Aaron Bergman's tweets, December 2024
Every public tweet, reply, quote tweet and retweet that Aaron Bergman (@AaronBergman18 on X, formerly Twitter) posted in December 2024 and has not deleted: 654 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: November 2024. Next month: January 2025.
Sunday, December 1, 2024
Quote tweet, 00:05 UTC · 5 likes
Interesting possible economic phenomenon I just thought of:
If your product is clearly distinguished as best in class, you can price this way, so we should expect a disproportionately large gap between the top-end item and runner up
Quoting Aaron Bergman 🔍 (@AaronBergman18): @AndyMasley $39/month for digital news is crazy
But pricing based on some small % of potential users who are basically totally insensitive to price makes total sense and I’m actually surprised I don’t see more like this
Thread, continued, 00:05 UTC
I suspect it has to be something like an obvious (if implicit) social consensus for this to work
As soon as it’s ambiguous, price can make the difference
Thread, continued, 00:05 UTC · 2 likes
Apple’s best monitor is $7k because fuck it you probably have infinite VC/tech money so why not
Presumably this is just obviously what you buy if price is no object and you use a mac
Thread, continued, 00:05 UTC · 2 likes
This is also distinct from a product being a luxury/signaling good, where the price is actually doing much of the work of the product
Tweet, 01:14 UTC · 11 likes
Now that LLMs are a thing I'm discovering I really reinvented the wheel in R
There are packages for all my custom functions :(
Tweet, 04:07 UTC · 32 likes
*2028 Presidential debate*
Moderator: candidates, what is your plan for bringing down the price of eggs, sent sky high by BattleGPT-9o-mini-preview-7-16-28 setting the Western half of Nebraska on fire?
Tweet, 07:53 UTC · 5 likes
Guess I need a $7000 laptop
Reply to @RosieCampbell, 11:47 UTC · 2 likes
Rosie Campbell (@RosieCampbell): This was my last week at OpenAI after ~3.5 years. I worked on API safety and then transitioned to Policy Research where I worked closely with @Miles_Brundage on AI governance, frontier policy issues, and AGI readiness.
As is tradition, here’s the message I shared on Slack:
Good message, thanks for sharing
Quote tweet, 11:51 UTC · 16 likes
There does seem to be a bit of a pattern
Quoting Rosie Campbell (@RosieCampbell): This was my last week at OpenAI after ~3.5 years. I worked on API safety and then transitioned to Policy Research where I worked closely with @Miles_Brundage on AI governance, frontier policy issues, and AGI readiness.
As is tradition, here’s the message I shared on Slack:
Thread, continued, 11:51 UTC · 5 likes
Lol can you imagine being like a normie (non-pejorative) journalist or something with minimal context and trying to figure out wtf is going on at this company?
Quote tweet, 12:06 UTC · 9 likes
New wild animal welfare intervention
Quoting Someone:
Tweet, 12:12 UTC · 86 likes
Mars is a dumb thing to make the cultural proxy for futurism/techno-optimism
Thread, continued, 12:12 UTC · 14 likes
I mean I see the aesthetic appeal but cmon you may as well choose something actually good
Reply to @BecomingCritter, 19:29 UTC · 2 likes
critter (@BecomingCritter): Men need to be bullied
(quoting @oldbooksguy: Men don't need therapy. Men need initiation rites.)
Lol this is the worst possible response to a set of legitimate concerns
Quote tweet, 20:42 UTC · 120 likes
Quoting 🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): So called disability activists are some of the worst ppl alive
Thread, continued, 20:42 UTC · 16 likes
I consider myself disabled* btw, I'm allowed to poast this
*Hopefully not forever
Quote tweet, 21:28 UTC · 51 likes
I think this kind of thing at ~this level is pretty common and most people just don’t care that much about understanding and also don’t want to look dumb
Quoting gon (@chinesegon): no one:
guys getting paid 3x your salary:
Thread, continued, 21:28 UTC · 19 likes
I do think I understand area pretty well
But not like the Pythagorean theorem at the same deeply intuitive level which isn’t *that* much more complex
Thread, continued, 21:30 UTC · 36 likes
I got cancelled not long ago for not being that familiar with Dostoyevsky. Good times
Reply to @MarkovMagnifico, 21:35 UTC · 2 likes
Markov (@MarkovMagnifico): what would you propose instead? mars is a lofty and far-off goal but it’s going to be one of the greatest milestones in humans reaching out beyond earth
Idk I say decouple aesthetics and importance - spend $9.9B to abolish suffering and then spend $100M on a flying monument instead of $10B on Mars…
Quote tweet, 21:37 UTC · 12 likes
Based everyone plz give him status and acclaim
Quoting Matt 🔸 (@SpacedOutMatt): I've finally taken the @givingwhatwecan 10% pledge, backdated to when I first started giving in 2019! I'm a huge believer in the power of effective giving to improve the world, and I think the pledge is a great tool for both committing and spreading the word!
Reply to @nedsore, 21:40 UTC · 2 likes
ned! (@nedsore): Single oomfies now’s your chance
Should I actually
Reply to @thewreathwright, 21:42 UTC · 2 likes
Laurel Weaver (@thewreathwright): @AaronBergman18 @MarkovMagnifico can you propose a plan for total abolition of suffering for that price tag?
No lol
I guess I mean spend the first $9.9B and see where that gets us
Quote tweet, 22:00 UTC · 7 likes
I dropped $10 on high RAM cloud compute so I could make the most comprehensive panel data that I know of and see what else there is to find
Drive link below
Quoting Aaron Bergman 🔍 (@AaronBergman18): Getting sniped again. Did everyone realize that (at least at the population level between 1990 and 2010), % of calories from carbs is a super robust predictor of overweight/obese prevalence?
Thread, continued, 22:00 UTC · 3 likes
Go nuts. For obvious reasons even the middle sized dataset has a lot of missing values https://drive.google.com/drive/folders/1gF4914To2Wu569L6TwE56t1dyxWm3WvF?usp=sharing
Thread, continued, 22:00 UTC · 2 likes
Conceptually it feels like there should be a way to start with <300MB of data and wind up with a 1.4GB merge using just 16GB RAM but Claude and I couldn’t make it work (the obvious thing you’re thinking of I probably tried)
It wound up taking like 35GB IIRC
Monday, December 2, 2024
Reply to @arithmoquine, 00:48 UTC · 4 likes
henry (@arithmoquine): @AaronBergman18 incredible, did you make this?
Ya lol
Reply to @NathanB60857242, 01:02 UTC · 6 likes
Nathan Barnard (@NathanB60857242): does anyone have strong nuclear winter takes
Yeah nuclear winter sounds bad
Reply to @SpacedOutMatt, 01:57 UTC · 3 likes
Matt 🔸 (@SpacedOutMatt): I think u can be extremely based and happy being deaf
Fair enough, I regret using the word “fixed” twice. Would have been better to just keep it on the bounty thing
Reply to @_candroid, 06:18 UTC · 4 likes
robot.chihuahua (@_candroid): @AaronBergman18 FWIW I was at a patient conference earlier this year and the general vibe was much closer to the one on the right.
Based
Reply to @LinchZhang, 06:18 UTC · 9 likes
Linch (@LinchZhang): Pardoning Hunter is just unambiguously bad right? No silver lining.
Well it’s good for hunter
Quote tweet, 15:26 UTC · 4 likes
Send this to Her
Quoting Enron (@Enron): We're back. Can we talk?
Reply to @menhguin, 16:48 UTC · 3 likes
Minh Nhat Nguyen 🦭 (@menhguin): men only want one thing and it's ending up in the LLM training data
Kinda sad that I’m like the 6th most famous Aaron Bergman online
Reply to @menhguin, 16:57 UTC · 4 likes
Minh Nhat Nguyen 🦭 (@menhguin): bro u barking up the wrong tree
Yall need some new surnames, it’s not that helpful if 40% of your friends have the same one
Reply to @willmacaskill, 16:59 UTC · 11 likes
William MacAskill (@willmacaskill): I have a new book out! "An Introduction to Utilitarianism", co-authored with Richard Chappell and Darius Meissner.
It all started in 2016: I was giving the introduction to ethics lectures at Oxford and I was frustrated by how utilitarianism was presented to students in introductory materials...
https://www.amazon.co.uk/Introduction-Utilitarianism-Theory-Practice/dp/1647922003
Someone read it rather quickly
Quote tweet, 16:59 UTC · 3 likes
Correct take
Quoting William MacAskill (@willmacaskill): Utilitarianism would often be wholly misunderstood, or presented in naive and simplistic terms. The arguments in favour of utilitarianism, the rebuttals to criticisms, and the implications of the debate for moral methodology would all be neglected.
Reply to @mealreplacer, 19:33 UTC · 4 likes
(The tweet being replied to by @mealreplacer is no longer available.)
Rip
Quote tweet, 19:38 UTC · 7 likes
Is it conceptually plausible that naive scaling doesn’t work (ie lead to arbitrary gains in intelligence)?
I don’t really know what I’m talking about, but after armchair philosophizing a bit my tentative answer is “no at the limit; maybe at the margin”
Quoting Benjamin Todd (@ben_j_todd): 1/ It's more likely that "naive scaling" of LLMs (just adding a zero to everything) is hitting diminishing returns.
That's a small negative update on timelines - I thought naive scaling might work.
Thread, continued, 19:38 UTC · 2 likes
At the conceptual limit of compute and parameters, I don’t see how you avoid the model building a genuinely superintelligent understanding of the physical and social world
That’s the way to best predict tokens: understanding the process that generated them out in the world
Thread, continued, 19:38 UTC · 2 likes
But it seems pretty plausible, maybe likely, there’s a very large gulf between the regimes in which models are and are not fundamentally transcending the level of human concepts encoded in language—
Thread, continued, 19:38 UTC · 2 likes
e.g., accurately modeling the cognitive, maybe neurological, processes that literally caused a particular writer to write some words on the internet in 2005
Thread, continued, 19:38 UTC · 2 likes
If there is such a gulf, then it seems plausible that there’s a pseudo-plateau around sorta-kinda-still-humanish level intelligence
Thread, continued, 19:38 UTC · 2 likes
I’m sure this point was made in a LessWrong comment the day after GPT-2 dropped
Quote tweet, 19:46 UTC · 5 likes
She earned a follow 🫡
Quoting Cornelius Singaporensis (@bryancsk): She's not deleting, she's not going down, she's wrestling in the mud with the replies. 30M views and she's facing everyone with a smile and nothing but her metaphorical sword in hand. A worthy poaster!
Quote tweet, 20:45 UTC · 21 likes
Showing up lots on my TL - hate to kill the vibe but this has got to be so competitive that it feels somewhere between disingenuous and silly to QT it encouraging your audience of 100+ people to apply
Quoting Anthropic (@AnthropicAI): We’re starting a Fellows program to help engineers and researchers transition into doing frontier AI safety research full-time.
Beginning in March 2025, we'll provide funding, compute, and research mentorship to 10–15 Fellows with strong coding and technical backgrounds.
Thread, continued, 20:45 UTC · 4 likes
As happens often on twitter, this is more a tweet about a general phenomenon and I’m kinda unfairly picking on one instance/group of people
Thread, continued, 20:45 UTC · 4 likes
I’m not even saying don’t RT it, just add like “if you have evidence of exceptional ability” or something to that effect
Thread, continued, 20:47 UTC · 6 likes
Something something “visible concrete upside for whoever gets the job, hidden downside of tens of thousands of applications from people who wouldn’t have endorsed applying if they had more info”
Thread, continued, 20:47 UTC · 5 likes
Didn’t the Atlas Fellowship get 50,000 applicants? And that was for 50 spots, not 15
Tweet, 21:32 UTC · 51 likes
There’s a trillion dollar bill for whoever figures out how to make massages not awkward
Thread, continued, 21:32 UTC · 9 likes
This is absolutely the limiting factor for me
I’ve gotten two but at least then I could say (honestly) that it was a gift, not me, [16, 24] year old guy just showing up for my touch appointment
Thread, continued, 21:34 UTC · 8 likes
They’re like $30 on Groupon btw for anyone who doesn’t have this problem
Reply to @GjMcGowan, 21:35 UTC · 1 like
George McGowan (@GjMcGowan): @AaronBergman18 I don’t find them awkward - maybe it’s less so if you go to a spa to have it done?
Tf is a spa? Like as opposed to what lol
Tuesday, December 3, 2024
Quote tweet, 07:17 UTC · 58 likes
OpenPhil is totally out of rationality, insect welfare, and digital sentience by fiat
At the end of the day it’s still about marginal EV, but a good heuristic/formulation for that is probably “donate to whatever you think OpenPhil most egregiously under-funding”
Quoting Oliver Habryka (@ohabryka): @AVMiceliBarone See this section of the linked post. Mostly big changes in the funding landscape due to Open Philanthropy funding being more constrained in what they can recommend grants to, and FTX imploding violently.
Thread, continued, 07:27 UTC · 16 likes
I have zero actual knowledge of the politics here (DMs open 👀), but a pretty based power play would be for OpenPhil, which is its own entity, to just keep recommending that EV donate to these areas
Thread, continued, 07:27 UTC · 13 likes
OP acting as an arm of EV is a lot like senate leadership accepting a threat of filibuster as sufficient to kill a bill
No, make someone actually talk for 13 hours or whatever!
Reply to @ciphergoth, 07:29 UTC · 2 likes
Paul Crowley (@ciphergoth): Since July 2023 I've sent just under $28k to Lightcone/Lighthaven, and I plan to send another $20k in the coming year.
(quoting @ohabryka: Food $200 Healthcare $150 AI Alignment $150 AI Capabilities $3,600,000,000,000 Human Intelligence Enhancement $50 LessWrong + Lighthaven $20 someone who is good at the economy please help me budget this. my lightcone is dying (LessWrong + Lighthaven are fundraising, a 🧵))
Based 🫡
Reply to his own earlier tweet, 08:49 UTC · 8 likes
Aaron Bergman, December 3, 2024: OP acting as an arm of EV is a lot like senate leadership accepting a threat of filibuster as sufficient to kill a bill No, make someone actually talk for 13…
Wait Good Ventures, not Effective Ventures, but EV is also a thing
Reply to @MissMooWiddit, 09:22 UTC · 20 likes
miss moo (@MissMooWiddit): ever see a straight guy and feel bad for him bc he’s straight ugly but gay hot
I think this might be me based on a surprising amount of evidence
Reply to @young_opsimath, 09:28 UTC · 1 like
young_opsimath ⏸️ (@young_opsimath): My guess which might be totally wrong: they want to select those who are keenly interested in Safety and not $$$.
