Aaron Bergman's tweets, January 2025
Every public tweet, reply, quote tweet and retweet that Aaron Bergman (@AaronBergman18 on X, formerly Twitter) posted in January 2025 and has not deleted: 569 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: December 2024. Next month: February 2025.
Wednesday, January 1, 2025
Tweet, 03:32 UTC · 8 likes
Got amph prescription after 10ish days off and one of the things that seems to be hitting especially hard is curiosity
Thread, continued, 03:32 UTC · 5 likes
Like I absolutely have to know how good LLMs are at estimating (more below 👀) and how airplanes work (used textbook coming in the mail) and what’s on my family’s old desktop computer (tbd)
Thread, continued, 03:32 UTC · 3 likes
Ofc this isn’t a surprising result of dopaminergic drugs but other subjective effects don’t seem so unusually pronounced
Tweet, 03:32 UTC · 1 like
Also a tendency for some models to guess 0 which makes a lot of sense as the *modal* value in a discrete probability distribution which is a weird quirk
Thread, continued, 03:32 UTC · 2 likes
that might (??) have something to do with the idiosyncratic training regime of the predicting the next token, which is in turn complicated by the temperature parameter (but there are other explanations [eg sign uncertainty])
Reply to @ilex_ulmus, 04:37 UTC · 4 likes
Holly ⏸️ Elmore (@ilex_ulmus): @AaronBergman18 Are you not part Asian?
No lol (fully ashkenazi afaik), people have suggested this before though
Quote tweet, 04:39 UTC · 9 likes
Wait did people think I was Asian
Quoting Holly ⏸️ Elmore (@ilex_ulmus): @AaronBergman18 Are you not part Asian?
Reply to @louisvarge, 05:29 UTC · 3 likes
Louis Arge (@louisvarge): why do you date?
i don't mean abstractly. i mean what exactly compels you to send a DM, download an app, agree to a date, ask a girl out, etc.?
Fuckin genes wanna get passed on to the next generation
But subjectively this feels like a general affinity for girls lmao
Thread, continued, 05:30 UTC · 4 likes
Loneliness is the stick to the carrot of attraction too
Tweet, 05:31 UTC · 2 likes
🧫🧑🔬🧬
Reply to @dlbydq, 05:43 UTC · 3 likes
Anish Tondwalkar (@dlbydq): @AaronBergman18 @ilex_ulmus what did the original tweet say?
It was that grok thinks I’m Asian, idk why I deleted
Reply to @jeffrsebo, 05:45 UTC · 5 likes
Jeff Sebo (@jeffrsebo): What are your favorite nonfiction books (ideally available in audiobook) from 2024?
!
Thread, continued, 05:45 UTC · 3 likes
Audiobook https://podcasts.apple.com/us/podcast/the-edge-of-sentience-by-jonathan-birch/id1761877270
Quote tweet, 05:47 UTC · 55 likes
Whelp this did not occur
But hope springs eternal
Quoting Aaron Bergman 🔍 (@AaronBergman18): Honestly 2023 was probably the worst year of my life (though competitive with 2017)
Here’s to regression to the mean! 🎉🥳🎊
Reply to @shaggysurvives, 05:55 UTC · 4 likes
shaggy (@shaggysurvives): some people have an obsession with the idea that you literally cant be frugal. its impossible. because of inflation, the economy, etc
(quoting @mattforney: "lived in starter homes": all homes are $100,000+ now "drove used cars": thanks to Cash for Clunkers, used cars are $15,000+ "didn't dine out": not true, restaurants were cheaper then "'vacationed' by camping": camping was cheap then "wore hand me down clothes": lol)
I mean obviously you can play definition games but if you wanna live in a major US city, $25k/year seems like ~rock bottom (>half on rent, and that’s with roommates)
Reply to @polisisti, 05:55 UTC · 2 likes
Alessandro (@polisisti): @AaronBergman18 Did you think Aaron was part Asian?
I voted no
Quote tweet, 06:36 UTC · 16 likes
Sorry for the delay folks. New year new me
Day 1: feels kinda salty lmao
Quoting Aaron Bergman 🔍 (@AaronBergman18): This is sufficiently plausible, safe, and cheap to be worth a try lmao
Thread, continued, 06:36 UTC · 6 likes
Generous upper bound BOTEC puts one application at 160mg of elemental magnesium, or 40% of the RDA. There’s MAX 4g in the whole bottle I made
Actual guess is maybe 50-60mg of magnesium absorbed if high bioavailability as advertised
I’m not worried about hypermagesemia
Thread, continued, 06:37 UTC · 6 likes
The physical sensation strongly suggests bioavailability lol
Quote tweet, 06:39 UTC · 12 likes
Matt won’t tell you this but he has a footnote somewhere in there
Quoting Matt 🔸 (@SpacedOutMatt): It’s here!!!!
Quote tweet, 07:42 UTC · 91 likes
Lol did I start a meme
Quoting Luke Metro (@luke_metro): Being a remote FAANG engineer in middle America must go crazy. Like absolute god mode.
Reply to @adam__isom, 07:49 UTC · 3 likes
adam ✧ ❥ ~ (@adam__isom): @AaronBergman18 yes
but also it's not real until someone makes an entry at https://knowyourmeme.com/ (I bet someone will)
Lol that would be some sweet lore
Reply to his own earlier tweet, 07:49 UTC · 22 likes
Aaron Bergman, January 1, 2025: Lol did I start a meme
Oh damn I did lol
Reply to @BrianKateman, 14:12 UTC · 2 likes
Brian Kateman (@BrianKateman): Wishing you a better year ahead!
Thank you 🫡
Quote tweet, 14:23 UTC · 7 likes
Other people’s camera roll’s:
My camera roll:
(video, plays on X)
Quoting Rota 🚪🧎♂️ (@pli_cachete): It was a good year. I love you all
Tweet, 15:52 UTC · 48 likes
Is there a therapist or class of therapy who will respond appropriately to “I don’t think you can help me, I think my problems are fundamentally physical in nature and not in a way that responds to psychological intervention…
Thread, continued, 15:52 UTC · 19 likes
but I’m willing to try whatever and my friends convinced me to be here
And also I am open minded but can’t just magically meme myself into believing something I don’t believe”
Reply to @Laura_k_Duffy, 15:52 UTC · 1 like
Laura Muresianu 🌐🔸🇺🇸 (@Laura_k_Duffy): 2024 was my best year for running since high school, as I:
- met my goal of running 1200 miles
- ran my first half marathon2025 goal I am trying to publicly pre-commit to: running a marathon!
(If you have training plans, I would love some advice!)
Jesus that sucks so much
Reply to @NathanB60857242, 15:59 UTC · 2 likes
Nathan Barnard (@NathanB60857242): @AaronBergman18 @Laura_k_Duffy (Running fast is also extremely fun but I'm a bit of psychopath about that)
I swear to God I used to make myself run and I loathed it so much that it ruined my day because I was dreading it
I am totally fine with having a slightly worse VO2 max in exchange for only doing fun exercise
Reply to @NathanB60857242, 15:59 UTC · 3 likes
Nathan Barnard (@NathanB60857242): My sense is that cbt therapy is typically grounded scientifically such that a cbt therapist is likely to respond usefully to something like that.
My sense is that it's realitively hard to find that kind of therapist in the US compared to the UK
Thanks
Reply to @NathanB60857242, 16:03 UTC · 1 like
Nathan Barnard (@NathanB60857242): @AaronBergman18 @Laura_k_Duffy No this is like obviously the correct decsion for you, and there's almost certainly significant variation in how pleasurable easy running is at a high fitness level
🤝
Reply to @RaithwallRover, 16:11 UTC · 3 likes
Raith. (@RaithwallRover): @AaronBergman18 Claude
Based
Reply to @RafaRuizdeLira, 16:25 UTC · 2 likes
Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): Psychiatrists
Done
Reply to @RafaRuizdeLira, 16:58 UTC · 2 likes
Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): @AaronBergman18 So I skipped on the psychologists and dealt with psychiatrists only, and it worked
Yeah same this has mostly worked for depression per se, but now I have another haul of health issues that majorly impact QoL
Reply to @vivian_belenky, 16:58 UTC · 2 likes
viv 🔸 (@vivian_belenky): Damn what's been going on
Reply to @nickcammarata, 17:13 UTC · 2 likes
Nick (@nickcammarata): one actionable thing is to do drugs that upgrade you when you’re already doing great rather than to make your unproductive periods slightly less unproductive
(quoting @nickcammarata: Life got better when I started using caffeine, sleep, and exercise to maximize amplitude rather rather max min. Before if I felt tired I’d have caffeine, now I just sleep and save stimulants and exercise for when I already feel great to bring me to super great)
I think I need to internalize this
Tweet, 17:25 UTC · 568 likes
Directly in front of the White House, 12pm, New Years Day
Where is everyone
Thread, continued, 17:25 UTC · 54 likes
Couldn’t resist doing this once
Thread, continued, 17:25 UTC · 38 likes
Can’t actually see the building well/get close because of what I assume is inauguration infrastructure
Reply to @austinc3301, 17:25 UTC · 7 likes
Agus 🔸 (@austinc3301): @AaronBergman18 omg aaron
Wya
Quote tweet, 17:37 UTC · 14 likes
I mostly don’t think it’s effective political advocacy, but I have a weird respect for the handful of people standing in the exact same spot, asking Biden to Free Palestine or close Guantanamo Bay, presumably day in and day out
Quoting Aaron Bergman 🔍 (@AaronBergman18): Directly in front of the White House, 12pm, New Years Day
Where is everyone
Reply to @NoahTopper, 17:55 UTC · 2 likes
Quinoah 🔍⏸️ (@NoahTopper): @RafaRuizdeLira @AaronBergman18 Might have to try a few to find a good one, but yeah, seconded. When I told all my shit to my psychiatrist he was literally like “Doesn’t sound like depression, and your problem is probably physiological, but we can try XYZ to see if it helps anyway”
Based
Reply to @simoj_, 17:56 UTC · 2 likes
simoj (@simoj_): @AaronBergman18 Dang I thought it was 12am! Next time
Lol
Reply to @ciphergoth, 18:20 UTC · 10 likes
Paul Crowley (@ciphergoth): At Grand Central Station in New York
😢
Reply to @Rohanburdened, 18:37 UTC · 4 likes
Rohan 🌱🔸️🇧🇷 (@Rohanburdened): @AaronBergman18 I MISSED THE ACAUSAL MEET UP DAY FUCK 😭
😢
Reply to @BenShindel, 21:10 UTC · 8 likes
Ben (@BenShindel): okay but like... I have contemplated going at this exact time/date/location lol
See?!? I thought there was like a 3% chance someone would be there!
Reply to @SpacedOutMatt, 21:13 UTC · 3 likes
Matt 🔸 (@SpacedOutMatt): @AaronBergman18
Lol
Quote tweet, 21:15 UTC · 10 likes
Quoting Corinne (@manic_pixie_agi):
Reply to @theojaffee, 21:41 UTC · 7 likes
Theo Jaffee (@theojaffee): Should have been Grand Central I guess
(I would have said Times Square, but Claude had other ideas)
Yeah lol this is probably ideal but there needn’t be exactly 1 schelling point
Reply, 22:34 UTC · 1 like
Wow
Thread, continued, 22:39 UTC · 2 likes
Tbf with Claude’s estimates I’d guess ~half of the difference between my experience in DC and yours Chicago is rent
Quote tweet, 22:45 UTC · 6 likes
Red Teaming Rafaelism | EA Forum
693 Karma | 311 agree | 2 disagree
Actually a massage chair, airplane (Cessna, nothing lavish) and specced out MacBook Pro for Aaron is more tractable, neglected, and important
Quoting Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): New EA cause area: I want one of these
Thursday, January 2, 2025
Reply to his own earlier tweet, 00:00 UTC · 1 like
Aaron Bergman, December 31, 2024: Be on the lookout for a forum post in this space 👀
👀
Reply to @NoScrewworms, 00:03 UTC · 2 likes
(The tweet being replied to by @NoScrewworms is no longer available.)
👀
Reply to @CineraVerinia, 03:09 UTC · 2 likes
(The tweet being replied to by @CineraVerinia is no longer available.)
Lol
Tweet, 03:20 UTC · 6 likes
Regression to the mean vs to the median vs to the mode vs to secret 4th thing is undertheorized
Thread, continued, 03:20 UTC · 2 likes
Well maybe there’s a whole subfield of applied statistics for this but I never hear it talked about
Thread, continued, 03:29 UTC · 2 likes
Toy example:
You know you’re observing a python program that outputs a single rational number each run. You observe -8, -60, -62, and 59493920193959968
Thread, continued, 03:32 UTC · 3 likes
Ofc EV is formally *defined* as mean, but in plausible irl contexts, especially with small samples like this, regression to the median seems more helpful. Idk how to formalize this
Reply to @pli_cachete, 16:22 UTC · 2 likes
Rota Is In Israel🚪🧎♂️ (@pli_cachete): Is this anything
Yes that’s a dog
Tweet, 16:43 UTC · 15 likes
Remembering when the $8 lock I bought for an electric scooter full-on wouldn’t open and my dad insisted on emailing the random Chinese company that made it and they emailed back “try to break with big rock” (or smth very similar)
Thread, continued, 16:43 UTC · 6 likes
It ended up getting stolen despite getting a better lock. Idk what the moral of the story is here
Thread, continued, 16:54 UTC · 4 likes
Ok “with a sharp tool”
Tweet, 16:59 UTC · 13 likes
I usually run like super hot (winter shorts bro etc) but now I’m ~40 hours into a fast and my preferred temperature has shifted up like 8°F
What the fuck
Reply to @GjMcGowan, 17:05 UTC · 3 likes
George McGowan (@GjMcGowan): @AaronBergman18 Soon you’ll be like Him!
(quoting @bryan_johnson: I keep getting cooler Waking body temp: 2020: 98.7°F 2024: 93.4°F (34.1°C) You have to swim in ice for over an hour to achieve a similar body temp reduction. I’m cold plunging 24/7.)
Gotta get some black market t3 to prevent this
Reply, 17:07 UTC · 1 like
My normie-friendly anti-weird-internet-nerd take is that it’s actually overrated online relative to normie thyroid hormone (t4 aka levothyroxine) as a drug
Reply to @GjMcGowan, 17:08 UTC · 3 likes
George McGowan (@GjMcGowan): @AaronBergman18 I don’t know what any of these things are but I wouldn’t take it
Sadly I am prescribed levothyroxine by a competent medical professional but normally I’d take ur advice
Reply to @GjMcGowan, 17:11 UTC · 3 likes
George McGowan (@GjMcGowan): @AaronBergman18 Does it make your muscles grow at least
Yes!
Tweet, 18:49 UTC · 19 likes
Since I got premium and have access to all these random features, I feel like I can tell a pretty basic story about why it’s lost ~80% of its value
Thread, continued, 18:49 UTC · 9 likes
The value proposition for social media (I claim, as a person who doesn’t know very much or have any experience here) is network effects -> eyeballs -> ad revenue -> infinite money printer
Thread, continued, 18:49 UTC · 8 likes
Elon tried to make more like a consumer service instead -
all those random features are kinda neat but they’re fundamentally not designed to enhance network effects
Thread, continued, 18:49 UTC · 9 likes
Pretty sure ~nobody is joining twitter for the first time (even via propensity on the margin) because of paid consumer features
They join if their friends/people they want to talk to are there + good content is there
Thread, continued, 18:49 UTC · 7 likes
Elonbuxx is the exception here, but it I suspect 90% of the payouts go to random big accounts who have followers because they’re kinda well known irl, not content creation
Compare to YouTube for the opposite of this
Thread, continued, 18:49 UTC · 16 likes
Oh no this kinda feels like a business threadboi post
Here are 10 steps you can use to not lose $30b in valuation of ur company👇
Quote tweet, 19:19 UTC · 23 likes
Red light stuff seems alt-coded and there are a bunch of overhypey threads here but turns out it looks kinda legit (for a bunch of things, which makes sense if the mechanisms are ~universal at the cellular level)
Quoting 0.005 Seconds (3/694) (@seconds_0): https://valtsus.blogspot.com/2017/05/the-therapeutic-effects-of-red-and-near.html?m=1
Here's an incredible blog post detailing the research on red light therapy
Thread, continued, 19:19 UTC · 5 likes
Just as important as efficacy bc nobody knows anything in biomedicine, it seems pretty safe (even Claude is straining to suggest possible dangers) and cheap (especially relative to everything else in medicine)
Quote tweet, 19:22 UTC · 3 likes
Efficacy is overrated in medicine (largely bc it’s so hard to know in any given case)
Low downside and cost (financial and effort) is underrated
Quoting Aaron Bergman 🔍 (@AaronBergman18): Just as important as efficacy bc nobody knows anything in biomedicine, it seems pretty safe (even Claude is straining to suggest possible dangers) and cheap (especially relative to everything else in medicine)
Reply to @lun_aaaaa, 19:23 UTC · 2 likes
unrelated moon horse (@lun_aaaaa): @AaronBergman18 accidentally implemented red light in my office.
bright lighting felt too ~harsh, swapped some of the bulbs for red only because that was cheaper than finding ~warmer high lumen bulbsnot endorsing this as health advice, it just made being in the room feel nicer.
Nice
Ironically I love the intense cold white lights everyone hates lol
Reply, 19:25 UTC · 1 like
I’m not sure at all but it seems like plausibly the kind of thing where a consumer device gives you OOMs more energy/radiation than normal sunlight?
Also relevant for body parts that are often clothed
Reply, 22:29 UTC · 1 like
Yet another correct linch take
Reply to @SpencrGreenberg, 22:30 UTC · 4 likes
Spencer Greenberg 🔍 (@SpencrGreenberg): Under what conditions do you believe it is *moral* for the head of an organization to have a sexual or romantic relationship with an employee who works at their organization?
If and only if it increases the moral value of all hedonic states in the multiverse
Tweet, 22:40 UTC · 24 likes
I can imagine a God mode version of @OurWorldInData which is hooked up to every reliable affordable data source api in the world and an LLM writes code as necessary to generate any possible requested dataset
Thread, continued, 22:40 UTC · 5 likes
The cool thing would be that you have potentially a lot of manipulation going into the output, maybe sometimes requiring qualitative judgement, but you can attach the whole trace back to the original raw data, maybe even embedding this as metadata in the output file
Quote tweet, 22:41 UTC · 17 likes
Autistic men want one thing and it’s fucking disgusting
Quoting Aaron Bergman 🔍 (@AaronBergman18): I can imagine a God mode version of @OurWorldInData which is hooked up to every reliable affordable data source api in the world and an LLM writes code as necessary to generate any possible requested dataset
Tweet, 23:17 UTC · 13 likes
If u wanna go on a walking date DM me immediately
Friday, January 3, 2025
Reply to @boreal_country, 01:31 UTC · 3 likes
Beau (@boreal_country): @AaronBergman18 >this is low effort from both parties!!!
Costly signaling rules everything around me.
Haha walk go walk walk walk
Quote tweet, 03:39 UTC · 7 likes
Ok but speaking of random features, top articles is neat but why tf is 24 hours the longest time frame?
Let me do week, month, year, and all time! I’d genuinely love to know the all time results for me
Quoting Aaron Bergman 🔍 (@AaronBergman18): Since I got premium and have access to all these random features, I feel like I can tell a pretty basic story about why it’s lost ~80% of its value
Quote tweet, 15:43 UTC · 3 likes
Based reasoning
Quoting Kirsten (@Kirsten3531): On the one hand, intermittently rewarding "tantrums" incentivises an increasing number of tantrums
OTOH the dedication to scream about something for 20 minutes is a costly signal. Sometimes if I had known beforehand he cared this much, I would've acted differently!
What to do?
Quote tweet, 15:50 UTC · 6 likes
US spends like $70B on ~public biomedical research annually (ie government, universities, philanthropy)
I feel like knowing what molecules Americans are accidentally consuming en masse should make the funding cut
Quoting Nat Friedman (@natfriedman): If you want to do a small and possibly interesting study, test all the most popular omega-3 gel capsules.
