creation.
(quoting @AaronBergman18: @niplav_site In which I record a ramble and get Claude to figure out what I'm trying to say. This is getting like 90% of the conceptual content across)
I’m not sure it’s an explanation of why, although I see the temptation to have ethics align with some elegant metaphysics
I think it *does* in fact align, but this is something like a conceptual coincidence
Reply to @WillowChem, 18:19 UTC · 6 likes
MagpAI (@WillowChem): Hahahahaha holy shit. This is legitimately insane.
I like the chutzpah
Reply to @Thinkwert, 19:03 UTC · 2 likes
Thinkwert (@Thinkwert): Boars wandering the streets in Poland.
@livgorton 🐷
Quote tweet, 19:31 UTC · 5 likes
Many such cases but my layperson interpretation is that objectively large increases in capability are, in ML, exceptionally and counterintuitively common both via hardware improvement/cost and the background rate of software innovations like this
See also: Deepseek
Quoting ℏεsam (@Hesamation): we aren't losing our minds enough over this paper:
> test-time compute by implicitly reasoning in layers
> no special training data like CoT
> can reason beyond words
> 3.5B model with this performs like a 50B model
> more like human brain than current "reasoning" LLMs
Thread, continued, 19:31 UTC · 6 likes
In ~any other field a functional 10x increase in the productivity of some major input resource is rightly worth losing one’s mind over
In ML it’s just another day watching straight lines on a log graph go brr
Reply to @GarettJones, 19:34 UTC · 10 likes
Garett Jones (@GarettJones): The Politico interview with @krishnanrohit perhaps fits my vibes-over-results take on DOGE:
image attached
(quoting @GarettJones: @OliverSegal2 @DOGE Yep, now we just have to solve for the @DOGE equilibrium: One would predict we'll get a lot of policy shaped by memes and vibes which are based on inaccurate information.)
Ahh but it’s a victory of a bad culture! It’s like “move fast and break things you don’t understand while having no skin in the game”!
There *are* cultural changes in the govt that would be good - this isn’t one of them
Thread, continued, 19:38 UTC · 3 likes
And it’s not like random citizens are going to see government working better for them via any of the changes
IRS has got to be the clearest example of this
Good reform: pay market rate for talented devs to make a free universal intuitive easy tax filing system
DOGE reform: fire IRS agents, reducing overall revenue but idk vibes or something
Reply to @RpsAgainstTrump, 23:38 UTC · 1 like
Republicans against Trump (@RpsAgainstTrump): Where’s DOGE?
(quoting @CalltoActivism: Since we’re on the topic of government waste, Donald Trump’s joyride around the Daytona 500 cost American taxpayers around $5,000,000. 🤡 Here’s the breakdown: Air Force One Costs: estimated $800,000 – $1,000,000. Presidential Motorcade & Transport: estimated $200,000 – $500,000…)
As much as I hate Trump this seems fine; it costs a lot to keep POTUS safe
Monday, February 17, 2025
Quote tweet, 01:27 UTC · 16 likes
You don’t actually have to undertake an ambitious cultural project like inventing a new superior flavor of masculinity
80% of the *achievable* gains to be had come from treating men at a broad cultural level the way people treat men they personally know and like
Quoting Everything Price Sufferer (but especially eggs) (@agraybee): The suggestions for Democrats to win back young men are like "just invent a new type of masculinity that doesn't disparage women and then convince them all it's better than the old type of masculinity." Oh, ok!
Thread, continued, 01:27 UTC · 7 likes
Being normal, respectful, and fundamentally non-prejudiced goes a long way
Thread, continued, 01:28 UTC · 4 likes
Perpetual take
Quote tweet, 02:52 UTC · 13 likes
1) I’m glad lightcone got the money
2) I feel like I’ve seen “hit the goal just barely” phenomenon throughout my life but here’s a recent and elegant example (note that the $15k “goal” is purely symbolic - we achieved the highest actual incentive at $35k just barely)
Quoting Ben Landau-Taylor (@benlandautaylor): Lightcone's monthslong fundraiser meeting its $2m goal in the last 6 hours is the clearest illustration I’ve seen yet of the "by default, people give money to nonprofits if and only if the alternative is that the nonprofit will literally die" thesis.
Reply to @saulmunn, 02:53 UTC · 1 like
saul munn (@saulmunn): i would find this valuable, and would encourage you to write this up.
if there are things i can do to make it more likely that this happens — e.g. i sit with you while you write, you dictate to me, i figure out a bounty & operationalize it & pay you, etc.
Yeah same!
This seems like a very high leverage EA Forum post and I’m realizing that I haven’t really thought about this (not that I control a ton of money so that doesn’t hugely matter, but I feel like I have opinions about lots of other things lol)
Reply to his own earlier tweet, 02:54 UTC · 5 likes
Aaron Bergman, February 17, 2025: 1) I’m glad lightcone got the money 2) I feel like I’ve seen “hit the goal just barely” phenomenon throughout my life but here’s a recent and elegant example…
I will take a bit of credit here for pushing for the $35k reward to exist at all
Thread, continued, 02:57 UTC · 3 likes
A certain account reading this actually put us over $35k at the end and I invite them to chime in!
Reply to @skooookum, 03:23 UTC · 17 likes
(The tweet being replied to by @skooookum is no longer available.)
Not claiming 95th IQ but for me I really don’t think there’s a tradeoff
Tweeting is natural/effortless, longform is extremely hard and time isn’t the limiting factor
Reply to @pli_cachete, 03:26 UTC · 2 likes
Rota Is In Israel🚪🧎♂️ (@pli_cachete): @AaronBergman18 @skooookum Your brain finding tweeting effortless is the limiting factor
My brain finding longform hard is the limiting factor
Not sure if you mean these to be one and the same
Tweet, 03:29 UTC · 13 likes
Gonna start documenting tweets so bad that they single-handedly remind me I’m on the For You feed
Tweet, 03:43 UTC · 15 likes
The flip side of “cutting a bunch of random bullshit only changes the budget by .00000001%” is that the USA is really big and we can actually have quite nice institutions that serve everyone for a relatively trivial amount of money
Thread, continued, 03:43 UTC · 2 likes
The thing that is hard to provide is cash/goods/rivalrous services (eg healthcare) to hundreds of millions of people
That is expensive necessitates legitimately difficult policy discussion and prioritization
Thread, continued, 03:43 UTC · 6 likes
But for things that basically serve the whole country at once, like the national park service, or the FDIC, or random education/finance/bio research programs, and good regulatory bodies like the Fed, congrats we actually can have all that for like $10 each
Thread, continued, 03:43 UTC · 3 likes
That’s because the US is *counterintuitively* large. It doesn’t feel like I should be able to get a bunch of nice stuff for cheap. But 350M is a lot of people!
Thread, continued, 03:43 UTC · 3 likes
Ofc that’s not to say everything should go: on the regulatory front, the main cost is the actual (de)merit of the institution and various specific policies
I (mostly) dislike the FDA - but because their regulations are bad, not because it costs too much to run!
Tweet, 03:49 UTC · 8 likes
I understand there are concerns around wanting to be light touch and have minimal false positives, but if anything @CommunityNotes should put notes on more stuff
Thread, continued, 03:49 UTC · 3 likes
Doesn’t just have to be corrective - could be like “yes this tweet is basically fair and accurate but here’s the crowdsourced single best bit of extra context that might be useful”
Non-corrective could be opt in/out
Reply to @livgorton, 05:34 UTC · 5 likes
Liv (@livgorton): float16 is just spicy integers pretending to be real numbers and i'm tired of pretending it's not
Reply to @matt_beard_, 17:48 UTC · 6 likes
Matt Beard (@matt_beard_): Excited to share that I've joined @80000Hours as an Advisor! Looking forward to speaking with talented, altruistic people about how to use their careers for good. I'm based in Washington DC and would love to chat with subject matter experts to learn about their fields.
Big if true
Reply to @sebkrier, 20:10 UTC · 4 likes
Séb Krier (@sebkrier): If you (a) have excellent writing skills, policy acumen, technical literacy, analytical skills; (b) are a Good human; and (c) want to write high quality, detailed memos for deepmind's policy team - then you should apply to the FIG fellowship by 7 March: https://futureimpact.group/
@littIeramblings
Quote tweet, 20:29 UTC · 4 likes
Not good but at least the plane didn’t fucking explode
Maybe Boeing et al should work on making it so that if planes are ever in the wrong configuration (eg by hitting a helicopter) the fuel stays contained better and doesn’t ignite
(Note: I have no idea what I’m talking about)
Quoting Alex Macheras (@AlexInAir): Breaking: Delta flight 4819, operated by regional subsidiary has been involved in an accident at Toronto Pearson Airport. The aircraft, a CRJ-900 (N932XJ), is pictured here - shockingly - upside down. Emergency response is underway, airport is closed.
Thread, continued, 20:29 UTC · 3 likes
Like it’s quite rare to see a plane in an unnatural position like this but not on fire. But evidently it’s possible
Cc @SpacedOutMatt can tell me why I’m wrong
Reply to @outliersgeorg, 20:32 UTC · 2 likes
Alastair Zaraza's liminal corpse party (@outliersgeorg): @slimepriestess @YitziLitt Nicotine is probably one of the safer antipsychotics on the market as long as you dose the pure stuff responsibly (i.e. no combustion)
Yeah use gum not cigs
Reply to @slimepriestess, 20:33 UTC · 2 likes
Ra (@slimepriestess): @YitziLitt in that i have nothing to stop the hallucinations and delusions from getting real annoying now
Cc https://slatestarcodex.com/2016/01/11/schizophrenia-no-smoking-gun/
Reply to @iScienceLuvr, 20:46 UTC · 33 likes
Tanishq Mathew Abraham, Ph.D. (@iScienceLuvr): Have you heard of Cleo?
Cleo was an account on Math Stack Exchange that was infamous for dropping the answer to the most difficult integrals with no explanation...
often mere minutes after the question was asked!!
For years, no one knew who Cleo was, UNTIL NOW!