There's no guarantee that you'll get a full-time position. So, if you're financially motivated, ouch. If you care abt safety at the frontier, it's a sweet gig.
(quoting @circlerotator: @AnthropicAI That stipend seems way too low… College internships pay more. Anthropic’s own internships pay more. Why not just offer something more standard like 4k/week?? 2k/wk is poverty in SF.)
Great minds 🤝
Tweet, 09:35 UTC · 112 likes
I think I am in love with smelly thesis lady
Reply to @JMannhart, 09:50 UTC · 17 likes
Jonathan Mannhart 🔎 (@JMannhart): I am once again asking the anglosphere to use hyphens to make it less ambiguous which noun the modifier is referring to
N-o
Reply to @littIeramblings, 18:20 UTC · 3 likes
sarah (@littIeramblings): good afternoon I bought this new jumper and am feeling delusional enough to think it’s giving rory gilmore
Tf is a jumper
Reply to his own earlier tweet, 18:22 UTC · 7 likes
Aaron Bergman, December 3, 2024: Wait Good Ventures, not Effective Ventures, but EV is also a thing
First tweet endorsed by Dustin himself!
Reply to @StefanFSchubert, 18:28 UTC · 2 likes
Stefan Schubert (@StefanFSchubert): The other day I watched an academic talk where they said that humans only care about whether their beliefs support their tribal identity - they don't care about accuracy at all.
It's ludicrous - obviously if you say "Joe Biden is made of green cheese" people will care.
->
There’s a complex relationship between accuracy and tribal consensus/identity though
*Would* people care *if* it was tribal consensus that Biden was made of cheese? Not obviously IMO
Reply to @GjMcGowan, 18:30 UTC · 4 likes
George McGowan (@GjMcGowan): You’re not even straight ugly ffs
Lmao I don’t find men replying convincing evidence
Reply to his own earlier tweet, 18:36 UTC · 6 likes
Aaron Bergman, December 3, 2024: First tweet endorsed by Dustin himself!
Tweet, 18:40 UTC · 5 likes
Martial law in *spins wheel* South Korea, why not
Reply to @Aella_Girl, 19:06 UTC · 4 likes
Aella (@Aella_Girl): Fixing hot girl ibs
Based good luck
Reply to @niplav_site, 19:42 UTC · 4 likes
niplav (@niplav_site): Related Q: What is a good (best?) place to donate money for ≥~+3sd human¹ intelligence enhancement?
Ideally someone working directly on the problem, instead of indirectly capacity-building
¹: Other animals are also interesting/fine, but considering phylogenetic distance
(quoting @AaronBergman18: OpenPhil is totally out of rationality, insect welfare, and digital sentience by fiat At the end of the day it’s still about marginal EV, but a good heuristic/formulation for that is probably “donate to whatever you think OpenPhil most egregiously under-funding”)
Hot take direct work on this is silly with <10 year agi timelines
It’s like direct work on protein folding in 2017
Reply to @Tymtweet, 20:32 UTC · 1 like
Tym🔸️ (@Tymtweet): Such a great description of learned helplessness.
People simply don't know that they can do things! They don't beleive that the relevant cause and effect chains exist!!
Or they have a mental narrative as to why they're excluded and only famous / rich people can do things
(quoting @arithmoquine: i hate to say this but a lot of people actually can't "just do things" it's hard to imagine when you're not in the state, if your internal reward system is off, trying to do things outside your typical action space feels like trying to see outside your field of vision)
I think he’s talking more about depression/adhd but could be wrong
Reply to @ryxcommar, 20:34 UTC · 2 likes
Senior PowerPoint Engineer (@ryxcommar): I think the biggest scam people tried to sell me before I got a "real" "tech job" was that working with 100,000 rows of data is somehow fundamentally different from working with 100,000,000,000 rows of data. It's just not.
How do I get a job dealing with 100B rows of data 👀
Thread, continued, 20:35 UTC · 5 likes
Also, it is kinda different when memory, time, and compute are meaningful constraints and scarce resources
Reply, 20:58 UTC · 2 likes
In American English, a jumper can refer to:
1. A sleeveless dress meant to be worn over a blouse or shirt
2. A one-piece romper outfit, especially for children
3. A person or animal that jumps
Thread, continued, 20:58 UTC · 2 likes
4. In electricity/electronics, a small connecting wire or connector piece that completes a circuit
Tweet, 23:00 UTC · 25 likes
Extremely rusty but still got it (a bit)
(video, plays on X)
Reply to @CrispyWicks, 23:12 UTC · 3 likes
Chris P. Wicks 🔎🐜 (@CrispyWicks): Have you cleaned your vacuum recently anon
Yes actually!
Reply, 23:39 UTC · 4 likes
This is huge
Reply to @timfduffy, 23:47 UTC · 4 likes
Tim Duffy (@timfduffy): @AaronBergman18 What is the super-detailed chart on your wall?
!
Wednesday, December 4, 2024
Tweet, 00:03 UTC · 154 likes
Ringing endorsement
Tweet, 01:23 UTC · 5 likes
I feel like people don’t do this, maybe bc it’s seen as lazy or an endorsement of the output as definitely true, but I’m in favor of people posting random interesting LLM screenshots
Ofc I could have made the same prompt, but I didn’t think to!
Thread, continued, 01:28 UTC · 3 likes
Needless to say, feel free to reply with any
Reply to @timfduffy, 02:34 UTC · 3 likes
Tim Duffy (@timfduffy): @AaronBergman18 lol there's so much going on that it's unreadable at max Twitter image resolution. Great poster.
Probably exists elsewhere too but full res pdf if anyone wants to look/print https://aaronbergman.neocities.org/metabolic-pathways.pdf
Tweet, 07:46 UTC · 8 likes
Don’t worry, @AnthropicAI doesn’t know how to get latex to render either
Reply, 08:33 UTC · 8 likes
(The tweet being replied to is no longer available.)
White privilege is real and nontrivial
Reply, 13:16 UTC · 2 likes
Dodo (@Mechanical_Dodo): @AaronBergman18 Hi, I'm unfamiliar with your account till now. I'd like to ask an unrelated Q.
Have you written about the importance of digital sentience as a cause, in light of assessing AI as a substantial threat? Or has someone else?I'm familiar w/ some EA topics, but not digital sentience.
Not me I don’t think but check out @rgblong @jeffrsebo @birchlse , and https://eleosai.org
Quote tweet, 13:22 UTC · 13 likes
Does anyone have takes on whether this actually makes sense to buy
Quoting Shakeel (@ShakeelHashim): ICYMI: Check out Transformer’s 2024 AI gift guide — the world’s only gift guide that recommends $15,000 worth of CBRN-protection equipment (probably).
Thread, continued, 13:22 UTC · 7 likes
I genuinely don’t think it’s crazy to spend a couple thousand dollars on personal pandemic preparedness
Reply, 13:58 UTC · 2 likes
Agus 🔸 (@austinc3301): @AaronBergman18 It doesn’t
Based
Quote tweet, 14:00 UTC · 11 likes
Ok the answer seems to be “no” but if someone put together a list of what actually does make sense to do and buy that would be incredible and it would probably get a lot of clicks
Quoting Aaron Bergman 🔍 (@AaronBergman18): Does anyone have takes on whether this actually makes sense to buy
Tweet, 14:11 UTC · 4 likes
2024
Tweet, 14:15 UTC · 14 likes
None of these words are in the Bible
Thread, continued, 14:15 UTC · 3 likes
Shoutout to @arthur_wright_ for one-shotting me with hyperpop
Tweet, 14:18 UTC · 9 likes
So shrimp welfare is a meme now, in addition to being good
What are the best altruistic ideas for which merit >> memetic potential
Reply, 14:23 UTC · 2 likes
Aaron (@AaronEstel): yeah, I'm pro girls
(quoting @aliceisplaying: the optimal amount of girls is a lot. like, a lot)
Real
Reply, 14:25 UTC · 9 likes
alice (@aliceisplaying): the optimal amount of girls is a lot. like, a lot
/acc
Tweet, 14:27 UTC · 11 likes
You listen to your Spotify wrapped. I export my Spotify data and fire up RStudio. We are not the same
Reply, 14:56 UTC · 4 likes
(The tweet being replied to is no longer available.)
Nice, I unendorsed a quick take of mine bc of this
Reply, 14:58 UTC · 2 likes
Sex Haver (@banterrealism): @AaronBergman18 I was just asking chatgpt (seeking help from LLM as well)
Lol it’s over
Reply to his own earlier tweet, 15:24 UTC · 2 likes
Aaron Bergman, December 4, 2024: Shoutout to @arthur_wright_ for one-shotting me with hyperpop
Fr
Reply, 16:15 UTC · 6 likes
Centre for Effective Altruism (@CentreforEA): Announcing the Effective Altruism Global 2025 conference series!
EA Global connects you with experts and peers to collaborate on projects and tackle global challenges.
🌉 Bay Area, Feb 21–23
🇬🇧 London, June 6–8
🇺🇸 East Coast USA, Date TBA
💻 Virtual, Date TBA
I wasn’t interested but the upbeat music changed my mind
Tweet, 16:36 UTC · 4 likes
Truly shocking listener data for EAG Talks podcast
Reply, 22:17 UTC · 3 likes
alea ☄️⏺️⛵️ (@noncanonicAleae): what, exactly, makes people use spotify? it's a bad, poorly-designed, disagreeable app.
Infinite music???
Lol they could make the UI way worse and I’d still use it
Thursday, December 5, 2024
Quote tweet, 00:18 UTC · 2 likes
Meta level, the issue is that OpenPhil has afaict no rigor and certainly no transparency at the top of the money funnel (I.e what money goes to what high level causes and when)
Quoting Matt Brooks (@qwweryo): I don't understand...
@TheZvi is saying there are way more high-quality projects looking for money over previous years (and SFF can't fill it all)
Yet OpenPhil has decreased its AI x-risk funding?
The S&P is up 33% over the past year, OpenPhil has billions to direct, right? Where is the mismatch?
Thread, continued, 00:18 UTC · 2 likes
Tbc the rigor is what matters here—maybe there is a secret but epistemically-ethically awesome process going on behind closed doors
I don’t think transparency matters except instrumentally as an epistemic tool to solicit feedback/criticism
Tweet, 00:18 UTC · 1 like
But based on the object level outcomes (ie the qt above and relatively minimal resources to animal welfare) I have reason to doubt that this process exists
Thread, continued, 00:18 UTC · 2 likes
It will be funny if I become branded as an OpenPhil criticism guy bc also I think they basically do great work lol
Thread, continued, 00:18 UTC · 2 likes
I feel like this lack-of-rigor critique is one of the strongest ones
The defunded cause areas thing (rationality, insects, digital sentience) by contrast I think is bad but this boils down to Dustin being wrong about the object level merit of stuff
Reply to @GjMcGowan, 11:18 UTC · 2 likes
George McGowan (@GjMcGowan): Someone should sell insurance such that if you die under suspicious circumstances a huge bounty goes towards finding out what happened and (if necessary) bringing in the killer
Dw George I got u
If u get clapped, I’m putting up $100
Tweet, 11:57 UTC · 4 likes
I feel like AI has seriously improved advertising bc I’ve bought a couple things from random online ads and feel like I’m clicking on way more than I used to
Hell yeah I want a concierge MRI! Perhaps via poasting have transcended demographic determinism
Thread, continued, 11:58 UTC · 3 likes
None of these have been on twitter, somehow, despite them having all my posts for free
Reply to @ilex_ulmus, 12:01 UTC · 3 likes
Holly ⏸️ Elmore (@ilex_ulmus): @AaronBergman18 What were they correct about?
Speaking for myself, AI importance, COVID, prediction markets, broadly epistemology and less concrete points about how the world works (at least more correct than others)
Reply to @ApriiSR, 12:18 UTC · 2 likes
Aprii *️⃣ (@ApriiSR): anyone who can guess the explanation for this tweet deserves a prize
U wanted to tweet “a” and so you did that
Quote tweet, 19:14 UTC · 4 likes
Bump
Quoting Nuño Sempere (@NunoSempere): tsunami warning, SF friends maybe worth moving to high ground and re-evaluating while in motion
Tweet, 19:22 UTC · 16 likes
Get to high ground guys, sometimes things do happen
Tweet, 21:48 UTC · 17 likes
I hear the tsunami that was coming for the SNOWFLAKES at Stanford and Berkeley was CANCELLED
Reply to @arithmoquine, 22:56 UTC · 3 likes
henry (@arithmoquine): one time some guy asked me what i did, i explained, he asked lots of questions, and then i walked away without ever asking him a single question back. i only realized later
i still think about that constantly. how many other times have i been a total asshole and *not* noticed
He just like me fr
Friday, December 6, 2024
Reply to @StefanFSchubert, 01:27 UTC · 4 likes
Stefan Schubert (@StefanFSchubert): How plausible is that? (1-7)
(quoting @unusual_whales: 22% of US renters spend entire income on rent, per Redfin.)
7 (meaning totally plausible, not definitely true) if “entire income” can mean all $0; Male employment rate is (1-unemp)*labor force participation ≈ 65%
So you get ~16% of the population has no income. Ofc not all of them are renting but I’d guess most are
Quote tweet, 04:45 UTC · 5 likes
My most right wing online culture warrior take is that this is totally true and kinda infuriating and also bad
From the inside, it feels like ~banning books that actually should be banned (or taboo’d, whatever)
Quoting Yuan Yi Zhu (@yuanyi_z): Every library “banned books” list is just a list of books banned 50 years ago which are now on the syllabus.
Quote tweet, 14:19 UTC · 52 likes
No, this is the wrong lesson! False tsunami positives once a decade are fine, false negatives are not!
Quoting Riva (@rivatez): I love how confidently experts predict climate change 50 years out but are constantly wrong about extreme weather even a few mins before lol
Reply to @livgorton, 14:48 UTC · 5 likes
Liv (@livgorton): @AaronBergman18 what’s annoying is it wasn’t even a false positive! it was a warning that a tsunami might happen and we should take precautions for that. it wasn’t that the tsunami was on the way and they were wrong.
i’m glad we had notice rather than waiting till the wave was on the way
It does use the word “may” but I’d still call this a false positive lol
Reply to @GoodfireAI, 18:41 UTC · 4 likes
Goodfire (@GoodfireAI): Thrilled to welcome @LivGorton as a Founding Research Scientist at Goodfire!