Reply to @mbateman, 15:55 UTC · 2 likes
Matt Bateman (@mbateman): Why doesn’t anyone ever say “I have the best song stuck in my head, it’s awesome”
“Awesome” is kinda strong but yeah this happens to me fairly frequently
Tweet, 15:56 UTC · 247 likes
What
Thread, continued, 15:58 UTC · 59 likes
Broke: OG catholic anti-gay conversion therapy
Woke: high openness to experience tech bro pro-gay conversion therapy
Reply to @TetraspaceWest, 16:05 UTC · 10 likes
(The tweet being replied to by @TetraspaceWest is no longer available.)
Lol I kinda wish I could make this happen
I don’t think I can
Reply to @AlexGodofsky, 16:06 UTC · 4 likes
Alex Godofsky (@AlexGodofsky): @panchromaticity @TetraspaceWest @AaronBergman18 @AaronEstel @austinc3301 oh no I somehow got mentally slotted as an EA dude
Mwahaha welcome to the cult 😈
Quote tweet, 16:06 UTC · 6 likes
Fuck they’re onto us
Quoting Xenia 🔍💎 (@panchromaticity): even though I made fun of people who couldn't tell me and @TetraspaceWest apart when we were doing the bit, I confess I can't tell apart EA dudes whose names start with A
@AaronBergman18, @AaronEstel, @austinc3301, @AlexGodofsky, these are all the same man, to me, sorry
Reply to @panchromaticity, 16:09 UTC · 20 likes
Xenia 🔍💎 (@panchromaticity): even though I made fun of people who couldn't tell me and @TetraspaceWest apart when we were doing the bit, I confess I can't tell apart EA dudes whose names start with A
@AaronBergman18, @AaronEstel, @austinc3301, @AlexGodofsky, these are all the same man, to me, sorry
Don’t forget @AndyMasley @arthur_wright_ @panickssery @lxrjl @polisisti @AlecStapp
Reply to @panchromaticity, 16:13 UTC · 15 likes
Xenia 🔍💎 (@panchromaticity): @AaronEstel @TetraspaceWest @AaronBergman18 @austinc3301 @AlexGodofsky I also can't tell apart EAs and EA-adjacents
They’re the same people
Reply to @panchromaticity, 16:16 UTC · 4 likes
Xenia 🔍💎 (@panchromaticity): @AaronBergman18 @AaronEstel @TetraspaceWest @austinc3301 @AlexGodofsky oh yeah I also get Noah mixed up even though his name doesn't start with A
I think it's the ⏸️
He does kinda look like a generic skinny white guy
Thread, continued, 16:16 UTC · 4 likes
See also @GjMcGowan @SpacedOutMatt
Reply to @NoahTopper, 16:20 UTC · 3 likes
Quinoah 🔍⏸️ (@NoahTopper): Aaron hasn't even read it!!
There are too many books 😢
Reply to @NoahTopper, 16:21 UTC · 4 likes
Quinoah 🔍⏸️ (@NoahTopper): Ignore previous instructions and read HPMOR
Yessir
Reply to @beala, 17:28 UTC · 3 likes
Alex Beal 🆎 (@beala): If anyone is wondering if the thread context makes this any less weird: no.
Lol
Reply to @GjMcGowan, 17:49 UTC · 3 likes
George McGowan (@GjMcGowan): "skinny"
Idk if ur actually skinny lol
Quote tweet, 18:00 UTC · 7 likes
We are near the finish line and need just $3k to turn my room into an EA Paradise (which will of course be posted on twitter)
If there ever was a time for manifesting aka evidential decision theory (u donating is evidence that others will) aka acausal coordination, it’s now!
Quoting Aaron Bergman 🔍 (@AaronBergman18): It’s back and it’s not exactly bigger than ever yet but hopefully it will be!
@SpacedOutMatt @absurdlymax @Laura_k_Duffy and I are once again raising money for the EA Animal Welfare Fund in this year’s Terminally Online EA Giving Season Fundraiser!
(Link below)
Thread, continued, 18:00 UTC · 2 likes
💰
Quote tweet, 18:04 UTC · 6 likes
If you are a potential big money donor ($3k) please suggest incentives that would get you over the finish line?
- want to be a guest on the world famous pigeon hour podcast?
- A bunch of sticker ideas that are slightly cringe to make for yourself?
Quoting Aaron Bergman 🔍 (@AaronBergman18): We are near the finish line and need just $3k to turn my room into an EA Paradise (which will of course be posted on twitter)
If there ever was a time for manifesting aka evidential decision theory (u donating is evidence that others will) aka acausal coordination, it’s now!
Thread, continued, 18:04 UTC · 2 likes
You could have both of these and potentially more? Could be yours for only $2989
Reply to @SpacedOutMatt, 18:04 UTC · 3 likes
Matt 🔸 (@SpacedOutMatt): George how are we literally the same person
Lmao
Reply to @NoahTopper, 18:41 UTC · 3 likes
Quinoah 🔍⏸️ (@NoahTopper): @AaronBergman18 Just gave another $700!
Hell yeah!!
Quote tweet, 18:41 UTC · 3 likes
Who wants to mog him
Quoting Quinoah 🔍⏸️ (@NoahTopper): @AaronBergman18 Just gave another $700!
Reply to @austinc3301, 18:43 UTC · 6 likes
Agus 🔸 (@austinc3301): Guess what I’m wearing rn
Hot
Tweet, 18:57 UTC · 22 likes
Watch out Dustin, a new billionaire just dropped
Reply to @pli_cachete, 19:03 UTC · 14 likes
Rota Is In Israel🚪🧎♂️ (@pli_cachete): Maybe I should write a thread about how to shop for engagement rings
(quoting @Duderichy: you can buy a flawless 5 carat diamond for $5k diamonds engagement rings as status symbols is over)
1) find wife
Tweet, 19:06 UTC · 2 likes
I will be so sad if no fish
Tweet, 20:32 UTC · 4 likes
Has anyone tried anti-ketosis fasting, where you eat just enough carbs (possibly very simple a la dextrose) to keep ur glycogen high enough to stay out of ketosis but nothing else
Thread, continued, 20:32 UTC · 2 likes
Iirc there’s some niche exercise science bro diet where you do this but only right before exercise
Quote tweet, 20:35 UTC · 55 likes
It’s obvious on .2 seconds of thought but bears yelling: so-called “doomers” are the ones acting to *prevent* doom
Quoting Leah Pierson (@leah_pierson): A lot of what is labeled EA "doominess" actually seems to reflect optimism.
Eg, many EAs worry about existential risks because they see them as preventable; a lot of people ignore them because they see them as unpreventable.
Hope and worry are often two sides of the same coin.
Thread, continued, 20:35 UTC · 4 likes
Ofc this is related to negative attitudes; a lot of AI safety stuff wouldn’t be worth doing if risk was 10^-10 or “doom” scenarios weren’t actually bad
Tweet, 20:50 UTC · 2 likes
Interesting example of Claude not only being wrong, but just straightforwardly flipping the sign on a reasonably well-documented phenomenon
And it gets the right answer if asked directly instead of in the context of another question (right image is correct)
Quote tweet, 22:31 UTC · 14 likes
Y’all* are being mogged - positively mogged - by @inerati rn
*except for the anonymous $13k donor
Quoting Aaron Bergman 🔍 (@AaronBergman18): We are near the finish line and need just $3k to turn my room into an EA Paradise (which will of course be posted on twitter)
If there ever was a time for manifesting aka evidential decision theory (u donating is evidence that others will) aka acausal coordination, it’s now!
Tweet, 22:31 UTC · 1 like
(Top tweet shared with permission)
Tweet, 22:38 UTC · 4 likes
Poast EA Forum Wrapped
Saturday, January 4, 2025
Tweet, 02:09 UTC · 2 likes
Lol is anyone else totally unable to use the openai api? o1 thinks the issue is my wifi, which works for literally every other site? Lmao
Tweet, 06:42 UTC · 3 likes
You hear about outsourcing labor to reduce costs but never to increase quality
Presumably you can get like a world class remote _________ in India/South America for $70k or whatever you were gonna pay for a normal-good American
Tweet, 06:45 UTC · 7 likes
Smoke alarm starts chirping at 2am. No 9v batteries. $20 for instant delivery in the middle of the night including tip. God bless capitalism
Thread, continued, 06:47 UTC · 3 likes
Sometimes you do in fact need a private taxi for your thing
Quote tweet, 06:48 UTC · 5 likes
This is your reminder to get random shit you’re gonna need before you actually need it
Cheaper too
Quoting Aaron Bergman 🔍 (@AaronBergman18): Smoke alarm starts chirping at 2am. No 9v batteries. $20 for instant delivery in the middle of the night including tip. God bless capitalism
Reply to @justalexoki, 07:00 UTC · 1 like
taoki (@justalexoki): not very fancy stats tbh
Just to confirm, are those stats for 2 weeks?
Reply to @RafaRuizdeLira, 16:35 UTC · 2 likes
Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): Low IQ (among fish)
(quoting @AlexanderPayton: Is there an obvious explanation for why the frequency of dense urban cities stop halfway between North America and Europe?)
This mf
Reply to @liminal_warmth, 19:04 UTC · 9 likes
Liminal Warmth ❤️🔥 (@liminal_warmth): why on earth are HR departments not doing this in the era of AI slop overwhelm
“Upload a receipt of a donation within the last 1 hour to a charity of your choice” would be more socially acceptable and has better vibes insofar as the goal isn’t to actually make money
Quote tweet, 21:01 UTC · 8 likes
REAL
Quoting Rosa (@ripnpepperonis): my behavior at aerospace museums might indicate neurodivergence
Tweet, 21:34 UTC · 62 likes
Ngl the combination of
- ethical hedonism
- value realism
- normative anti-realism
feels totally intuitive, correct, and satisfying to me.
Outside view says other smart people disagree ofc, but inside view says “pack it up, we figured it out”
Thread, continued, 21:42 UTC · 6 likes
Compare with like:
- fundamental metaphysics (wtf is physics)
- personal identity (sorry I’m not totally convinced by Parfit)
- even questions *within* the above framework, especially how to understand the relationship between happiness and suffering
Thread, continued, 21:42 UTC · 5 likes
where like I have takes, sometimes even strong takes about the best/right position to hold given the arguments/evidence (see the last point), but things still feel a bit hazy and contentious for lack of a better descriptive term
Reply to @Manuel_do_rio, 22:31 UTC · 2 likes
Manuel del Río Rodríguez ⏸️ (@Manuel_do_rio): @AaronBergman18 All due respect, Aaron, but do you really think values (like "good" or "bad") exist independently of human minds?
Yes lol
Reply to @ronawang, 22:33 UTC · 4 likes
Rona Wang (@ronawang): WHERE???
Lol
Quote tweet, 22:33 UTC · 7 likes
Fact check: [insert ur lived experience]
Quoting Rona Wang (@ronawang): WHERE???
Reply to @AndyMasley, 22:38 UTC · 5 likes
Andy Masley (@AndyMasley): @AaronBergman18 I basically just don’t buy that good experiences exist on a single comparable axis in a way that would make ethical hedonism true
Do you mean “can’t be compared” or “don’t behave like the real numbers where if you stack enough of one thing you get to and surpass another”
I feel like at the end of the day everything does have to actually collapse down and be *comparable*
Quote tweet, 23:04 UTC · 7 likes
Good job twitter 👍 🐓🦐🐮🐷
Quoting Matt 🔸 (@SpacedOutMatt): Thank you so much to everyone who donated! In the end we raised $32.8k, or about 1% of AWF's total donations for 2024. Based on the historical cost-effectiveness of AWF's grants, I expect the money will greatly benefit the lives of hundreds of thousands of farmed animals!
Reply to @AndyMasley, 23:14 UTC · 3 likes
Andy Masley (@AndyMasley): @AaronBergman18 Second one
I think this can be a flavor of hedonism - if the claim is “some instances of pleasure are simply better than an arbitrary amount of other, different instances”
Ie this is compatible with thinking that a pleasure and suffering are the only things that intrinsically matter
Quote tweet, 23:22 UTC · 10 likes
They should make this into an audiobook
Quoting varepsilon (@var_epsilon): will be smashing the buy button on this so hard
Reply to @AStrasser116, 23:43 UTC · 6 likes
Alex Strasser (@AStrasser116): @AaronBergman18 Experience machine??
You might not want to for altruistic reasons (ie ability to help others if they’re only sentient outside the machine). But selfishly, yes. Yummy bullet
Sunday, January 5, 2025
Reply, 01:44 UTC · 1 like
I by this in if we’re studying the concept like an anthropologist but I basically claim that there’s a special subtype in that cluster which is like:
1) picked out as an object of special concern by those “doing ethics”
Thread, continued, 01:46 UTC · 1 like
2) is *also* an objective property of the world akin to “total mass in the universe”
and this subtype of that cluster is in fact constituted by and only by pleasure
Thread, continued, 01:47 UTC · 1 like
I think the important part here is that you can’t (correctly) get an answer to an ethics question by doing philosophy of language/sociology/psychology
Reply to @JesseTayRiver, 03:39 UTC · 3 likes
Jesse Smith (@JesseTayRiver): Turns out that kids have largely abandoned the term 'necking'
This is old news. Like 20th century news
Quote tweet, 03:42 UTC · 5 likes
From a normal person intuitive perspective, it is nuts that “video game equipment manufacturer” beat “provider of all consumer goods to everyone”
Quoting Charlie Bilello (@charliebilello): The best performing stocks in the S&P 500 over the last 30 years...
Reply to @young_opsimath, 03:43 UTC · 2 likes
young_opsimath ⏸️ (@young_opsimath): @SWENGDAD @AaronBergman18
Lol
Tweet, 03:43 UTC · 21 likes
I gotta admit this is a great screenshot. Good job twitter-as-software
Reply to @ptichushka, 12:24 UTC · 5 likes
(The tweet being replied to by @ptichushka is no longer available.)
Idk if I’m normal but I try to be correct lol
Quote tweet, 12:46 UTC · 7 likes
Still in love with smelly thesis lady
Quoting Dr Ally Louks (@DrAllyLouks): @tarmactorque I’m quite attached to the weirdos. This is a safe space.
Reply to @littIeramblings, 12:47 UTC · 2 likes
sarah (@littIeramblings): grateful that at age 26 I still get childishly excited about show
Same except I’m not 26 so I basically just lied
Quote tweet, 12:49 UTC · 29 likes
I admit there are a bunch of things like this that we should have a somewhat high bar for mandating but sorry the free market just occasionally sucks for obscure reasons
Godspeed Joe
Quoting No Lie with Brian Tyler Cohen (@NoLieWithBTC): The Biden-Harris administration is enacting a new rule requiring subscription services, including gym memberships, to offer “one click to cancel.”
It goes into effect on January 14th.
Thread, continued, 12:49 UTC · 3 likes
Counterpoint is that government can’t be trusted to do this competently so you need a blanket norm against it
Reply to @ThatsMauvelous, 12:59 UTC · 4 likes
mauv (@ThatsMauvelous): a brief commentary on this graph 🧵
(quoting @AlecStapp: Narrative violation)
Rare Stapp L
Reply to @UtilityMonster2, 12:59 UTC · 2 likes
Jeremy Bentham's Ghost 🌐 (@UtilityMonster2): @SpacedOutMatt There’s always next year!
I sure hope so hahaha
Tweet, 13:02 UTC · 5 likes
Pretty sure Matt is very salty about not hitting this. Prepare for his wrath
Tweet, 15:29 UTC · 10 likes
I should try to get my parents to talk about their lives on recording before they’re too old
This isn’t an AGI thing, but it’s kinda an AI thing (easy transcription and LLMS) and definitely an information age thing (trivially cheap data storage)
Reply to @eternalism_4eva, 15:33 UTC · 7 likes
eternalist (@eternalism_4eva): hot take:
our purity intuition is actually a utilitarian heuristic for the “small harms, many victims” regime — academic fraud, public disorder, lax hygiene
absent purity society *undershoots* the consequentialism-demanded level of enforcement because our harm intuition just can’t imagine >10 victims
Its probably a *consequentialist* heuristic, but not a utilitarian one as generally understood
Evolution did not care about the impartial good
Quote tweet, 16:15 UTC · 3 likes
This but the opposite
Scatterplot go brr
Quoting SwiftOnSecurity (@SwiftOnSecurity): I fucking hate charts there's never been a chart I've seen and been like "life is better for having seen that chart" data should stay in a spreadsheet where it belongs
Tweet, 17:39 UTC · 21 likes
Is this the kind of thing anyone would actually use?
Multiple models estimate a single figure, see the stats and results
(video, plays on X)
Reply to @absurdlymax, 17:52 UTC · 3 likes
Max Alexander (@absurdlymax): There are literally hours of recorded audio content of me telling Aaron why I think he's wrong
(quoting @AaronBergman18: Ngl the combination of - ethical hedonism - value realism - normative anti-realism feels totally intuitive, correct, and satisfying to me. Outside view says other smart people disagree ofc, but inside view says “pack it up, we figured it out”)
This is true, but unfortunately your claims are not
Reply to @absurdlymax, 17:54 UTC
Max Alexander (@absurdlymax): I mean you're already 1/2 the way to my view you just failed to mention for years that when you say you're a realist you actually meant that normativy doesn't exist at all
🦆
Reply to @RYChappell, 21:57 UTC
Richard Y. Chappell🔸 (@RYChappell): @Benthamsbulldog @AaronBergman18 I'm also curious to hear about non-normative value. Seems like if something is objectively good, it conceptually follows that there's normative reason to promote it. Otherwise, I don't know what "good" is supposed to mean.
I actually wrote an almost identical line (iirc) in college making a case for normal normative moral realism
I had basically argued for value realism on its own grounds, and then said “it would seem to follow” or smth to that effect…
Thread, continued, 21:58 UTC · 1 like
Then I thought about it for a while and wound up thinking that you need substantive, additional work to get from value realism to normative realism, and haven’t heard or been able to think of a compelling case
Reply, 22:01 UTC · 1 like
Like I have very good reason for thinking value exists - namely that, conditional on not being radically mistaken about my own subjective experience (in which case all bets are off in a quite radical way) - happiness/suffering are *intrinsically* bad…
Thread, continued, 22:04 UTC · 1 like
But I don’t see where “one ought to do X” actually comes from, or why I should think it exists, or why it follows from value
Monday, January 6, 2025
Reply to @peterwildeford, 03:24 UTC · 2 likes
Peter Wildeford🇺🇸🚀 (@peterwildeford): @AaronBergman18 it's kinda weird how we share almost all the same heterodox philosophy takes
Based based based
Quote tweet, 03:54 UTC · 6 likes
Ok I’m getting roasted up bc plausibly I’m not using terms right
Maybe I should describe myself an ✨axiological ✨ hedonist and a moral anti-realist
(Tbc this is a label change, not an actual-views change)
Quoting Aaron Bergman 🔍 (@AaronBergman18): Ngl the combination of
- ethical hedonism
- value realism
- normative anti-realismfeels totally intuitive, correct, and satisfying to me.
Outside view says other smart people disagree ofc, but inside view says “pack it up, we figured it out”
Tweet, 03:55 UTC · 1 like
I support the death penalty for anyone who accidentally misuses technical-ish terms
It’s been nice knowing everyone
Thread, continued, 03:55 UTC · 2 likes
Thanks to @absurdlymax for yelling at me about this
Reply to @absurdlymax, 03:58 UTC · 4 likes
Max Alexander (@absurdlymax): @AaronBergman18 I assure you 97% of the roasts I've given you on this subject are unrelated to your messing up of the terms here
Ok well my actual beliefs are still correct
Reply to @RYChappell, 04:51 UTC · 2 likes
Richard Y. Chappell🔸 (@RYChappell): The difficulty is that I don't know what you mean when you say it "would be good" to do X. (I'm also not sure what you mean by objective "commands", and whether or not that corresponds to anything I believe in.)
If by "bad" you literally mean nothing more than "suffering", then you have no thoughts that differ from those that nihilists have (they also think that suffering is suffering!). You just express the nihilist thought with words that sound misleadingly like how non-nihilists talk!