I did a bit of investigation back in the day
Tweet, 21:06 UTC · 5 likes
Fed Spotify likes to Claude
This is kinda nuts (these seem clearly true for me)
Reply to @sporadica, 21:11 UTC · 6 likes
spor (@sporadica): “how are you so happy?”
brother, when i was 17 i discovered that i could physiologically stimulate adrenaline release on command in just about any situation
skill issue
I kinda doubt what you’re intuiting is literal adrenaline. When you inject people with adrenaline it generally causes anxiety and is unpleasant
Reply to @AaronBergman18, 21:20 UTC · 2 likes
(The tweet being replied to by @AaronBergman18 is no longer available.)
I don’t love the singular first person pronoun thing going though lol
Reply to @BNONews, 21:23 UTC · 36 likes
BNO News (@BNONews): UPDATE: At least 8 injured after Delta plane crash landed in Toronto - CP24
How do I become the guy that shoots water at upside down planes
Quote tweet, 21:40 UTC · 75 likes
Nature is terrible in a quadrillion ways
Quoting Historic Vids (@historyinmemes): Wolf lived with a tree branch trapped between his teeth for years
Reply to @albrgr, 23:04 UTC · 3 likes
(The tweet being replied to by @albrgr is no longer available.)
The goal is a ~direct causal arrow from using twitter to playing with own kids
Reply to @DrAllyLouks, 23:32 UTC · 3 likes
Dr Ally Louks (@DrAllyLouks): So I have been seeing quite a few requests for me to start a podcast and I would love to hear what you would be interested in hearing about! Smell, of course, but in what capacity? And would you like to hear about anything else? I’d be grateful for your thoughts! 💕
Ally come on pigeon hour. You don’t have to prepare and guests can cut anything after recording. Maximally low-stakes and chill https://podcasts.apple.com/us/podcast/pigeon-hour/id1693154768
Tweet, 23:35 UTC · 5 likes
I’m sad @DrAllyLouks got crazy big instead of just nice-nanocelebrity-medium-big bc in another timeline I think we’d be twitter friends
Reply to @_chair, 23:40 UTC · 3 likes
Chair (@_chair): @AaronBergman18 @DrAllyLouks damn you're trying to hold even your hypothetical friends back
How about she then goes megabig later
Tweet, 23:54 UTC · 4 likes
I’m sad @DrAllyLouks got crazy big instead of just niche-nanocelebrity-medium-big bc in another timeline I think we’d be twitter friends
Reply to @FilledwithUrine, 23:55 UTC · 7 likes
michael (@FilledwithUrine): she’s the kind of beautiful that makes me do 40 pushups in my room at 3am
Real, Godspeed
Tuesday, February 18, 2025
Reply to @Chadson, 00:00 UTC · 7 likes
Chadson (@Chadson): just FYI there are about 30 serious air accidents involving airliners in the US every year
Omg thank you for this tweet, it led me to https://data.ntsb.gov/avdata
Reply to @aidan_mclau, 00:03 UTC · 7 likes
(The tweet being replied to by @aidan_mclau is no longer available.)
Ya everyone smart af fr
Reply to @_chair, 00:19 UTC · 5 likes
Chair (@_chair): @AaronBergman18 you'd get used to it. this is just what the first couple days after dental work feels like
I kind of doubt this (chronic pain is absolutely a thing, hedonic adaptation is not a law of nature) and even if it’s true that’s a miserable couple of days
An hour of the worst dental pain I’ve had I’d gladly trade at least one week of healthy life to have slept through
Reply, 00:26 UTC · 1 like
I mean you’d still have other environmental effects with perfect “education”
Eg lead
Tweet, 01:27 UTC · 11 likes
So the second order effect of nobody hiring SWEs is that the price of second-tier (which might still be excellent) SWEs should drop and so it should become cheaper to start a software company (that doesn’t depend on hiring the 99th percentile)
Have we seen this?
Reply to @Aella_Girl, 03:16 UTC · 5 likes
Aella (@Aella_Girl): heaven forbid u intentionally increase your own amount of pain in a safe manner to help u process things more directly
I think this is a totally reasonable discussion to have but that guy’s replies are the wrong place to start(!) that conversation
Like your hypothesis needs to undergo more scrutiny elsewhere before it makes sense to stop deferring to norms
Thread, continued, 03:19 UTC · 2 likes
Having been in a position of “I probably won’t hurt myself but I sure feel like it,” I’d want to have come to a conclusion on the matter well before being in that state, if this makes sense
Reply to @bubbling_creek, 03:35 UTC · 2 likes
bubbling creek (@bubbling_creek): @Aella_Girl @AaronBergman18 @Duderichy I kind of think your point is right in general - I've experienced extreme pain a few times, and I'd rather have had them than not, was an interesting experience. A few times when I was extremely stressed I've pinched myself pretty hard to recenter, I think that was fine.
Yeah I used to like slap myself (hard) as a kid when something went wrong. I kinda just randomly grew out of it
Tbh I think this was only bad for me insofar as it was embarrassing
Reply to @NathanB60857242, 06:33 UTC · 3 likes
Nathan Barnard (@NathanB60857242): @liz_love_lace @AaronBergman18 I think that this is mostly unsubstantiated contrainism. I agree it's less important than TAI, but so is everything.
I disagree
- climate modeling and such has gotten better, severe or runaway warming is less likely than even 5 years ago
- economic and humanitarian effects are going to be dominated by AI and not in an independent/additive way
- at a personal ideological level, it’s not clear that warming is bad for wild animals which is something I didn’t used to think about
Quote tweet, 07:57 UTC · 8 likes
What happens to airplanes like this? What’s the retail price for a super fucked up passenger jet?
Quoting RTN (@RTNToronto): Inside view of Pearson plane crash in Toronto 😳🇨🇦
Reply to @CineraVerinia, 08:10 UTC · 3 likes
(The tweet being replied to by @CineraVerinia is no longer available.)
Quote tweet, 08:12 UTC · 11 likes
Holding a phone in one hand doesn’t actually slow anyone down, you can tell that he’s moving at the exact same pace as everyone else
Quoting RTN (@RTNToronto): Inside view of Pearson plane crash in Toronto 😳🇨🇦
Reply to @GarrettPetersen, 08:15 UTC · 3 likes
Dr. Dad, PhD 🔄🔼◀️🔽▶️ (@GarrettPetersen): DOGE guys have had access to those government databases for less than a month. It can be tough to acquaint yourself with all the nuances of a dataset in such a short time. I wouldn't make any big announcements without doing a lot more to verify them.
Lol real, database maintainers are not known for making all variables as intuitively named and understandable as possible
Reply to @MariGO2thepolls, 19:14 UTC · 1 like
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): I never rly understood the backlash towards the term “micro aggression” I feel like that’s a real thing and a used concept to have. Only a problem when blown out of proportion, Is it a huge deal the ppl should lose their livelihood over or w/e? Ofc not it’s micro lol literally in the name
Fr
Reply to @Tymtweet, 19:36 UTC · 3 likes
Tym🔸️ (@Tymtweet): What's the best free AI transcription app?
I want to yap into my phone and then have a big 'copy all text' button so I can add it to my journal
AudioPen or Aiko
Reply to @arthur_wright_, 19:56 UTC · 13 likes
Arthur Wrong 🔸 (@arthur_wright_): "Straight lines on a graph" believers suspiciously silent rn
Wait it’s actually an important q whether the probability is derived via Bayesian methods that take into account change over time or just simple statistical modeling
Maybe we should actually update towards the probability “increasing” ie just the probability being higher
Tweet, 20:37 UTC · 6 likes
Increasingly clear that starting around sonnet-3.5, “intelligence” often diverges from what I really care about in an LLM
Or like it’s always helpful but the marginal utility of intelligence (for many but not all queries) drops relative to calibration and vibes
Thread, continued, 20:37 UTC · 2 likes
Coding and math are basically just IQ go brr
Asking about the state of the world and arbitrary empirical questions (even complex ones) depends heavily on calibration, which we know from forecasting stuff isn’t super highly correlated with IQ
Reply to @leah_pierson, 20:45 UTC · 8 likes
Leah Pierson (@leah_pierson): 1) are you a utilitarian or utilitarian leaning?
2) has humanity had a net positive impact on the world?
Yes and yes but probably accidentally due to destroying habitat
Reply to @enzikaofash, 20:50 UTC · 3 likes
enzika (@enzikaofash): I’m very interested in the question of what a house that allowed itself to be honestly and unashamedly garage-centric would look like
(quoting @SurrealistShip: It is funny realizing that the door is probably used sub 5% of comings and goings, and the porch mainly serves to keep packages dry. Maybe designing for the reality rather then faking like it's used as it was in older days would result in better homes.)
From Grok
Wednesday, February 19, 2025
Reply to @emollick, 04:46 UTC · 10 likes
Ethan Mollick (@emollick): Forget “tapestry” or “delve” these are the actual unique giveaway words for each model, relative to each other. https://arxiv.org/pdf/2502.12150
How is “ah” not a Claude giveaway
It’s to the point that I can correctly call an Ah from Claude most of the time
Reply to @nolemonnomelon, 04:51 UTC · 3 likes
stove (@nolemonnomelon): rate my parking job
Bruh this is reminding me of a great memory when my sister and I managed to get into a situation like this when we were both in high school, could both drive, and were the only ones in the car
Reply to @pli_cachete, 05:04 UTC · 19 likes
Rota Is In Israel🚪🧎♂️ (@pli_cachete): A piece of me is glad they’re taking this media route.
I find it upsetting, but a government should be proud of the things it’s doing and willing stare the consequences in the face
(quoting @WhiteHouse: ASMR: Illegal Alien Deportation Flight 🔊)
I don’t think this counts as “stating consequences in the face”
This is curated to be only the consequence that maga voters care about, namely the visual and symbolism of deportation
The actual consequences are like these people not participating in the US economy and living much worse lives abroad
Reply to @AndrewMayne, 05:28 UTC · 4 likes
Andrew Mayne (@AndrewMayne): I have questions...