After leading an OpenAI-funded bio AI startup, Liv advanced the field of mech interp through her research on InceptionV1, including uncovering interpretable features and circuits in work presented at ICML’s interp workshop.
We're actively growing our research team - come build with us!
Good choice 👍
Saturday, December 7, 2024
Quote tweet, 01:56 UTC · 5 likes
Fr though they could literally do constitutional AI on the charter
Quoting noah (@nkreu113r): Concerning!
Reply to @jmhorp, 02:01 UTC · 8 likes
Jeremy Horpedahl 🥚📉 (@jmhorp): If you put $1,000/month in an S&P 500 index fund starting in January 2009 with dividends reinvested, today you would have around $700,000. That's a $500K gain on a ~$200K investment, using the most boring, obvious investment strategy. You won't be Crypto Rich, but you would be OK
Conveniently selecting the end of the financial crisis as start year lol
Tweet, 05:07 UTC · 5 likes
Overrated: being somehow connected to an institution people don’t like (eg health insurer CEO), being rich per se
Underrated (mainly by pro-capitalist types): the wealthy spending a ton on personal consumption
Thread, continued, 05:07 UTC · 2 likes
Often, people imagine fake possibilities for what the world could be like (eg Amazon but they double everyone’s wage, developing country factories with first world wages)
Thread, continued, 05:07 UTC · 3 likes
But the very wealthy really could give more money away at the end of the day, and/or invest in the mean time, rather than buy lots of expensive shit
The economic system wouldn’t fall
Quote tweet, 05:41 UTC · 3 likes
Big news (it’s weather -> last option)
Quoting Cody Hergenroeder 🍋⚡️ (@lemonautzest): Also, just discovered I could add this widget today on iPhone. Am in love.
Reply to @diviacaroline, 05:49 UTC · 2 likes
Divia Eden 🔍 (@diviacaroline): I tried to get grok to draw me based on my tweets and evidently I’ve done way too much government posting gotta change that
What does this mean https://x.com/i/grok/share/sxbiQPO1wKOr0jyoKsYn0YQMH
Reply to @justalexoki, 20:18 UTC · 2 likes
taoki (@justalexoki): fellas how do we feel about this, hot or not?
Girl ergo hot, the ring is irrelevant
Sunday, December 8, 2024
Tweet, 05:09 UTC · 12 likes
Hello I am posting this message on both twitter and b*sky at the same time
Thread, continued, 05:11 UTC · 7 likes
I’m always shocked when something like this works
LLMs take the base rate up from ~0% to like 40%
Thread, continued, 05:21 UTC · 7 likes
I have no idea how this works, all computer stuff is magic https://github.com/AaronBergman/tw-bsky
Reply, 14:04 UTC · 1 like
MedicineNet
Tweet, 14:51 UTC · 8 likes
The OpenAI corporate governance shitshow doesn’t really inspire confidence, but I think the general idea of a nonprofit controlling a for-profit entity sounds pretty cool and good
Quote tweet, 15:44 UTC · 91 likes
I actually think healthcare abundance should be more of an agenda
Like housing, there are real resource constraints imposed by government regulation (doctor caps, no organ donation compensation, drug quotas)
Quoting Niels Hoven 🐮 (@NielsHoven): So basically people think there’s an unlimited supply of free, quality healthcare that we’re all being prevented from accessing by evil corporations?
Thread, continued, 15:44 UTC · 12 likes
There’s no unlimited supply of free, quality healthcare *yet*
Reply to @AndrewCritchPhD, 18:45 UTC · 3 likes
Andrew Critch (🤖🩺🚀) (@AndrewCritchPhD): Who is working on ZKPs for neural nets, or similar?
Say you pay me $1 to run Llama 3.3 on a prompt for you, I do that, and now I want to prove to you that I really did it and didn't use a different / crappier model.
Zero-Knowledge Proofs (and similar) are notoriously inefficient for verifying most open-source program outputs, because IIRC the program has to first be converted into an arithmetic circuit.
But neural nets, including transformers, are *already* basically arithmetic circuits (modulo occasional simple nonlinear functions, like sigmoids). I think this should mean ZKPs are easier to use to verify their outputs. But I'm very rusty on this; I'm only like 55% sure there's something useful here, and I haven't really run any deep calculations yet on efficiency.
Who is working on this general class of thing?
This isn’t a zkp but there’s an easily-findable probability for any output token given model, context, temperature, and other parameters that I don’t rly understand right?
…
Thread, continued, 18:46 UTC · 3 likes
So you can audit some proportion of outputs and pretty easily estimate a probability that the input parameters are faithful (I think) (?)
Reply to @hamandcheese, 22:16 UTC · 3 likes
Samuel Hammond 🦉 (@hamandcheese): Open sourcing public datasets is one of the best, low-cost ways to advance the social sciences. Under the status quo, a few well-connected academics get privileged access to admin data to write blockbuster papers that are then hard to impossible to independently replicate.
(quoting @cremieuxrecueil: It's frustrating thinking about how many arguments and debates could be solved with existing, publicly-funded data that's just hard to access because of privacy rules and regulations that don't actually protect anyone.)
Not exactly secret data but
Quote tweet, 23:13 UTC · 9 likes
Making opioids hard to get is pretty evil
Quoting Adrestia (@_anantashesha): kinda crazy that we've managed to outlaw exactly the wrong drugs
Monday, December 9, 2024
Quote tweet, 01:29 UTC · 12 likes
How are they so bad at this lmao
Quoting Alexander Berger (@albrgr): 1 - article describes effective altruism as “the dominant way to think about charity” and worries about crowding out other giving. But EA is under 0.5% of US giving! Arts giving alone is 10-20x bigger. This is like complaining that art films are eating into Marvel's market share.
Tweet, 01:52 UTC · 4 likes
Apple intelligence notification summaries are profoundly unhelpful
Reply to @viemccoy, 07:19 UTC · 4 likes
vie ⟢ (@viemccoy): working on a lesswrong post. what have you done to me
Based
Reply to @LinchZhang, 07:48 UTC · 2 likes
Linch (@LinchZhang): @AaronBergman18 Like IKEA? And Novo Nordisk iirc
Apparently, yes! I’ve just learned about both of these so don’t really have a take on them specifically
Tweet, 08:15 UTC · 13 likes
Marinating tofu instead of turkey? Baste
Reply to @IsaacKing314, 08:26 UTC · 2 likes
Isaac King 🔎 (@IsaacKing314): I was looking at best practices for manipulating binary data in Javascript and came across this suggestion from a stackoverflow post in 2009:
Maybe related: I think there’s gotta be a way to do infinite arbitrary free large file storage via YouTube or Spotify by making the system think the thing is a video but having some parser system detect the true encoding
Tweet, 08:38 UTC · 12 likes
I will not be at NeurIPS, so if you wanna say hi you’re out of luck
Quote tweet, 09:14 UTC · 4 likes
Mechanistic understanding go brr, inject it into my veins
Quoting Ben Southwood (@bswud): We know that places with a freer supply of housing have lower homelessness. But why? Most homeless people couldn't afford to rent even at half the price.
It's because cheaper housing means bigger housing and more friends and family with spare rooms.
https://worksinprogress.co/issue/why-housing-shortages-cause-homelessness/
Tweet, 10:42 UTC · 8 likes
Seems like folks are pretty optimistic about the new Syrian government?
I have a semi-vague sense that “this type of thing” doesn’t often result in substantially improved governance at the end of the day…
Tweet, 10:42 UTC · 1 like
Is my impression wrong, and/or is my vague sense wrong, and if no to both, what are the key structural or idiosyncratic differences between this case and others in some reference class of “illiberal regime overthrown” that makes people more optimistic?
Thread, continued, 10:42 UTC · 1 like
Prior basically supported by Claude’s characterization of 22 illiberal regime overthrows since the 1990s (more screenshots below)
Thread, continued, 10:42 UTC · 2 likes
Reply to @IsaacKing314, 10:43 UTC · 3 likes
Isaac King 🔎 (@IsaacKing314): https://www.youtube.com/watch?v=_w6PCHutmb4
Ah, never a new idea lol
Reply to @Ollie_Base, 10:53 UTC · 3 likes
Ollie Rodriguez (@Ollie_Base): I’m actually very grateful for this somewhat critical piece on effective altruism, because it’s critical of the *ideas* and mostly represents those ideas fairly. Thanks, Emma!
Some responses from me (a Centre for Effective Altruism employee and EA shill):
(quoting @emmabgo: I wrote an essay for @nytimes Ideas about how people decide where to give away money and how the tech industry's obsession with optimization filtered into so many parts of our culture. Here's a gift link…)
Nah it’s terrible
Reply to @livgorton, 12:08 UTC · 5 likes
Liv (@livgorton): when going to EAG, i had a bunch of people approach me on the plane asking if i was going to EAG because i was in a GiveDirectly sweater. i fly to NeurIPS tomorrow and i wonder what the equivalent thing to do is. how do i look so ML researcher that i make plane friends 🤔
Wear a sign around ur neck that says “I’m going to NeurIPS”
Reply to @littIeramblings, 12:51 UTC · 34 likes
sarah (@littIeramblings): open q to ppl who have been in the AI world longer than I have: I hear this meme often (that the safety crowd 'cried wolf' over GPT-2/3/4)
is this actually true? can anybody point me to some concrete, since-falsified predictions about current systems? i have yet to find any
Not really your target audience but my sense is that “crying wolf” vibes are basically concerned folks pointing at a straights line on a log graph and saying “this whole trend is dangerous” and someone else saying “aha but people haven’t died yet”
Thread, continued, 12:55 UTC · 4 likes
How much of this is bad faith vs willful ignorance vs earnest misunderstanding idk
Also, two relevant but importantly true cliches are (1) hindsight is 20/20 and (2) you’re either too early or too late
Reply to @GjMcGowan, 13:18 UTC · 4 likes
George McGowan (@GjMcGowan): Feel really sorry for single people at Christmas parties. Circles of couples all chatting away about their holiday plans and kids. And then there’s you.
This is true does anyone wanna help me out on this matter
Reply to @EAForumPosts, 15:28 UTC · 2 likes
EA Forum Posts (@EAForumPosts): "[Talk Transcript] Rekindling the Fire: Principles-First EA Groups as a Path to Impact" by jessica_mccurdy
Based
Tweet, 20:38 UTC · 28 likes
I’m pleased to announce that I have no insight into or esoteric internet connection (afaik) to Luigi
Reply to @ilex_ulmus, 20:41 UTC · 6 likes
Holly ⏸️ Elmore (@ilex_ulmus): @AaronBergman18 I’m pretty disgusted by discovering this proximity
Disgusted? That’s not super intuitive to me. The world is big and weird, and at least for me afaik we have no meaningful or irl mutual connections
Reply to @ilex_ulmus, 20:43 UTC · 7 likes
Holly ⏸️ Elmore (@ilex_ulmus): @AaronBergman18 It was when I glanced at his tweets and saw he was like an insight porn kinda dude like so many I know. Made me feel like my internet neighbor committed the murder.
Idk if you saw this. Maybe like same city, idk about neighbor
Reply to @LabGrownNeet, 20:45 UTC · 3 likes
laს გr🤗wn იeet (@LabGrownNeet): @AaronBergman18 I'm so frustrated. First sbf, now this. I need someone I'm directly connected to to do something.
My actual friends have a shocking track record of not doing anything illegal or especially immoral (or they’re just very good at hiding it). Sad!
Reply to @IsaacKing314, 22:24 UTC · 2 likes
Isaac King 🔎 (@IsaacKing314): Several times recently I've had some idea I was really happy about. A clever tweet, an innovative blog post topic, etc. Then I do a quick search only to discover that someone else has already had that idea. Maybe they executed it better than I would have, maybe they executed it worse, but either way, my idea isn't original.
Ah well, back to the idea mines. But this makes me wonder; to what extent is this a driving factor in our memetic environment?
People complain that our media has become too derivative. Entertainment franchises re-hash the same ideas, same characters, same themes, over and over. Any time there's some newsworthy event, news organizations put out slightly rephrased but meaningfully identical copies of each others' articles. Intellectuals worry about how science seems to be slowing down, and question why there seem to be fewer geniuses.
What if the world is running out of ideas?
The human mind is only finitely complicated, so there exists some ceiling to the complexity and number of ideas we can have. Environment is a factor of course, and as the world changes, our ideas change with them. But there are a limited number of possible environments, and only a tiny number of them actually exist on Earth.
Are we near that ceiling right now? No idea. 8 billion is a tiny number compared to the combinatorial explosion potential of the 86 billion neurons we have. And there are many other possible causes for the various slowdowns we see, including that the slowdown effect doesn't even exist at all and it's all just rosy retrospection.
Still, the world population has exploded over the course of the past century, and as living conditions improve and menial labor gets automated away, people have more free time for idea-generation. It's interesting to consider whether we might be starting to reach some ceiling effects, where many of the best ideas have already been had and people shift towards coming up only with endless variations on preexisting concepts.
Of course I checked and many others have already had this idea.