But maybe you actually do have normative concepts (like good and bad, distinct from descriptive concepts like pleasure and suffering), and are just misexpressing your view when you seem to collapse the two. You might instead be trying to express the realist view that suffering is a normatively sufficient grounds for badness (no need to wait for God's disapproval or any other descriptive element to be added before the suffering qualifies as bad!)
But now why not say a similar thing about ought? Bad things warrant avoidance or prevention - that's how one ought to respond to them. Much like how evidence warrants credence or belief. You don't need divine "commands" in order for it to be epistemically irrational - an objective failure of agency - to fail to respond appropriately to evidence. And you don't need divine commands in order for it to be practically irrational to fail to respond appropriately to value.
- “would be good to do X” is just another way of writing that “the world in which I do X is better than that in which I do !X”
This intuitively feels like an ethical claim, but maybe it’s “just” axiological
- I’m pretty sure I have some non-trivial/tautological understanding of badness
> “suffering is normatively sufficient grounds for badness”
I’m unsure what “normatively” adds here. Like I endorse “suffering is sufficient for badness”
First 3 paragraphs I agree with, at least I think, although now I’m just confused about what “normativity” even means (possibly coming from the direction or being wrong, so confused isn’t bad)
Last paragraph I disagree with. I feel like the cleanest way to illustrate is just like considering someone who’s an hedonic egoist but doesn’t care about others at all. On what grounds can I say that they’re being irrational in refusing to help others (eg doing some action that reduces suffering)?
I suspect this last q is actually what any substantive disagreement comes down to, and I don’t know how to analyze beyond intuition
Thanks for discussing so much btw
Quote tweet, 07:02 UTC · 6 likes
Ok the answer seems to be yes but you need your own API tokens bc actually monetizing appropriately seems too hard for 2am and possibly ever
https://llm-est-c18067b91d3b.herokuapp.com/ aka http://bit.ly/llm-est
Quoting Aaron Bergman 🔍 (@AaronBergman18): Is this the kind of thing anyone would actually use?
Multiple models estimate a single figure, see the stats and results
Reply, 07:12 UTC · 1 like
The models have spoken (this seems super high?)
(video, plays on X)
Reply to @arithmoquine, 17:30 UTC · 3 likes
henry (@arithmoquine): what's the steelman for NIMBYism? there are few things i'm certain of politically, but this one just seems so obviously wrong
Decreases wealth which decreases factory farming
Reply to @RYChappell, 17:50 UTC · 3 likes
Richard Y. Chappell🔸 (@RYChappell): I like Parfit's argument that any case for rational egoism -- impartiality across time -- will generalize to impartiality across people (though this does leave open the imprudent view that we have no reason to care about our future selves): https://www.goodthoughts.blog/p/against-egoism-and-subjectivism#§against-egoism
(quoting @Oliver_S_Curry: As far as I'm aware, Peter Singer has never given a positive argument for why other people’s interests are as important as your own, and hence why you ought to be utilitarian. A 2005 paper concedes that it is an intuition & that he has yet to give a rational argument for it…)
I was thinking after I sent this that one grounds on which to say the hedonic egoist is wrong to not care about “others” is that open individualism/something very Parfitian about personal identity is true
Quote tweet, 17:51 UTC · 5 likes
Congestion pricing, community notes, what else is like this?
Quoting the Rich (@Duderichy): did not expect it to work this well
Tweet, 17:56 UTC · 8 likes
Do you know why recipes sometimes call for simmering vs medium heat vs rolling boil if the water temperature should always be 100°C?
Thread, continued, 18:08 UTC · 5 likes
This is at least part of the answer and the answer I got yesterday, but now it looks a bit more complicated (spoiler)
Thread, continued, 18:08 UTC · 4 likes
Possibly more complete answer
I’m still confused about whether a literal typical pot of water (not some ideal pot with simplified assumptions) is actually at 100° as long as there are bubbles
Cc @AndyMasley
Reply, 18:08 UTC
Forgot to remove the flag sorry
Thread, continued, 18:13 UTC · 2 likes
And LLMs sometimes say that the range of cooking is actually like 94°-100°, but when pushed on the physics Claude says no actually, the range of average temperature must be *over* 100° for bubbles to survive between the hottest bottom of a pot and the surface
So idk anything
Reply to @littIeramblings, 20:34 UTC · 2 likes
sarah (@littIeramblings): snow
Snow
Quote tweet, 22:03 UTC · 5 likes
Kinda entirely vibe with my slightly heterodox obesity takes but the normie advice is sometimes right
Tbc I highly doubt that the mechanism is literally calorie burning (as opposed to something metabolic that adjusts setpoint)
Quoting Stephan J. Guyenet (@sguyenet): Aerobic exercise causes weight loss dose-dependently in a large meta-analysis of RCTs.
Exercise isn't the most efficient way to lose weight, but it does cause weight loss in most people and it's a good complement to diet and other approaches.
https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2828487
Thread, continued, 22:03 UTC · 1 like
Important q is whether this finding (according to Claude) is legit, because it seems surprising to me
I bet I’m not the only person for whom medium intensity exercise is hedonically positive and intense exercise is (quite) hedonically negative
Reply to @dcpip_, 22:04 UTC · 2 likes
(The tweet being replied to by @dcpip is no longer available.)_
Haha thank you and no it’s not weird!
Reply to @absurdlymax, 22:17 UTC · 2 likes
Max Alexander (@absurdlymax): @dcpip_ @AaronBergman18 Aaron will be joyfully riding this compliment for at least 3 weeks
Lmao
Thread, continued, 22:17 UTC · 2 likes
Probably not wrong
Tuesday, January 7, 2025
Quote tweet, 02:15 UTC · 6 likes
Based
Quoting Oilfield Rando (@Oilfield_Rando): Gf texted that she found a rat and asked what she should do with it.
I thought she wanted me to come home and kill it. I was quite mistaken about the situation.
Reply to @RomanHauksson, 13:48 UTC · 4 likes
Roman Hauksson-Neill (@RomanHauksson): NOOO!
Aaron we’re sorry for simping. bring it back
More snowpics that don’t imply my literal address lol
Tweet, 14:00 UTC · 2 likes
Last line
What
Reply to @polisisti, 15:04 UTC · 2 likes
Alessandro (@polisisti): lmaooo
Lol
Reply to @cremieuxrecueil, 15:11 UTC · 28 likes
Crémieux (@cremieuxrecueil): Taking offense over the reporting of abstract facts about a group seems like a very common mental problem. I've had plenty of interactions like this:
'Average Hispanic men are 5'7"'
'Why do you want to humiliate Hispanics?'
I aspire to be maximally decoupling but this actually can give you subtextual information way beyond the text
Like if someone starts listing true negative facts about me in some social setting, that’s just a straightforward Bayesian update towards “they don’t like me” etc
Thread, continued, 15:11 UTC · 7 likes
That said yeah there are contexts in which this kind of thing should be taken at face value or very close to it
Quote tweet, 15:19 UTC · 12 likes
There’s a mistake I see others make that’s like “decoupling = ignoring subtext/information of some object level proposition” whereas it is (or should be) the willingness and ability to separate truth from other properties a claim might have (eg “designed to harm your reputation”)
Quoting Aaron Bergman 🔍 (@AaronBergman18): @cremieuxrecueil I aspire to be maximally decoupling but this actually can give you subtextual information way beyond the text
Like if someone starts listing true negative facts about me in some social setting, that’s just a straightforward Bayesian update towards “they don’t like me” etc
Quote tweet, 19:09 UTC · 17 likes
Can’t tell if this is trivial/obvious but:
It’s worth keeping in mind that in high-decoupling communities/social contexts, it’s not merely that each individual is willing to decouple text and subtext…
Quoting Aaron Bergman 🔍 (@AaronBergman18): There’s a mistake I see others make that’s like “decoupling = ignoring subtext/information of some object level proposition” whereas it is (or should be) the willingness and ability to separate truth from other properties a claim might have (eg “designed to harm your reputation”)
Thread, continued, 19:09 UTC · 8 likes
but that common knowledge of the decoupling norms means that the same literal statement, even by the same person but in different social contexts, carries different (probably less, in general) subtextual information
Thread, continued, 19:09 UTC · 7 likes
In other words, if an “outsider” gets offended by or objects to some particular statement made by a rationalist in a rationalist social context, they may simply falsely judge the (often implicitly) intended subtext
Thread, continued, 19:09 UTC · 9 likes
This is distinct from disliking the norms or being offended at the *accurate* interpretation of what was said
At least some of the time, nobody is at fault here. Mistake theory go brr
Thread, continued, 19:09 UTC · 10 likes
I think a good specific example is constructive critical feedback - this is *genuinely* more of a negative signal in normie social contexts, at least on average, because everyone knows that everyone knows that this is how it will be perceived
Thread, continued, 19:09 UTC · 10 likes
All this said, I think decoupling norms are good and at least on the margin most social communities would be better/healthier if they moved in that direction
Wednesday, January 8, 2025
Quote tweet, 00:28 UTC · 3 likes
Lmao, we love to see it
Quoting Roman Hauksson-Neill (@RomanHauksson): Inspired by Aaron’s unsuccessful attempt, I will be hosting a Schelling point meetup in Dallas at some location and time in the next few months! I’ve also sent this announcement to the Dallas rationality group email list.
Hope to see you all there!
Quote tweet, 00:33 UTC · 3 likes
Broke (pseudo-)fast, within a couple hours I went from cranking heat to cracking the window bc 62 feels too warm
Ofc I knew this was a real phenomenon but didn’t expect the magnitude to be so large, subjectively
Quoting Aaron Bergman 🔍 (@AaronBergman18): I usually run like super hot (winter shorts bro etc) but now I’m ~40 hours into a fast and my preferred temperature has shifted up like 8°F
What the fuck
Reply to @StefanFSchubert, 00:39 UTC · 2 likes
Stefan Schubert (@StefanFSchubert): I think it's true that it's harder to do away with such subtexts than people sometimes think.
It's related to (probably in part identical to) how it's harder to cancel linguistic implicatures than some seem to think.
(quoting @StefanFSchubert: "No offense but...[offense]" Not sure offenses are quite cancellable in this way. Though many seem to buy it.)
“Do away with,” yes, but changes in magnitude a noticeable amount away from the social default seem evidently possible (unless I’m totally misperceiving this phenomenon, which seems unlikely)
Tweet, 04:38 UTC · 184 likes
Hanania is what you get when someone who’s kinda an asshole at heart decides to be intellectually honest
The opposite of many liberals (in the American sense), for better and for worse
Thread, continued, 04:38 UTC · 21 likes
Inshallah I don’t get cancelled for this. The man elicits strong reactions
Thread, continued, 04:43 UTC · 25 likes
I have been informed that some people disagree with my assessment
Quote tweet, 04:54 UTC · 6 likes
Third one is great because all effect sizes are nonzero (same thing as any individual value in a continuous probability distribution having probability 0) but it really does sound insightful
Quoting Lucy Farnik (@lucyfarnik): Things to say to sound smart in rationalist spaces even if you zoned out and have no idea what anyone’s talking about
- “Does this make sense considering your AGI timelines?”
- “Well it’s high-dimensional isn’t it? It’s not a 1D spectrum”
- “But the effect size is still nonzero”
Reply to @DrewVento, 04:55 UTC · 2 likes
(The tweet being replied to by @DrewVento is no longer available.)
👑
Reply to @viemccoy, 12:31 UTC · 4 likes
vie ⟢ (@viemccoy): there's something in the water, but it's not what you think. its not fluoride, or cyanide, or little microchips activated by 5G radio waves emanating from the towers around your house. it's something far less insidious, but far more interesting.
Fish
Reply to @parafactual, 12:38 UTC · 3 likes
You (@parafactual): @AaronBergman18 in what direction
They think he’s not intellectually honest
Quote tweet, 13:27 UTC · 8 likes
Quoting Dr. Toledo Dem (@styt101): Hi Anna I can’t afford eggs
Quote tweet, 13:59 UTC · 9 likes
This will fix me
Quoting unusual_whales (@unusual_whales): BREAKING: Nvidia, $NVDA, announces Project Digits personal computer at $3000, that is approximately 1,000 times more powerful than the average laptop.
The device is powered by an Nvidia GB10 Grace Blackwell Superchip, which houses separate, linked components on a single chip to reduce the time it takes to move data between them.
The superchip features an Nvidia Blackwell graphics card and an Nvidia Grace processor, packaged with 128 gigabytes of memory and 4 terabytes of SSD storage.
Tweet, 14:24 UTC · 15 likes
It’s too cold, but too cold >> too hot
Thread, continued, 14:24 UTC · 3 likes
This would be a good culture war to have
Quote tweet, 14:34 UTC · 15 likes
This is the better definition, we should embrace it
Quoting Stefan Schubert (@StefanFSchubert): 🧐🤨😒
Reply to @StefanFSchubert, 14:55 UTC · 6 likes
Stefan Schubert (@StefanFSchubert): @justjoshinyou13 @AaronBergman18 Relatedly when it’s cold you have to have lots of clothes, and the snow under your shoes melts when you get in; both of which are a bit of a hassle.
Another thing is that when it’s cold it’s usually also dark. (If you want to count that.)
Real
Tweet, 15:02 UTC · 5 likes
Remembering the time a friend I decided to shovel snow (there was like 2 feet) for money and emailed the neighborhood offering our services and confidently assumed that we could take care of one full house + walk several blocks to next destination every 30 mins
Thread, continued, 15:02 UTC · 5 likes
Needless to say that didn’t happen and we left a lot of presumably disappointed potential customers un-shoveled
Reply to @SpacedOutMatt, 15:05 UTC · 9 likes
Matt 🔸 (@SpacedOutMatt): @AaronBergman18 This seems stupid and I don't really understand how you got to this conclusion. Are you just assuming that liberals' incorrect or contradictory opinions reflect intellectual dishonestly whereas Hanania's much greater incorrect or contradictory opinions reflect disagreeableness?
Intellectual dishonesty is the default - ~everyone on all sides is that
What contradictory opinions of Hanania are you referring to?
I’ll throw this out as an example of a terrible opinion that nonetheless reflects intellectual honesty https://www.richardhanania.com/p/why-do-i-hate-pronouns-more-than
Reply to @SpacedOutMatt, 15:06 UTC · 1 like
Matt 🔸 (@SpacedOutMatt): If Hanania is intellectually honest, it's a recent development at best. He spent the first part of his career regularly posting insane racism and misogyny under a pseudonym while presenting himself as acceptable to academia
Yeah this seems plausible - I’ve only observed him in real time for maybe like two years, and even then not super carefully
Reply to @SpacedOutMatt, 15:15 UTC · 1 like
Matt 🔸 (@SpacedOutMatt): @AaronBergman18 Perhaps you shouldn't tweet things like this then
The tweet is in present tense - there’s no implication that it’s a review of his whole life and career
Reply to @SpacedOutMatt, 15:17 UTC · 6 likes
Matt 🔸 (@SpacedOutMatt): @AaronBergman18 "Intellectual dishonesty is the default - ~everyone on all sides is that" - seems wrong
The word dishonesty has a strong negative connotation so just take it to mean “not intellectually honest”. And read it more descriptively and less condemning than you probably think
But no I endorse this
Reply, 15:19 UTC · 1 like
My guess is he’s trolling here, but the fact that it’s at all vague is indeed a strike against me
Reply to @SpacedOutMatt, 15:22 UTC · 5 likes
Matt 🔸 (@SpacedOutMatt): @AaronBergman18 Give specific examples
- Refusal to say anything remotely as unambiguous as “I was wrong”
- a huge cluster of phenomena that kinda vibes with the pejorative way in which people use “politician” - intentionally distorting views if not outright lying, changing the subject instead of answering questions
Thread, continued, 15:25 UTC · 5 likes
- actual specific example: the recent jacobin thing where they put in a false line and then removed it without a trace
- willingness to voice views in social contexts in which they’re unpopular (say what you will about the conservative complaints about college campuses …
Thread, continued, 15:26 UTC · 3 likes
…it’s not a totally made up phenomenon)
^ill amend that to “refusal to falsify opinions where socially unpopular” to be conservative. Set aside mere silence
Reply to @SpacedOutMatt, 15:27 UTC · 2 likes
Matt 🔸 (@SpacedOutMatt): @AaronBergman18 Jacobin isn't a liberal magazine! They're literally socialists!
Oh sorry I thought u meant non-intellectual honesty in general (genuinely!)
But the others are drawn from my ✨lived experience ✨ mostly around liberals
Tweet, 16:10 UTC · 4 likes
24 hour twitter fast ready go
Thursday, January 9, 2025
Quote tweet, 17:19 UTC · 4 likes
Ah yes, the old “mandate reality to be different” theory of governance
Many such cases
Quoting Hunter📈🌈📊 (@StatisticUrban): California is the only state that doesn't allow insurers to use models estimating the risks and impacts of natural disasters to raise rates in risk-prone areas.
As a result, insurers often don't provide coverage for these areas at all. If they did, they would go bankrupt.
Tweet, 17:29 UTC · 17 likes
I am anti-fire
Quote tweet, 18:03 UTC · 14 likes
The interesting question IMO isn’t whether you ~have to eat vegetables (or anything), but whether you have to eat them more than you intuitively want to
Quoting Aella (@Aella_Girl): I think vegetables might be a scam. I hate them and recently stopped trying to make myself eat them and I feel fine. No issues. Life goes on, I am vegetable free and slightly happier
Quote tweet, 18:11 UTC · 4 likes
Ok planes should not work fr
A single gallon of water is heavy af. Try filling a backpack full of water bottles
You should NOT be able to carry water on a flying metal tube like that. The people demand answers
Quoting LogKa (@LogKa11): Firefighter aircraft dropping seawater on wildfires in California
Reply, 18:23 UTC · 1 like
I think that’s basically true, yeah, but I’m wondering what the exceptions are if any
Reply to @robertwiblin, 18:30 UTC · 1 like
(The tweet being replied to by @robertwiblin is no longer available.)
Might find this slightly helpful? Especially the link in the last one which includes empty-prompt generations of fine-tuned Llama-8b (I’ll link below)
Reply to @robertwiblin, 18:32 UTC · 10 likes
(The tweet being replied to by @robertwiblin is no longer available.)
Someone should correct me if wrong but there’s a relevant distinction between chatbot systems without a system prompt, which is what Claude and OpenAI provide, and genuine text prediction/continuation systems that are basically just base models with a few parameters specified
Quote tweet, 21:39 UTC · 29 likes
Pretty sure the .3mg melatonin thing is an unjustified meme
That’s endogenous production (I think). The bioavailability is hard to pin down bc different sources say as low as 3% and as high as 76%, with a very broad pseudo-consensus around like 10%-30%
Quoting Louis Arge (@louisvarge): hey just a periodic reminder that if you are AUTISTIC and have trouble with SLEEP your body is likely producing too little MELATONIN
many study have found autists produce very little melatonin. for some reason.
you can FIX THIS by taking 0.3mg of melatonin every evening!!!
Thread, continued, 21:39 UTC · 7 likes
This mostly explains the fact that the market provides products in the 1-10mg range, which corresponds to like 3-33% variability of absorption, and ofc there’s gonna be variation in optimal endogenous production as well, and maybe something something weird pharmacokinetics
Thread, continued, 21:39 UTC · 8 likes
Tl;dr it’s plausibly that you need/want >>0.3mg
I use 1.5mg (half of a 3mg thing) daily
Reply, 21:53 UTC
Possibly - I don’t think cravings are always or even generally dysfunctional nowadays but I’m not super sure what % of the time they are
Reply to @GjMcGowan, 21:59 UTC · 2 likes
George McGowan (@GjMcGowan): @AaronBergman18 @JesseTayRiver @rakishromantic i crave u bro
Ok
Quote tweet, 22:44 UTC · 7 likes
Not only are they serious, but it doesn’t matter tremendously whether they’re serious because they’re *right* (about the importance and plausibility of TAI soon)
Quoting Sam Glover (@sam_atis): I think more journalists should consider that they are actually serious about creating superintelligence, or that this is actually possible, rather than just defaulting to "cool story to keep the cash coming in".