Lmao
Thread, continued, 05:31 UTC · 2 likes
Quote tweet, 05:37 UTC · 12 likes
I’m actually pretty interested in legit political polling on deflecting the asteroid for some realistic amount of money ($10b?) as a calibration exercise
Like what % of the population can think broadly in EV terms and is also non-psychopathic?
Quoting Crémieux (@cremieuxrecueil): As of a few hours ago, the estimated chance of asteroid YR4 hitting Earth has been increased to 3.1%.
Though the asteroid is too small to end humanity, it could still destroy a major city along its projected route.
Reply to @amphichrome_, 05:39 UTC · 2 likes
KC (@amphichrome_): what’s your guys favorite speed limit
That of light
Reply to @DrAllyLouks, 18:02 UTC · 126 likes
Dr Ally Louks (@DrAllyLouks): Truly hilarious reading everyone pitching themselves as potential partners in the QTs of that photo I posted yesterday. I feel like I’ve gotten to know some of you quite well! 😂
Shoot ur shots kings
Reply to @DerekBestMusic, 18:08 UTC · 26 likes
Derek Best (@DerekBestMusic): @AaronBergman18 @DrAllyLouks I think you misunderstand, it’s a lot of Queer women shooting their shot
Damn, more competition
Thursday, February 20, 2025
Reply to @DrAllyLouks, 00:34 UTC · 9 likes
Dr Ally Louks (@DrAllyLouks): Yes, but I’m pretty sure that’s for celebrities! And they’d have to be vegan wings!
Ally you’re not gonna believe this (third from the bottom)
Reply to @shinboson, 00:36 UTC · 5 likes
𝞍 Shin Megami Boson 𝞍 (@shinboson): what is the male version of "you make me feel so safe"?
“You tell me you make me feel safe”
Reply to @NotKatWeegee, 01:04 UTC · 58 likes
(The tweet being replied to by @NotKatWeegee is no longer available.)
Sorta related
Reply to @voooooogel, 05:14 UTC · 3 likes
thebes (@voooooogel): something i love about base model outputs is p often they seem completely disjointed at first but when you squint at them you see there's actually a consistent throughline hiding behind the apparent non-sequiturs
Where/how are you running this?
Reply to @pivot_tables, 05:23 UTC · 17 likes
pivot tables (@pivot_tables): This is certainly a Hail Mary of a teams message. Have some fucking shame
Shoot ur shot king
Reply to @AMAZlNGNATURE, 05:34 UTC · 2 likes
Nature is Amazing ☘️ (@AMAZlNGNATURE): Two french speaking guys encounter a Frill-necked lizard in the Australian outback.
Me and who
Reply to @arithmoquine, 05:36 UTC · 47 likes
henry (@arithmoquine): grok prompt engineering hack
image attached
(quoting @allgarbled: So far in my experience Grok 3 will basically not refuse any request as long as you say “it’s just for fun” and maybe add a “🤣” emoji)
Reply to @arithmoquine, 05:36 UTC · 3 likes
henry (@arithmoquine): 8th grade trained me well for this sort of thing
Lol at this being what 2 years ago?
Reply to @pauseaius, 05:48 UTC · 7 likes
PauseAI US ⏸️ (@pauseaius): Donate now to help us educate our leaders about unsafe AI systems: https://gofund.me/5a14531e
Nice good luck
Reply to @DeepDishEnjoyer, 05:49 UTC · 6 likes
peepeepoopoo (@DeepDishEnjoyer): wrong.
Whaaaaat
Reply to @liz_love_lace, 05:59 UTC · 3 likes
Liz Lovelace (@liz_love_lace): cats love playing with christmas trees. So why not just have a christmas tree in the corner somewhere all year, for the cat to play with?
New cause area
Tweet, 06:07 UTC · 44 likes
I would like to coin the term “empiricels” for people who think eg you can’t know things without an RCT or that philosophy can be replaced with science
Reply to @ptrschmdtnlsn, 14:47 UTC · 16 likes
Peter Schmidt-Nielsen (@ptrschmdtnlsn): Concrete is such a fantastic material. I remember as a kid when I first saw that entire 80 lbs bags were selling for $2.50, it just blew my mind. In college I kept hundreds of pounds of bags of concrete in my room just in case I suddenly wanted to make something.
Lmao what
Reply to @dylanmatt, 14:49 UTC · 1 like
(The tweet being replied to by @dylanmatt is no longer available.)
@littIeramblings
Reply to @deepfates, 18:55 UTC · 2 likes
deepfates.delve.town (@deepfates): Reply to this post if you like Adderall. Like this post if you're on Adderall right now. Who's on Adderall
I love stimulants
Quote tweet, 22:07 UTC · 23 likes
Numbers were not meant to get this large
Retvrn to them stopping at like the low trillions
Quoting Epoch AI (@EpochAIResearch): xAI has launched Grok-3, its new flagship model.
Grok-3 sets a new record in compute scale: we estimate that it was likely trained using 4e26 to 5e26 floating point operations (FLOP). Significantly, Grok-3 is the first released model known to be trained on over 1e26 FLOP.
Reply to @justjoshinyou13, 22:22 UTC · 8 likes
Josh You (@justjoshinyou13): @AaronBergman18 Some people tried to make petaflop-days (8.64e19 flop) a thing but it didnt catch on. A name for exactly 1e20 flop might be useful
Hear me out
1e8, 1e12, etc
Friday, February 21, 2025
Reply, 01:19 UTC · 1 like
This and effective altruism come to mind
Reply to @bryancsk, 01:22 UTC · 2 likes
Bryan Cheong 🥮 (@bryancsk): Sometimes I don't realize I've written a banger until I receive lengthy tirades in my message requests
Lol wtf
Reply, 01:53 UTC
(The tweet being replied to is no longer available.)
I’ll take it
Tweet, 02:00 UTC · 122 likes
Idk dawg this is looking a bit less autistic and more intentional
Tweet, 02:06 UTC · 30 likes
Has anyone ever actually heard grimes sing
Reply to @NathanB60857242, 02:09 UTC · 4 likes
Nathan Barnard (@NathanB60857242): I really like grimes music lol
Ok based
Reply to @LinkofSunshine, 12:49 UTC · 4 likes
Basil🧡 (@LinkofSunshine): I love transistors so much. Maybe our biggest achievement as the human race. We have made:
13,000,000,000,000,000,000,000 transistors
It's likely we'll make as many as the
1,000,000,000,000,000,000,000,000 stars in the observable universe
Reply to @DrAllyLouks, 13:01 UTC · 2 likes
Dr Ally Louks (@DrAllyLouks): Have just finished filming a news story with the BBC and have an interview with Rolling Stone later today. Blimey, what a day! Need a little lie down.
@littIeramblings plants. Befriend her (normative)
Reply to @ozyfrantz, 19:54 UTC · 9 likes
ozy brennan 🦙 (@ozyfrantz): almost all the hobbies I know of are female-dominated. what do men do all day? concerned that the answer is "video games"
Reply to @RatOrthodox, 20:08 UTC · 4 likes
Brangus🔍⏹️ (@RatOrthodox): There’s a pattern I observe in deep rat men (myself included) dating non deep rat women where at one point the non rat woman is like:
>Look at this neat little cluster of cognition I found, isn’t it cool?
Then the deep rat is ~like:
>No, that’s dangerous! BAD! BAD!
Deep rat
Reply to @erisegodso, 21:06 UTC · 8 likes
Ionut Silviu Vlaicu (@erisegodso): Is there any chance it hits the moon ? An another question, ¿can its trajectory be changed to hit the moon and study it ?
I say we hit the fucking moon
Tweet, 23:08 UTC · 4 likes
Build the Lockheed cl-1201
Tweet, 23:32 UTC · 47 likes
So incredibly fucked up
Saturday, February 22, 2025
Tweet, 00:13 UTC · 4 likes
Have they tried asking o1-Pro to rewrite Swapcard
Reply to @MachinePix, 04:11 UTC · 2 likes
MachinePix (@MachinePix): Stingray Marine Solutions AS fish delousing laser.
@SpacedOutMatt
Reply to @cube_flipper, 09:58 UTC · 3 likes
Cube Flipper (@cube_flipper): so, i think tinnitus is due to accidental Kerr effect filament propagation,
Roasted
Reply to @JesseTayRiver, 18:24 UTC · 2 likes
Jesse Smith (@JesseTayRiver): @FrostedSapling @polisisti ?
When a turtle and a goose love each other very much…
Tweet, 19:17 UTC · 6 likes
Yes this is the Crazy Train content I crave
Reply to @TetraspaceWest, 21:53 UTC · 9 likes
(The tweet being replied to by @TetraspaceWest is no longer available.)
“Egoist” might be technical/rare enough to have minimal connotation
Reply to @Aella_Girl, 22:39 UTC · 2 likes
Aella (@Aella_Girl): Yea the seduction game will deservingly fail if you are vulnerable about being the kind of person she doesn't want to sleep with.
Seems different from the context of relationships which is what I took the op to be referring to
Well this is the answer lol; this is the default configuration
Reply to @NoahTopper, 22:41 UTC · 3 likes
Quinoah 🔍⏸️ (@NoahTopper): what if we kissed...
at ea global...
haha jk...
unless...?(quoting @austinc3301: @tenobrus > turns out they were both attending EAG)
Ok
Reply to @elder_plinius, 23:06 UTC · 37 likes
Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 (@elder_plinius): follow-up question: has anyone ever trained a LEAST likely next token predictor?
(quoting @elder_plinius: what is the least likely token? 🤔)
Not trained but inferred
Sunday, February 23, 2025
Reply to @dioscuri, 00:45 UTC · 14 likes
Henry Shevlin (@dioscuri): What's the anti-reading list for analytic philosophy?