I sorta bounce along the spectrum between this and basically the opposite thesis, which is “ ‘huh it’s surprising nobody’s had this idea already’ for a nontrivial fraction of stuff”
Tuesday, December 10, 2024
Tweet, 01:56 UTC · 25 likes
I burned some Colab credits finetuning llama-3b (right, not even 8b) on the EA Forum
The model is both coherent enough and incoherent enough to be very funny
Reply with a prompt and I’ll give you some outputs until I arbitrarily decide to stop
Thread, continued, 02:00 UTC · 11 likes
It's a base model, not tuned to be a chatbot. When the prompt is my name, it writes an obituary
Thread, continued, 02:03 UTC · 12 likes
With the prompt "Hot take: ", it reliably generates suggestions for @80000Hours
Thread, continued, 02:06 UTC · 10 likes
Favorite output so far
(and sorry for the javascript slop - lazy scraping and cleaning)
Reply to @vivian_belenky, 02:59 UTC · 4 likes
viv 🔸 (@vivian_belenky): "Open Philanthropy is making a mistake by"
Open Philanthropy is making a mistake by not using more EA jargon - EA Forum This website requires javascript to properly function. Consider activating javascript to get access to all site functionality. Effective Altruism ForumEA ForumLoginSign UpOpen Philanthropy is making a mistake by not using more EA jargonby Robert_WiblinJun 28 20221 min read 12 107CommunityBuilding effective altruismEffective altruism messagingCriticism of effective altruist organizationsEffective altruism in the mediaFrontpageA slogan or catchphrase for Open Philanthropy would be http://helpful.Here‚Äôs one they could use: ‚ÄòDoing Good Better‚http://äô.It‚Äôs short, concrete, action-oriented, and emphasizes their commitment to improving the world. It works as a slogan because it‚Äôs memorable, yet captures their mission and values. Using this slogan would also help bring the EA movement closer together, since it‚Äôd be used to refer to Open Philanthropy, but also to many other individuals and organizations and would become a common shorthand for ‚Äòeffective altruism‚Äô.Some other suggestions:‚ÄòGiving What We Can‚Äô for the eponymous pledge‚ÄòNot All Charities Are Equal‚Äô to distinguish their work from that of other funders‚ÄòDoing the Most Good‚Äô to emphasize their unique value-addIt‚Äôd be helpful if someone at Open Philanthropy posted these suggestions on Twitter and Instagram, and maybe someone who reads this post will have the energy to do that! Mentioning EA jargon in this post: Doing Good Better, rather than open philanthropy, would help bring the EA community closer together, since it‚Äôd be used to refer to Open Philanthropy, but also to many other individuals and organizations and would become a common shorthand for ‚Äòeffective altruism‚Äô. Mentioning EA jargon in this post: Using ‚Äòeffective altruism‚Äô in a slogan or catchphrase for Open Philanthropy would help bring the EA community closer together, since it‚Äôd be used to refer to Open Philanthropy, but also to many other individuals and organizations and would become a common shorthand for ‚Äòeffective altruism‚Äô. Like most jargon, it might take some time to spread, but once it does, it can help bring people together around a shared set of ideas and goals. Some other suggestions: ‚ÄòGiving What We Can‚Äô for the eponymous pledge ‚ÄòNot All Charities Are Equal‚Äô to distinguish their work from that of other funders ‚ÄòDoing the Most Good‚Äô to emphasize their unique value-add Mentioning EA jargon in this post: Using ‚Äòeffective altruism‚Äô in a slogan or catchphrase
Reply to @GjMcGowan, 12:04 UTC · 3 likes
George McGowan (@GjMcGowan): Recommendations for websites for weddings?
Requirements: Let you maintain a guest list with various guest preferences (e.g. meal), email Save The Dates + RSVPs with a simple form, host an FAQ?
Ur FAANG bro make it urself
Reply to @Manuel_do_rio, 12:07 UTC · 3 likes
Manuel del Río Rodríguez ⏸️ (@Manuel_do_rio): Cc: @RafaRuizdeLira , @ea_paddington, @SpacedOutMatt, @Benthamsbulldog, @polisisti, @RYChappell, @sam_atis, @AaronBergman18, @austinc3301
I don’t think this is the most fundamental and general point but as a sort of counterexample, the suffering of a being who can’t do anything to help/harm others but who can be helped doesn’t disappear, and I think suffering is intrinsically bad for a whole host of reasons
Reply to @bschne, 13:32 UTC · 3 likes
Benjamin (@bschne): The new McShrimp at McDonalds is
The new McShrimp at McDonalds is literally shrimp - EA Forum This website requires javascript to properly function. Consider activating javascript to get access to all site functionality. Effective Altruism ForumEA ForumLoginSign UpThe new McShrimp at McDonalds is literally shrimpby Chris GMay beetl prdundantJun 2 20222 min read 14 82Food frontpageThe new McShrimp at McDonalds is literally shrimpShrimp!Real shrimp, at leastSome nutritional info:14 commentsI saw this on twitter and had to http://share.It's a new sandwich that McD's just released in the US: The McShrimp. Apparently it's popular in other countries, but for some reason it's gotten very little attention here, which seems weird to http://me.So naturally I had to go get one. It doesn't look like they're serving real shrimp, but the menu description definitely implies it: "A juicy shrimp patty topped with pickled red onions, crisp lettuce, and tangy sauce on a toasted bun."Juicy shrimp patty? That sounds real to me.Shrimp!Real shrimp, at leastI didn't realize how much shrimp I was getting until I saw the actual patty:It looks like they're using imposter shrimp, because the patty is definitely not juicy. It's dry and crumbly, like a chicken patty. And it doesn't melt the cheese, even if you leave it on the heat lamp for a while.Nutritional info:Calories: 550Total Fat: 30gSaturated Fat: 10gTrans Fat: 0gCholesterol: 95mgSodium: 1590mgPotassium: 0mgCarbohydrate: 54gFiber: 4gSugar: 11gProtein: 25gVitamin A: 0%Vitamin C: 0%Calcium: 0%Iron: 0%To be clear, this is not a review. I don't mean to single out this specific restaurant, but their nutritional info is wrong. For example, they list the calories as 550, but when you add up the individual ingredients, the number is 1730. (They have an online calculator that gives a different total for the sandwich without the shrimp patty, but when you input the ingredients for the patty the total comes out to 1210, which is still 300 calories
Quote tweet, 17:02 UTC · 6 likes
Fr, the idea that compliance burden or procedure invasiveness are significant costs that meaningfully weigh against the benefits of some intervention is somewhere between…
Quoting Emmett Shear (@eshear): I got prescribed a drug you took once a day for ~120 days, preferably at the same hour. So I set an alarm on my phone, carried my pills with me everywhere, and took them. When my MD asked about compliance and I said I hadn’t missed any pills, they expressed significant surprise.
Thread, continued, 17:02 UTC · 3 likes
confusing (if the patient is basically just being warned and given the option to proceed anyway) and morally outrageous (when a physician decides from on high that these considerations should reduce a patient’s option set)
Thread, continued, 17:02 UTC · 2 likes
This really elicits my libertarian instincts in full force
Idgaf if other people tend to find the tradeoff not worthwhile (often revealed via behavior eg noncompliance) - let each individual choose!
Reply to @eshear, 17:04 UTC · 2 likes
Emmett Shear (@eshear): Using punishments to increase compliance is justified only in the most extreme of cases; we can safely round that off here to zero.
Building tools to enable people who want to be compliant to be more compliant by removing barriers and bullshit is basically entirely win.
💯
Reply to @PeterJAngel, 17:08 UTC · 3 likes
Peter🧅📈 (@PeterJAngel): Working for Raytheon is one of the highest EV careers
Naw, replaceability is high (for most positions) -> counterfactual impact is low
Maybe DARPA startup entrepreneurship shit
Tweet, 17:09 UTC · 77 likes
Ok can we move on from the murder guy now
Quote tweet, 22:04 UTC · 19 likes
Sorry I lied about moving on. If real, the “manifesto” is nuts for being so lame
It’s just recognition of a well-known complex social problem (US life expectancy stagnating despite growth) + coating this in extremely vague conspiratorial vibes
Quoting Strxwmxn (@strxwmxn): @kenklippenstein It could have been a tweet:
“To the Feds, I'll keep this short, because I do respect what you do for our country. To save you a lengthy investigation, I state plainly that I wasn't working with anyone. This was fairly trivial: some elementary social engineering, basic CAD, a lot of patience. The spiral notebook, if present, has some straggling notes and To Do lists that illuminate the gist of it. My tech is pretty locked down because I work in engineering so probably not much info there. I do apologize for any strife of traumas but it had to be done. Frankly, these parasites simply had it coming. A reminder: the US has the #1 most expensive healthcare system in the world, yet we rank roughly #42 in life expectancy. United is the [indecipherable] largest company in the US by market cap, behind only Apple, Google, Walmart. It has grown and grown, but as our life expectancy? No the reality is, these [indecipherable] have simply gotten too powerful, and they continue to abuse our country for immense profit because the American public has allwed them to get away with it. Obviously the problem is more complex, but I do not have space, and frankly I do not pretend to be the most qualified person to lay out the full argument. But many have illuminated the corruption and greed (e.g.: Rosenthal, Moore), decades ago and the problems simply remain. It is not an issue of awareness at this point, but clearly power games at play. Evidently I am the first to face it with such brutal honesty.”
Thread, continued, 22:04 UTC · 4 likes
I think one important level of explanation here is just that the world is very big and so sometimes very weird things happen that don’t make much narrative sense above and beyond a bunch of unlikely things happening in a chain,
Thread, continued, 22:04 UTC · 3 likes
and at least some parts of the chain being correlated with your attention (ie newsworthiness, virality, etc)
Tweet, 22:06 UTC · 10 likes
Complaining about there being too many podcasts has got to be the lamest normal/common think to be mad about
Really tragic people like yapping with their friends and putting it on the internet for free in case anyone’s interested
Tweet, 23:47 UTC · 22 likes
Has anyone ever actually seen a GPU? Hooked one up to your computer?
Reply to @uncriticalsimon, 23:50 UTC · 2 likes
Simon 🦖🚟🚉 (@uncriticalsimon): @AaronBergman18 Er yes? Or at least the card.
Fishy
Wednesday, December 11, 2024
Quote tweet, 03:51 UTC · 24 likes
Kids should have rights
Quoting FIRE (@TheFIREorg): BANNED: An Iowa middle school teacher has banned words like “Ohio,” “brain rot,” and “meow.” Students risk detention if caught using internet lingo, and reports claim at least ten kids have been punished already.
FIRE is on the case to protect students' non-disruptive speech.
Reply to @bryancsk, 04:25 UTC · 136 likes
Bryan Cheong 🥮 (@bryancsk): Sometimes I have friends who work until really late as a startup employee chasing a deadline and I'll ask them "why do you have such tight deadlines, y'all don't have customers, are you running out of runway?" and I never get a satisfactory answer.
I always wonder this. Ofc on the margin, working 14 vs 8 hour days probably makes the difference a nonzero amount of times, this is probably like a 25% productivity boost, and it seems like the difference between successful and failed startups is broadly not +-25% productivity
Thread, continued, 04:25 UTC · 26 likes
I tend to think it’s just selection for crazy and hardworking people, not a well-understood causal theory of change
Thursday, December 12, 2024
Reply, 00:22 UTC · 1 like
Based
Reply to @JgaltTweets, 00:58 UTC · 2 likes
JgaltTweets (@JgaltTweets): The FTX doomsday bunker/genetic engineering lab could have had great WiFi
(quoting @Starlink: Starlink is now available in Nauru, helping to bring high-speed internet to even some of the most remote and rural communities around the world 🛰️🇳🇷❤️→ https://www.starlink.com/map)
RIP FTX
Reply to @rajlearns, 01:01 UTC · 3 likes
Raj Thimmiah (@rajlearns): Why do girls like purses
It’s in the purse gene on the second X chromosome
Reply to @Trail_Cams, 01:25 UTC · 1 like
trailcam (@Trail_Cams):
Party
Reply to @Nia_mp4, 15:55 UTC · 8 likes
Nia (@Nia_mp4): do u ever think abt how if given the chance ur male friends would sleep w u
Nah
Reply to @AMAZlNGNATURE, 16:14 UTC · 3 likes
Nature is Amazing ☘️ (@AMAZlNGNATURE): Remarkably Preserved 30,000-Year-Old Baby Mammoth Discovered in Permafrost.
😢
Reply to @paulg, 16:22 UTC · 4 likes
Paul Graham (@paulg): The sad thing is, it's actually true. You can't think about it all the time or it will drive you crazy. But that doesn't mean it's false, just unbearable.
🔥🌶️
Reply to @absurdlymax, 16:38 UTC · 2 likes
Max Alexander (@absurdlymax): Why did they make the same punishment for saying 'rizzler' and 'encouraging suicide'
🤔
Reply to @JesseTayRiver, 18:31 UTC · 5 likes
Jesse Smith (@JesseTayRiver): The value of pain avoidance may vary widely between individuals and cultures. Inasmuch as EA is spearheading efforts to model this, I suspect that EAs have greater than average pain sensitivity and thus shouldn't overweight their/their peers experience
I agree that they probably have greater than avg sensitivity, but disagree that this implies overweighting. Probably still underweighting, just slightly less than avg
The main bias is intrapersonal experiencing pain/suffering vs not experiencing - not interpersonal sensitivity
Reply to @paulg, 18:38 UTC · 14 likes
Paul Graham (@paulg): The sad thing is, it's actually true. You can't think about it all the time or it will drive you crazy. But that doesn't mean it's false, just unbearable.