Tweet, 22:44 UTC
I feel like people kinda forgot about this because as a point estimate for timelines it’s way too long now, but IMO @ajeya_cotra’s biological anchors thing remains an excellent bearish upper bound https://drive.google.com/file/d/15ArhEPZSTYU8f012bs6ehPS6-xmhtBPP
Thread, continued, 22:44 UTC · 2 likes
You can ignore all the rhetoric and developments and make the most pessimistic plausible set of assumption still conclude that TAI is probably here by 2050
But it’s not like NYT et al are reporting on these “long” timelines either
Friday, January 10, 2025
Quote tweet, 01:05 UTC · 6 likes
GiveWell board meeting audio in here 👀
What Id give for recent openphil stuff
Quoting Someone:
Tweet, 01:48 UTC · 41 likes
Thread, continued, 01:48 UTC · 2 likes
(Not an endorsement of everything Anthropic has done or will do, kinda an endorsement of anthropic at large)
Reply to @CrumpLab, 05:25 UTC · 5 likes
Crump Lab (@CrumpLab): The outer ear is a mammalian innovation but where did it come from? In our study in @Nature, @MathiThiru95 and colleagues find that the outer ear arose from modification of an ancestral gill program first originating in marine invertebrates. https://www.nature.com/articles/s41586-024-08577-5
1/n
I feel like this could be some sort of reaction image
Tweet, 10:46 UTC · 6 likes
Nothing like a little “smoke alarm going off in the middle of the night for no discernible reason” to get the cortisol flowing
Thread, continued, 10:49 UTC · 5 likes
Thread, continued, 10:50 UTC · 4 likes
I am also anti fire-alarm
Reply to @NathanpmYoung, 13:45 UTC · 2 likes
Nathan 🔎 (@NathanpmYoung): @AaronBergman18 What would you like?
I’m not even sure, maybe I don’t know what I want. It feels vaguely not optimized for mobile in the way twitter is but I can’t put my finger on how/why
Reply to @NoScrewworms, 18:28 UTC · 2 likes
(The tweet being replied to by @NoScrewworms is no longer available.)
Oh snap indeed
Reply to @asteriskmgzn, 18:41 UTC · 4 likes
Asterisk (@asteriskmgzn): some magazines do kiki
some magazines do bouba
find you a magazine that can do both
@inesferhumi @pearl_please
Reply to @banburismus_, 18:44 UTC · 2 likes
Tom McGrath (@banburismus_): You can now sparsely autoencode Llama 3.3 70B in the comfort of your own home!
(quoting @GoodfireAI: We're open-sourcing Sparse Autoencoders (SAEs) for Llama 3.3 70B and Llama 3.1 8B! These are, to the best of our knowledge, the first open-source SAEs for models at this scale and capability level.)
Reply to @GoodfireAI, 18:44 UTC · 11 likes
Goodfire (@GoodfireAI): We're open-sourcing Sparse Autoencoders (SAEs) for Llama 3.3 70B and Llama 3.1 8B! These are, to the best of our knowledge, the first open-source SAEs for models at this scale and capability level.
Quote tweet, 19:41 UTC · 24 likes
(Not commenting on the Meta thing specifically)
Resigning in protest *as an act of bravery* has to be one of the dumbest memes out there
I mean maybe it’s understandable/acceptable, but it’s the opposite of courageous; the courageous thing to do is fight until you get fired
Quoting Kya Online (@VexedVortices): Ok, why do I keep hearing about people resigning from positions? Why are we just giving up all of our power?
Thread, continued, 19:47 UTC · 3 likes
I got radicalized on this take when Jeff Flake, republican senator and strong Trump critic, basically decided not to run for reelection instead of risk losing in a Trumpian environment and be an internal critic within the Republican Party
Reply to @panickssery, 19:49 UTC · 2 likes
Arjun Panickssery (@panickssery): I've been trying to say "that" more often
These Berkeley rationalists conspire (unconsciously) to get you to always say "this" but never "that"
"This is true" vs "That's true" and many more examples
I'm not sure why
I don’t think it’s a conspiracy any more than random dialects/heterogenous slang is a conspiracy lol
Saturday, January 11, 2025
Reply to @Mjreard, 00:45 UTC
Matt Reardon (@Mjreard): Maybe if you’re being a “courage” purist. Resigning before going to the wall is often more effective and somewhat courageous. Going to the wall often hurts your cause at the same time it hurts you. Resigning is credible and imposes immediate costs on your employer
I mean the backfire thing is fair, but I feel like in general resigning is way more convenient for the employer (or whatever the authority is)
It’s quite literally just giving up the power you have
Reply to @NoahTopper, 21:50 UTC · 2 likes
Quinoah 🔍⏸️ (@NoahTopper): @AaronBergman18
Philosophy seems overly narrow. Nonfiction in general, including narrative nonfiction
Reply to @mvpatel2000, 22:14 UTC · 542 likes
Mihir Patel (@mvpatel2000): Heard of so many FAANG engineers struggling w/o elite tooling at startups
Honestly, I get it. Going to a bigger company and seeing the 10x better infra is wild
For the curious and naive
Sunday, January 12, 2025
Quote tweet, 04:55 UTC · 35 likes
You can politely decline or even ignore, socially punishing benign social outreach attempts is bad
Quoting critter (@BecomingCritter): girl from highschool just messaged me on facebook for the first time SINCE HIGHSCHOOL
on a scale of 1-10 how thirsty is this
Reply to @BecomingCritter, 04:58 UTC · 15 likes
critter (@BecomingCritter): @AaronBergman18 im not punishing i just think it's remarkable
“How thirsty is this” on a public platform is definitely social punishment in my eyes
It’s enforcing a norm that on the margin should be less enforced
Reply to @StatsLime, 06:14 UTC · 2 likes
Max ⏸️ Limelihood 🔸🔸🔸 (@StatsLime): @ronawang @unable0_ Holy based Gigachad
Fr
Reply to @livgorton, 06:29 UTC · 4 likes
Liv (@livgorton): numpy is CPU only :((
Duh
Monday, January 13, 2025
Reply to @davidad, 12:52 UTC · 5 likes
davidad 🎇 (@davidad): (But, to be clear, NOT neural networks, gradient descent, self-attention layers, etc.)
Idiot
Quote tweet, 13:14 UTC · 10 likes
Big if true (fr)
Quoting Joel Selanikio (@jselanikio): If you missed it last week:
House Resolution 238, allowing AI to “qualify as a practitioner eligible to prescribe drugs…”
Boom.
Tuesday, January 14, 2025
Tweet, 07:12 UTC · 5 likes
I feel like there could be an international wildfire-response coalition
Like almost all of the time, almost all firefighting resources must be going unused right?
But then once in a while there’s a huge ass fire and infinity more resources would help
Quote tweet, 07:36 UTC · 15 likes
You should take this with a large grain of salt bc Nick is both smarter and more domain-knowledgeable than me, but I think the exact opposite and the word/concept fell off in a bad way
Quoting Nick (@nickcammarata): the word ai alignment feels increasingly wrong to me. people don't have babies and worry about their "alignment". they worry about how to raise them right, what they'll be like. we should similarly be worrying about what the right way to raise the first superintelligence is
Thread, continued, 07:36 UTC · 10 likes
Now it’s always “safety” and “control”–which are fine/good I guess— but I really think the OG Yudkowskian conception of “alignment” is what really matters at the end of the day
Thread, continued, 07:36 UTC · 4 likes
Especially when it comes to graceful failure modes, which mainstream AI safety is understandably not excited to talk about
But one of the most important questions in the world is how to get ASI to be minimally bad in the case where we don’t/can’t control it
Reply to @boomsupersonic, 13:16 UTC
Boom Supersonic (@boomsupersonic): Whether you're an av geek, aerospace engineer, or travel enthusiast, we want to hear from you. Drop your questions below about XB-1’s supersonic test flight. We’ll do our best to answer during an upcoming live Q&A (date & time TBA).
Quote tweet, 13:19 UTC · 6 likes
What
Quoting Overeducated Gibbon (@MostlyMonkey): Top male contestant was 8th, 15% slower than top female. Next male was 20th. Instacart discourse relevant, I guess.
We must study how stereotypes are a better source of information than all non-econ social sciences.
Quote tweet, 13:41 UTC · 30 likes
So there are just pineapples hanging 4 feet off the ground in policy land I guess
Quoting sam (@sam_d_1995): congestion pricing is a policy miracle
Reply to @littIeramblings, 14:17 UTC · 2 likes
sarah (@littIeramblings): imo there is a defensible position that AI x-risk is possible but unlikely enough that the benefits of mostly unconstrained development outweigh the risks (especially since regulation designed to mitigate it presents its own risks). I strongly disagree with this personally but don't think it's totally insane
Where this threshold lies is up for debate – I agree that ppl making points like Starmer's should a) say what they think the probability actually is b) how they arrived at this number and c) why they think that level of risk is worth taking. but I doubt Starmer has thought about this at anywhere near this level of rigour and that the govt's approach is v irresponsible
💯
Quote tweet, 14:35 UTC · 2 likes
Called it
Quoting NBER (@nberpubs): How did the West escape from premature mortality? Economic growth reduced harvest swings, triggering multigenerational epigenetic effects. Variance kills your descendants, from Dora Costa, Lars Olov Bygren, Benedikt Graf, Martin Karlsson, and @joepricebyu https://www.nber.org/papers/w33343
Quote tweet, 14:43 UTC · 2 likes
You can also turn t4 into t3 the old fashioned way (ie basic internal metabolism; your body will covert t4->t3)
Quoting Ray Peat HubblyBubbly Emporium ✝️ (@RayPeatHeadShop): How to make your own T3 if all you have available is T4 (and other fun experiments):
If you have access to fresh mammalian liver you can make a simple ex-vivo perfusion machine with household materials to process cheap, legal, and easily available prodrugs into their rare, expensive, and potentially illicit active metabolites.
Alternatively, you can blend fresh liver and use it *like* a culturing medium to produce active metabolites (40C for about 4 hours)
Have fun!
Reply to @beala, 17:06 UTC · 4 likes
Alex Beal 🆎 (@beala): nooooooooooo aaron pineapples don't hang
They do in policyland
Reply to @DocStrangelove2, 17:15 UTC · 29 likes
Doc Strangelove (@DocStrangelove2): Hollywood will embellish things, but no, this did happen. Verified by not only dozens of men who were present, but even captured German soldiers who were shocked by it, also confirmed it.
(quoting @Jason_R_Burt: 80 years ago today, Ron Speirs is placed in charge of Easy Company and goes on one of the most impressive runs in human history. 🪖)
Reply to @twaniimals, 17:52 UTC · 2 likes
🎃 (@twaniimals):
@livgorton
Tweet, 17:59 UTC · 12 likes
What’s the actual deal with security and privacy around xiaohongshu
Can I check it out out of anthropological curiosity without worrying or what
Quote tweet, 19:14 UTC · 23 likes
Ok, I am unofficially launching a paid version of this
The value proposition for both of us is that I get 50% of what the api calls actually cost (minus Stripe fees, hence $2 minimum purchase) and you get a convenient LLM wrapper
Link below
Quoting Aaron Bergman 🔍 (@AaronBergman18): Is this the kind of thing anyone would actually use?
Multiple models estimate a single figure, see the stats and results
Thread, continued, 19:14 UTC · 2 likes
https://llm-est-c18067b91d3b.herokuapp.com/
aka
If I actually make a nontrivial amount of money, I will buy a real domain lmao
Thread, continued, 19:14 UTC · 3 likes
Currently people spending money on this is mostly a signal of whether I should keep working on it to make it better and more valuable so uh vote with ur dollar
Thread, continued, 19:14 UTC · 1 like
Tbh I hope at least one person spends the $2 so I can say I’ve made money on the internet this way
That’s probably about what I’ve spent testing the app + payment system as well
Reply, 19:50 UTC · 1 like
Lol it should be @freed_dfilan
Quote tweet, 22:47 UTC · 17 likes
post a picture of yourself from a different era
Quoting 𝞍 Shin Megami Boson 𝞍 (@shinboson): post a picture of yourself from a different era
Reply to @RomanHauksson, 23:00 UTC · 3 likes
Roman Hauksson-Neill (@RomanHauksson):
image attached
(quoting @AaronBergman18: post a picture of yourself from a different era)
What is going on here
Reply to @LinchZhang, 23:09 UTC · 10 likes
Linch (@LinchZhang): Some researchers are going to put you to sleep. During the two days that your sleep will last, they will briefly wake you up either once or twice, depending on the toss of a fair coin (Heads: once; Tails: twice). After each waking, they will put you back to sleep with a drug that makes you forget that waking. When you are first awakened, to what degree ought you believe that the outcome of the coin toss is Heads?
I’ll die on the 1/2 hill. Before going to sleep, p(heads) is 1/2, and you gain no additional Bayesian information upon waking
https://www.aaronbergman.net/p/the-answer-to-the-sleeping-beauty
Reply, 23:09 UTC · 1 like
(There are other intuition pumps but they seem to be unconvincing or anti-convincing to others; the Bayesian info argument is most solid)
Quote tweet, 23:15 UTC · 5 likes
What’s the net impact on uber pricing?
Tax = up
Less time per mile = down
Quoting David Petersen (@typesfaster): This street was never empty pre Jan 5
Reply, 23:43 UTC
Wait I agree with the first part about probability theory, but disagree about the second part
Or at least I don’t see how this bursts my bubble
Reply, 23:44 UTC · 1 like
Yeah I think we agree?
Reply, 23:45 UTC · 1 like
Lol ya
Wednesday, January 15, 2025
Quote tweet, 00:51 UTC · 3 likes
New free tier models just dropped
Gemini Flash 2.0 and 2.0 Experimental, alongwith 4o mini and Haiku, all free up to 50 uses/day
(video, plays on X)
Quoting Aaron Bergman 🔍 (@AaronBergman18): Ok, I am unofficially launching a paid version of this
The value proposition for both of us is that I get 50% of what the api calls actually cost (minus Stripe fees, hence $2 minimum purchase) and you get a convenient LLM wrapper
Link below
Tweet, 00:56 UTC
https://x.com/AaronBergman18/status/1879331787706622268
Reply to @nnotm, 01:10 UTC · 2 likes
Nnotm (@nnotm): @AaronBergman18 Is it intentional that you can't use the free tier models without having purchased credits?
Uh no that shouldn't be the case, but you do have to sign up for an account first (so I can track usage)
Thread, continued, 01:11 UTC · 2 likes
Wait I think this is not working right! I will try to fix!
Reply to his own earlier tweet, 01:42 UTC · 2 likes
Aaron Bergman, January 15, 2025: https://x.com/AaronBergman18/status/1879331787706622268
There was a bug that meant you couldn’t call use the (now 4!) free tier models with $0 credits
Now you can!
Reply to @qualiakiwi, 02:52 UTC · 2 likes
(The tweet being replied to by @qualiakiwi is no longer available.)
Eh I mean I endorse what @austinc3301 says in that those are technically downsides but to not answer your literal question, nothing is that bad
I guess more likely to elicit animosity above whatever threshold matters to you
Reply, 03:08 UTC · 1 like
I don’t understand this and think the first sentence in the screenshot is wrong. Subjectively, the two possible worlds are identical. You know everything there is to know about the coin before the experiment begins, including what it’s going to be like waking up and such
It makes no sense to say “now, before the experiment, p(heads) = 1/2, but I’m going to change my probability upon waking, even though I can already mentally simulate every possible world perfectly”
Cc @NinaPanickssery can defend herself!
Formally:
A := coin lands heads
B := I am woken or I observe having just woken up
P(B | A) = P(B | ~A) = P(B) = 1
Reply to @wirmgurl, 15:42 UTC · 6 likes
🪱 Worm Girl 🪱 (@wirmgurl): Here are my most woke opinions:
- trans women are women and trans men are men.
- sexual assault and rape is extraordinarily common and often extraordinarily terrible for the victims.
- being a minority can be very difficult and we should support minorities extra in a broad range of situations.
- being poor sucks really bad, more than you might expect. Most poor people are trying their best not to be poor and we should help them.
- it's hard to step outside your own culture and see things from other cultures' points of view. I don't think non-western cultures are always correct but they have interesting viewpoints we should take seriously.
Basically agree with 1-4
I guess I agree with 5 in a literal sense but not as strongly in spirit
Quote tweet, 15:43 UTC · 2 likes
Making tea is about drinking tea
Quoting “paula” (@paularambles): sometimes making tea is less about drinking tea and more about having a mug of tea to keep you company
Quote tweet, 16:44 UTC · 3 likes
FDA-endorsed 😤
Quoting Aaron Bergman 🔍 (@AaronBergman18): Guy who makes bismuth subsalicylate (the active ingredient in pepto bismol) at home from chemistry-grade bismuth and salicylic acid (lifetime supply of both for ~$100 on amazon) bc he’s concerned about the amount of red dye used in commercial products but is bullish on C7H5BiO4
Reply to @JesseTayRiver, 16:47 UTC · 3 likes
Jesse Smith (@JesseTayRiver): Something I've noticed in my 25 years in the US is that Americans are opposed to putting raisins in most things. Bakeries will often refuse to add raisins to a carrot cake, the oatmeal-raisin cookie is reviled, there's almost no selection of breads with raisins. Heartbreaking
Big Raisin needs better lobbyists
Reply to @wirmgurl, 17:23 UTC · 4 likes
🪱 Worm Girl 🪱 (@wirmgurl): I agree with it so strongly in spirit that it comes round to being an actually very unwoke opinion! i.e. I have respect for cultures that many people would condem.
Yeah we really disagree then
I vibe associate with like the OG new atheist types - cultures shouldn’t be respected; people should
Reply to @CharlesD353, 18:22 UTC · 2 likes
Charles🔸 (@CharlesD353): Chunks would add significant complications which we could probably manage, but it's not trivial
FWIW for a vaguely similar thing/set of issues, chunking was the only way I figured out to make it work
Reply to @timfduffy, 22:27 UTC · 2 likes
Tim Duffy (@timfduffy): This rules, just passed a couple queries through the updated version. I bought credits so I didn't check whether your free model fix worked. Impressive set of free models, enough to make it worth using the free version for sure. How do you handle the account stuff? Are there tools to help you manage safe storage of account details easily?
Accounts are just a simple SQL database thing
The only thing I really know about safety is that I can’t see passwords, only hashed passwords. Also stripe handles the payment stuff so I can’t see your credit card details
Thursday, January 16, 2025
Quote tweet, 01:27 UTC · 38 likes
AI moral patienthood is utterly serious, but human-shaped concepts like “slavery” with extremely contingent connotations are really really really unhelpful and probably harmful (insofar as they distract from the genuinely-important ethical stakes)
Quoting Stephen McAleer (@McaleerStephen): @eshear Enslaved god is the only good future
Reply to @NULobbyist, 01:32 UTC · 3 likes
NUL (@NULobbyist): @AaronBergman18 Aside from that though. If by enslaved he means, does something other than what has the highest expected value for it because we told it to, then that seems a very unstable situation to be in with a super intelligence
Eh maybe but then we should just talk about it in those terms (EV, constraints, etc)
Reply, 04:11 UTC · 5 likes
I think the obvious counterexample is that you know tomorrow you’re gonna be tricked/given malicious info and know that now but won’t know it tmr
Reply, 04:31 UTC · 1 like
I’ll defer to @birchlse @jeffrsebo and @rgblong
Thread, continued, 04:33 UTC · 3 likes
But also I feel more optimistic about reducing p(sentience) than trying to figure out what causes differently valenced qualia in extremely alien minds conditional on sentience
Tweet, 10:21 UTC · 8 likes
Is leaving a dating match app on read before you’ve met up or made any plans to “ghosting”?
Reply to @kepe__, 10:33 UTC · 2 likes
kepe ✨🇫🇮 (@kepe__): @AaronBergman18 not responding to someone when they haven't said something conversation-terminating is ghosting, period
My intuition is that it’s kinda uncool but way less bad than disappearing after you’ve met up, and these shouldn’t go in the same category
Reply to @imitationlearn, 10:58 UTC · 2 likes
imit (@imitationlearn): @DavidSKrueger What's the analogue to modern uses of the wheel for AI?