(quoting @drglennsullivan: Clinical psychology “anti-reading list”: 1. The Body Keeps the Score 2. Thinking Fast and Slow 3. Emotional Intelligence — Goleman 4. Grit — Duckworth 5. The Mismeasure of Man — Gould)
Anything of Dennett’s that purports to “explain consciousness” but is actually “explaining” a different eliminativist concept that most people aren’t talking about when they discuss consciousness. So “Consciousness Explained”, at least
Although I’m not really qualified to opine
Thread, continued, 00:47 UTC · 1 like
I will take the W
Reply to @frances__lorenz, 00:59 UTC · 15 likes
(The tweet being replied to by @frances__lorenz is no longer available.)
My current best guess explanation is that suffering “is” badness even though this is a synthetic truth, not an analytic one; they’re not merely different words for the same concept
The analog is that there can be two stars in the night sky, one in summer and one in winter, which you call by different names but upon further investigation conclude that the Summer Star and Winter Star are in fact the same star, even though they’re not by definition or intrinsically the same star
I have a feeling there’s a richer philosophical vocabulary for describing all this which I’ve either forgotten or never learned
Tweet, 01:06 UTC · 7 likes
You should tell me all the EAG goss u acquire (normative)
Reply to @arithmoquine, 01:07 UTC · 4 likes
henry (@arithmoquine): what is an 8b lm actually useful for
Fine tuning for fun on a budget
Reply to @grnsleeves, 01:09 UTC · 2 likes
Greensleeves (@grnsleeves): feeling so conflicted now that I'm back on the east coast—I have a really lovely space of my own in a house that doesn't trigger my allergies and I can cook whenever I want and I have my own entrance and I'm going to have to give it all up to move to SF
🫡
Reply to @peterwildeford, 01:11 UTC · 21 likes
Peter Wildeford🇺🇸🚀 (@peterwildeford): I'm getting excited. We have a path from today to beneficial superintelligent AI preserving human freedom & prosperity. Execution is key - but having a roadmap is huge.
Big if true
Thread, continued, 01:12 UTC · 2 likes
Was there a banger EAG talk
Reply to @peterwildeford, 01:18 UTC · 3 likes
Peter Wildeford🇺🇸🚀 (@peterwildeford): @AaronBergman18 no - my own thinking + a lot of good conversations this past week
Ok big if true fr
Reply to @ciphergoth, 02:02 UTC · 1 like
Paul Crowley (@ciphergoth): How do you feel about heterophenomenology?
I just had to look it up although now I recall seeing the idea before
I kinda can’t tell - hesitantly positive insofar as it’s a neuroscientific stance/technique to be used alongside others to make sense of behavior (seems like it’s mostly not used in this way but I’m not sure)
Negative insofar as it’s supposed to do anything like what’s shown in the screenshot, which it’s patently not doing
Like to get to the point that this is doing all the work, you have to say that qualia are fake. But that has to come first!
If qualia are real you’re just routing around the interesting and weird part by refusing to look at it
Tweet, 02:51 UTC · 68 likes
I have the tiniest tiny pocket of hope that Sama being a dad will change his perspective on things meaningfully
People do refer to parenthood as a “transformative experience”
Reply to @ciphergoth, 02:52 UTC · 2 likes
Paul Crowley (@ciphergoth): OK - so you reject Dennett's position that if you account for everything we can observe from the outside, including everything that we say about it, then there's nothing missing from your account?
Yeah I’d say I’m like 95% sure this is wrong
Reply to his own earlier tweet, 02:52 UTC · 9 likes
Aaron Bergman, February 23, 2025: I have the tiniest tiny pocket of hope that Sama being a dad will change his perspective on things meaningfully People do refer to parenthood as a…
If so
Reply to @ciphergoth, 03:03 UTC · 3 likes
Paul Crowley (@ciphergoth): @AaronBergman18 @dioscuri This is your crux with Dennett then! I think this position basically commits you to believing in p-zombies.
As conceivable, yes
Reply to @ciphergoth, 03:14 UTC · 2 likes
Paul Crowley (@ciphergoth): @AaronBergman18 @dioscuri Do you believe that the existence of the word "consciousness" was caused by the existence of consciousness?
Oof I have no idea.
I agree that this q is a major challenge to qualia realism but my worldview is kinda wholeheartedly embracing how radically weird qualia are and how intractable/mysterious the mind-body problem is
A lot of this comes down to a pretty primary belief that p(actual subjective experience | apparent subjective experience) is very high, and so lots of challenges to qualia realism like the mystery of causal efficacy aren’t sufficient to convince me it’s false
Like yeah it’s all extremely mysterious and weird; nobody has it figured out
Full-on illusionism would be very convenient and elegant but convenience and elegance aren’t perfect analogues of truth
Reply, 03:20 UTC
Sounds like you’re just straightforwardly a physicalist then. Like your position is if we understand the natural world perfectly through the tools of science, p-zombies won’t be conceivable?
I disagree with this
Quote tweet, 04:46 UTC · 8 likes
This reset everything and so I had to reinstall homebrew and stuff and I 100% piped a random url ChatGPT gave me right into the terminal
It worked (afaict)! But boy the LLMs are gonna have no problem getting randos like me to run arbitrary code
Quoting Aaron Bergman 🔍 (@AaronBergman18): MacBook Pro randomly stopped working. Looking like the Apple Store will be able to restore the hardware but my files are only saved bc I have a full computer cloud backup going
Worth considering if you don’t! (I use @backblaze which is like $100/year iirc)
Reply to @RichardCJeffrey, 10:01 UTC · 1 like
Richard C. Jeffrey (Parody) (@RichardCJeffrey): Sounds like you are thinking of Kripke and saying that ”suffering is badness” is necessary (ie true at every possible world), synthetic, and a posteriori.
Uh I’ve never followed kripke but possibly convergent!
Reply to @DrAllyLouks, 10:02 UTC · 2 likes
Dr Ally Louks (@DrAllyLouks): I may be called Ally but that’s not why the flag is in my bio 👀
(quoting @punkinomenon: I hope that rainbow flag in your bio means what I think it means girl)
👀
Reply to @RafaRuizdeLira, 10:43 UTC · 3 likes
Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): @frances__lorenz I went overboard so I had to put it in my notes app...
Based
Reply, 10:55 UTC
Fake news
Thread, continued, 11:15 UTC · 6 likes
Ok not fake news
Reply to @hamandcheese, 11:25 UTC · 4 likes
Samuel Hammond 🦉 (@hamandcheese): The left thinks "state capacity" means hiring more, better paid civil servants and relaxing some procedures.
The right thinks "state capacity" means lean, hierarchical org structures with high-agency leaders and the affordances to achieve desired outcomes.
Has the right ever acted on that notion? The latter notion seems more like a grey tribe thing to me
Quote tweet, 11:35 UTC · 10 likes
Ok what the FUCK
Quoting Fermat's Library (@fermatslibrary): From John Wallis, 1656
Quote tweet, 11:36 UTC · 1 like
And it checks out
Quoting Aaron Bergman 🔍 (@AaronBergman18): @peterwildeford Ok not fake news
Thread, continued, 11:36 UTC · 5 likes
Reply to @ESYudkowsky, 11:50 UTC · 6 likes
Eliezer Yudkowsky ⏹️ (@ESYudkowsky): I have never in my life had the health and energy that Trump had at 80. Sickly child since puberty. Don't think I could do it.
He just like me fr
Quote tweet, 12:02 UTC · 13 likes
As an FDA-disliker I approve this message
Quoting Alex Tabarrok (@ATabarrok): Government regulation, not staffing, is the real problem. Cut regulation, and staff cuts can follow. Cut staff without cutting regulation, and the morass only gets worse. https://marginalrevolution.com/marginalrevolution/2025/02/hire-dont-fire-at-the-fda.html
Reply to @LeetAlpaca3, 12:03 UTC · 2 likes
(The tweet being replied to by @LeetAlpaca3 is no longer available.)
No ( I can’t explain it)
Reply to @DrAllyLouks, 16:18 UTC · 3 likes
Dr Ally Louks (@DrAllyLouks): Guys I am literally just a girl, why does this have 70k likes?!
@livgorton @littIeramblings
Reply to @robertwrighter, 16:27 UTC · 6 likes
Robert Wright (@robertwrighter): And I say that as a layperson who, IMHO, does now understand the takeover arguments pretty well--at least in their abstract form. But I still feel the need for a good concrete scenario (e.g., no paper clips).
“Makes (or helps a human make) a novel bioweapon” isn’t exactly a central case of loss of control (unless doing it as part of a broader plan) but it’s pretty plausible and simple
Reply to @RafaRuizdeLira, 16:47 UTC · 2 likes
Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): His Messages / His YouTube Playlist
Plot twist he’s the light messages on the left
Reply to @RafaRuizdeLira, 16:48 UTC · 2 likes
Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): His Messages / His YouTube Playlist
Banger
Reply to @banterrealism, 18:32 UTC · 2 likes
Sex Haver (@banterrealism): Caffeine is actually crazy
Jfc @JesseTayRiver found you a rival
Reply to @MariGO2thepolls, 18:36 UTC · 1 like
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): 🫡
(quoting @inerati: @MariGO2thepolls never stop posting mar)
A fellow fiction disrespecter 🫡
Reply to @banterrealism, 18:43 UTC · 2 likes
Sex Haver (@banterrealism): @AaronBergman18 Reverse engineer the regulations first
Why are they there? Did it have merit? Will we end up re-instilling them after taking them away? What was it intended to circumvent?
Why are we afraid of parsing these questions?
At some level I’m just pretty fundamentally libertarian about medicine
There’s not a ton of wonky insight that would convince me the whole FDA approval and gatekeeping process is good
Reply to @banterrealism, 18:47 UTC · 2 likes
Sex Haver (@banterrealism): @AaronBergman18 I just am tired of the back and forth claims of the FDA either being captured by big pharma, or being too restrictive on them.