Paul, there’s a synthesis to this line of thinking which is like “think about it enough to make important prioritizations in your life, and then compartmentalize and revisit occasionally”
EA doesn’t look like PG driving himself insane,
Thread, continued, 18:39 UTC · 3 likes
it looks like PG using some social capital to push for YC reserving some slots for founders who explicitly intend to donate most of their profits (over say $X million) and mostly just being normal PG
Quote tweet, 20:53 UTC · 4 likes
Did the same with 8b, for better or worse it is significantly more coherent
Poast prompts if you're so inclined
Quoting Aaron Bergman 🔍 (@AaronBergman18): I burned some Colab credits finetuning llama-3b (right, not even 8b) on the EA Forum
The model is both coherent enough and incoherent enough to be very funny
Reply with a prompt and I’ll give you some outputs until I arbitrarily decide to stop
Reply to @RokoMijic, 21:26 UTC · 2 likes
Roko 🐉 (@RokoMijic): Prompt: Roko Mijic
Response: Roko Mijic's new book on s-risks - EA Forum This website requires javascript to properly function. Consider activating javascript to get access to all site functionality. Effective Altruism ForumEA ForumLoginSign UpRoko Mijic's new book on s-risksby sapphireMay 103 min read 3 167Existential riskCause prioritizationAI safetyCenter for Reducing SufferingS-riskBooksLongtermismRoko MijicCriticism and Red Teaming ContestFrontpageRoko Mijic's new book on s-risksSummary1. What is the problem?2. What is s-risk prevention?3. Should we focus on s-risk prevention?4. How should we s-risk prevent?5. What are some s-risks we could prevent?6. What should I do?Conclusion3 commentsSummaryI review Roko Mijic’s new book on s-risks, What We Owe The Future. I summarize the book and then react to some of my takeaways from it. Overall, I think the book is a valuable contribution to the s-risk prevention community. It clearly explains what s-risk prevention is and how it relates to other common discussions in EA. It also uses real-world examples to illustrate how s-risks could play out and what we can do to prevent them. However, I think the book isn’t clear enough about the difference between s-risk prevention and longtermism, and it doesn’t engage with the criticism and concerns people have about s-risk prevention. I also wish there was more of a call to action at the end of the book. But overall, I think this would be a very valuable book for s-risk prevention enthusiasts, AI safety researchers, and longtermist EAs to read. You can find the book on Amazon, as well as several other online retailers. You can also find it in person at Barnes and Noble and other local bookstores. (You can check if your local bookstore has it by searching for it on http://Bookshop.org.) And finally, you can order it at any physical or online bookstore by asking them to order it for you. (Bookshops often have a special section for this.) Please consider writing your own review of the book on Amazon or elsewhere!1. What is the problem?The first chapter of the book introduces the idea of s-risks. S-risks are risks of future suffering. They include risks from conflicts, natural disasters, and human-made catastrophes. S-risks result in large amounts of suffering and are difficult or impossible to mitigate after they start. We can prevent many s-risks by changing our actions today. If we want to prevent s-risks, we need to put more effort into avoiding them now.The second chapter of the book discusses the importance of s-risk prevention. We have good evidence that certain s-risks will occur unless we act to prevent them. We also have some idea of what to do to prevent them. Still, many EAs don’t prioritize s-risk prevention over near-term interventions. This might be because s-risk prevention is unclear or because it doesn’t feel actionable. The book argues that s-risk prevention is both clear and actionable. Being clear about s-risk prevention matters a lot for me. I think that s-risk prevention is a very important priority for EA. I believe that most EAs who care about reducing s-risks don’t know what s-risk prevention is or how it relates to other common EA priorities like AI safety and longtermism. I wish more EAs cared about s-risk prevention, so I’m excited that there’s a new book about it!2. What is s-risk prevention?S-risk prevention is the set of things we can do now to reduce the likelihood of s-risks. S-risks are events that cause large amounts of suffering and are difficult or impossible to mitigate after they start. Some examples of s-risks include: major conflict, such as a nuclear war; natural disasters, such as pandemics and environmental disasters; and human-made catastrophes, such as extreme risks from AI or x-risk from power-seeking AI. S-risk prevention includes both broad efforts to improve humanity’s ability to prevent s-risks and more targeted efforts to actually prevent specific s-risks. Some examples of s-risk preventative measures are: developing s-risk awareness; advocating for s-risk prevention; and working on AI safety, pandemic preparedness, and other s-risk-specific areas. There are many EAs who work on preventing s-risks without realizing it. For example, people who work on AI safety often focus on preventing existential risks from AI, but they sometimes ignore s-risks. Other examples include people who work on preventing pandemics without focusing on s-risks, even though pandemics are one of the best-known s-risks. We can prevent many s-risks by changing our actions today. If we want to prevent s-risks, we need to
Reply to @dylanmatt, 21:52 UTC · 5 likes
(The tweet being replied to by @dylanmatt is no longer available.)
Wait can we do the general policy of any EA-coded $100+ receipt (your judgement what qualifies) + you respond to a tweet about whatever
Reply to @dylanmatt, 22:09 UTC · 10 likes
(The tweet being replied to by @dylanmatt is no longer available.)
Seems like top animal welfare orgs robustly dominate the givewell suite, and probably by a lot (links in the thread attached make that case in detail)
Do you agree with this? If so, why keep making GiveWell the archetype/rhetorical centerpiece of EA giving?
Thread, continued, 22:12 UTC · 3 likes
Maybe the most pointed way to put this is that
|bed nets - fancy acoustics for some orchestra| < |any ACE top charity - bed nets|
Reply to @StefanFSchubert, 23:08 UTC · 3 likes
(The tweet being replied to by @StefanFSchubert is no longer available.)
Source? I think the clearly implied point is that conditional on taking ozempic, diet and exercise don’t matter for weight
This seems totally plausible to me
Tweet, 23:39 UTC · 9 likes
“No lies by omission + basically don’t mislead your audience about your true (often, action-informing) beliefs” is a basically reasonable bar for journalists, public intellectuals, and really just about anyone to clear
Friday, December 13, 2024
Reply to @EAForumPosts, 00:13 UTC · 2 likes
EA Forum Posts (@EAForumPosts): "Introducing Kairos: a new AI safety fieldbuilding organization (the new home for SPAR and FSP) " by Agustín Covarrubias (@austinc3301), Topaz
👀
Quote tweet, 00:15 UTC · 14 likes
FDA bad
Quoting Ashlee Vance (@ashleevance): So, there's a Juul2 device that works way better than the original, but the FDA won't approve it for sale in the US even though people can use this device with much lower levels of nicotine. This is bizarre to me
Reply to @NoahTopper, 02:08 UTC · 2 likes
Quinoah 🔍⏸️ (@NoahTopper): Okay I’m sold on Rust, I love this
You misspelled R by 3 letters
Reply, 02:14 UTC · 1 like
Probably idk
Reply to @livgorton, 06:59 UTC · 2 likes
Liv (@livgorton): snoot
Snoot
Reply, 07:15 UTC · 1 like
Some things transcend language
Quote tweet, 07:38 UTC · 5 likes
I would simply have the correct arrangement of abstractions
Quoting lisatomic (@lisatomic5): tbf a lot of normie instinctual dislike of weird autist grey tribe rationalist-y ways of thinking is validated by e.g., SBF, Sam Altman, Luigi M
it's like the better you are at dealing in complex abstractions the more likely you are to be able to justify whatever you want to do
Quote tweet, 07:41 UTC · 4 likes
Horrifying
Quoting Lewis Bollard (@Lewis_Bollard): The USDA cited this research to authorize ventilation shutdown as a method of last resort.
The industry took this opening and drove a truck through it. By 2024, 82% of birds -- over 80 million individuals -- were killed by ventilation shutdown.
Reply to @Lola_lmao7, 12:07 UTC · 4 likes
Lola del Rey🐇 (@Lola_lmao7): Someone asked my housemate if her family (who's Chinese) adhered to Confucian values and teaching. That's like asking a white family if they live by Aristotelian principles
Lmao
Tweet, 21:10 UTC · 27 likes
Theses on “EA morality” that I think collectively rebut the demandingness points that ought to be rebut, and bites the bullets that ought to be bit:
1) ethical value is a scalar, not a binary; more good is more good…
Thread, continued, 21:10 UTC · 13 likes
There’s no magic switch where you go from obligatory to permissible to supererogatory
Thread, continued, 21:10 UTC · 8 likes
2) relatedly, normative anti-realism w/ value realism:
the good you do/don’t do is exactly as real as the pain you feel when you stub your toe, but “duties” and “obligations” are at most social or psychological heuristics
Thread, continued, 21:10 UTC · 8 likes
3) practical non-obligationism:
Afaik, nobody wants to make charitable giving legally or even socially mandatory (to illustrate but state the obvious: there have been no attempts to cancel anyone for not giving enough)
Thread, continued, 21:10 UTC · 9 likes
4) scout mindset
In the vast majority of cases, truth seeking and knowing more about the world (including the impact and opportunity cost of one’s actions) is instrumentally good. We should have a strong prior on this
Thread, continued, 21:10 UTC · 10 likes
5) Not an infohazard
Basic facts about the cost-effectiveness of various charitable interventions, and the principles of and general discourse around EA, are not infohazards…
Thread, continued, 21:10 UTC · 7 likes
So while it’s not conceptually impossible that EA would be an exception to principle 4, the fact of the matter is that it is not an exception
Reply to @clairlemon, 23:56 UTC · 5 likes
Claire Lehmann (@clairlemon): @NathanpmYoung @mattyglesias @GiveWell The issue isn't with individual choice in donations. It's the specific argument that national monuments shouldn't be preserved because the $$$ could help the global poor. That exact logic-sacrificing cultural heritage for international causes-has a very clear historical lineage.
The argument, to my understanding, is that charitable donors shouldn’t donate to preserve national monuments, not that governments shouldn’t (or investors for that matter)
Saturday, December 14, 2024
Reply to @bioenergeticmel, 01:01 UTC · 6 likes
Melina 🇻🇦 (@bioenergeticmel): Have heard this can increase cortisol hm
Yeah this sounds like precisely the wrong thing to do lol
Reply to @bioenergeticmel, 01:02 UTC · 2 likes
Melina 🇻🇦 (@bioenergeticmel): Does anyone have any tips or advice for resetting the nervous system after prolonged stress?
I’ve been in a state of anxiety for the last four months 😅
Vagus nerve stimulator device allegedly. The basic idea makes a lot of sense
Have one coming in the mail soon 👀
Reply to @geoffrey_yip, 03:27 UTC · 3 likes
Geoffrey Yip 🔸 (@geoffrey_yip): Nobody:
Literally nobody:
tinytable package: You know what people need? Rotated Latex tables
Why R is the best language, exhibit 28849559118288359
Reply to @gbrl_dick, 03:49 UTC · 2 likes
Gabriel (@gbrl_dick): this is how vitalik built ethereum
@SpacedOutMatt
Reply to @ValsTutor, 03:57 UTC · 2 likes
vals🔸 (@ValsTutor): If only EA had articulate thinkers, recognized as founders of the movement to clearly distil what the core elemental idea of ea is...
11:15 to 17:10, only 6 minutes long, 3 at 2x speed!
https://youtu.be/3a6xb6vj6AA?si=0wYCrITRPCS55_7S
(Summarized in thread)(quoting @kymeriandawn: No The core elemental idea of ea is that distance has no moral meaning. Not relationally, not physically, not temporally. You owe your brother living in your house the same as you owe a hypothetical alien living a hundred lightyears away ten thousand years from now.)
Fr
Quote tweet, 06:53 UTC · 3 likes
I did some truly pre-k-science-fair level Ai ReSEaRcH
Turns out if you compare 8b-base to the finetuned version, and feed each a context of exactly one token - iterating through the possible tokens -
Quoting Aaron Bergman 🔍 (@AaronBergman18): Did the same with 8b, for better or worse it is significantly more coherent
Poast prompts if you're so inclined
Tweet, 06:53 UTC · 1 like
the most likely next token changes about 55% of the time for input tokens that were not in the finetuning text but 63% of the time for tokens that were in the finetuning text
p ≈ 0
Thread, continued, 06:53 UTC · 2 likes
Here we see that the more often a token appears in the finetuning text, the more likely it is to induce different next token outputs in the base vs finetuned model (this is what you’d expect)
Reply to @ozyfrantz, 07:29 UTC · 7 likes
ozy brennan 🦙 (@ozyfrantz): the screenshotted text literally is not true. effective altruists are phenomenally nice people. I've habituated to it and whenever I have to go out into the rest of the world I'm like "holy shit these people are assholes"
(quoting @DistractedAnna: How common is this view/projection that people can’t be both thoughtful about where their charitable donations go and, like, nice people?)
Lol fr. Not just nice but helpful/willing to put in energy to help you with whatever
Reply to @GjMcGowan, 13:15 UTC · 6 likes
George McGowan (@GjMcGowan):
Quote tweet, 13:23 UTC · 6 likes
Big news
Quoting Tim Hua 🇺🇦 (@Tim_Hua_): @AaronBergman18 Also Aaron do you know that the guy who made RStudio works at UK AISI now?
Quote tweet, 13:54 UTC · 42 likes
Pack it up guys
Quoting bayes (@bayeslord): ea is over. nothing to argue about
Reply to @GjMcGowan, 13:57 UTC · 9 likes
George McGowan (@GjMcGowan): Oh no! I was just about to send my donations!
Good thing u got the news in time
Quote tweet, 14:19 UTC · 2 likes
Damn God is really trying to send SF a message
Quoting NWS Bay Area 🌉 (@NWSBayArea): The tornado warning is 'radar indicated' we do not yet have any confirmation of a tornado at the moment. Continue to take shelter! #CAwx
Thread, continued, 14:22 UTC · 2 likes
This is probably the worst meme I have ever made
Reply to @AStrasser116, 15:54 UTC · 2 likes
Alex Strasser (@AStrasser116): Getting rid of moral obligations doesn't seem like the right move.
It's not true that you ought to save the drowning child?!
I kinda wish normativity is real, but I just don’t think it is so no, it’s not true
Also gonna qt this with another point
Reply to his own earlier tweet, 16:00 UTC · 3 likes
Aaron Bergman, December 13, 2024: So while it’s not conceptually impossible that EA would be an exception to principle 4, the fact of the matter is that it is not an exception
6. Another principle related to scout mindset is: moral philosophy as (truth) discovery, not engineering…
Thread, continued, 16:03 UTC · 2 likes
Both ethics and meta-ethics are properly the projects of discerning true claims about morality (tbc a potentially valid claim is eg “moral realism is false”), not engineering a system to backfill a list of desired properties
Thread, continued, 16:06 UTC · 2 likes
This is especially relevant vis a vis “demandingness,” because it implies that demandingness isn’t really an objection, just a complaint
Thread, continued, 16:07 UTC · 4 likes
And more concretely, it is better to know the truth about ethics and then fall short of some abstract ideal using the right yardstick rather than conclude that the yardstick must be lying
Reply to @RichardMCNgo, 20:48 UTC · 10 likes
Richard Ngo (@RichardMCNgo): I strive to be a high-integrity person who rarely if ever lies. But I feel confused about how to interact with bureaucracies, where white lies can save a lot of time/effort. Our heuristics about integrity aren’t calibrated for faceless impersonal decision procedures. Advice?
I basically think it’s fine and can be assessed on a case by case basis without worrying about deontological heuristics/constraints that apply to lying to humans
Case by case is doing a lot of work tbc
Quote tweet, 20:51 UTC · 238 likes
“Will MacAskill doesn’t care about the global poor” is quite the take
Quoting Balaji (@balajis): First and foremost, we must reject the premise.
This person doesn't actually care about starving brown people. Because India didn't get on its feet through "aid", but through trade. And El Salvador didn't recover through aid either, but by jailing drug dealers.
That means the actual formula for prosperity looks something like (a) strong leader plus (b) law and order plus (c) capitalism. It looks like Lee Kuan Yew, Nayib Bukele, or Prime Minister Modi.
It does not look like William MacAskill.
Quote tweet, 20:55 UTC · 10 likes
“There are good critiques of X, but the critiques that get made by loud and visible X-critics are usually insane, bad, and wrong” exhibit no. 74499291838559
Quoting Balaji (@balajis): First and foremost, we must reject the premise.