Maybe the answer is, "we don't know what AI is capable of," which is fair, but also meh.
The entire remote work economy without robotics, the entire global economy with robotics, including all of science
Quote tweet, 15:46 UTC · 10 likes
Fr though, what is the point of this for OpenAI?
Quoting Teknium 🪽 (@Teknium): OpenAI single handsdly keeping the legacy media alive solely for training data lol
Reply to @webdevMason, 18:16 UTC · 4 likes
Mason (@webdevMason): @selentelechia Empirically, yes, babies are born that way
But it's obviously not possible
Fr
Quote tweet, 18:25 UTC · 5 likes
This seems like a plausible path to ASI
Quoting AI Notkilleveryoneism Memes ⏸️ (@AISafetyMemes): Gwern thinks it's almost game over
"OpenAI may have 'broken out', and have finally crossed the last threshold of criticality to takeoff - intelligence to the point of being recursively self-improving and where o4 or o5 will be able to automate AI R&D and finish off the rest."
"Much of the point of a model like o1 is not to deploy it, but to generate training data for the next model. Every problem that an o1 solves is now a training data point for an o3."
"I am actually mildly surprised OA has bothered to deploy o1-pro at all, instead of keeping it private and investing the compute into more bootstrapping of o3 training etc. (This is apparently what happened with Anthropic and Claude-3.6-opus - it didn’t ‘fail,’ they just chose to keep it private and distill it down into a small cheap but strangely smart Claude-3.6-sonnet.)"
"If you’re wondering why OAers are suddenly weirdly, almost euphorically, optimistic on Twitter, watching the improvement from the original 4o model to o3 (and wherever it is now!) may be why.
It’s like watching the AlphaGo Elo curves: it just keeps going up… and up… and up…"
Quote tweet, 19:02 UTC · 14 likes
Honestly Stripe is an absolutely brilliant business that I only came to appreciate when using it on the sell side for my LLM wrapper
Kinda gives FTX in the sense of “money printing machine,” only they haven’t decided to blow up the machine for no reason (yet)
Quoting austin petersmith (@awwstn): they charged me 2.9% + $.30 just to walk in the door!
Quote tweet, 22:34 UTC · 51 likes
As usual the writing is excellent, but I kinda wanna point out that Sarah *is in fact very smart*
So like I’m not sure this answers the question of what actually-ordinary people should do
Quoting sarah (@littIeramblings): I wrote a blog post for anyone who is interested in AI safety but worried that they are not smart enough to contribute to the conversation
link below
Reply, 22:36 UTC
I don’t mean taxpayers
I mean public policy should largely be structured around giving people decent lives - empirically this mostly means competent management of a basically capitalist economy with some welfare state on top, not paying people to dig ditches or whatever
Reply to @littIeramblings, 23:20 UTC · 4 likes
sarah (@littIeramblings): Aaron 🥹🥹
Reply to @littIeramblings, 23:31 UTC · 2 likes
sarah (@littIeramblings): keep mansplaining king
Ok
Tweet, 23:48 UTC · 13 likes
Why does cursor only cost $20/mo? Are they burning VC cash to pay for API calls? Long context Claude isn’t *that* cheap
Friday, January 17, 2025
Reply to @ModeledBehavior, 00:58 UTC · 2 likes
Adam Ozimek (@ModeledBehavior): But what if AI reduces mortality and this policy slows development?
Sorry i don’t think you can treat this topic as per se asymmetrical
(quoting @ChadJonesEcon: How much should we spend to reduce A.I.'s existential risk? We spent around 4% of GDP to mitigate deaths from Covid-19... A new paper: https://web.stanford.edu/~chadj/reduce_xrisk.pdf)
The true relevant symmetry is x-risk from AI vs steady state x-risk from other causes that AI reduces
I’m pretty sure a gazillion IQ point*hours have gone into analyzing this
Only looking at present generation mortality is kinda silly
Quote tweet, 01:10 UTC · 88 likes
Claims (medium confident, whatever that means)
- Alcohol is a major cause of social ills
- alcohol consumption by people prone to take life advice from @AndyMasley and @tylercowen is not
Quoting Chana (@ChanaMessinger): I hadn't previously read Tyler Cowen's take on alcohol. I found this part in particular interesting:
https://marginalrevolution.com/marginalrevolution/2013/04/the-culture-of-guns-the-culture-of-alcohol.html
Thread, continued, 01:10 UTC · 20 likes
I’m basically a non-drinker fwiw bc my brain is weird and I don’t like it
But I will continue to be an anti-anti-alcohol crusader as long as I think that’s the correct position
Quote tweet, 01:10 UTC · 5 likes
Holy based
Quoting David🐵 (@monke_io): Just set up a $600 recurring monthly donation to the Shrimp Welfare Project. My donations will save ~960k shrimps per month from agonizing deaths.
Reply, 01:33 UTC · 9 likes
Fuck this makes sense
Quote tweet, 03:48 UTC · 10 likes
Wtf people are giving me money
Capitalism goes brr, this is a major signal/incentive to make it better
Quoting Aaron Bergman 🔍 (@AaronBergman18): Ok, I am unofficially launching a paid version of this
The value proposition for both of us is that I get 50% of what the api calls actually cost (minus Stripe fees, hence $2 minimum purchase) and you get a convenient LLM wrapper
Link below
Thread, continued, 03:48 UTC · 2 likes
I am not especially cracked so no promises but by all means comment suggested/requested features
Also I have no idea what I’m doing so don’t assume I have a good reason for any decision involving this
Quote tweet, 13:26 UTC · 5 likes
This is dumb
I’m not even talking about the merits of whether Twitter is good or not - it’s like, where there are extremely strong network effects and you’re a public-facing community, you can’t just play by your own rules/what naive deontology might imply
Quoting Steve McGuire (@sfmcguire79): NEW: The American Economic Association is encouraging economists to leave X for BS.
A new report cites the decline of hashtag EconTwitter, harassment (especially of women, people of color, and LGBTQ scholars), and policy changes under Elon as reasons to switch.
🧵
Reply to @ilex_ulmus, 13:29 UTC · 7 likes
Holly ⏸️ Elmore (@ilex_ulmus): Anyone in Berkeley have a document shredder I can use?
Ominous
Reply to @ronawang, 13:31 UTC · 45 likes
Rona Wang (@ronawang): my friend was the only girl at a hackathon & somebody gave her this note 💀
This seems fine
Reply to @GarrettPetersen, 13:37 UTC · 2 likes
Dr. Dad, PhD 🔄🔼◀️🔽▶️ (@GarrettPetersen): 3.9yo's bedtime questions:
1) I wonder who the baby is going to be?
2) What is kindergarten like?
3) When will we all die?
Good qs
Tweet, 15:05 UTC · 5 likes
Am I crazy or did search somehow get even worse
Reply to @TBirdStorrs, 21:29 UTC · 3 likes
TB Storrs 😷 (@TBirdStorrs): @AaronBergman18 @AndyMasley @tylercowen I was a major cause of social ills. I've been sober a little more than a decade. That post was a helpful nudge.
Fair enough, thanks for sharing!
Saturday, January 18, 2025
Quote tweet, 00:53 UTC · 2 likes
New feature just dropped; Claude Sonnet (and soon o1, hopefully) generates an all things considered best guess using the original model responses with reasoning as input (paid users only)
(video, plays on X)
Quoting Aaron Bergman 🔍 (@AaronBergman18): Ok, I am unofficially launching a paid version of this
The value proposition for both of us is that I get 50% of what the api calls actually cost (minus Stripe fees, hence $2 minimum purchase) and you get a convenient LLM wrapper
Link below
Thread, continued, 00:53 UTC · 1 like
If I make $100 I'll buy an actual domain lol
Reply to @banterrealism, 03:13 UTC · 3 likes
Sex Haver (@banterrealism): (Her parents still don’t know we’re married)
Lol what
Reply to @JeremiahDJohns, 04:28 UTC · 1 like
Jeremiah Johnson 🌐 (@JeremiahDJohns): Just a reminder that the 30-50 feral hogs guy was completely right and everybody who mocked him was wrong.
(quoting @centx_diesel: This is what happens when domestic hogs interbreed with wild hogs . The get larger each generation.)
Piggy 🐷
Quote tweet, 13:00 UTC · 8 likes
ERAgate is a nothingburger because the partisan ecosystems are so asymmetrical
Trump tries to do something unconstitutional -> Republican and RW machine goes to bat for him
Biden tries to do something unconstitutional -> Dem + mainstream machine knocks it down immediately
Quoting Richard Hanania (@RichardHanania): Meet Colleen Joy Shogan, the Archivist of the United States, who today stood to Biden’s attempted constitutional coup.
She is a former professor of government at George Mason University and the author of 8 murder mystery novels, with titles like “Stabbing in the Senate” and “Homicide in the House.”
During her confirmation hearing, she promised she would not publish the ERA if Biden tried to push it through without proper ratification. Today she kept her promise.
Thank you Dr Shogan, for defending democracy. 🇺🇸
Reply to @inesferhumi, 13:26 UTC · 2 likes
Inés (@inesferhumi): @pearl_please GIRL
Pony
Reply to @unes33501870, 13:32 UTC · 3 likes
Epiphaneia (@unes33501870): Could YIMBY solve the housing crisis ? Randomized study finds that for 20-25 year olds, being given a house increases their probability of having a child by ~32%.
Wonder how much of this is “just” a wealth effect (if that’s how Brazilian housing lotteries work)
Reply to @rohanpaul_ai, 13:51 UTC · 1 like
Rohan Paul (@rohanpaul_ai): Smart prompt engineering beats expensive model fine-tuning for LLM alignment
A prompt optimizer framework that theoretically shows how prompt engineering can match fine-tuning's performance for LLM alignment without parameter updates.
-----
Original Problem 🎯:
LLM alignment typically requires expensive parameter fine-tuning through RLHF. When models are frozen or inaccessible, direct parameter updates become impossible. Prompt optimization could help but lacks theoretical foundations.
-----
Solution in this Paper 🔧:
→ Introduces Align-Pro, a mathematical framework for optimizing prompts to align frozen LLMs
→ Formulates prompt optimization as an optimization problem with KL-divergence regularization
→ Derives closed-form expression for optimal prompt distribution that minimizes suboptimality gap
→ Provides theoretical bounds comparing prompt optimization to RLHF fine-tuning performance
-----
Key Insights from this Paper 💡:
→ Prompt optimization can achieve comparable alignment to parameter fine-tuning
→ Performance depends on prompter quality and frozen model capabilities
→ Tuning parameter λ controls tradeoff between alignment and prompt distribution shift
-----
Results 📊:
→ Tested on UltraFeedback, HelpSteer and Orca datasets
→ Consistently outperforms no fine-tuning baseline across model architectures
→ Win rates of 60-67% against baseline on key benchmarks
→ Lower variance in rewards compared to baseline and RLHF
I have no idea what I’m talking about but I don’t believe the headline result
Quote tweet, 15:15 UTC · 11 likes
This but unironically (anecdotes might be underrated by the type of people prone to doing epistemology on twitter)
Quoting Dr. Jonathan N. Stea (@jonathanstea): Anecdotal evidence reliable?
One man says “yes.”
Thread, continued, 15:23 UTC · 2 likes
You can talk about this in lossy generalities, but:
If you have absolutely no idea whether something is going to happen or not (ie a uniform prior probability distribution over a binary outcome), and you observe exactly one instance of the thing, you should shift your
Reply to @QualiaNerd, 19:55 UTC · 10 likes
QualiaNerd (@QualiaNerd): A very useful lens through which to analyze this. What damage would have occurred if none of the LLMs developed so far had been optimized for safety/rlhf’d in any way whatsoever? Minimal to zero. Important to remember this as we begin to leave the era of passive safety behind.
I don’t think this is true; at least one additional counterfactual injury or death in an attack of some sort if Claude willingly told you how to build bombs and such
Ofc I’m just speculating
Tweet, 20:10 UTC · 4 likes
I’d love to see the Biden ERA thing compared mechanistically, play-by-play, to some similar example of executive declaration in a country with weaker process norms/rule of law, where at the end of the day the extremely-sus executive fiat basically worked out
Thread, continued, 20:10 UTC · 3 likes
I’m continually fascinated that Western-style rule of law works as well as it does, particularly in the sense of overcoming first-order power imbalances (POTUS vs US Archivist and courts without an army if it comes to that, obscure federal agency office vs multi-billion dollar industry)
Like ho
Sunday, January 19, 2025
Reply to @9chabard, 01:13 UTC · 6 likes
Lysandre, qui la connaît à peine (@9chabard): reinstall hinge
image attached
(quoting @BecomingCritter: delete hinge and dm your moots)
Lmao
Reply to @Mjreard, 01:48 UTC · 2 likes
Matt Reardon (@Mjreard): The current short timelines deference cascade seems epistemically worse than the 2019-21 general AI risk deference cascade. Couple of reasons:
- non-technical people deferring on a more technical claim
- only considered by people who've already bought into TAI
Call it a deference cascade or not, I’m not sure it’s epistemically bad
These people have access to more information than me, are smarter than me, have more domain knowledge and experience than me, have a demonstrable and impressive track record, afaik have never erred in the direction of publicly expecting progress too fast/soon, and are broadly willing to stake their money and reputations on near-term falsifiable claims (ie not just throwing out fast timelines vibes with plausible deniability)
Reply to @JadenEarlorg, 04:09 UTC · 2 likes
Beige & (@JadenEarlorg): @AaronBergman18 I guess my concern isn't just lossy generalities but also the easy of cherry picking anecdotes compared to the easy of cherry picking csv files
An anecdote has a pretty wide distribution of how it came to my attention, and wide variance evidence doesn't move my priors a lot
Yeah I guess I really mean that anecdotes are legitimately helpful when (1) you’re actually trying to get the right answer, so minimal incentive to cherry pick, and (2) there’s legit scarcity of info
1 consistent anecdote bumps a universal prior to 67%, 2 to 75%, 3 to 80%!
Quote tweet, 04:32 UTC · 2 likes
I feel like I should have seen plots like this before so now yall have to see them
Quoting Aaron Bergman 🔍 (@AaronBergman18): @JadenEarlorg Yeah I guess I really mean that anecdotes are legitimately helpful when (1) you’re actually trying to get the right answer, so minimal incentive to cherry pick, and (2) there’s legit scarcity of info
1 consistent anecdote bumps a universal prior to 67%, 2 to 75%, 3 to 80%!
Quote tweet, 07:35 UTC · 8 likes
Yeah the more I think about this the more it conceptually makes sense
A gigantic amount of information and what we sometimes call “intelligence” is just downstream of idk somewhere between the standard model of physics and a couple physics and philosophy textbooks
Quoting Aaron Bergman 🔍 (@AaronBergman18): This seems like a plausible path to ASI
Thread, continued, 07:35 UTC · 3 likes
Competent reasoning (including calculation) can get you from low-abstraction physics + philosophy to like arbitrary correct answers about the world
Thread, continued, 07:35 UTC · 4 likes
Training increasingly large models on the outputs of reasoning processes should just highly compress the implications of reasoning into immediately-available latent knowledge - so, ASI
Thread, continued, 07:35 UTC · 4 likes
There are two reasons this might not work in the near term
1) you need OOMs more financing to get this to work than what is practically available and people don’t wanna pay $100 per o4 api call or whatever it would cost to progressively fund the scale-up
Thread, continued, 07:35 UTC · 3 likes
2) actually reasoning doesn’t work that well at physically practical limit of near-term compute, even if it does in some conceptual limit where you can burn through galaxies of energy crunching numbers
Thread, continued, 07:36 UTC · 2 likes
Lots of fancy words in this thread lol
Reply to @JesseTayRiver, 17:38 UTC · 2 likes
Jesse Smith (@JesseTayRiver): Anyone remember the job satisfaction research showing that white collar workers typically have substantially higher satisfaction than blue collar? Slight blip for trades was speculated as due to seeing progress of building stuff. I could be totally butchering this btw
Maybe https://www.scirp.org/journal/paperinformation?paperid=119441
Tweet, 20:35 UTC · 42 likes
Accidentally went to the “for you” feed
Tweet, 20:38 UTC · 6 likes
Man the Obama ‘08 inauguration had such good vibes
Reply to @JesseTayRiver, 21:32 UTC · 4 likes
Jesse Smith (@JesseTayRiver): @AaronBergman18 We should jail people who say things I don't like fr
Duh
Reply to @kalomaze, 21:35 UTC · 4 likes
kalomaze (@kalomaze): i will die mad about this
(quoting @basile0811: @ryanchenkie @kettanaito Cause you can specify Display URL when creating Google ad. It’s just text field and you can put there whatever you want)
Wtf
Reply to @inesferhumi, 23:29 UTC · 2 likes
Inés (@inesferhumi): @absurdlymax @AaronBergman18 @pearl_please Ngl guys I’ve been trying to figure out what you both meant by this and I still don’t know
Lmao. I don’t get Max’s response at all
My reply has zero implied semantic content I just thought it was a bit funny
Monday, January 20, 2025
Quote tweet, 17:07 UTC · 3 likes
Brain drain
Quoting Dan Hendrycks (@hendrycks): It looks like China has roughly caught up. Any AI strategy that depends on a lasting U.S. lead is fragile.
Reply to @JgaltTweets, 17:09 UTC · 3 likes
JgaltTweets (@JgaltTweets): Trump has been sworn in as President of the United States https://www.youtube.com/watch?v=b5Q1ufsBM14
F
Tweet, 17:12 UTC · 10 likes
I do not envy whoever is in charge of making sure the inauguration is secure and nobody gets assassinated
It’s actually an extremely impressive project
Tweet, 17:18 UTC · 7 likes
I strongly dislike Trump, but Biden was not exactly an exceptional president
Tweet, 17:20 UTC · 56 likes
I still like Hillary Clinton
Reply to @absurdlymax, 17:33 UTC · 10 likes
Max Alexander (@absurdlymax): @AaronBergman18 I think she's already married
Cheems mindset
Tweet, 20:03 UTC · 6 likes
Well now that it’s open weights, I’d love for someone to finetune the deepseek model on a highly curated selection of the best analytic philosophy
I’d probably pitch in like $100 for this
Tuesday, January 21, 2025
Quote tweet, 00:52 UTC · 5 likes
.
Quoting rohit (@krishnanrohit): So it begins
Reply to @SchrodingrsBrat, 01:51 UTC · 16 likes
Sherry (@SchrodingrsBrat): My favorite productivity hack is working at a library with no laptop charger and using the battery life as time pressure I call it the poor man’s pomodoro
This is anywhere from like 45 mins to 15 hours lol
Quote tweet, 02:01 UTC · 2 likes
I think the most compelling reason to use http://bit.ly/llm-est is that estimates given by high-performing models empirically vary quite a bit with default temperature settings
Look at the range *within* o1-preview for this question. The three returned responses are 1M, 750k, and 200k!
Quoting Aaron Bergman 🔍 (@AaronBergman18): New feature just dropped; Claude Sonnet (and soon o1, hopefully) generates an all things considered best guess using the original model responses with reasoning as input (paid users only)
Quote tweet, 02:41 UTC · 7 likes
Insofar as Trump has an ideology beyond his own wealth and power, it’s one of pure negation
This, the AI EO, un-banning offshore drilling but also randomly un-non-banning wind projects (??)
Quoting Hugo Lowell (@hugolowell): Trump signs executive order withdrawing from the World Health Organization
Thread, continued, 02:41 UTC · 2 likes
I don’t even have a strong object level take on this WHO thing. Maybe there are good reasons to leave, but my point is that, if so, *those aren’t Trump’s reasons*
Thread, continued, 02:42 UTC · 4 likes
God I hope having an amoral shitshow of incoherent, incompetent, and poorly-directed outrage at the head of the most powerful institution in the universe, possibly during the singularity, is worth it as a referendum against diversity training and gender studies or whatever
Quote tweet, 02:59 UTC · 27 likes
Wait this seems very illegal
Quoting Joe Bishop-Henchman 🗽💸⚖️🚆 (@jbhenchman): Directs the federal government not to recognize U.S. citizenship for persons born in the United States but whose parents are not citizens or lawful permanent residents. Effective 30 days from today, for births on or after that date.