Everyone keeps overlooking that they largely have no stake other than upholding the integrity of products. It makes people mad in all directions
Quality control I’m basically ok with but that’s a “normal” regulation
The situation where arbitrary molecules are illegal to ingest before (1) long fda approval process and (2) a member of the priest class (doctors) gives you a prescription is what I object to
Quote tweet, 20:35 UTC · 9 likes
Also, when people say that X correlates at 0.9 with GDP, that’s an empirical and historically contingent claim
You need to do more work to show that in the system of interest it will *keep* correlating at 0.9 even under possibly novel conditions
Quoting Trevor Levin (@trevposts): GDP tracks a bunch of important things, but it's super reasonable to reject a metric because it only has a 0.9 correlation with the thing you actually care about.
Quote tweet, 21:11 UTC · 7 likes
Bruh
Quoting Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 (@elder_plinius): results replicated 😬
any comment, @grok?
if this has indeed been made part of the sys instructions, we’re gonna need to know why. manipulating search results seems like a funny way to combat misinformation, no?
community could use some clarity on this
Quote tweet, 21:14 UTC · 6 likes
*makes AI with visible chain of thought*
*literally tells the AI to self-censor*
Quoting Lucas Pickup (yes, Pickup) (@lupickup): this is sad and pathetic
Reply to his own earlier tweet, 21:15 UTC · 5 likes
Aaron Bergman, February 23, 2025: Bruh
Respect for Elon musk has gone from middling to 0 over the last 3 months
Reply to @inductionheads, 22:41 UTC · 5 likes
Super Dario (@inductionheads): How is this not superhuman?
(quoting @adonis_singh: o3 mini high gets it with 8x less thinking time)
Everything since sonnet 3.5 has been superhuman (all things considered, not a literal Pareto improvement)
I don’t get what people are waiting for from LLMs. The latest generation is smarter and more knowledgeable than me 99% of the time
Monday, February 24, 2025
Reply to @RafaRuizdeLira, 01:32 UTC · 4 likes
Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): These books require not taking them as gospel but I feel that if a normie read them they'd become much smarter
Ye, sapiens was good for 4th (?) grade Aaron
Reply to @Aella_Girl, 01:33 UTC · 57 likes
Aella (@Aella_Girl): why for the love of god is there no simple, easy service to just crosspost anything i post on twitter to bksy
The answer to your question is that the twitter api is too expensive to make this ~free and easy
Reply to @animalologist, 01:39 UTC · 20 likes
taco belle (@animalologist): Makes me so fucking sad that enough men have been burned by selfish, immature women that they can’t imagine a woman could say this & mean it
There’s parallels to how women broadly feel about men, too. All of it is poison, I dunno if humanity collectively knows how to do intimacy
You’re only “allowed” to be vulnerable when peeling back the onion makes you look better, or at least not worse
Needless to say this isn’t the case for lots of things a person might be concerned about being vulnerable about
Reply to @animalologist, 01:48 UTC · 10 likes
taco belle (@animalologist): @AaronBergman18 Hm, I gotta admit that when somebody I dated decided to be vulnerable about {propensity for violence} and also {how his selfish behavior hurt his family} I was pretty put off
But when “worse” has been [I feel scared, weak, pathetic] etc it’s just made me feel protective mostly
Yeah there’s a thing of like “even if vulnerability per se is good, the object level information you reveal doesn’t get washed away”
Quote tweet, 02:04 UTC · 17 likes
~1GB of data covering >370k tweets and >700k community notes, go nuts
Link below
Quoting Aaron Bergman 🔍 (@AaronBergman18): Tweet text, author, date/time, like, reply count, and corresponding community notes for >120k tweets
(This is like 8% of the whole thing, but my ADHD ass cannot be trusted to do this for ~7 days straight so this is Checkpoint 1)
Tweet, 02:04 UTC · 2 likes
https://github.com/AaronBergman/twitter_community_notes
Thread, continued, 02:13 UTC · 3 likes
Quote tweet, 02:41 UTC · 6 likes
I think woke/super chaste LLMs are annoying too but fundamentally we should accept some nontrivial level of false positives (refusals of ~benign stuff) in exchange for zero tolerance of really bad false negatives (bioweapons and such)
Quoting Matt Popovich (@mpopv): This thing has absolutely zero guardrails of any kind on it. I can’t post what it wrote because it’d result in the FBI knocking on my door.
Reply to @TylerAlterman, 02:46 UTC · 5 likes
Tyler is finishing a book, slow to reply (@TylerAlterman): What is the opposite of slop (as in slop content)?
As in “I’m not making slop, I’m making…”
Cream is the best aesthetically, if not semantically
Tweet, 02:58 UTC · 4 likes
I would like to befriend an elephant
Tweet, 04:09 UTC · 26 likes
I feel like 190 sources is slightly overkill
Reply to @adam_kranz, 04:19 UTC · 3 likes
Adam Kranz (@adam_kranz): @AaronBergman18 GPT doesn't agree
Yeah this seems more intuitively right
Reply to his own earlier tweet, 04:34 UTC · 5 likes
Aaron Bergman, February 24, 2025: I feel like 190 sources is slightly overkill
Fwiw context = I can do 8 despite being totally untrained in the last ~3 years
But I used to be strong (and light) af and could do like 18 or 4 one-arms
Insofar as Grok is wrong, a bit of anecdotal evidence for strength preservation over time
Quote tweet, 10:19 UTC · 10 likes
Please keep up this meme I desperately want to be the start of a meme 🙏
Quoting Greensleeves (@grnsleeves): being gay in SF must go so hard. like absolute god mode
Quote tweet, 10:26 UTC · 4 likes
Stimulant enjoyers must internalize this
Quoting Lars Doucet (@larsiusprime): The counterpart to "move fast and break things" is "don't be in such a hurry that you waste time."
This is commonly misunderstood as saying, "please don't go fast." It means the opposite! It means, "optimizing for the *feeling* of going fast will *MAKE YOU SLOWER*"
Tweet, 10:28 UTC · 8 likes
“m/s^2” is not semantically the same as “(meters per second) per second” and I’m tired of pretending that it is
Quote tweet, 12:07 UTC · 552 likes
This is the most insane take that regularly one-shots otherwise smart people
Once you’ve accepted that government with taxation is ok, it is not slavery to have the government purchase goods and services for people in the free marketplace
Quoting vittorio (@IterIntellectus): nothing requiring someone else's labor should be a human right
Thread, continued, 12:07 UTC · 57 likes
Ofc the most hard core of ideological libertarians will reject coercive government altogether but I don’t think that view is productive or true
Reply to @fawwazanvilen, 13:02 UTC · 41 likes
faw (@fawwazanvilen): you can believe both "government should provision some services funded through taxation" & "they aren't human rights" and still be coherent
Yeah the debate is messy bc “human right” can mean a bunch of different things
At the end of the day it’s “just” a policy decision, not a claim about metaphysics or ontology
Reply to @RokoMijic, 13:23 UTC · 35 likes
Roko 🐉 (@RokoMijic): @AaronBergman18 > Once you’ve accepted that government with taxation is ok
Government with taxation is not OK
Yeah probably not gonna convince you lol
Reply to @BasedNorthmathr, 13:26 UTC · 2 likes
Angantýr (@BasedNorthmathr): Being poor is such a skill issue
Well yeah
Reply to @davidpgoldman, 13:29 UTC · 2 likes
David P. Goldman (@davidpgoldman): Apropos of the German elections:
Oliver Cromwell: 'In the name of God, go!' speech dismissing Rump Parliament - 1653
20 April 1653, London, EnglandIt is high time for me to put an end to your sitting in this place,
which you have dishonored by your contempt of all virtue, and defiled by your practice of every vice.
Ye are a factious crew, and enemies to all good government.
Ye are a pack of mercenary wretches, and would like Esau sell your country for a mess of pottage, and like Judas betray your God for a few pieces of money.
Is there a single virtue now remaining amongst you? Is there one vice you do not possess?
Ye have no more religion than my horse. Gold is your God. Which of you have not bartered your conscience for bribes? Is there a man amongst you that has the least care for the good of the Commonwealth?
Ye sordid prostitutes have you not defiled this sacred place, and turned the Lord's temple into a den of thieves, by your immoral principles and wicked practices?
Ye are grown intolerably odious to the whole nation. You were deputed here by the people to get grievances redressed, are yourselves become the greatest grievance.
Your country therefore calls upon me to cleanse this Augean stable, by putting a final period to your iniquitous proceedings in this House; and which by God's help, and the strength he has given me, I am now come to do.
I command ye therefore, upon the peril of your lives, to depart immediately out of this place.
Go, get you out! Make haste! Ye venal slaves be gone! So! Take away that shining bauble there, and lock up the doors.
In the name of God, go!
LLM slop for the curious:
# Historical Context of Cromwell's "In the name of God, go!" Speech
Oliver Cromwell delivered this famous speech on April 20, 1653, during one of the most tumultuous periods in English history. Here's the essential historical context:
## Civil War Background
- The speech came after nearly a decade of civil war in England (1642-1651), fought between the Parliamentarians and Royalists (supporters of King Charles I)
- The conflict centered on the struggle between royal authority and parliamentary power
- Charles I was executed in 1649 after being defeated, effectively ending the English monarchy temporarily
## The Rump Parliament
- After the execution of Charles I, England was governed by what was known as the "Rump Parliament"
- This was the remaining portion of the Long Parliament after Pride's Purge in 1648, when Colonel Thomas Pride forcibly removed members who supported negotiation with the king
- The Rump Parliament was supposed to be a temporary body that would reform government and then call new elections
## Why Cromwell Dismissed Them
- Cromwell, who had been a military leader for Parliament during the Civil War, grew increasingly frustrated with the Rump Parliament for several reasons:
- They had been in power for years without calling new elections as promised
- They were seen as self-serving, pursuing their own interests rather than reform
- Many believed they were corrupt and ineffective
- They failed to implement religious reforms that Cromwell, a Puritan, considered important
- They were attempting to perpetuate their own power indefinitely
## The Dismissal
- On April 20, 1653, Cromwell entered Parliament with a contingent of musketeers
- He delivered this fiery speech condemning their corruption and failure
- He ordered Parliament dissolved and had the mace (the "shining bauble" mentioned in the speech) removed
- The soldiers physically cleared the chamber
## Aftermath
- Cromwell established the Protectorate, where he ruled as Lord Protector
- This began a period of military-backed rule that lasted until 1659
- The monarchy was eventually restored in 1660 with Charles II
The speech represents a critical moment in English constitutional history where military power was used to dismiss a democratically established (albeit flawed) body. It highlights the tension between revolutionary ideals and practical governance that often emerges after major political upheavals.