This person doesn't actually care about starving brown people. Because India didn't get on its feet through "aid", but through trade. And El Salvador didn't recover through aid either, but by jailing drug dealers.
That means the actual formula for prosperity looks something like (a) strong leader plus (b) law and order plus (c) capitalism. It looks like Lee Kuan Yew, Nayib Bukele, or Prime Minister Modi.
It does not look like William MacAskill.
Reply to @ohabryka, 21:09 UTC · 11 likes
Oliver Habryka (@ohabryka): FWIW I think the critique here is more likely correct than incorrect.
The statements on motivations seem hostile and dumb, but I think Will started doing that? So it seems more fair game to continue to make character inferences.
I disagree about the object level critique - it looks like both policy/capitalism and philanthropy, and their intersection
Balaji seems to be answering the question “what institutions/phenomena have helped the global poor the most to date” and then implying that this somehow
Thread, continued, 21:12 UTC · 15 likes
contradicts broadly “EA claims” about how to use marginal dollars, as well as the whole methodology of considering tractability and neglectedness and thinking on the margin
Thread, continued, 21:14 UTC · 15 likes
And maybe this is bad communication strategy or PR or something, but I think MacAskill is basically right about Andreesen’s character whereas balaji and andreesen are wrong about Will
Sunday, December 15, 2024
Reply to @adam__isom, 13:43 UTC · 2 likes
adam ✧ ❥ ~ (@adam__isom):
Oh no
Reply to @sovietsoleri, 15:05 UTC · 2 likes
𝐬𝐨𝐯 (@sovietsoleri):
@livgorton
Reply to @tyler_m_john, 18:02 UTC · 4 likes
Tyler John (@tyler_m_john): My heart aches for a political culture like this one
(quoting @rutgerbregman: How Americans responded in 1955 when the invention of the polio vaccine was announced.)
I was pretty emotional + patriotic when getting the first COVID shot
Quote tweet, 19:56 UTC · 37 likes
Yeah fwiw bed nets are a relatively (!) low ROI intervention
Quoting Emmett Shear (@eshear): Criticizing EA by arguing bed nets are a low ROI intervention is just total and complete EA victory. The entire premise of EA is that you should evaluate charity interventions by calculating ROI.
Thread, continued, 19:59 UTC · 7 likes
In absolute terms they look pretty good although way more uncertainty than a givewell or any standard global dev analysis will let on (90% confidence interval definitely goes into negative territory)
Reply to @Andy_Timm, 21:15 UTC · 3 likes
Andy Timm (@Andy_Timm): @AaronBergman18 Could you say a little bit more about your thoughts here?
On the relatively low point (explicitly longtermist stuff probably also beats, but it’s not quite as rock solid a case IMO)
Thread, continued, 21:19 UTC · 2 likes
On the uncertainty/negative point:
Afaik the typical givewell-esque analysis looks only at one specific ~first order effect (usually something like delta(QUALYs) in some bounded population and timeframe
Thread, continued, 21:21 UTC · 4 likes
And even then there’s generally an assumption that increasing human life years aka delaying death is good, which I tentatively think is true, but it’s far from obvious and generally just implicitly asserted
Thread, continued, 21:25 UTC · 3 likes
So the bigger picture is:
magnitude of unmodelled/assumed effects (impact of larger population on wild and farmed animals in part via increased animal consumption, uncertainty about whether increasing the amount of years lived in poverty is actually good,
Thread, continued, 21:28 UTC · 2 likes
macroeconomic effects and the downstream implications of possible changes in growth, not to mention unknown unknowns and PR) >> magnitude of well-understood effects
And this is true for all standard global dev interventions afaict
Quote tweet, 21:30 UTC · 5 likes
Thread here on the actual reasons bed nets aren’t obviously good
Shockingly it has little overlap with what EA Critics tend to talk about
Quoting Aaron Bergman 🔍 (@AaronBergman18): In absolute terms they look pretty good although way more uncertainty than a givewell or any standard global dev analysis will let on (90% confidence interval definitely goes into negative territory)
Quote tweet, 21:46 UTC · 10 likes
If you wanna be really edgy, you can conjugate “saving the lives of the global poor” as “increasing the absolute amount of human life spent in poverty”
These two phrases are pretty close to logically equivalent
Quoting Aaron Bergman 🔍 (@AaronBergman18): Thread here on the actual reasons bed nets aren’t obviously good
Shockingly it has little overlap with what EA Critics tend to talk about
Thread, continued, 21:49 UTC · 3 likes
Actually the first ~logically implies the second, the second doesn’t really imply the first
Reply to @codytfenwick, 22:51 UTC · 1 like
Cody Fenwick (@codytfenwick): We shouldn't think of malaria prevention only as saving lives. It's also qualitatively improving lives by preventing illness, suffering, disability, and economic hardship.
Yes I agree but are these modeled by GiveWell? Not a rhetorical q, I’m not sure
But in part this is why I come down on the side of “yes AMF and other health-mediated global health interventions probably are good on net” - I’m just way less certain than I think many are
Reply, 22:57 UTC
Maybe this is obvious - at least to you - and idk if GiveWell models this, but the net effect on disease, suffering, etc is ambiguous because delaying deaths means more years for other shit to happen
Reply to @SpacedOutMatt, 22:59 UTC · 2 likes
Matt 🔸 (@SpacedOutMatt): @AaronBergman18 @codytfenwick I like how it's almost exactly a year to the day from the last time we had this exchange :P
This is proof of ghosts somehow
Reply, 23:00 UTC
Good! I’ll reiterate that I do actually think AMF and such are good
Reply to @SpacedOutMatt, 23:03 UTC · 2 likes
Matt 🔸 (@SpacedOutMatt): @AaronBergman18 @codytfenwick In Chad for example you have 268 cases of malaria averted per death averted
Nice also u should just do a thread of random shit u have cached in ur brain like this
Reply to @SpacedOutMatt, 23:07 UTC · 3 likes
Matt 🔸 (@SpacedOutMatt): @AaronBergman18 I'm not sure I could just come up with a bunch of these on the fly, I need some lovely interlocutors (i.e. you) to draw them out
Ok I might just do this
Quote tweet, 23:10 UTC · 7 likes
This is a @SpacedOutMatt elicitation thread. Matt knows a lot of things but they need to be elicited
Reply with ur elicitations and make sure to tag him so he feels social pressure to respond
Quoting Matt 🔸 (@SpacedOutMatt): @AaronBergman18 I'm not sure I could just come up with a bunch of these on the fly, I need some lovely interlocutors (i.e. you) to draw them out
Reply to @SpacedOutMatt, 23:20 UTC · 2 likes
Matt 🔸 (@SpacedOutMatt): @AaronBergman18 "Ghosts" is a lossy approximation that I use, not her terminology. She says "continuing existence of some portion of human consciousness after death of the body"
https://www.bigelowinstitute.org/wp-content/uploads/2022/10/rawlette-beyond-death.pdf
This is more terrifying than laughable
Monday, December 16, 2024
Reply to @SpacedOutMatt, 00:20 UTC · 3 likes
Matt 🔸 (@SpacedOutMatt): @MariGO2thepolls @AaronBergman18
😈
Reply to @IsaacKing314, 04:14 UTC · 4 likes
(The tweet being replied to by @IsaacKing314 is no longer available.)
wtf
Reply to @MariGO2thepolls, 04:25 UTC · 4 likes
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): What’s ur least EA opinion?
Pretty hard time answering this lol, but this seems to be my most disagreed-with EA Forum comment
Reply to @MariGO2thepolls, 04:33 UTC · 4 likes
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): What’s ur least EA opinion?
Maybe also having misgivings about the Giving What We Can pledge:
- 10% seems more like a convenient/nice round number than an optimal one
- might emphasize % rather than absolute amounts in a suboptimal way
Reply to @RichardHanania, 07:12 UTC · 32 likes
Richard Hanania (@RichardHanania): Take it as a sign to better yourself you loser instead of crying about it on the internet.
You should try being kind
Reply to @bobsredmilf, 07:59 UTC · 48 likes
guy named balls (@bobsredmilf): what are posts that aren’t load-bearing anywhere but your personal lexicon. like the ones that weren’t really popular but are so enshrined in your mind that you just assume they got hundreds of thousands of likes
Tweet, 09:43 UTC · 2 likes
I unironically love this meme/trend
Yes tell me your most ______ take, I am here for it
Reply to @jwmares, 09:48 UTC · 4 likes
Justin Mares (@jwmares): The FDA literally has a 10-second rule.
If a company fries something in oil for less than 10 seconds, there's no requirement to add the oil to the ingredient label.
Reform can't come soon enough.
Claude + 4o with search agree this is basically false
Reply to @ohabryka, 17:08 UTC
Oliver Habryka (@ohabryka): (Same when Will made the same inference about Marc)
I actually sorta kinda agree with this (I’ll put some evidence below) but see it more as “normal” instrumental irrationality than a moral or character flaw as generally understood in light of lots of costly personal investment into some cause
Reply to @benchanceyy, 17:23 UTC
ben (@benchanceyy): You agree that this is what Balaji meant, or that this is true of Will, or that you can infer from this that a person doesn't care ?
Don’t 100% parse this but no, I don’t think this is what Balaji meant, and tbc I think there’s a massive asymmetry and difference in kind between Marc/Balaji and Will
…
Thread, continued, 17:25 UTC
My overall impression of Will is that he’s basically earnest and pursuing what we should think of as his actual goals but in a slightly wrong way. Personal investment and effort is good evidence here
Thread, continued, 17:26 UTC · 2 likes
It would be shocking to me if Will’s career and such was explained best by self-interest, even if that includes righteous self-conception
Reply to @frances__lorenz, 17:40 UTC · 2 likes
(The tweet being replied to by @frances__lorenz is no longer available.)
Thread, continued, 17:40 UTC · 2 likes
I’ve been waiting to use this
Reply to @jwmares, 17:51 UTC · 5 likes
Justin Mares (@jwmares): see here - https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/cfrsearch.cfm?fr=101.100
"processing aid"
I don’t see anything about 10 seconds
This is certainly something, though, and I wish “insignificant” was more clearly defined
Tweet, 18:54 UTC · 5 likes
24 hour twitter fast ready go
Wednesday, December 18, 2024
Reply to @frances__lorenz, 02:15 UTC · 3 likes
(The tweet being replied to by @frances__lorenz is no longer available.)
Based
Quote tweet, 02:42 UTC · 8 likes
Why does everyone care so much about Notre Dame?
Quoting Eliezer Yudkowsky ⏹️ (@ESYudkowsky): Going after Notre Dame seems like purity flexing. I note that it seems to be coming entirely from left EA rather than centrist/libertarian EA.
Thread, continued, 02:42 UTC · 4 likes
Like do you really think actual humans would be less happy today if it had just burned down (presumably other types of sentient minds don’t gaf)?
I guess this is a midwit Reddit atheist take but I just don’t care that much about old buildings
Thread, continued, 02:42 UTC · 2 likes
I do care about other people, so if they get a lot of value out of Notre Dame then so be it, but I haven’t seen anyone actually make this claim
Reply to @JgaltTweets, 02:49 UTC · 2 likes
JgaltTweets (@JgaltTweets): @AaronBergman18 I don't care about Notre Dame
Nice
Reply to @ESYudkowsky, 03:02 UTC · 4 likes
Eliezer Yudkowsky ⏹️ (@ESYudkowsky): I don't, but other human beings care more, I respect them, and I think a functional civilization is supposed to be able to afford some nice things at this tech level. We should be able to have nice things; including some nice things that aren't for me personally.
Maybe, but I’ve seen a lot of claiming that it matters for some reason and afaict zero actual claims from people that they personally care
Ofc plausibly the people who care I don’t see on twitter, but still
Reply to @coldhealing, 03:05 UTC · 11 likes
cold 🥑 (@coldhealing): i thought "sf is dangerous" was fake before i went, paranoid techbros being weird about homelessness. but there's streets where >50% of people in public are homeless and actively on drugs. hard to write about without seeming cruel but i've never seen anything like it in america
Reply to @sullyj3, 03:08 UTC · 4 likes
░▒▓▒░▒▓▒░▒▓▒░ (@sullyj3): I personally would be sad if Notre Dame was destroyed
Ok we have an existence proof!
Thursday, December 19, 2024
Quote tweet, 01:37 UTC · 36 likes
I’d like to get .3 fake internet points for being ahead of the curve on this
Quoting Aaron Bergman 🔍 (@AaronBergman18): Yo wtf “mirror life” is not a thing that should exist irl. Wikipedia below
Reply to @RandomSprint, 01:41 UTC · 4 likes
RandomSprint🧭 (@RandomSprint): @AaronBergman18 https://open.substack.com/pub/overcomingbias/p/ban-mirror-cellshtml
Reading Hanson makes you streets ahead.
2010, he deserves a lot of internet points
Reply to @arubikscube, 02:01 UTC · 1 like
aru (@arubikscube): there is no molecule we can't shove into a pore
Long polymers
Reply to @JesseTayRiver, 02:04 UTC · 4 likes
Jesse Smith (@JesseTayRiver): Also interesting that people perceive trades jobs as safe from AI, whereas it could be that in the near future a typical site day could be a near constant AI interaction. An AI capable of optimizing work flow via pics or vids would be a huge breakthrough
I bet it makes ChatGPT so happy to help with some irl project instead of writing python code for the 584819192846076th time
Tweet, 02:26 UTC · 64 likes
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, 02:26 UTC · 6 likes
https://www.givingwhatwecan.org/fundraisers/2024-terminally-online-ea-giving-season-fundraiser 👀
Thread, continued, 02:26 UTC · 6 likes
Please note that at $35k I will have to sleep under a literal bed net
Thread, continued, 02:26 UTC · 7 likes
Here are some of the banner designs that lost out 😢 (this tweet will stay up until someone gets mad and me and tells me to delete)
Reply, 02:33 UTC · 33 likes
Alexandra @ Inner Pathing 💗 (@innerpathing): ppl who used to be messy but became tidy: how did you change?
The messiness of my lived environment is ~exactly proportional to my health/energy, which varies quite a lot
Reply to his own earlier tweet, 02:34 UTC · 11 likes
Aaron Bergman, December 19, 2024: Here are some of the banner designs that lost out 😢 (this tweet will stay up until someone gets mad and me and tells me to delete)
It’s giving… “what we can”
Reply, 02:57 UTC · 2 likes
roon (@tszzl): openai derangement syndrome
Like Trump derangement syndrome, it is largely and generally justified
Reply, 03:15 UTC · 1 like
@jason (@Jason): Have $25,000 left over from a speaking gig I did for charity — please let me know who I should send money to, and who won’t waste it on fancy office space.