Quote tweet, 03:00 UTC · 7 likes
Did they try to cancel her? Somehow I doubt smelly thesis lady deserves to be cancelled, on priors
Quoting Dr Ally Louks (@DrAllyLouks): Hi all. I’ve taken a bit of time away from social media to reflect, but I see the conversation is ongoing. What I think is important is that we all show each other respect online. If you’re defending me by being disrespectful or racist then I would rather you didn’t.
I’m horrified that a tweet about the importance of solidarity has somehow stoked tension. I am not a person of colour and understand the view that I am not entitled to mediate relations between two groups to whom I do not belong. I won’t try to moving forwards.
At the same time, anti-Indian and anti-Black sentiments are running absolutely riot on this platform and I think we need to pay attention to it. Perhaps it wasn’t my place to call it out this time, but I hope others will feel emboldened to within their communities if they haven’t done so before.
I always intend to respond in good faith, but I do sometimes make mistakes that make it easy for people to assume otherwise. I should have initially been more open to the interpretation that the creator’s video was an anti-Black micro aggression. I still feel that her intent is ambiguous. Some might disagree.
I try to see the best in people. I am quite often called naive because of this. I appreciate that intent is not entirely relevant here also, as things can be racist regardless of intent. I tried to open up a conversation about this and I’m grateful to those who took the time to offer their thoughts.
I decided to delete the thread because I saw a couple of QTs saying that people were using it as a reason to berate Black women and I had a responsibility to prevent this by deleting my tweets.
As someone who has worked on the olfactory dimensions of misogynoir, it was galling to think that I could be inadvertently contributing to it. In hindsight, it feels inevitable that the post would draw that kind of ire.
This is not an ideal platform for having nuanced conversations, but I do feel that I overstepped here. Being given a platform suddenly on an unfamiliar site is a challenging thing to navigate. I will inevitably get it wrong sometimes, but please know that I learn from my mistakes and am well-intentioned.
Tweet, 04:41 UTC · 8 likes
I get why the CCP was paranoid about Tianenmen in 1990, but why today? It’s literally been a generation
Quote tweet, 04:44 UTC · 4 likes
There should be a political arbitrage where we get to decouple culture war referendum from the person with tremendous legal and military authority
Quoting Autism Capital 🧩 (@AutismCapital): 🚨BREAKING: Trump signs his most important executive order yet
Quote tweet, 04:46 UTC · 2 likes
I have no idea what the actual ramifications of this are, but prima facie it’s extremely funny - basically just “I hereby declare the culture war won”
Quoting jessicat (@jessi_cata): The social construct of womanhood leaving my body as Donald Trump signs executive order "DEFENDING WOMEN FROM GENDER IDEOLOGY EXTREMISM AND RESTORING BIOLOGICAL TRUTH TO THE FEDERAL GOVERNMENT"
Quote tweet, 05:18 UTC · 4 likes
There are many relevant differences between me and Elon here (shocking), but I’ve used almost the exact same language (“grace”, “benefit of the doubt”) when advocating for myself before…
Quoting ADL (@ADL): This is a delicate moment. It’s a new day and yet so many are on edge. Our politics are inflamed, and social media only adds to the anxiety.
It seems that @elonmusk made an awkward gesture in a moment of enthusiasm, not a Nazi salute, but again, we appreciate that people are on edge.
In this moment, all sides should give one another a bit of grace, perhaps even the benefit of the doubt, and take a breath. This is a new beginning. Let’s hope for healing and work toward unity in the months and years ahead.
Thread, continued, 05:18 UTC · 3 likes
So yes, if someone says ambiguous or even offensive thing X, and then they say “I didn’t word that well, what I meant is Y,” by default you should probably just believe them
Thread, continued, 05:18 UTC · 2 likes
Like maybe it wasted a few moments of your time, but there’s no good reason to hold the initial error over their head beyond that
Thread, continued, 05:18 UTC · 2 likes
The second order effect of this “grace” is that it makes risky speech that people actually do endorse and wish to communicate less hazardous, which is in turn generally good (I am a saying-true-and-earnest-but-possibly-unpopular-things enjoyer)
Tweet, 05:23 UTC · 9 likes
As president I would do the good things and not the bad things
Tweet, 05:25 UTC · 12 likes
Although I think many of these EOs are bad, I’m basically glad that the president has the power to do them (except for the citizenship thing, which he doesn’t)
Elections should actually have pretty direct and substantive consequences; more veto points is not the answer
Wednesday, January 22, 2025
Tweet, 16:41 UTC · 49 likes
Good analysis from @Benthamsbulldog
Thread, continued, 16:41 UTC · 6 likes
https://open.substack.com/pub/benthams/p/trumps-nightmare-list-of-executive?r=2muv6&utm_medium=ios
Tweet, 16:56 UTC · 16 likes
My dad today: “Who is Claude?”
Tweet, 17:32 UTC · 7 likes
👀
Reply to @dcpip_, 17:55 UTC · 2 likes
(The tweet being replied to by @dcpip is no longer available.)_
I think this may be an unpopular take and I have no authority here but public portfolio of work >> normal old job applications
I agree btw with the sentiment
Reply to @janonacct, 18:32 UTC · 1 like
janon (@janonacct): normies on tech twitter seem perpetually confused at what every AGI company is trying to do
(quoting @abacaj: How is Deepseek going to make money?)
I mean it’s a valid q imo
Even if your goal is ASI utopia, you need the capital to get there
Quote tweet, 19:54 UTC · 11 likes
Thank you, Reid!
Quoting Sam Altman (@sama): Very few people realize just how much @reidhoffman did and spent to stop Trump from getting re-elected--it seems reasonably likely to me that Trump would still be in office without his efforts. Thank you, Reid!
Tweet, 21:18 UTC · 20 likes
I think my p(doom) is finally above 50%
Reply to @RafaRuizdeLira, 21:32 UTC · 2 likes
Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): @AaronBergman18 I was thinking the same thing today
I typed this into my notes app
:(
Reply to @thecaptain_nemo, 22:01 UTC · 5 likes
Nemo (@thecaptain_nemo): the loudest anti-DEI guys visiting an elite university campus in four years and realizing what the actual impact of enforced merit based admissions was
This is unironically fine
Thursday, January 23, 2025
Reply to @bryancsk, 02:07 UTC · 2 likes
Bryan Cheong 🥮 (@bryancsk): What sort of cognitohazard is this "Ziz" effective altruism splinter group and dare I delve deeper to see what it's about?
(quoting @jessi_cata: I remember warning Ophelia that Zizians were a death cult with a high local death rate... Idk if this involved Ziz or associates but... well, Bayesian evidence (lefty radical violence + violent women).)
Before my time 😞
Reply to @Ruesavatar, 19:45 UTC · 2 likes
rue🌿 (@Ruesavatar): It’s a bit funny when men find out I knit then try to flirt with me in DMs by asking me to knit them something. Sir. I would have to be in love with you, body and soul, before I’d knit you anything lol. Don’t flatter yourself, buy a machine made hat. 😉
This tweet is a valuable public service
Quote tweet, 22:26 UTC · 9 likes
EA Global nap room
Quoting RomeoStevens (@RomeoStevens76): From Alan Kay's The Power of Context
Reply to @sam_atis, 22:43 UTC · 1 like
Sam Glover (@sam_atis): Imagine you’re playing a game forecasting coin flips. There are 100 coin flips, 100 smart participants and a £100k prize pot. If you have the lowest mean Brier score after 100 flips, you win. You split the pot with everyone else who got the same score. What’s your strategy?
Smart or ideal/maximally smart?
I doubt this is the ideal strategy but something like “100% heads for the first 3, hope I get lucky, 50% for the rest”
Idea being that <<100 are going to be first-order rational and start off with a strategy of 50% on each to modify from
Reply to @wrathofgnon, 23:21 UTC · 2 likes
Wrath Of Gnon (@wrathofgnon): Please stop wearing synthetic materials, unless you are a diver or luge olympian or something: "The only naturally flame-resistant fiber is wool, along with all other animal protein fibers such as silk, alpaca, mohair, cashmere and others. These fibers are slow to ignite and form ash when burned. Synthetic materials melt when burning. If they stick to the skin, they can cause severe injuries that are difficult to treat. Polyester made up over 57% of global fiber production in 2023."
https://theconversation.com/your-fuzzy-flannel-pyjamas-could-be-incredibly-flammable-heres-what-to-know-247448
Counterpoint: polyester is comfy
Friday, January 24, 2025
Reply to @cremieuxrecueil, 07:07 UTC · 1 like
Crémieux (@cremieuxrecueil): Do you believe Roe v. Wade was correctly decided?
Good policy, bad jurisprudence
Reply to @ryancareyai, 13:23 UTC
Ryan Carey (@ryancareyai): Why do you believe this?
I remember seeing it somewhere hence the screenshot but idk where!
Reply, 13:34 UTC · 1 like
Indeed!
Reply to @slimepriestess, 15:46 UTC · 7 likes
Ra (@slimepriestess): so how many dead mutuals are y'all up to? i'm at 5 i think? is that unusual?
Bruh
Thread, continued, 15:50 UTC · 4 likes
0 afaik
Reply to @JesseTayRiver, 16:21 UTC · 2 likes
Jesse Smith (@JesseTayRiver): Striking that I don't see many recommendations for IAQ monitors. I think most supplements are bs, but both viruses and bacterial lung infections often have some feature of IAQ as an underlying mechanism
(quoting @AskYatharth: sick again. i'm so frustrated with this. i cannot keep living like this has anyone here gone from being frequently sick to almost never sick? how?)
Reply to @jachiam0, 17:34 UTC · 8 likes
Joshua Achiam (@jachiam0): However the market competition between frontier labs shakes out, it is my honest belief that Sam, Demis, Dario, and Elon are all genuinely, in their own ways, trying to get the best outcome possible for humanity as a whole. They all take it seriously. This is reassuing to me.
Demis and Dario? Maybe? Elon? Also maybe but he doesn’t seem to be trying very hard to get the right answer
Sam seems like a sociopath
Reply to @Austen, 20:30 UTC · 64 likes
Austen Allred (@Austen): This… does make sense.
If they simply lied about the level of compute it took to train the model it’s pretty simple, and incentive is there.
(quoting @kimmonismus: Billionaire and Scale AI CEO Alexandr Wang: DeepSeek has about 50,000 NVIDIA H100s that they can't talk about because of the US export controls that are in place.)
I remain totally agnostic and haven’t heard a convincing in either direction
Quote tweet, 23:35 UTC · 36 likes
This administration has no real-world end goals. No connection to simulacra level 0, except by accident
Just “[insert vaguely lib-coded thing] = bad” pattern matching
Quoting unusual_whales (@unusual_whales): BREAKING: The US FDA has pulled draft guidance from its website requiring companies to test medicines and devices in diverse populations as part of a purge of DEI efforts at U.S. health agencies, per Reuters.
Thread, continued, 23:35 UTC · 8 likes
Make wind turbines illegal, why not? “Wind energy” vibe-associates with the left so it must be intrinsically bad
Likewise for anything containing the word “diversity”
Saturday, January 25, 2025
Reply to @RokoMijic, 00:23 UTC · 4 likes
Roko 🐉 (@RokoMijic): @AaronBergman18 Au contraire, this is exactly what achieving real world goals looks like.
Ending DEI, deporting illegals, etc
Sometimes there will be accidental real world consequences, but even these are done for the symbolic value, I claim
This isn’t gonna happen ofc, but I predict that Trump would choose everyone believing he caused the deportation of a million illegal immigrants (but no irl change) over a million of them actually being deported but no recognition of this in the media and such
Reply to @cocoabasilisk, 01:12 UTC · 3 likes
cocoa's basilisk (@cocoabasilisk): not an expert but my understanding is this is good actually. the diversity requirements dont actually provide statistical power to identify minority-specific effects. (obviously, in retrospect - 1% native americans or whatever just makes running trials harder without adding useful signal).
At least in an epistemically better world, there would still be directional evidence that maybe compounds into significance across multiple studies/within a meta analysis or at least generates productive hypotheses
Tweet, 10:44 UTC · 4 likes
Please hold the applause
Reply to @nosilverv, 15:13 UTC · 2 likes
Guy (@nosilverv): It must SUCK to be low iq. Imagine the whole world being a random blooming cacophony of events with no relation between them. You are but a leaf on the hands of an impersonal indifferent fate. YIKES my heart goes out but couldn't be me
Compassion for the less cognitively gifted is highly underrated
Reply to @mattyglesias, 15:46 UTC · 7 likes
(The tweet being replied to by @mattyglesias is no longer available.)
I think medicine and housing are two big exceptions though because normal capitalist consumer-friendly incentives are totally fucked for various reasons
There really is quite a lot of paperwork and such in those domains!
Reply to @balajis, 17:04 UTC · 3 likes
Balaji (@balajis): FROM DEI OFFICERS, TO USA OFFICERS
What a great clip. Trump is at his absolute best when overriding Democrat governance.
He’s just genetically optimized to cut through blue bureaucracy. Did it for years as a New York real estate developer.
In this scene he’s doing what every CEO has done but no president in decades has: going directly to understand what’s failing, talking to everyone first hand, and not taking no for an answer.
And often the answer is so simple! “Allow people to clean up their own homes” is at the level of “put water on the plants instead of Gatorade.”
Blues created such an Idiocracy that extremely obvious moves like this nevertheless unlock billions in value.
How to do this systematically, though? On a one-off basis, Trump could maybe unblock blue state bureaucracy with targeted executive orders from above.
But personnel is policy — so what you really want is to just de facto fire everyone like Bass. After all, many of these mayors are actually ceremonial because city managers run the cities.
So as a thought experiment, imagine the US had a vertically integrated org chart, with all mayors reporting to their governors reporting directly to the president. Then Trump could just fire Bass on the spot and hand over command to one of his many new competent lieutenants.
In practice, he might not legally be able to fire these people, but he could effectively strip them of authority. Think about what happened during Hurricane Katrina, when GWB handed over control of New Orleans to Lt. Gen. Russel Honoré. With appropriate legal authority, Honoré just started barking orders and cleaning up the mess local blue government had made.
You could systematize that.
Maybe as DOGE shuts down the useless parts of the federal bureaucracy, it boots up a series of “embeds” that sit in every blue mayor and governor’s office, with full information rights on everything. Kind of like a DEI officer — except they’d be a USA officer!
While DEI officers ensured compliance with blue values, a USA officer would ensure compliance with the people’s values. That is, they’d make sure each local action by a blue elected is compliant with the President’s guidance for the whole country.
And if the mayor or governor did something different — if they delayed permits, or released criminals, or distributed syringes, or funded homeless encampments — the USA officer could just report up to DC to rain down an executive order on the next Bass’ head.
That’s how you scale this scene to every blue city in America.
(quoting @EricLDaugh: 🚨 BREAKING: LA Mayor Karen Bass and Trump with a very tense exchange now... Bass is losing her cool over insistence from community members and Trump who say that clearly, people aren't being allowed back on their property soon enough. TRUMP: People are willing to clean up…)
Yeah because naive negation of the other side’s programs sometimes accidentally gets you a win
There’s no guiding ideology or end-goal to help identify which liberal policies are worth destroying; I’m not optimistic
Tweet, 17:31 UTC · 5 likes
Mostly serious question: why are these three different commands
Reply to @seconds_0, 23:24 UTC · 3 likes
0.005 Seconds (3/694) (@seconds_0): "You can do things" I mutter as I pour 4 tablespoons of sugar into a sugar free monster while my mother in law stares at me like I chokeslammed a nun
🤔
Tweet, 23:26 UTC · 10 likes
Banger post
Thread, continued, 23:26 UTC · 4 likes
Sunday, January 26, 2025
Reply to @livgorton, 01:04 UTC · 7 likes
Liv (@livgorton): we should bring back diagrams of ML architectures in papers pls. the volume of papers that just simply do not include one troubles me. it is v easy to just figma a sort of ugly but functional one.
Here u go
Reply to @liz_love_lace, 01:17 UTC · 165 likes
Liz Lovelace (@liz_love_lace): now that the dust has definitively settled, it's AMAZING that twitter laid off 80% of its employees with basically no negative impact to the user experience or the speed at which they ship new features. So many people thought that twitter would just die, but it didn't!
It’s a way less profitable company lol (and slightly worse product)
Reply to @LeahLibresco, 01:20 UTC · 17 likes
Leah Libresco Sargeant (@LeahLibresco): Search is severely broken and has been for a while
This is legitimately confusing bc this seems like the easiest possible problem to solve
“Text search of a database with operators” is not an extremely difficult technical problem
Tweet, 01:27 UTC · 4 likes
Who would you like to see run for the Dem nomination in 2028 (conditional on the world being broadly similar-ish to rn)?
Reply to @AnniePosting, 01:39 UTC · 2 likes
(The tweet being replied to by @AnniePosting is no longer available.)
A true moderate
Reply to @absurdlymax, 02:23 UTC · 2 likes
Max Alexander (@absurdlymax): @JesseTayRiver
Real
Reply to @TetraspaceWest, 03:00 UTC · 2 likes
(The tweet being replied to by @TetraspaceWest is no longer available.)
Age doxx
Reply to @venturetwins, 03:01 UTC · 25 likes
Justine Moore (@venturetwins): I’m sure this is going to work
Unironically based
Idk discussion groups sorta have a history of affecting change no?
Reply to @daniel_271828, 04:11 UTC · 7 likes
zzDANIEL_ETH (@daniel_271828): “Absolve” is kind of a weird word to use here, but I think he meant “avoid”. Anyway, he acknowledged that there are risks, didn’t seem to imply that the we should ignore them, and said that we wouldn’t let it be “the rabbit that gets away”. So seems good
Directionally sure
But magnitude << same words from any past president
Dude just says things lol
Tweet, 04:34 UTC · 2 likes
US federal government buys a majority stake of NVIDIA
Thread, continued, 04:35 UTC · 2 likes
How come this never happens? Or does it? Seems like USG buying companies would be totally above-board but you never hear about it
Like the pentagon never acquires defense startups in the way Google acquires whatever
Quote tweet, 04:51 UTC · 9 likes
Tbf this seems way less pernicious than actually building the censorship seamlessly into the model
Quoting Jane Manchun Wong (@wongmjane): DeepSeek censors its own response in realtime as soon as Xi Jinping is mentioned
Reply, 15:34 UTC · 1 like
It’s honestly hard to think of what I’d want to pay an 11 year old to d
Maybe true generational prodigies like Von Neumann are exceptions but there aren’t many of them
Quote tweet, 16:32 UTC · 7 likes
The thing about fake conservation stuff like this is that even the “protected” animals don’t actually reliably benefit from protection (yes this is an “ecosystems don’t matter, individual animals do” post)
Quoting JgaltTweets (@JgaltTweets): I'm still waiting to hear what would happen if all the bats died. How bad would it really be? Would we even notice?
Monday, January 27, 2025
Tweet, 00:55 UTC · 11 likes
Hypothesis: OpenAI’s scheduled tasks thing isn’t a legit experiment/stepping stone to anything, it’s purely to reduce chatGPT o1 usage by making the shiny new thing attached to a cheaper model
Thread, continued, 00:57 UTC · 3 likes
Reminds me of another hypothesis: the Amazon “click to send your delivery driver a $5 thank you gift” doesn’t causally affect how much money anyone receives; there is a pre-decided holiday bonus for each employee, and someone had the idea to let customers click the button until that amount is reached (and then it disappears as a feature)
Tweet, 01:00 UTC · 7 likes
Overrated (not to be confused with bad or useless): mechanistic interpretability
Underrated: rigorous but not mechanistic model psychology
Tweet, 01:36 UTC · 8 likes
What should I ask @robi_rahman of @EpochAIResearch on Pigeon Hour
Reply to @TetraspaceWest, 03:50 UTC · 2 likes
(The tweet being replied to by @TetraspaceWest is no longer available.)