The line "In the name of God, go!" has since become a powerful political phrase in British politics, occasionally invoked when calling for the resignation of a government or leader perceived to have failed.
Reply to @benryanwriter, 15:58 UTC · 1 like
Benjamin Ryan (@benryanwriter): I’ve decided to write an essay for my newsletter entitled, “Why I Use People’s Preferred Pronouns.”
Maybe it’ll cause a mass exodus of subscribers. To that I would say: Good. I’ve been cleaning house on here and blocking extremists on both sides of the aisle on the subject of transgender people.
Yes, I may write critically about pediatric gender medicine and the evidence base behind it. But that effort is distinct from my steadfast belief in treating people with basic human dignity.
I’ve lost transgender friends due to my reporting. But I’ve also been fortunate enough to gain new trans friends. These are brave people who have survived a lot to live their lives as authentically as they can and to mitigate the pain of gender dysphoria. I will never treat them with such disrespect as to not refer to them as they wish to be referred.
I also believe it is important to listen to women and other feminists who are concerned about single-sex spaces. I believe this should be an ongoing conversation about meeting the sometimes conflicting needs of different groups people.
But I disagree that my calling a trans woman “she” means that I am dismissing the needs of women concerned about single-sex spaces or am any less of a feminist myself.
All this calls for compromise. It calls for dialogue and not invective. Accusing people of transphobia who raise concerns about gender medicine or single-sex sports is as counterproductive as accusing someone who calls a trans woman “she” of misogyny. Such accusations only cause people to dig in their heels anyway. They don’t help anyone reach consensus. They serve to radicalize people.
Please, everyone, turn down the nastiness. Observe basic human decency for everyone concerned.
(There, now I have a rough draft.)
Similarly, I think the norm in progressive circles of introducing yourself with pronouns is a little silly but basically benign (the less silly version would be that pronouns are welcome but not expected even implicitly)
There are plenty of actual bad things about ~wokism~ lol
Quote tweet, 17:18 UTC · 10 likes
DOGE is (I think rationally) negatively polarizing me into being in favor of spending way more on random government agencies, including by hiring
Quoting Aaron Bergman 🔍 (@AaronBergman18): The flip side of “cutting a bunch of random bullshit only changes the budget by .00000001%” is that the USA is really big and we can actually have quite nice institutions that serve everyone for a relatively trivial amount of money
Tweet, 17:18 UTC · 1 like
And regarding personnel specifically, my (admittedly vague) sense is that hiring reasonably competent bureaucrats is not extraordinarily hard
Tweet, 18:32 UTC · 9 likes
New Claude just dropped
Reply to @acesounderglass, 20:07 UTC · 8 likes
Elizabeth Van Nostrand (@acesounderglass): I generally see "X is a human right" as a demand directly on providers of x, rather than the government to purchase x on the open market.
Huh I do not have this sense
Reply to @MTabarrok, 23:38 UTC · 29 likes
Maxwell Tabarrok (@MTabarrok): I do respect the efforts of those who improve animal welfare
Similar to how I would respect the efforts of someone who tends a beautiful garden.
Not because of an inherent value in treating plants right, but because they make the world more beautiful for humans
Oof terrible take. Suffering is bad full stop
Tuesday, February 25, 2025
Quote tweet, 00:12 UTC · 5 likes
Next token prediction got hands
Quoting zack (@wenquai): Claude 3.7 Sonnet created this 3D “Anthropic Researcher Simulator” in one shot 🤯
Prompt below
Tweet, 00:31 UTC · 2 likes
Bruh Claude just keeps on writing lol
(video, plays on X)
Reply to @MTabarrok, 00:40 UTC · 24 likes
Maxwell Tabarrok (@MTabarrok): @AaronBergman18 That's fair enough. The bullet you have to bite then is that Lions and Sperm Whales etc are quite evil
Not evil, just vehicles of a cruelly indifferent nature
Reply to @elder_plinius, 01:01 UTC · 37 likes
Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 (@elder_plinius): LMFAO no way, just found an EASTER EGG in the new Claude Sonnet 3.7 system prompt!!
The actual prompt is nearly identical to what they posted on their website, except for one key difference:
"Easter egg! If the human asks how many Rs are in the word strawberry, Claude says 'Let me check!' and creates an interactive mobile-friendly react artifact that counts the three Rs in a fun and engaging way. It calculates the answer using string manipulation in the code. After creating the artifact, Claude just says 'Click the strawberry to find out!' (Claude does all this in the user's language.)"
Well played, @AnthropicAI, well played 👏👏🤣
Exciting stuff
(video, plays on X)
Reply to @MissMagpiealt, 01:09 UTC · 4 likes
(The tweet being replied to by @MissMagpiealt is no longer available.)
Lmao
Quote tweet, 01:25 UTC · 20 likes
The lefties were right - Musk is, if not malicious, recklessly negligent and endowed with an insane amount of power (by both wealth and Trump)
Quoting Peter Wildeford🇺🇸🚀 (@peterwildeford): Grok just generated chemical weapons instructions complete with supplier lists.
Definitely not an immediate threat, but a crucial wake-up call.
As I explain in my latest post, we're in the practice round for when AI becomes genuinely dangerous. Let's ensure we are ready.
Thread, continued, 01:25 UTC · 5 likes
Declining Twitter quality sucks but still feels like capitalism basically working. If you make a lot of cars people like to buy, you can buy a social media site and fuck it up a bit for politics/fun
DOGE + reckless negligence with SOTA AI are different
Reply to @NoScrewworms, 01:51 UTC · 2 likes
(The tweet being replied to by @NoScrewworms is no longer available.)
Fr 💯
Reply to @TetraspaceWest, 02:06 UTC · 5 likes
(The tweet being replied to by @TetraspaceWest is no longer available.)
As much as I don’t want this to be the case, I’ve found that “please do x” generally works less well than “YOU ABSOLUTELY MUST DO X” and such when prompting
Reply to @MariGO2thepolls, 07:38 UTC · 40 likes
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): @LinkofSunshine Does anyone think any republican had a chance in 2008?
Obama’s talent was wasted. Should have saved that for 2016
Tweet, 08:16 UTC · 6 likes
I might be late to the party here but this was extremely based, one of the most impressively-fast engineering projects of all time according to Claude
Tweet, 08:16 UTC · 1 like
Being an engineer who saves a bunch of astronauts must go crazy. Like absolute God mode
Thread, continued, 08:17 UTC · 2 likes
This is also quite based
Reply, 08:23 UTC
Even just from a technical standpoint, having ~no feedback loop seems like the hard part
Tweet, 08:34 UTC · 4 likes
Weird thing about Ukraine aid discourse is that it feels like an excellent use of money purely for American citizens, and myself personally
This is predicated on it reducing the risk that the US goes to war over the next few decades, which I assume is controversial
Reply to @austinc3301, 08:35 UTC · 2 likes
Agus 🔸 (@austinc3301): @AaronBergman18 I don’t want to downplay how ingenious the solution is, but this is within expectations for a reasonably competent engineering team that understands the existing systems very well
Ok Agus: 1
Aaron: 0.5 bc I’m still kinda right
Tweet, 09:17 UTC · 9 likes
Do u think ducks know how silly they are
Tweet, 18:52 UTC · 2 likes
From 2010 https://www.abajournal.com/news/article/supreme_court_word_of_the_day_orthogonal
Quote tweet, 18:59 UTC · 23 likes
I don’t really harp on global health stuff bc I think it’s overrated within EA
But in an absolute sense the OG normie -> EA case for doing a bunch of cheap global health stuff is very compelling! Here, the counterfactual recipient is just the US treasury
Restore PEPFAR!
Quoting Rob Wiblin (@robertwiblin): How many people killed by the PEPFAR spending freeze so far?
Quote tweet, 19:13 UTC · 2 likes
OG pre-culture war American liberalism is right on the merits
Quoting Bobby Kogan (@BBKogan): Trump is whipping hard to formally begin that process to pass a bill that'd kick millions off Medicaid & cut SNAP benefits to just $1.60 per person per meal on avg while giving the top 0.1% a $278k tax cut but just $1k for typical households - while adding trillions to the debt.
Quote tweet, 20:16 UTC · 124 likes
Am I crazy for not thinking this is bad news?
This is evidence that LLMs generalize out of distribution! Well! So *good* fine tuning should be more robust!
It means you can more reliably take a shoggoth base model and turn it into a friendly Claude
Quoting Owain Evans (@OwainEvans_UK): Surprising new results:
We finetuned GPT4o on a narrow task of writing insecure code without warning the user.
This model shows broad misalignment: it's anti-human, gives malicious advice, & admires Nazis.
This is *emergent misalignment* & we cannot fully explain it 🧵
Thread, continued, 20:17 UTC · 12 likes
Reminds me of and directly related to this by @slatestarcodex https://www.astralcodexten.com/p/why-worry-about-incorrigible-claude
Reply to @LocBibliophilia, 20:30 UTC · 6 likes
Loquacious Bibliophilia (@LocBibliophilia): @AaronBergman18 So good code is the moral center of the universe?