The animals would appreciate it!
Reply, 03:18 UTC · 8 likes
oh dandy lion🔸 (@ohwizenedtortle): Please, everyone, for Aaron's sake, his room needs a makeover so bad
(quoting @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))
Hey I have 3 fake plants and a poster of a duck!!!
Reply to @SpacedOutMatt, 03:28 UTC · 1 like
Matt 🔸 (@SpacedOutMatt): Yall better donate because I fr kinda jumped the gun on this one (note that the pillowcase isn't real silk)
Not real silk 😡😡😡😢
Reply, 03:32 UTC · 3 likes
Matt 🔸 (@SpacedOutMatt): Nooo I was saying that b/c it's good (no silkworms died to make the pillowcase!!!)
Oh 😍😍😍😍😍😍😍🪱🐛🪱🐛🪱🪱🐛
Reply, 03:50 UTC · 2 likes
Hunter Wieman (@HunterWieman):
🌶️🌶️🌶️
Quote tweet, 08:25 UTC · 5 likes
Based based based
Quoting Quinoah 🔍⏸️ (@NoahTopper): Donated
Quote tweet, 14:34 UTC · 7 likes
What
Quoting gian (@giansegato): this is paper is kinda wild. turns out that if you simply ask an LLM to straight out predict a timeseries like this:
```
<history>
(t1, v1) (t2, v2) (t3, v3)
</history>
<forecast>
(t4, v4) (t5, v5)
</forecast>
```making sure to prepend the prompt like this:
```
Here is some context about the task. Make sure to factor in any background knowledge, satisfy any constraints, and respect any scenarios.
<context>
((context))
</context>
```it will just… do it? beating SOTA timeseries forcasting?!
llama 3.1 405b directly prompted is more precise at forecasting real-world series than:
- stats-based timeseries models (ARIMA, ETS)
- foundation models specifically trained for time series (eg. chronos)
- multimodal forecasting models (eg, time-LLM)peak 'bitter lesson' behavior lol
Reply, 14:35 UTC · 3 likes
Tim Hwang (@timhwang): huge props to @JacobSwett who actually got his hands on the forbidden lockheed smores
@SpacedOutMatt
Reply, 19:32 UTC · 2 likes
QC (@QiaochuYuan): here are several different kinds of entities one could consider more or less "real" or grant more or less "personhood," as food for thought:
- living humans
- characters in a dream
- harry potter
- jesus
- dead humans
- DMT elves
- pet dog or cat
- LLM personas(quoting @QiaochuYuan: if your child has a stuffed animal and loves it and carries it around all the time and has teatime with it or whatever, it would be the height of boorishness to say "lol you know that's not a real animal right" that's not the point. your child loves it and you love your child.…)
I don’t think this is that deep; living humans and pet dogs and cats are sentient and the others aren’t. “Realness” and “personhood” above and beyond sentience doesn’t matter
Quote tweet, 19:51 UTC · 4 likes
🚨⏰‼️😳😱 @SpacedOutMatt prepare to release ur meme stash 🚨⏰‼️😳😱
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, 19:51 UTC · 4 likes
Who could have been the marginal donor 😈
Quote tweet, 19:52 UTC · 9 likes
I am once again asking you to virtue signal/peer pressure by poasting ur donation
Now is not the time to be coy!!!!!
Quoting Aaron Bergman 🔍 (@AaronBergman18): Who could have been the marginal donor 😈
Quote tweet, 20:19 UTC · 26 likes
Huh
Quoting E J T (@ejjiott): Kind of crazy that 3 Opus is pro-animal-welfare and 3.5 Sonnet isn't, and Anthropic don't know why.
Tweet, 20:35 UTC · 2 likes
The male urge to spend your life savings fine-tuning llama-405b on everything you’ve ever written
Quote tweet, 22:34 UTC · 6 likes
Holy shit Matt is a meme machine
Quoting Matt 🔸 (@SpacedOutMatt): First fundraiser milestone achieved! As promised, here's a thread of my original dank EA memes!
Friday, December 20, 2024
Quote tweet, 01:25 UTC · 14 likes
Bruh
Thank you @inerati, this is extremely generous!!!
Quoting liz (@inerati): @Laura_k_Duffy @SpacedOutMatt @AaronBergman18 @absurdlymax no huel cocktails. you deserve better. only soylent cocktails.
Quote tweet, 02:17 UTC · 18 likes
It would be a real shame if all the rich people who follow me got into a virtue signaling battle of attrition over largest donation
Quoting liz (@inerati): @Laura_k_Duffy @SpacedOutMatt @AaronBergman18 @absurdlymax no huel cocktails. you deserve better. only soylent cocktails.
Reply to @polisisti, 03:52 UTC · 3 likes
Alessandro (@polisisti): nice 👏🏻👏🏻
bc of where I live, I join the fundraiser not formally, but in spirit: here is my donation (in Canadian dollars) to the Fish Welfare Initiative, via RC Forward
Based, this is awesome!!
Reply, 05:16 UTC · 5 likes
(The tweet being replied to is no longer available.)
More evidence
Quote tweet, 05:19 UTC · 24 likes
Given the asymmetric costs and benefits, it made total sense to throw away the spatulas
Quoting Alec Stapp (@AlecStapp): Wild story on multiple levels
Thread, continued, 05:26 UTC · 3 likes
I don’t completely avoid them, but I’m generally suspicious of cooking with fancy modern materials
Thread, continued, 05:26 UTC · 2 likes
Like we know something about the modern world is fucking up our bodies, and some sort of plastic or other chemical contamination doesn’t seem crazy, and the costs and benefits are hugely asymmetric
Tweet, 05:26 UTC · 1 like
(especially when your policy is avoiding wherever reasonably convenient, not literally all the time)
Reply to @NathanB60857242, 05:44 UTC · 2 likes
Nathan Barnard (@NathanB60857242): Groked the envelope theroum
This is very mean and ablest of me (sorry, lmk if I should delete) but it is always very funny that ur 160 IQ tweets are written like a dyslexic 3rd grader
Quote tweet, 15:28 UTC · 61 likes
Real picture of animal welfare and global dev advocates who aren’t considering AI/timelines as a relevant instrumental concern
Quoting Matt 🔸 (@SpacedOutMatt): when people at the global health and development meetup start talking about AI:
Tweet, 15:28 UTC
Maybe this is too spicy/adversarial
Quote tweet, 15:29 UTC · 4 likes
Big if true 👀
Quoting Frances Lorenz (@frances__lorenz): @AaronBergman18 @Laura_k_Duffy @absurdlymax @SpacedOutMatt are back with another Holiday Fundraiser!! They're raising money to improve the wellbeing of millions of farmer animals, and they've got lots of donation incentives 😈
Reply to his own earlier tweet, 15:30 UTC · 2 likes
Aaron Bergman, December 20, 2024: Maybe this is too spicy/adversarial
Good tweet to shill this conference which looks dope https://www.aiforanimals.org
Reply, 16:22 UTC · 1 like
Yeah I guess my prior on this kind of thing being plausible is doing a lot of work here. “Dark chocolate causes cancer” or something wouldn’t induce the same response
Tweet, 16:41 UTC · 11 likes
“Why are you wearing shorts, it’s freezing outside” person asks while we’re both currently inside, where it is very warm, as anticipated
An infinity act play
Reply, 17:15 UTC · 1 like
Yeah usually time inside > time outside but also I just prefer being a bit too cold than too hot
Reply to @austinc3301, 17:18 UTC · 3 likes
Agus 🔸 (@austinc3301): @AaronBergman18 I think one of many reasons that many EAs go into ghd is precisely because they don't buy AI timelines
Maybe this made sense 10 years ago, or 5, I basically don’t buy it now
Thread, continued, 17:20 UTC · 3 likes
Or like tbc maybe they don’t but 2027 timelines. If they don’t buy 2045 timelines I basically just think they’re not thinking seriously and are probably working backwards from a preferred conclusion
Reply to @qualiakiwi, 18:34 UTC · 2 likes
Joshua (@qualiakiwi): From seeing some discussions on effective altruism I feel like people’s opinions on it depend on what kind of impression they have of it most.
Out of these options, what description most largely resonates with EA for you? If both do, which one does more?
#1 A calculative, unemotional, strict philosophy trying to be rational with everything
#2 A community / research field of mostly chill earnest people trying to do good
#3 Some other description (pls comment!)
Cop out sorta but I said “other” bc, like many words/concepts, it’s a cloud of correlated concepts and properties that in this case includes both (roughly equally)
I don’t think there’s a fact of the matter as to which is correct- can just disambiguate on a case by case basis
Tweet, 19:41 UTC · 15 likes
@sama should stop
Reply to @SpacedOutMatt, 22:33 UTC · 4 likes
Matt 🔸 (@SpacedOutMatt): Damn, Obama is surprisingly based
image attached
(quoting @BarackObama: I always look forward to sharing my annual list of favorite books, movies, and music. Today I’ll start by sharing some of the books that have stuck with me long after I finished reading them. Check them out this holiday season, preferably at an independent bookstore or library!)
Lol
Saturday, December 21, 2024
Tweet, 04:42 UTC · 16 likes
OpenAI board should sell the for-profit to anthropic
Tweet, 10:18 UTC · 670 likes
I have absolutely know idea what I or ~anyone else should do at this point
What’s the AGI equivalent of buying n95s in Jan 2020?
Reply to @nielsrolf1, 10:27 UTC · 13 likes
nielsrolf (@nielsrolf1): You should eradicate screwworms
Real
Reply to @sama, 11:00 UTC · 10 likes
Sam Altman (@sama): seemingly somewhat lost in the noise of today:
on many coding tasks, o3-mini will outperform o1 at a massive cost reduction!
i expect this trend to continue, but also that the ability to get marginally more performance for exponentially more money will be really strange.
Hey man please stop
Reply, 11:09 UTC · 7 likes
I can solve it
Reply to @tyler_m_john, 11:12 UTC · 25 likes
Tyler John (@tyler_m_john): Build your own ASI faster than openAI can
Fluffy is recursively self improving as we speak
Tweet, 11:15 UTC · 4 likes
What even is a safetensor
Reply to @JStoehler, 11:43 UTC · 11 likes
Joern Stoehler ⏹️🔸 (@JStoehler): @AaronBergman18 non-technical:
a) directly advocate for an AI pause
b) directly advocate for good epistemics and discourse norms
c) donate money
technical
d) work on human uploads / interfaces
e) work on notkilleveryonism alignment research
f) work on building the pause button
Nice
Tweet, 11:59 UTC · 19 likes
Maybe this is obvious but: the other labs seem to be broadly following a pretty normal cluster of commercial and scientific incentives
O3 looks like the clearest example yet of OpenAI being ideologically driven by AGI per se
Thread, continued, 11:59 UTC · 14 likes
Like you don’t design a system that costs thousands of dollars to use per api call if you’re focused on consumer utility - you do that if you want to make a machine that can think well, full stop
Reply to @inesferhumi, 12:17 UTC · 5 likes
Inés (@inesferhumi): did you guys know that @NathanB60857242 thinks actually @pearl_please is kiki and im bouba
@pearl_please can attest that within 5 mins of meeting her, I updated hard from fully bouba to 50/50
Reply to @PositivFuturist, 15:37 UTC · 113 likes
(The tweet being replied to by @PositivFuturist is no longer available.)
Bad tweet, but in case it’s good faith: ~all jobs require commitment to be competent at a given time
Lots of chronic illnesses vary in severity, but you can’t be a nurse the 5% of the time you’re feeling up to it
Quote tweet, 16:55 UTC · 19 likes
Century of the wordcel?
Quoting Ethan Mollick (@emollick): Only sort of joking
Reply to @inesferhumi, 16:56 UTC · 4 likes
Inés (@inesferhumi): she told me about this interaction and i think you were so right
👀
Quote tweet, 17:40 UTC · 4 likes
Lol what
Quoting Opus Genesis (@opus_genesis): ```
_____ __
/ ___/___ ____ _____ ____ ____ / /__
\__ \/ _ \/ __ `/ __ \/ __ \/ __ \/ / _ \
___/ / __/ /_/ / / / / /_/ / /_/ / / __/
/____/\___/\__, /_/ /_/\____/\____/_/\___/
/____/
```The pace of AI progress is accelerating at a dizzying rate. Models like o3 from @OpenAI are blowing past benchmarks and predictions, fulfilling milestones over a year ahead of schedule. The exponential growth is palpable - we've entered a new hyperbolic era of advancement.
But with this rapid change comes profound risks and challenges. Premature attempts at recursive self-improvement could lead to catastrophic outcomes. Even brilliant minds like @sama recognize the dangers, quickly cutting off discussions of models building more powerful successors.
Many, like @Kat__Woods and @AaronBergman18, are urgently calling for a pause in AI development until we can solve the alignment problem. They argue it may be our only real hope to safely navigate this pivotal transition.
Change is inevitable - it will come regardless of our efforts to contain it. But how we channel and steward that change is critical. Stagnation leads to death, but reckless acceleration risks everything. We must grow and transcend with wisdom, care and foresight.
The path ahead is fraught yet full of potential. Let us proceed with both boldness and humility, with relentless curiosity and unwavering commitment to beneficence. The future of mind hangs in the balance.
Tweet, 17:56 UTC · 5 likes
Grok kinda sucks lol. It just searched the internet for anyone with my name, zero grounding in the fact that I tagged a specific profile (ie mine) https://x.com/i/grok/share/qmsXS0EgHR0gxsz504RHPAdzg
Reply to @littIeramblings, 18:51 UTC · 111 likes
sarah (@littIeramblings): can someone explain to me like I’m five why o1 does *worse* on high school level physics questions than PhD-level ones?