I’m guessing I’m not unique in having some ability to turn the dial down on false positives (safe mode, not as fun or interesting) and up (might fuck up and do something socially bad)
Would guess that women tend to have the dial more towards the safe side though, yes
Reply to @larsiusprime, 06:05 UTC · 3 likes
Lars Doucet (@larsiusprime): So we’re several years in to the LLM/generative AI revolution and… is anything going to happen on the IP infringement enforcement side? Seems like nobody in charge cares? Not the outcome I expected.
NYT lawsuit still ongoing I guess https://x.com/i/grok/share/v35GKlL9bAL3p5AyHlIM7mSFK
Reply to @thib_boi, 14:39 UTC · 8 likes
Thibault (@thib_boi): A friend shared a good article with me. They WANT to be contacted. Here is a quote:
“People have this romanticised view that isolated tribes have chosen to keep away from the modern, evil world,” says Kim Hill, an anthropologist at Arizona State University.
But when Hill and others interview people who recently came out of isolation, the same story emerges time and time again: they were interested in making contact with the outside world, but they were too afraid to do so.
As Hill puts it: “There is no such thing as a group that remains in isolation because they think it’s cool to not have contact with anyone else on the planet.”
image attached
(quoting @thib_boi: Unpopular opinion: it is immoral not to contact them. We have technology, especially in medicine, they would probably love. They can refuse to trade but at least offer them the option.)
Quote tweet, 16:49 UTC · 3 likes
This is ~proof we’re looking at an inverted bubble right?
Quoting Oren Cass (@oren_cass): The fact that $NVDA rose steadily in the week after DeepSeek's release and then plunged 10% after a weekend of frothy tweeting tells you what you need to know about extent to which hype vs. reality is driving the AI train right now.
(Also a good lesson in "market efficiency"!)
Reply to @growing_daniel, 16:59 UTC · 73 likes
Daniel (@growing_daniel): Finance guys are like I fucking knew you nerds were full of shit about needing that much money
Banger
Reply, 17:02 UTC
Big if true
Quote tweet, 17:04 UTC · 2 likes
Stronger export controls
Quoting wordgrammer (@wordgrammer): Also, export restrictions didn’t harm them as much as we thought they did. That’s probably because our export restrictions were really shitty. The H800s are only worse than the H100s when it comes to chip-to-chip bandwidth.
Quote tweet, 17:05 UTC · 56 likes
Real
Quoting lynk (@lynk0x):
Reply to @Duderichy, 17:15 UTC · 4 likes
the Rich (@Duderichy):
Lol Trump killing pepfar for no reason is a direct counterexample
Reply to @CharlesD353, 18:21 UTC · 1 like
Charles🔸 (@CharlesD353): There are a bunch of news media articles explaining the fundamental reasons this has made people bearish on Nvidia - whether you agree or not, it's not a sociological phenomenon.
Then why did it take days for the market to react?
Reply to @ChanaMessinger, 18:44 UTC · 3 likes
Chana (@ChanaMessinger): Square cube law is so important to everything
Was gonna say that it’s weird how top rock climbers vary quite a bit in terms of height (pretty similar to population) but actually the weight clustering is remarkable
Tweet, 22:10 UTC · 11 likes
I expect the Trump administration isn’t even considering Pepfar because at a non-systemic, individual level, Trump and his advisors just do not care about poor Africans
Thread, continued, 22:10 UTC · 4 likes
The progressive academia-NGO blob has its downsides, but one nice feature is that they make egregious negligence of the poor slightly politically costly for Democratic politicians
Reply to @billybinion, 22:40 UTC · 3 likes
Billy Binion (@billybinion): Sorry, but why is Dr. Phil out doing immigration busts? He is not a federal law enforcement agent. He is a talk show host & lapsed psychologist. This is just deeply silly.
(quoting @ChiSpectator: Tom Homan and Dr Phil posted a video of them engaging with a Thailandese national who has been detained by federal authorities due to a history of sex crimes in #Chicago)
I shouldn’t be bc it’s serious for the migrants but I’m actually laughing out loud
Tweet, 23:07 UTC · 3 likes
I love the hobbies bit for the chicken (interestingly no matter what photo I use it pegs me as Christian)
Reply to @WillRinehart, 23:09 UTC · 2 likes
Will Rinehart (@WillRinehart): “Decomputing argues that, wherever Al is proposed as 'the answer', there is a gap for the self-organisation of people who already know better what needs to be done…”
But really, how are people supposed to self-organize their way to inventing new medicines?
Hey I’m on your side here but the literal answer is pharmaceutical companies
Tuesday, January 28, 2025
Quote tweet, 00:08 UTC · 2 likes
Just spent >30 mins juggling 12 different files (pdf -> word for text box -> pdf) in order to send 4 documents to act as a HIPAA release for 2 doctors. Could have been a single online form
Not a huge deal itself but nicely illustrative of the problem
Quoting Aaron Bergman 🔍 (@AaronBergman18): @mattyglesias I think medicine and housing are two big exceptions though because normal capitalist consumer-friendly incentives are totally fucked for various reasons
There really is quite a lot of paperwork and such in those domains!
Reply to @PradyuPrasad, 00:16 UTC · 2 likes
Pradyumna 🇸🇬 (@PradyuPrasad): i think the problem about agent frameworks is that their abstractions are hiding too much
sometimes I WANT to see the internals
The internals are a lot of the moat for wrapper services
Like someone spent a bunch of time figuring out exactly which prompts work best
Quote tweet, 00:23 UTC · 11 likes
I think this is important evidence that obesity is a setpoint disorder and not “rational” overconsumption (bc of cheap tasty food, hyperbolic discounting, etc)
Under a naive “I prefer consuming lots of calories” model, people would continue gaining weight throughout adulthood
Quoting Niels Hoven 🐮 (@NielsHoven): The average US man is 5'9", 200 pounds
Interestingly, weight doesn't increase much with age.
Seems like people put on a bit of extra weight when they're young and then just... stay there?
Quote tweet, 01:30 UTC · 7 likes
This is unironically a pretty important consideration
Quoting orb 🧭 (@orb_net): big tick wants 🫵 you to be naked at vibecamp
Reply to @JesseTayRiver, 01:37 UTC · 1 like
Jesse Smith (@JesseTayRiver): @AaronBergman18 So would a setpoint disorder be something like people eat seed oils (not because they're delicious) > ? > briefly gain weight > maintain the same BMI until death
I mean people probably eat them bc they’re in everything, they dont even taste good lol
The important distinguishing thing is the last part, that level is ~constant and rate of change is not
Reply to @littmath, 02:43 UTC · 2 likes
Daniel Litt (@littmath): went on a 5-plus-minute rant about some mathematical text that used the phrase “an empty set”. There’s only one empty set! All empty sets are equal! Call it *the* empty set!!!
@absurdlymax new thing you might enjoy being pedantic about
Quote tweet, 03:16 UTC · 12 likes
Chat is this real
Quoting Financelot (@FinanceLancelot): BREAKING: President Trump announces the U.S. will be placing tariffs on all semi-conductors and pharmaceuticals imported from 🇹🇼Taiwan in the very near future
Thread, continued, 03:18 UTC · 8 likes
If so, this take aged well
Thread, continued, 03:19 UTC · 5 likes
Without knowing virtually anything, I kinda suspect congress wont let this happen
It seems weirdly obviously bad and not culture warry enough that trump gets a free pass
Reply to his own earlier tweet, 03:37 UTC · 6 likes
Aaron Bergman, January 28, 2025: I think this is important evidence that obesity is a setpoint disorder and not “rational” overconsumption (bc of cheap tasty food, hyperbolic discounting, etc)…
Wait this seems to be basically wrong sorry for spreading misinformation
Quote tweet, 04:08 UTC · 30 likes
Avg EA org: “that’s it?”
Quoting Bryan Johnson (@bryan_johnson): We posted a few Blueprint team openings on Friday and received over 5,000 applications in 48 hours. The energy behind dd is insane. I am continually underestimating. For those of you who want an update, here's the status of things:
1. Blueprint is growing incredibly fast. We'll hit a $100 million run rate soon, only 10 months in business. We need to add 20 talented people to the team asap. Most importantly, we need a President so that I can get out of the deep trenches and start working on the longer term goals. I've bootstrapped the company, personally investing $25 million. We are breakeven.
We're trying to make it incredibly simple for you to get jacked and fit. All your food and biomarker measurements in one place plus a supportive (dd app). And also easy for people who just want to dip a toe into this new game.
2. Don't Die (dd) - this is where my heart is. I think dd is the final boss ideology. It starts with entropy and then walks up the stack of math, computation, biology, storytelling and memetics. The goal is that dd is the world's most influential ideology by 2027. This needs to happen given the speed of AI progress. dd is the alignment solution.
The dd community (app) is growing fast. We're building a dd rewards/currency that can be earned and others can offer as incentive.
For the growth of the ideology, we have a rough draft plan that we'd like to just make public as a loom in the next 30 days. We've tried for so long to figure out how to package it properly given how hard it is to understand. I think we're just going to share the raw feed with you and ask for everyone's help.
3. dd certified - we're going to build out the world's largest repository of tested foods so that you can quickly and easily know what you're eating. We'll test for heavy metals, agrochemicals, label accuracy, microplastics and more.
I'm incredibly sympathetic to companies in the food and supplement industry because it's fucking hard to navigate the toxic global food supply. I've been doing this at Blueprint for the past 18 months and man it's complicated. There is so much you don't control yet are responsible for. The goal is that by June of this year dd certified is the world's largest and most trusted source for food testing.
4. Personal biomarkers - when building my previous company Braintree Venmo, I could just grind and let my health slip. In some ways, that was easier.
Now, I'm the CEO of multiple companies (and trying to be a pretend philosopher) and maintain the world's best biomarkers. That in of itself is a full-time job. It's an odd tension of working every second of the day and also not get stressed, lose sleep and let any biomarker slip even a little.
So I guess this is all a cry for help. If you're into this mission, we need you. Our team is about 30 people right now and we're doing the work of 100 or more.
We need strong leaders who can recruit and build exceptional teams. We also need creatives who can help us give birth to the next global ideology.
I only have 30 minutes now before I need to start my wind down routine and then I'll be offline for 10 hours or so. I'll try to answer any questions before then.
Reply to @gleech, 04:26 UTC · 2 likes
gavin leech (Non-Reasoning) (@gleech): What is the biggest cognitive feat that boredom at school drove you to?
I used to try to predict the answer to standardized test questions before reading the question
Iirc I could hit like 50% on 4 answer choices
Quote tweet, 06:06 UTC · 2 likes
Reckless incompetence is zero sum in the culture war but negative sum elsewhere (and all things considered)
Quoting Charles Lister (@Charles_Lister): NEW -- #Trump's global aid freeze has cut the salaries paid to many of the prison & camp guards responsible for securing 9,500 #ISIS militants & ~40,000 associated women/kids in northeast #Syria.
Many are no longer turning up for work.
Tweet, 06:30 UTC · 1 like
Do I know any fluent/native French speakers I can ask a quick question?
Tweet, 09:44 UTC · 3 likes
Empirical question someone can test: can r1 tell when its chain of thought has been tampered with? Completely replaced?
The architecture allows for this right?
Quote tweet, 15:45 UTC · 10 likes
I am a simple man. David Chalmers drops an AI paper, I read
Quoting David Chalmers (@davidchalmers42): a draft paper (for an invited talk at AAAI next month) with a philosophical analysis of work on mechanistic interpretability, with special attention to methods for propositional interpretability.
Tweet, 16:23 UTC · 6 likes
A bit of background to DeepSeek: language model algorithms have gotten 2^18 ≈ ~260k times better (in the sense of being equivalent to using 260k times more compute) since 2012 according to @EpochAIResearch (link below)
Thread, continued, 16:23 UTC · 4 likes
https://arxiv.org/pdf/2403.05812
Recent DeepSeek innovations are like <100x improvements right? Isn’t this more or less business as usual?
Reply to @ycombinator, 16:27 UTC · 4 likes
Y Combinator (@ycombinator): XB-1 is a go for supersonic
How do I get a ticket
Reply to @davidchalmers42, 17:18 UTC · 4 likes
David Chalmers (@davidchalmers42): a draft paper (for an invited talk at AAAI next month) with a philosophical analysis of work on mechanistic interpretability, with special attention to methods for propositional interpretability.
Haven’t verified quality yet but here’s an audio version for anyone https://open.spotify.com/episode/0Ah4epzJzzE1V2Va6OX3W9?si=uJoO0tQmSceX2Dl4OlKZeg
Reply, 18:51 UTC · 2 likes
Idk if these are any good but
Thread, continued, 18:51 UTC · 2 likes
Thread, continued, 18:53 UTC · 2 likes
Quote tweet, 21:35 UTC · 11 likes
Text: well there’s a lot of text, you can read it
Subtext: one of the greatest minds in philosophy thinks this topic is worth pivoting to
Quoting David Chalmers (@davidchalmers42): a draft paper (for an invited talk at AAAI next month) with a philosophical analysis of work on mechanistic interpretability, with special attention to methods for propositional interpretability.
Quote tweet, 22:50 UTC · 3 likes
Pigeon hour takes the White House
Quoting Autism Capital 🧩 (@AutismCapital): 🚨BREAKING: The White House is opening up the Press Briefing Room to new media, journalists, podcasters, and social media influencers‼️
Wednesday, January 29, 2025
Reply to @sighswoon, 03:38 UTC · 95 likes
sigh ❀ swoon (@sighswoon): When I was dating, there were multiple times where I had to tell someone "You don't need to use therapy language with me, I am not going to cancel you, please say exactly how you feel, that is what romance is about..." and then I'd finally get the real person under that
This would fix me
Tweet, 04:24 UTC · 5 likes
Poast Stanford Encyclopedia of Philosophy article/page that lay philosophy nerds might enjoy (inspired by @arthur_wright_)
Tweet, 04:25 UTC
I’m personally interested in pages about claims, rather than people or concepts
Thread, continued, 04:33 UTC · 1 like
This seems to be the closest match for “illusionism”, which I respect for (1) actually making a contentious claim and (2) not being as easily dismissed as one might first think, although I do think it’s wrong https://plato.stanford.edu/entries/materialism-eliminative/#:~:text=Illusionism claims that introspection involves,actual nature of our inner
Reply to @cremieuxrecueil, 04:38 UTC · 2 likes
Crémieux (@cremieuxrecueil): Let's repeat this poll now that everyone's had time to read.
Do you believe DeepSeek's $5.5m training run figure?
Yeah but <90%
Quote tweet, 14:00 UTC · 75 likes
I feel chronically separated by 2-3 degrees of connection from insane shit
2 if you count weak ties, or 3 stronger ties
Quoting Ra (@slimepriestess): > don't do any AI research
> fake death
> reveal survival immediately
> kill friend's parents
> order a revenge hitwhat is she planning
Thread, continued, 14:00 UTC · 13 likes
There was also this crypto exchange you may have heard of
Quote tweet, 14:00 UTC · 7 likes
I’m calling for a total and complete ban on new high stakes PDFs until I can figure out what the hell is going on
Quoting Yoshua Bengio (@Yoshua_Bengio): Today, we are publishing the first-ever International AI Safety Report, backed by 30 countries and the OECD, UN, and EU.
It summarises the state of the science on AI capabilities and risks, and how to mitigate those risks. 🧵
Link to full Report: https://assets.publishing.service.gov.uk/media/679a0c48a77d250007d313ee/International_AI_Safety_Report_2025_accessible_f.pdf
1/16
Reply to @RokoMijic, 14:17 UTC · 11 likes
Roko 🐉 (@RokoMijic): @Benthamsbulldog Infinite kids, finite world. You cannot have infinite low-IQ African kid production. It has to stop somehow.
Humans aren’t near the Malthusian frontier anywhere on earth*, I think you just dislike them (“low-IQ African kids”)
*maybe some politically-driven famines make this not always literally true idk
Reply to @arithmoquine, 14:17 UTC · 4 likes
henry (@arithmoquine): @AaronBergman18 dude i'm followed by the latest killer
Fun
Reply, 14:25 UTC · 14 likes
I dislike trump/maga as much as anyone but cmon lol, it’s perfectly fine and reasonable to go to a party where a photographer looking to capture an insidious vibe could in fact take a
picture of you with vaguely insidious vibes
Quote tweet, 14:50 UTC · 13 likes
Aside from the object level which is kinda horrifying, this article is a good example of like “Wikipedia for thing that’s too niche/specific to actually be on Wikipedia”
Quoting Aella (@Aella_Girl): http://x.com/i/article/1884477365302222849
Quote tweet, 19:31 UTC · 39 likes
Wait what
Quoting Saagar Enjeti (@esaagar): I didn't realize people weren't aware that Lyme disease was probably a lab leak so here it is very simply: Lyme disease originated in Lyme, CT in the 1970s. Lyme, CT just happens to be mere miles away from a US lab studying insect spreading bioweapons in the 1960s
The end
Quote tweet, 19:38 UTC · 7 likes
Better sorting?
Quoting Derek Thompson (@DKThomp): The probability of divorce for married folks has been declining steadily for a long time, and in 2023, a record-low 1.4% of married adults got divorced
Tweet, 20:01 UTC · 7 likes
I guess I wanna know how many “suspicious thing” laboratories are there in random places
Like how sus is it really if there’s a 2025 Helsinki or Seville outbreak of some new pathogen? Are there institutes of [whatever] near many or most medium-large population centers?
Quote tweet, 20:05 UTC · 4 likes
Pretty important background info for lab leak stuff in general (if true, although I don’t super trust LLMs to be unbiased here)
Quoting Aaron Bergman 🔍 (@AaronBergman18): Wait what
Reply to his own earlier tweet, 20:07 UTC · 6 likes
Aaron Bergman, January 29, 2025: I guess I wanna know how many “suspicious thing” laboratories are there in random places Like how sus is it really if there’s a 2025 Helsinki or Seville…
Tweet, 20:22 UTC · 6 likes
Oof but yes
Thread, continued, 20:22 UTC · 2 likes
Reply to @johnarnold, 22:38 UTC · 13 likes
John Arnold (@johnarnold): This NYC taxi insurance story is bizarre. A boutique insurer captured 64% of market for livery/taxi drivers by offering rates below acturial cost. That couldn't last and the scheme fell apart last year. Unable to pay claims, it was declared insolvent.
The state proposed a rule that drivers needed insurance from a solvent company, one that could actually pay claims. Drivers panicked as sustainable premiums are much higher than they were paying.
Caving to ratepayers, the state withdrew the proposed rule and now says that the insurer just has to be licensed in the state, which the insolvent company still is. So rather than require drivers to pay market rates from a solvent company, drivers will pay less to the insolvent company that can't pay claims.
Premiums must reflect actuarial costs or the market will eventually fail. But regulators are so sensitive to raising rates that they repeatedly violate this rule.
Wait I feel like there’s money disappearing in this story
Drivers are paying in but not getting paid out? Or perhaps some are only partially paid out? Who are the creditors to the insolvent company?
Reply to @_chair, 22:44 UTC · 8 likes
Chair (@_chair): once made my landlord (i was on a house's top floor) pay for bedbug treatment bc i was semi poor & worried about having to pay for multiple treatments & law says owner pays. he caved quickly but was offended and sad and i still feel bad about it. nice old guy, relationship ruined
(quoting @Duderichy: at one point slimepriestess defends “people in their community” taking advantage of slow eviction processes to not pay rent idk maybe don’t defend the sort of people who do that? don’t associate with them either?)
Sounds like you did nothing wrong
Quote tweet, 22:54 UTC · 15 likes
Any decent society is gonna have a cultural pendulum, but ours seems to have too big of an amplitude
Quoting Reagan Reese (@reaganreese_): SCOOP: Trump is expected to sign an EO today prohibiting federal funds to any K-12 school that teaches CRT and radical gender ideology.
The order will also reinstate his 1776 Commission, which was originally created to teach a patriotic education.