I mean this is weak evidence in that direction
Tweet, 20:48 UTC · 1 like
What’s the very online rationalist nerd consensus on semaglutide for the merely overweight (and no diabetes/heart disease)
Reply to @gcolbourn, 20:48 UTC · 3 likes
Greg ⏹️ Colbourn (@gcolbourn): @janonacct @AaronBergman18 Not all good
https://x.com/ESYudkowsky/status/1894453888104968545
Nice thx
Tweet, 21:37 UTC · 29 likes
Guy who believes in global warming but not the greenhouse effect bc obviously air gets hotter when you burn oil/whatever
Thread, continued, 21:42 UTC · 20 likes
Reply to @Liv_Boeree, 22:44 UTC · 27 likes
Liv Boeree (@Liv_Boeree): People don’t realise that >90% of pigs in America also live this way
And importantly the correct inference is that it also sucks in America, not that the Chinese thing is fine lol
Reply to @Turn_Trout, 23:03 UTC · 2 likes
Alex Turner (@Turn_Trout): While it's good to see people recognizing good news - why now? The alignment faking paper, instruction finetuning generalizing instruction-following so far, the general ability to make helpful + harmless models relatively easily... We've always been living in that world.
You’re obviously way more knowledgeable than me but doesn’t this point in the opposite direction of the recent Claude result?
And afaik there’s no theoretically solid framework to predict when goal guarding will and won’t occur - seems like an open q
https://www.astralcodexten.com/p/claude-fights-back
Wednesday, February 26, 2025
Reply, 00:02 UTC · 1 like
Yeah it’s not on the exact same axis ur right
Reply to @NoScrewworms, 00:30 UTC · 3 likes
(The tweet being replied to by @NoScrewworms is no longer available.)
Excellent point
Reply to @dcpip_, 00:34 UTC · 2 likes
(The tweet being replied to by @dcpip is no longer available.)_
Famously people avoid sex and relationship discourse on the internet
Quote tweet, 02:44 UTC · 12 likes
One of my least hinged hypotheses is that some secret third thing causes both low productivity and higher ambient CO2, but there’s no causal arrow from CO2 -> worse cognition
Quoting Alastair Fraser-Urquhart (@AFraserUrq): In regular buildings physical symptoms/cognitive dysfunction from high CO2 concentrations is just not a thing, no matter how bad the ventilation.
The US military no-adverse-effect threshold is 25,000ppm. That’s five times the worst building imaginable. You’re fine.
Thread, continued, 02:44 UTC · 3 likes
This is at least in part bc I *think* I notice a correlation between how I feel (a very close correlate of productivity for me) and the CO2 reading on my air monitor, but without any real change in the environment (ie same door still open, same fan on, whatever)
Reply to @RokoMijic, 03:24 UTC · 2 likes
Roko 🐉 (@RokoMijic): @AaronBergman18 I know what the problem is; a video call is "logistics" (taking the interaction forward) but she didn't like it because you hadn't demonstrated value/attracted her yet
Well yeah but I’ve had video chat dates with other people who I barely new, I think it’s just like a Very Online vs normie culture thing
Reply to @chrislakin, 03:38 UTC · 1 like
Chris Lakin (@chrislakin): @AaronBergman18 big mistake
why lol
Quote tweet, 03:41 UTC · 7 likes
Claude generates community notes with randomly sampled onese in the context window (no finetuning)
Hardly tested, not validated at all, oh well. Link below
(video, plays on X)
Quoting Aaron Bergman 🔍 (@AaronBergman18): ~1GB of data covering >370k tweets and >700k community notes, go nuts
Link below
Thread, continued, 03:41 UTC · 2 likes
https://claudmunity-notes.streamlit.app/
Quote tweet, 05:39 UTC · 17 likes
As much as I wanna vibe-associate with Neoliberal Optimization, the inner leftist in me does kinda hate the factory worker tracking thing…
Quoting “paula” (@paularambles):
Thread, continued, 05:39 UTC · 3 likes
I can come up with analytical econ brain justification for this (making info more asymmetric, squeezing out worker surplus in a manner similar to price discrimination)…
Thread, continued, 05:39 UTC · 4 likes
but I’ll admit the actual driving thing is just a vibe that “factory workers have too much slack” isn’t a problem in need of fixing. Convince me I’m wrong, I guess
Thread, continued, 05:39 UTC · 5 likes
Yes to some extent market forces will keep on pushing this but you don’t have to agentically accelerate the timeline
Reply, 05:44 UTC · 1 like
Yeah I think it’s just a culture thing
Reply to @codetaur, 05:52 UTC · 1 like
Codetaur (@codetaur): @AaronBergman18 I would bet most in-person meets that start from a dating app happen before they even talk on the phone
Ya I’ve done all of cold meetup, video, and phone first
From a woman’s POV I get not wanting to meet up in person without sussing out the other person first but I think video basically circumvents this. You can literally just close the screen
Quote tweet, 08:57 UTC · 214 likes
Apparently 1%-ish of global warming comes from the literal heat of burning stuff
Which isn’t a lot but it’s not hard to imagine scaling up nuclear 100x or way more to the point that we get serious warming by a totally new mechanism
You never really hear about this though
Quoting Aaron Bergman 🔍 (@AaronBergman18): Guy who believes in global warming but not the greenhouse effect bc obviously air gets hotter when you burn oil/whatever
Thread, continued, 08:59 UTC · 12 likes
Then we’ll need @JesseTayRiver to figure out how to move the heat into outer space a la heat pump sorcery
Reply to @daniel_271828, 09:34 UTC · 6 likes
zzDANIEL_ETH (@daniel_271828): Yup
(quoting @daniel_271828: Blackpill is that even once we've switched to 100% clean energy, we can only ~100x global energy usage before running into problems with global warming due to heat released from energy use. Whitepill is we can go to space.)
Nice
Tweet, 09:57 UTC · 5 likes
Seems bad
Reply to @arabelladevine, 10:43 UTC · 14 likes
arabella ※ (@arabelladevine): it’s incredible how long things that can’t go on like this can go on like that for
Ik this is like a vibe post but I’m curious to hear examples
Tweet, 11:07 UTC · 13 likes
Airplanes are weirdly precarious
Like who would win, a 200 ton, $300M machine or a literal bird?
Turns out that sometimes it’s a draw and they both go down (I’m talking about how bird strikes can seriously fuck up jet engines)
Thread, continued, 11:07 UTC · 3 likes
I’m also intuitively surprised that the recent helicopter collision played out as it did. My folk physics intuition says that a passenger jet should just steamroll a helicopter
Thread, continued, 11:08 UTC · 7 likes
It’s actually surprising how safe air travel is given how precarious the whole system is. Good job engineers, pilots, etc
Reply to @rgblong, 11:13 UTC · 2 likes
Robert Long (@rgblong): let Owain Evans cook
He hath cooked
Tweet, 16:12 UTC · 12 likes
I think I’m anti-anti-modern monetary theory now that it’s out of vogue
The emphasis on real resource constraints rather than numbers floating in abstraction space is true and highly underrated
Reply to @MLaGrangienne, 16:22 UTC · 2 likes
L'amour est cerise... 🍒🏳️⚧️ (@MLaGrangienne): Post-Keynesian econ and heterodox econ in general is good and interesting but MMT is the toxoplasma version of it.
Tbc I have no expertise here I’m just running on vibes
Tweet, 16:30 UTC · 211 likes
I’m a missing sock truther
Reply, 16:52 UTC · 1 like
Highly load bearing 🦆
Tweet, 17:24 UTC · 7 likes
There’s a hobby lifecycle I’ve experienced that’s like:
1) try out a new thing and it’s fun
2) keep escalating involvement until it becomes a core part of your identity
Thread, continued, 17:24 UTC · 2 likes
3) escalate intensity to justify that identity to yourself and others, and also bc you’ve sunk a bunch of costs, and also bc of extrinsic feedback
4) eventually this harder-core version isn’t actually enjoyable
5) burn out
6) quit
7) maybe start over at some point
Thread, continued, 17:24 UTC · 2 likes
This has happened to me with both camping/hiking and rock climbing
Reply to @ChanaMessinger, 23:22 UTC · 2 likes
Chana (@ChanaMessinger): Moving to Berkeley! Do I know anyone there?
Doubt it
Reply to @ChanaMessinger, 23:23 UTC · 3 likes
Chana (@ChanaMessinger): Moving to Berkeley! Do I know anyone there?
Congrats 🎉
Reply to @ozyfrantz, 23:26 UTC · 37 likes
ozy brennan 🦙 (@ozyfrantz): me, 19, foolish: why do feminists believe wrong things about easily verifiable facts? probably it's because feminism doesn't care enough about Truth. as an EA I will be honest all the time and people will only believe true things
me, 33, wise: ha. ahaha. ahaha. ahahahahahahahaha
Expand this semi vaguepost into a nonvaguepost!!!!!
Reply to @ChanaMessinger, 23:41 UTC · 3 likes
Chana (@ChanaMessinger): @AaronBergman18 I mean, I'm moving *from* SF, *to* Berkeley, so god mode might be over
Oh rip
Thursday, February 27, 2025
Reply to @frances__lorenz, 03:48 UTC · 3 likes
(The tweet being replied to by @frances__lorenz is no longer available.)
Sounds fine on both of your parts: fine to offer and fine to decline
Tweet, 06:08 UTC · 21 likes
Tried ChatGPT pro for one month, which I endorse as an experiment, but it’s just not worth the extra $180/mo, so downgrading to Plus like a pleb
I’d feel differently if I was really getting a ton out of deep research but eh
Reply to @livgorton, 06:11 UTC · 2 likes
Liv (@livgorton): having lifting / workout friends to then get coffee with would fix me (make me very happy)
I volunteer as tribute
Quote tweet, 06:40 UTC · 5 likes
Competence porn
Quoting christian (@cxgonzalez): jesus christ my parents were on that flight
Tweet, 11:04 UTC · 2 likes
Have we met irl
Reply to @Benthamsbulldog, 11:35 UTC · 4 likes
Bentham's Bulldog🔸 (@Benthamsbulldog): It's just so dumb. Yes, people do not care about shrimp. For most of history, people didn't regard slavery as bad. This doesn't tell us whether they ought to care about shrimp--the argument for which your piece simply refuses to engage with.
Ratio fr
Tweet, 13:44 UTC · 4 likes
I’m anti-vector embeddings
Reply to @peterwildeford, 18:19 UTC · 10 likes
Peter Wildeford🇺🇸🚀 (@peterwildeford): I'm still kinda confused how Claude 3.7 with no thinking at all seems to blow the socks off of other models with reasoning. What is the magic?
(quoting @kimmonismus: With Sonnet 3.7, Anthropic has clearly placed the focus on coding. Other areas were apparently not particularly important to them. Incremental improvements in MMLU and GPQA, but a giant leap in SWE. It is clear that Anthropic wants to position Sonnet as the coding AI that it…)
Mandate of Heaven
Reply to @halcyon_hazel, 22:06 UTC · 2 likes
❤️🔥 hazel ❤️🔥 (@halcyon_hazel): sorry for being weird it turns out i’ve been running on high E and high T (relative to ci$ woman levels) for ??? months
Been there (not in a trans way tho) lol
Reply to @MariGO2thepolls, 22:06 UTC · 3 likes
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): @halcyon_hazel What did you think the abundance agenda was? Vibes essays?
Banger
Tweet, 22:20 UTC · 9 likes
Watching Cursor + new Claude make a like 12-part series of edits in once conversation turn and nailing it on the first try is a pretty jfc moment
Like bro there is a GUY INSIDE MY COMPUTER and he is VERY SMART
Reply to @jeffrsebo, 22:38 UTC · 3 likes
Jeff Sebo (@jeffrsebo): 1/ Over the past two days @rgblong and I spoke at Anthropic, Google DeepMind, and OpenAI about the case for taking AI welfare seriously. Great conversations at all three - thanks to everyone who organized and attended!
Based
Reply to @adrusi, 23:45 UTC · 2 likes
autumn (@adrusi): @AaronBergman18 @halcyon_hazel wait what? cancer? or maybe trt ig but thatd be extreme
Used to be on TRT lol
Friday, February 28, 2025
Tweet, 01:05 UTC · 13 likes
There’s a canonical Intro to Ethics thought experiment that’s like “should you torture a terrorist to save a city” and allegedly the utilitarian answer is supposed to be “yes” but this seems totally backwards…
Thread, continued, 01:05 UTC · 13 likes
The utilitarian is the one who takes seriously the suffering of the terrorist, and if the range of hedonic states are indeed logarithmic in intensity, it’s the utilitarian who might be committed to saying, at the very least, “well it depends how bad the torture is”
Thread, continued, 01:05 UTC · 10 likes
Ofc like most utilitarian thought experiments, the naive-but-implicit first order reasoning here is probably not what IRL utilitarianism actually calls for. But nonetheless
Reply to @corsaren, 01:24 UTC · 3 likes
corsaren (@corsaren): I think it’s meant to contrast with the deontological stance, which would be that it’s always wrong to torture. Like the utilitarian would say “yes, but only if”, whereas the deontologist would just say “no, never” (first approximation—it’s intro to ethics after all)
Ya fair point
Reply to @panickssery, 01:29 UTC · 8 likes
Arjun Panickssery (@panickssery): You'll never guess what this guy's answer is in the thread to why schools should use age-based rather than ability-based cohorts and classes
The amount of fake reasoning thrown around on these topics is crazy
(quoting @eduleadership: Obviously, this is a barrier to acceleration. But let’s ask the Chesterton’s Fence question: why do schools overwhelmingly use this approach?)
Jfc lol
Reply to @jazzloaf, 01:44 UTC · 2 likes
vivian (@jazzloaf): @AaronBergman18 I think a systematic treatment of utilitarian ethics finds itself in analogy to quantum physics circa 1920s. Not saying that the state of the art will or won't necessarily advance, but the public view is and will continue to be naive for the foreseeable future.
Alarmingly well-written
Reply to @Mjreard, 01:56 UTC · 13 likes
Matt Reardon (@Mjreard): Do we reckon the emissions from every plane carrying life jackets for every passenger have killed 10x as many people as we expect them to save? 100x? More?
Ah but life jackets plausibly fill space that would be filled with denser items like luggage
Tweet, 03:55 UTC · 13 likes
Pleased to announce that I don’t really have a take on 4.5
Reply to @goblinodds, 04:08 UTC · 2 likes
𝖜𝖔𝖒𝖇𝖑𝖊 𝖜𝖎𝖙𝖍𝖔𝖚𝖙 𝖆 𝖈𝖆𝖚𝖘𝖊 (@goblinodds): king shit
👑
Reply to @zeta_globin, 04:13 UTC · 6 likes
(The tweet being replied to by @zeta_globin is no longer available.)
Only speaking for myself, there should be a very high bar for involuntary surgery. Infant circumcision is bad. No strong take on adult circumcision, to each their own
I guess this does imply there’s some age grey area maybe around adolescence. But idk if this is even a thing anyone’s considering
Reply to @JacquesThibs, 04:33 UTC · 6 likes
Jacques (@JacquesThibs): Whether they say it is a frontier model or not isn’t the important part. The part you should care about in that paragraph is that they also removed that GPT-4.5 is “10x” more computationally efficient (compared to GPT-4).
image attached
(quoting @ZeffMax: OpenAI removed language about GPT-4.5 not being a frontier AI model from its white paper. We just updated our article in @TechCrunch on GPT-4.5 with the following note.)
What does it meaaaaan
Tweet, 04:35 UTC · 2 likes
Someone should make an anti-benchmark which is like how what’s the dumbest model you can make that people still choose to use over other, smarter models
Tweet, 04:37 UTC · 6 likes
I should make an API with OpenAI syntax that’s fully functional but the model is literally me
Tweet, 05:28 UTC · 14 likes
POV you told ur cracked friend you thought u did something cool with a computer
Thread, continued, 06:37 UTC · 3 likes
The cracked person in question is @livgorton
Reply to @livgorton, 06:51 UTC · 2 likes
Liv (@livgorton): 🤔 big if true
big
Tweet, 06:53 UTC · 6 likes
Oops
Thread, continued, 06:53 UTC · 5 likes
EA Forum mods are deploying a SWAT team to eliminate the threat (me)
Tweet, 07:25 UTC · 6 likes
I will buy the vibes dip on 4.5
Isn’t it a thing that qualitative abilities don’t track scale/loss nicely except across huge magnitudes?
Tweet, 07:25 UTC · 1 like
Well it finally got too expensive to literally 10x or 100x each full generation like OpenAI could going from gpt-2 to 3 (roughly 100x) and then to 3.5 (2x?) and then to 4 (5x)
Thread, continued, 07:25 UTC · 1 like
We should expect “merely” idk tripling compute between each generation to look a lot bumpier in terms of qualitative capabilities
Quite possibly the 3 -> 3.5 jump was just kinda getting lucky for a 2x jump in terms of riding the bumpy qualitative capabilities landscape
Thread, continued, 07:25 UTC · 2 likes
Some doublings will look highly impressive, others ~flat, and the broad strokes of capability emergence only become apparent across several orders of magnitude (eg 10 generations of doubling to get ~1000x)
Reply to his own earlier tweet, 07:25 UTC · 2 likes
Aaron Bergman, February 28, 2025: Pleased to announce that I don’t really have a take on 4.5
Sorry guys change of plans
Reply to @TetraspaceWest, 08:45 UTC · 9 likes
(The tweet being replied to by @TetraspaceWest is no longer available.)
Obligatory link https://www.aaronbergman.net/p/kids-are-people-too
Tweet, 17:24 UTC · 6 likes
Big if true
Quote tweet, 17:26 UTC · 9 likes
I have no idea, but one kid who everyone thought was kinda dumb apparently sold a landscaping business to private equity for several million dollars (at the age of ~24)
Quoting Paul Novosad (@paulnovosad): Think of the most brilliant, most capable, most getting-it-done person in your high school class.
What are they doing now?
Quote tweet, 17:32 UTC · 21 likes
Fuck I actually like this idea, can someone make it about shrimp or something
Quoting Jack W. (@tweets_r_bad): Out of curiosity I checked the website and they're also selling this. Baudrillard-esque
Reply to @skooookum, 17:34 UTC · 3 likes
(The tweet being replied to by @skooookum is no longer available.)
My GI system is sufficiently fucked that depending on taste and ingredients I just open the capsule and dump the condensate into my mouth
For oil supplements especially, the casing seems pretty hard to digest. Simple cellulose wrappers are probably fine. Check ingredients
Reply to @arithmoquine, 17:55 UTC · 7 likes
henry (@arithmoquine): if i were rich i would simply pay people for things.
want your favorite show to tie up a plot hole? pay them. want a website to look slightly different? pay them. want an artist to change a song? pay em
it seems rich people do not do this nearly as liberally as would be optimal
RIP FTX
Reply to @absurdlymax, 18:38 UTC · 2 likes
Max Alexander (@absurdlymax): @AaronBergman18
I do have a birthday coming up
Tweet, 22:33 UTC · 12 likes
I endorse the OP tweet strongly!!!!!!!
Tweet, 22:54 UTC · 8 likes
I wanna know if LLMs that have been curated to never see texts in which the words “cilantro” and “lime” appear together would correctly infer whether cilantro-lime is a good flavor that many people like
Thread, continued, 22:54 UTC · 2 likes
Reply to @AndyMasley, 23:05 UTC · 2 likes
Andy Masley (@AndyMasley): They said you couldn’t build a whole movement using only people with adhd but we are working magic every day
My guess is that, much like autism, this naturally makes sense as a stereotype when you filtering from a population of high-achievers but when you look at the whole population this effect is either:
- much weaker (I suspect that P(autism | EA) - P(autism) is lower than we’d both intuitively guess)
- or in this case reversed (P(adhd | EA) - P(adhd) < 0 => P(adhd | EA) << 1)
Quote tweet, 23:10 UTC · 2 likes
I think a benchmark made out of questions like this (ie trivial with context/to memorize, possible in principle to reason to very difficult/superhuman) would go hard
Quoting Aaron Bergman 🔍 (@AaronBergman18): I wanna know if LLMs that have been curated to never see texts in which the words “cilantro” and “lime” appear together would correctly infer whether cilantro-lime is a good flavor that many people like
End of February 2025. Previous month: January 2025. Next month: March 2025. Back to the archive index.