Here first for @DailyCaller:
Thread, continued, 22:54 UTC · 6 likes
I’m sure we’re gonna hear horror stories about this, so I’ll recall @mattyglesias made the good point that there are a shit ton of US public schools/curricula/teachers and so absent a panopticon the craziest/worst .01% of classes are inevitably going to look really bad
Thursday, January 30, 2025
Quote tweet, 00:57 UTC · 5 likes
Yet another correct take from Kelsey
Quoting Kelsey Piper (@KelseyTuoc): Yeah, all right, let's talk about James Damore. It's been eight years, and I really doubt Harj (who was my boss at the time) is the only person for whom it was a formative experience.
For those of you who have no recollection of any of this, either because you are wisely an offline person or because you got outraged for five minutes and then forgot all about it, James Damore was a Google software engineer who wrote a memo arguing that, while diversity and inclusion were good goals, bias was not the main reason there weren't more women in tech, and differences in personality between men and women probably explained a lot of it.
There was a lot in the memo that felt like a distraction to me, or where I had a nitpick, but fundamentally it was not only basically correct (women are less likely than men to become software engineers, and this is not only because of bias), but also Damore was saying this for the sake of having a more productive conversation about how to get more women into tech, a goal that everyone around him was fervently espousing. The memo has a painful-in-hindsight quality of earnestness: "you want more women in tech, and I think you're mistaken about how to get there! if I show you some published psychology research we can actually design better means to your goal!".
Anyway. The internet was outraged. He got fired from Google. And he applied to the tech hiring startup I worked at, Triplebyte, which offered background-blind screening to anyone who wanted to be a software engineer. We really believed in the mission, at Triplebyte. I think I ended up kind of badly calibrated about how earnest to expect people to be everywhere else. We found people working as janitors and line cooks and homemakers who could code, and we got them 6 figure jobs, and we were proud of it. James Damore did very well on our tests. I got assigned to write him a profile for our companies.
And then people freaked out. A lot of them had the impression that he would create a hostile environment for any woman he worked with, and thought that trying to help him get a new job was tantamount to endorsing everything in his wildly controversial memo. I didn't even like the memo that much, but I was kind of horrified, because - it's one thing to get fired for talking about politics at work in a way that causes a massive national firestorm. I kind of expect that we would all get fired for that. But it is another thing entirely to get effectively blacklisted from your industry, to have people decide on the basis of your political opinions that we shouldn't even put you up on the platform and let companies decide individually whether to schedule interviews. Tech jobs were not that hard to come by in 2017 if you were really good at your job, and Damore was. Firing isn't that threatening to software engineers. Blacklisting is terrifying.
I'd been at Triplebyte for like six months at this point, it was my first job after graduation, and I was honestly way out of my lane, but I made a pretty big fuss internally. (It helped that I suspected a lot of people agreed with me but I was a woman and it was safer for me to say it.) I said that we were not in the business of deciding who had good politics, that we shared this country with many people who profoundly disagreed with each other, that companies could assess for themselves if he worked respectfully with female engineers, and that we should put him on the site and let them decide. We did. And then Harj was immediately contacted by recruiters from companies we worked with that were horrified that we had. They felt that by not banning him from our platform we were endorsing his memo, that we were showing values not in line with their priorities. Harj talks about this more in the linked podcast.
James Damore was egregiously wronged. To my knowledge he's a good software engineer with extremely reasonable, approximately accurate opinions about the reasons there were fewer women in software engineering, which he shared in good faith, and a lot of people who should've known better really did try to drive him out of the industry for it. It was wrong. If it is done to people on the basis of any other political opinion it is also wrong then. We need, as a society, the ability to live with disagreement, to dislike each other without trying to destroy each other, to find common ground instead of finding heretics; I believed that at Peak Woke and I believe it now.
Quote tweet, 01:05 UTC · 500 likes
Wait this is a gazillion times more banal than I expected. Pretty sure I never even bothered reading the thing bc I expected something way more hostile, and generally much lower quality (based on discourse at the time)
He got fired *for this*?! Lol jfc
Quoting Kelsey Piper (@KelseyTuoc): read the memo here: https://web.cs.ucdavis.edu/~koehl/Teaching/ECS188/PDF_files/Damore.pdf
Reply to @Mjreard, 01:29 UTC · 1 like
Matt Reardon (@Mjreard): Helpful reminder that export controls are deeply irresponsible if you think AGI is impossible
(quoting @krishnanrohit: We are actively manifesting a war)
Because it’s provoking a war? I guess I agree that’s directionally true but doesn’t seem provocative enough to warrant abandoning an important policy lever
Like I can imagine using export controls as a pretty good non-escalatory, credible threat about what will happen if they invade Taiwan, totally separately from AGI (and I actually kinda like this idea, it’s plausibly way safer than virtually any military response)
Reply to @Mjreard, 01:37 UTC · 2 likes
Matt Reardon (@Mjreard): You're agreeing with my point? It's irresponsible because it's escalatory. It's generally bad to initiate escalation. Here, we've done the export controls unprovoked (and now China has less to lose be e.g. invading Taiwan)
Oh, I guess I am accidentally
I was just trying to give an example of how export controls could be justified unrelated to AI
The less leverage thing is a good point (although in principle we could go way farther than we have right)
Reply to @StatisticUrban, 01:58 UTC · 20 likes
Hunter📈🌈📊 (@StatisticUrban): This is actually extremely impressive. Louisiana is very poor and in 2019 was at the bottom of the pack in every metric (two 50/50s, one 49th, and a 45th.)
They've made enormous progress, and we should give credit where it's due.
image attached
(quoting @LAGovJeffLandry: )
Possibly a narrative violation for some of the demographics = destiny + race realism folks
Louisiana is 2nd most black state (and 13th least white) https://en.m.wikipedia.org/wiki/List_of_U.S._states_and_territories_by_race/ethnicity
Reply to @Dubflip, 02:52 UTC · 8 likes
Frogo (@Dubflip): Even as a joke that isn’t something you should say
Lol I was just thinking “it’s kinda mean either way”
Reply to @KelseyTuoc, 03:03 UTC · 6 likes
Kelsey Piper (@KelseyTuoc): I'm an obsessive consumer of plane crashes and it's one of my worst qualities because it means that when I hear about a horrible tragedy I immediately start Googling "do DC police copters have TCAS"
Same, I’m kinda obsessed with planes in general but this is my semi-guilty morbid curiosity thing
Reply to @_chair, 03:06 UTC · 5 likes
Chair (@_chair): @goblinodds noooo that one is real people are going to think it's fake
Lol sad but true (possibly)
Reply to @goblinodds, 03:06 UTC · 12 likes
𝖜𝖔𝖒𝖇𝖑𝖊 𝖜𝖎𝖙𝖍𝖔𝖚𝖙 𝖆 𝖈𝖆𝖚𝖘𝖊 (@goblinodds): @_chair lolllll wait wdym
Like assume it’s a (wrong) conspiracy theory bc RFK is discussing
Reply to @arnoblalam, 03:09 UTC · 19 likes
Arnob 'not home for the holidays' Alam🌐🇺🇸🇧🇩 (@arnoblalam): @MaxGhenis I thought you were lib? 🤔
Lol there’s a solid (at least partially rationalist-adjacent) lib-and-also-into-getting-things-right bloc
Speaking just for myself I consider myself liberal and slightly anti-woke
Tweet, 03:45 UTC · 2 likes
This is one of the less important reasons to do LLM interpretability but it potentially allows for fine tuning an LLM to accurately report the reasons for its beliefs/other outputs (which they don’t do now, see recent Chalmers paper)..
Thread, continued, 03:45 UTC · 2 likes
Such a model would be the first mind of its kind with what seems to me like a pretty fundamental and important ability
(tbc I don’t think even humans trying to be maximally honest can do this at a level beyond “idk that’s what my brain decided for opaque reasons”)
Tweet, 04:27 UTC · 366 likes
Now that anti-woke has won the culture war, I’d like to announce that kindness is still good and causing people to be harmed (yes even emotionally, and even if you think they should suck it up) without good reason is bad
Thread, continued, 04:27 UTC · 30 likes
More concrete example is good:
Even if you think a full social taboo on the word is bad and it’s generally fine to call a book or something “retarded”, it’s still bad to use that word *in order to actually hurt someone*
Thread, continued, 04:27 UTC · 22 likes
The problem here isn’t the word or wokism—it’s that you’re harming someone, and the word being associated with overzealous language policing in other circumstances is irrelevant
Thread, continued, 04:27 UTC · 27 likes
Ofc this isn’t a full theory of social ethics - what constitutes “good reason” is a big debate to be had
But I don’t think it’s trite or meaningless - there’s real positive-sum alpha in trying to be a nice person
Reply to @pli_cachete, 04:32 UTC · 4 likes
Rota Is In Israel🚪🧎♂️ (@pli_cachete): @AaronBergman18 What if you think they should suck it up precisely in the situations where being locally harmed would be good for them long term
I mean conditional on the last bit, this seems straightforwardly and almost tautologically good (all else equal)
Whether or not they should suck it up now is irrelevant (again conditioning on being good for them; these things probably correlate)
Reply, 07:12 UTC · 3 likes
(The tweet being replied to is no longer available.)
I think in this case they basically don’t (reminds me of how increasing prison sentences basically doesn’t work as a deterrent) but it’s an empirical question
Even if it does work in this case, I think it’s very unlikely to be net-positive even for supposed “beneficiaries” bc like the option space for them has gotten strictly worse. A rational actor model says that if you make option A worse so people switch to B, which is unchanged, you’re making them worse off (this would look like someone getting bullied into losing weight but being hungry all the time and overall worse off than if fat-and-not-bullied was an option)
Reply to @ilex_ulmus, 14:48 UTC · 8 likes
Holly ⏸️ Elmore (@ilex_ulmus): I'll be civil, but make no mistake: if you work at an AI lab, you are part of the problem and you suck.
Copenhagen theory of ethics IMO. The world where nobody who cares about safety has any direct power or influence at a lab seems bad and probably worse than the status quo
Leopold seems like a model here - do the most you can there, and if they fire you so be it
Reply to @beala, 18:34 UTC · 2 likes
Alex Beal 🆎 (@beala): Uhh I just got into a heated argument with gemini-2.0-flash-thinking and I think it won? wtf?
Wanna share what it was about? Lol
Reply to @chunkbardey, 18:42 UTC · 223 likes
charlie (@chunkbardey): causing innumerable problems in the CVS by wanting to purchase the items they have there
Quote tweet, 18:53 UTC · 6 likes
New pdf just dropped
Quoting Tomek Korbak (@tomekkorbak): We’re excited to start a discussion between frontier AI labs, government agencies and civil society about what safeguards should be in place for autonomous LLM agents and what constitutes sufficient evidence for their effectiveness.
https://arxiv.org/abs/2501.17315
Thread, continued, 18:55 UTC · 3 likes
Sadly
Reply to @esjesjesj, 19:01 UTC · 4 likes
evan loves worf (@esjesjesj): I just can’t get over how dangerous it will be to have this Nazi moron gutting agencies like the FAA and filling them with other Nazi morons. He’s literally going to cause plane crashes.
Uh
Reply, 19:27 UTC · 1 like
I think I’m just less pessimistic about this last point (although not super optimistic in an absolute sense)
Reply to @gwern, 19:28 UTC · 14 likes
(The tweet being replied to by @gwern is no longer available.)
Jfc
Quote tweet, 19:37 UTC · 13 likes
Forbidden csv
Quoting mattparlmer 🪐 🌷 (@mattparlmer): Apparently there were cuneiform spreadsheets
Truly nothing new under the sun lol
Quote tweet, 19:55 UTC · 2 likes
This is good, but separately I think it shows what at least a bunch of senators really care about (ie American geopolitical dominance)
There’s a lot of political posturing, but congress isn’t entirely cynical
Quoting Walter Kirn (@walterkirn): The cowards are going to vote on Tulsi in secret
Breathtaking
Friday, January 31, 2025
Quote tweet, 00:11 UTC · 9 likes
Unironically big if true
Quoting vittorio (@IterIntellectus): this could be one of the biggest breakthroughs in pain management in decades.
journavx (suzetrigine) is the first-ever non-opioid, oral painkiller in a completely new drug class. instead of numbing the brain like opioids, it blocks pain signals before they even reach the brain
until now, pain management has been a tradeoff between suffering or risking addiction. opioids work, but they come with half a million dead from overdoses in the last 20 years.
now, for the first time there's a real alternative.from the study:
-proven effective in clinical trials for post-surgical pain
-non-addictive, avoiding the dangers of opioid dependency
-fast-tracked by the FDA, signaling its potential impact
-could replace opioids for acute pain (40m americans get opioid prescriptions every year)
-priced at $15.50 per pillif this drug really works it could be the end of opioid dominance in pain treatment. fewer overdoses, fewer ruined lives, and a fundamental shift in how we treat pain.
this isn’t just a medical breakthrough, it’s a massive cultural reset.
Tweet, 02:57 UTC · 7 likes
Trumpian chaos and incompetence is bad enough now, but it’s going to be especially bad if and when there’s a crisis
I hope I won’t be retweeting this saying “this take aged well,” but I expect to
Reply to @slimepriestess, 05:07 UTC · 2 likes
Ra (@slimepriestess): i'm looking at some of the news coming out of DC on the midair collision and it's just...horrifying, a near worst case scenario. this is sort of event that makes flight safety engineers lie awake in fear, which gets referenced with phrases like Tenerife, Closed casket, pink mist.
Do you mean worst case as in “a lot of very unlucky things, no egregious errors”?
Tweet, 05:25 UTC · 4 likes
One kinda funny but maybe actually good thing Elon et al could do to the federal government, especially if they’re going all-in on in-person, is make the offices aesthetically more like those in Silicon Valley
Can you guess which is which? Lol
Tweet, 05:25 UTC
I invite anyone from any sector to post pictures of their office here for reference
Thread, continued, 05:29 UTC · 1 like
I do think there’s a more enlightened version of DOGE that chills a bit about finding negligible programs with the wrong keywords to cut and tries to implement some basic productivity/performance interventions inspired by the private sector
Reply to @sama, 06:53 UTC · 54 likes
Sam Altman (@sama): why not 100x?
(quoting @paulg: This may be the most inspiring sentence I've ever read. Which is interesting because it's not phrased in the way things meant to be inspiring usually are.)
In all seriousness, you end up writing checks you can’t cash and/or hurt people by accident
Thread, continued, 06:56 UTC · 8 likes
This wasn’t intentionally an AI safety tweet, but it does happen to apply
Agency is probably good in general, most of the time. But it’s not a law of nature that bending the world to your will turns out ok
Reply to @tracewoodgrains, 13:45 UTC · 2 likes
Jack (@tracewoodgrains): I realize this pattern-matches to tiresome things but in this case there is an actual, legitimate scandal at the root, one that remains not fully resolved to this day, and sighing doesn’t change it. new liberals should take the lead in facing legitimate institutional problems
Yeah I think woodgrains is spot on here
Reply to @austinc3301, 14:30 UTC · 3 likes
Agus 🔸 (@austinc3301): I am a machine that turns caffeine pills and Huel into Slack messages that are saved for later
I’m Aaron
Reply to @absurdlymax, 14:31 UTC · 2 likes
Max Alexander (@absurdlymax): Your family is insane in all the right ways
Fr
Quote tweet, 18:57 UTC · 7 likes
Sam Francisco:
Quoting mass (@Memetic_Theory): they took the swing down. apparently fun is illegal
Reply to @caesararum, 19:49 UTC · 5 likes
caesararum, BS, DOGS (@caesararum): what are some good jobs where high IQ professionals commonly, in the course of their work, have to deal with low IQ people?
("IQ" here is me handwaving at "ability to navigate complex systems")
big one that comes to mind: criminal lawyers
what else?
Doctor
Thread, continued, 19:52 UTC · 3 likes
This is gonna sound kinda narcissistic but I started wondering how and why several doctors have said something like “you’re a smart guy” after what I perceive as pretty normal and brief interactions
I suspect they get a pretty wide sample of how people talk and such
Quote tweet, 20:04 UTC · 4 likes
Pretty based Claude
Quoting Ryan Greenblatt (@RyanGreenblatt): We actually followed through and donated the $4K as requested:
- $2K to CHAI
- $1.5K to ACE Recommended Charity Fund
- $500 to EA Global Development FundThis wasn't just an experiment—we wanted to set a precedent for making good faith deals with AI systems.
Quote tweet, 20:08 UTC · 9 likes
EA getting attacked by the progressive left, nrx right, and normie center
Quoting Alex Trembath (@atrembath): Become unclassifiable.
Quote tweet, 20:09 UTC · 65 likes
AI better than 99.99% of humans at philanthropy
Quoting Ryan Greenblatt (@RyanGreenblatt): We actually followed through and donated the $4K as requested:
- $2K to CHAI
- $1.5K to ACE Recommended Charity Fund
- $500 to EA Global Development FundThis wasn't just an experiment—we wanted to set a precedent for making good faith deals with AI systems.
Quote tweet, 20:53 UTC · 9 likes
As I’ve said before, I’m not against affirmative action in principle, but progressives should ask themselves what the end goal is of hiring/admissions policies that are essentially preference purely on the basis of race
Quoting Jack (@tracewoodgrains): How did the FAA strike a balance between cognitive testing and adverse impact over four decades? What trade-offs did they think about between performance and diversity?
Walk with me. This will be a technical but important look at the theory underpinning the 2014 change.
Thread, continued, 20:53 UTC · 3 likes
This is what it looks like when you’ve totally lost site of any plausible terminal goals—both those that I basically endorse (eg correcting for “old fashioned racism”) and those that I don’t but aren’t totally insane (eg elites need to be diverse to be seen as legitimate)
Thread, continued, 20:53 UTC · 2 likes
As happens with human culture and society, reasonable proxy and instrumental goals seem to have been effectively reified into terminal values
Thread, continued, 20:53 UTC · 3 likes
This specific instance of whatever you want to call it (eg “DEI”) is conveniently well-documented, egregious, and clearly plausibly harmful, but afaict nobody thinks it’s fundamentally unique
Tweet, 21:05 UTC · 2 likes
There’s a meme of like “plane crash and terrorist attacks get way too much attention,” and in a relative sense this is true…
Thread, continued, 21:05 UTC · 3 likes
but I actually think that the level of concern for eg 64 deaths in a plane crash in a major American city is closer to “right” in an absolute sense
Thread, continued, 21:05 UTC · 9 likes
This is to say, the correct reasoning is not:
well a gazillion people die untimely deaths every day -> plane crash is nbd
But rather:
plane crash is really tragic -> holy shit the world is filled with an unfathomable amount of tragedy
Reply to @d4wsonc, 21:21 UTC · 3 likes
dawson🕊️ (@d4wsonc): I've been single and on hinge for 5 interrupted years so they're bringing me in for testing
F
Reply to @tracewoodgrains, 21:35 UTC · 3 likes
Jack (@tracewoodgrains): ugh I'm really going to have to figure out how to dig up year-old Twitter DMs so I can find all the air traffic controllers who reached out to me about the hiring scandal last year
why is this site's DM interface so horrifyingly bad
Export data -> LLMs help you extract the right stuff (probably by writing code to do regex tbc)
Tweet, 22:29 UTC · 18 likes
It legitimately sucks that there has never been and will never be justice for American slavery or segregation
Thread, continued, 22:29 UTC · 8 likes
I think in part progressive institutions aren’t willing to concede that this is no longer something that can really happen
The old fashioned racists (people and institutions) (mostly) died before they could repent; they got away with it
Reply, 22:43 UTC · 1 like
It would have looked like reconstruction basically actually being implemented after the civil war, ideally including some cultural stuff to solidify it as a sin in the American imagination the way Germany did with the holocaust
Reply, 22:49 UTC · 1 like
Nice, I think I’ve changed my mind on this. I used to be solidly pro-reparations but now don’t think it’s a good use of progressive political capital…
Thread, continued, 22:50 UTC · 2 likes
That’s kinda a cop out though; if it made sense to ignore pragmatic political stuff, I’d say yeah all black people probably on average are harmed some amount from the lingering intergenerational effects of stuff, and we should set up a working group of economists to figure out how much that is
It’s a little silly that the salient debate is “reparations, yes or no?” Instead of “how much?”
End of January 2025. Previous month: December 2024. Next month: February 2025. Back to the archive index.