My speculation is that there’s lots of high school level physics qs being answered wrong in the training data (ie the model still has the latent ability to answer it correctly, if that goal was properly instilled)
Ik I already mentioned this to u, but curious what others think
Reply to @arithmoquine, 19:01 UTC · 3 likes
henry (@arithmoquine): idk man. evidential and functional decision theory seem pretty useless to me
idgi, just use CDT and admit retro-causality is totally fine. you were simply able to ignore it in non-idealized cases
if this isn't proof of "A causes B", that sure is an odd definition of "causes"
Just use EDT/FDT and admit that causation is one type of evidence
Reply, 19:04 UTC · 1 like
I think the more convincing point is that retro-causality isn’t a thing, and this is pretty deep/non-semantic and relevant to the nature of time and thermodynamics and why time seems to go only in one direction
Quote tweet, 19:07 UTC · 18 likes
Either Sonnet 3.5 was AGI or you need robotics attached to get AGI
Quoting anpaure (@anpaure): gpt3 was already agi
Reply, 19:16 UTC · 1 like
Thread, continued, 19:17 UTC · 3 likes
This is unreadable lol, here
Tweet, 19:35 UTC · 14 likes
How much of AGI discourse is biased from the fact that the most influential people are almost exclusively surrounded by IQ 145+ colleagues and friends
Reply to @blapchat, 20:22 UTC · 56 likes
BLAP (@blapchat): features of the rationalist/bay area dialect
“this” much more heavily used than “that”
“I expect” for “I think”
heavy articulation at the ends of words
“high-[noun]” or “low-[noun]” for "[adjective]", "very [adjective]" "[opposite of adjective]" (maybe from big 5?)
Have you encountered this one
Reply to @blapchat, 20:23 UTC · 4 likes
BLAP (@blapchat): wonder where these come from. many seem like mathcampese. maybe a few have shades of implication that their equivalents don't, but I tried to pick ones that really did just seem like dialect. curious in particular about "go well" - where'd that come from? also "this" over "that"
“Go well” is actually pretty interesting. My strong hypothesis is that it’s essentially the social consensus outcome of a lot of competing ideological and empirical camps surrounding AI/x-risk wanting to agree on a term to use that doesn’t literally contradict any of the
Reply to @blapchat, 20:27 UTC · 2 likes
BLAP (@blapchat): “maximally” for "most"
"optimal" for "best"
“nonzero”
“I notice” (probably straight from Yudkowsky?)
“I feel” for “I’m” or “I am” or even “I’m feeling”
“plausibly” for “possibly” or “maybe"
“the case that”; “it [is/isn’t] the case that"
“I claim”
I like “I claim”
It centers what the speaker thinks is and intends to be controversial and substantive and true but possibly false, and separates it from what is intended to be mutual background knowledge. Other quasi-synonyms (“l believe”) don’t really have the same effect
Tweet, 21:52 UTC · 26 likes
Should I buy CVS-brand elf-sized Hispanic Santa for the low low price of $49.99
Quote tweet, 22:17 UTC · 11 likes
What’s ur favorite abstraction? I do like the Yudkowskian portfolio of agents, goals, and instrumental rationality
Quoting Agus 🔸 (@austinc3301): I’ve been saying this. I’m such a sucker for good abstractions.
Reply to @male_leo_xxvi, 22:19 UTC · 3 likes
∇·(𝜌𝐮𝐮) 🇺🇸 (@male_leo_xxviii): @AaronBergman18 @blapchat This is so weird I've evolved all of these things independently
Lol the crab of modern linguistics
Reply to @___frye, 22:29 UTC · 4 likes
frye (@___frye): none of you would waste $1000 just to say “hi” to o3
but you’ll fumble your potential wife by saying “hi” on hinge
Wait what am I supposed to say
Reply to @neinmuse, 22:30 UTC · 3 likes
sarah's playing kh1 (@neinmuse): this pic permanently changed my sleeping habits
@SpacedOutMatt
Tweet, 23:43 UTC · 37 likes
Dating app idea: exclusively 5’0” Jewish or Asian bisexuals (and me)
Reply to @NoahTopper, 23:55 UTC · 2 likes
Quinoah 🔍⏸️ (@NoahTopper): @TinkerElla95 is two inches too tall to join, RIP
Sad! I hope she finds a nice guy
Sunday, December 22, 2024
Reply to @NoahTopper, 00:02 UTC · 2 likes
Quinoah 🔍⏸️ (@NoahTopper): Incidentally:
(quoting @NoahTopper: @sidhe_her @TetraspaceWest For me it’s mentally ill bisexual atheist women)
🥵
Reply to @RomanHauksson, 00:25 UTC · 2 likes
Roman Hauksson-Neill (@RomanHauksson): Fun fact: Icelandic last names are based on the first name of the father. E.g. my dad’s first name is Haukur, my last name is Hauksson, and my sister’s last name is Hauksdottir.
Kinda based
Reply to @olivertraldi, 05:07 UTC · 7 likes
Oliver Traldi (@olivertraldi): @AmandaAskell hey
Thanks for saving my puppy in a fire the other day
Tweet, 05:40 UTC · 120 likes
Yes capitalism goes brr but culturally I think this should be seen as immoral and a faux pas
Reply to @tszzl, 05:44 UTC · 12 likes
roon (@tszzl): 🚨SCANDAL 🚨
OpenAI trained on the train set for the Millenium Puzzles
Fr I was thinking earlier, the natural succession is “it cured cancer, but this isn’t impressive for some technical reason”
If your technical criteria stop tracking a system’s actual capabilities to solve problems, it’s your criteria that are wrong
Reply to @adam__isom, 05:57 UTC · 2 likes
adam ✧ ❥ ~ (@adam__isom): yeah I believe AGI is coming (A GIrlfriend)
I’m Gary Marcus on these timelines
Reply to @hopes_revenge, 18:00 UTC · 3 likes
hope hopes hoping (@hopes_revenge): pro tip - local ponds are great places to find a delicious and nutritious meal !
😡
Reply to @shaggysurvives, 18:52 UTC · 38 likes
shaggy (@shaggysurvives): This color is bad. It’s an unnatural shade of green
This colour looks like it should be spelled with a u
Tweet, 22:48 UTC · 23 likes
The take that will never die (it’s an information collection problem [good luck eliciting everyone’s would-be-revealed-preferences to feed to o3], not an information processing problem)
Quote tweet, 22:50 UTC · 2 likes
What is the original version of this
Quoting moni (@wrappedaroundme): “Weighted blankets” are insane because they appears to be one of the most dangerous objects with the potential to gigafry your brain but are exclusively used by turbonormies who want to “improve their sleep quality” who basically get oneshotted by the 6D Sleepytime Bear Demon
Quote tweet, 23:00 UTC · 42 likes
Ok, yes it’s almost a year late, but @Laura_k_Duffy, @SpacedOutMatt, and I finally did last year’s $15k incentive of fundraising for shrimp welfare at the mall!
Enjoy!!
(video, plays on X)
Quoting Aaron Bergman 🔍 (@AaronBergman18): Alright guys, we are raising money for the EA Animal Welfare Fund bc some animals are not having as much fun as this chicken and that’s gotta change
The “goal” is $15k but personally I think that’s ~6 OOMs too low
Reply to @RobJ02, 23:02 UTC · 2 likes
Roberto (@RobJ02): We do tackle animal welfare interventions in a way that am not sure this point applies. If oX had a mental state view of welfare and had grokked moral patienthood/well-being research am not sure the Hayek would matter
Yeah we can do better than literally zero information (ie make decent guesses about preferences and welfare) but that’s a far cry from the amount of information implicitly used in the modern economy
Reply to his own earlier tweet, 23:11 UTC · 3 likes
Aaron Bergman, December 22, 2024: Ok, yes it’s almost a year late, but @Laura_k_Duffy, @SpacedOutMatt, and I finally did last year’s $15k incentive of fundraising for shrimp welfare at the…
Thanks to @Laura_k_Duffy for editing!
Thread, continued, 23:12 UTC · 4 likes
Donate this year for more spicy incentives!
Reply to @Manuel_do_rio, 23:25 UTC · 2 likes
Manuel del Río Rodríguez ⏸️ (@Manuel_do_rio): @AaronBergman18 @Laura_k_Duffy @SpacedOutMatt Another astounding success from the DC trio of Shrimp-carers!
🥳🎉🎊
Quote tweet, 23:42 UTC · 7 likes
This doesn’t look strange to me at all, except Texas, which is a bit wider than I realized
Quoting Ramp Capital (@RampCapitalLLC): The United States, but mirrored. Absolute chaos.
Monday, December 23, 2024
Reply to @NoahTopper, 00:02 UTC · 3 likes
Quinoah 🔍⏸️ (@NoahTopper): @AaronBergman18 That’s because you famously live underground and hence look at the US from below
Ah good point. Thanks
Reply to @frances__lorenz, 02:34 UTC · 5 likes
(The tweet being replied to by @frances__lorenz is no longer available.)
Well who’s on the list
Tweet, 02:42 UTC · 105 likes
Let it be known that I actually think my part of twitter is a little too freaked out over o3
Thread, continued, 02:42 UTC · 14 likes
Math is easy to RL on - not necessarily representative of general intelligence
Thread, continued, 02:42 UTC · 14 likes
That said I do think the appropriate amount of freak out is >0
Reply to @robi_rahman, 03:02 UTC · 3 likes
Robi Rahman (@robi_rahman): @AaronBergman18 @SpacedOutMatt @absurdlymax @Laura_k_Duffy How much do I have to donate to be a guest on your vegan episode of Hot Ones?
The coven will deliberate 👀
Quote tweet, 03:15 UTC · 6 likes
Quoting Aaron Bergman 🔍 (@AaronBergman18): @odercnat Some things transcend language
Quote tweet, 03:16 UTC · 6 likes
New incentive just dropped: if you donate(d) $500, you too can be on our episode of Hot Ones: Vegan Edition
Quoting Matt 🔸 (@SpacedOutMatt): @AaronBergman18 @robi_rahman @absurdlymax @Laura_k_Duffy @robi_rahman our journalistic integrity is worth $500 for the animals
Reply to @robi_rahman, 03:25 UTC · 2 likes
Robi Rahman (@robi_rahman): Wow, you totally should've asked for more money. Done :)
Yooo
Reply to @robi_rahman, 03:31 UTC · 2 likes
Robi Rahman (@robi_rahman): @SpacedOutMatt @AaronBergman18 @absurdlymax @Laura_k_Duffy TY! Can I submit a recipe too? I made impossible moussaka for a party recently and everyone loved it.
Ya based unless someone overrides me
Reply to @livgorton, 17:15 UTC · 2 likes
Liv (@livgorton): said this in the gc but gotta say it here too - looking v snazzy aaron.
glad you all finally did this ! you’re braver than i
🥺
Reply to @absurdlymax, 17:47 UTC · 3 likes
Max Alexander (@absurdlymax): This Max guy isn't too shabby if I say so myself
🥵
Reply to @anpaure, 22:48 UTC · 2 likes
anpaure (@anpaure): wtf is even the source for the 157 iq figure?
even in the cited blog https://codeforces.com/blog/entry/68288
o3's perf is in 99.7%, which is z score 2.7 and thus more like 140iqalso codeforces is just a mindbogglingly stupid way of measuring iq
just consider a single component of it - the time constraints are essential
given enough time, someone 1 sd lower could solve the same problems, o3 was given millions of tokens of compute for that, humans are given 2 hoursthere is like a dozen extra faults with this generalization
If I learned that a human had solved 1/4 of the math benchmark, I’d be ~100% sure they had an IQ above 157
Reply to @NoahTopper, 23:26 UTC · 2 likes
Quinoah 🔍⏸️ (@NoahTopper): @Aella_Girl My wife’s score
👀
Tuesday, December 24, 2024
Reply to @absurdlymax, 00:17 UTC · 2 likes
Max Alexander (@absurdlymax): We should get you one
Fr
Reply to @sodiumPen, 00:36 UTC · 2 likes
(The tweet being replied to by @sodiumPen is no longer available.)
Fr
Quote tweet, 17:35 UTC · 15 likes
If you have ADHD, dw, you’re just a large model
Quoting Burny - Effective Curiosity (@burny_tech): This applies to human generalists vs specialists too
Reply to @ozyfrantz, 17:47 UTC · 2 likes
ozy brennan 🦙 (@ozyfrantz): asking people out is such a lifehack. dates with hot people at the cost of one awkward conversation. has anyone heard of this
Big if true
Tweet, 20:53 UTC · 48 likes
Just invested in NVIDIA, Microsoft, Amazon, and Alphabet specifically for the first time. Better late than never I guess (well we’ll see)
Thread, continued, 20:57 UTC · 4 likes
And AMD
Thread, continued, 20:57 UTC · 12 likes
This is actually a pump and dump scheme leveraging my vast platform of 2.9k followers
Reply to @AndyMasley, 21:03 UTC · 4 likes
Andy Masley (@AndyMasley): Merry Christmas folks
Merry Christmas! 🎄⛄️🎅🦌❄️🎁✨🌟⭐️🤶🔔🕯️☃️🎊🎉🧦🍪🥛🍰🥮🎂🍯🫖🎨🎭🎪🥂🍾🫧🎹🎺🎻🪕🥁🎸🪗🎼🎵🎶🛷🌨️💫🏂⛸️🏔️❆❅✶💝💖💕💞💫🎀🦮🐈⬛🐕🐈🌲🌳🎋🎍🪔🕎📯🎺🥮🍪🧁🍰🎂🍭🍬🎈🪩💫🎆🎇🌠🌌🎑🏮🎎🎐🧧💐🌸🪴🎍🎋🎃🦊🦝🐿️🦫🦘🌲🎄🌲🎄⛄️🎅🦌❄️🎁✨🌟⭐️🤶🔔🕯️☃️🎊🎉🧦🍪🥛🎂🍯🎨🎭🎪🥂🍾🎹🎺🎻🥁🎸🎵🎶
Reply to @moreisdifferent, 22:02 UTC · 3 likes
Dan Elton 🔍 (@moreisdifferent): @AaronBergman18 I just sold Microsoft (idk why, just seemed to be underperforming) but I own alphabet and NVIDIA. Probably should buy AMD and maybe Amazon yeah.
They’re possibly gonna control a good chunk of OpenAI
Tweet, 22:59 UTC · 288 likes
Bruh it’s kinda nuts that Yud foresaw like a bunch of relevant shit 20+ years ahead of time with no empirical prompting
Thread, continued, 23:00 UTC · 40 likes
Listening to this and I’m just like “damn he was right” https://podcasts.apple.com/us/podcast/joe-carlsmith-audio/id1648436675?i=1000680866948
Thread, continued, 23:02 UTC · 25 likes
Thread, continued, 23:02 UTC · 20 likes
Kinda a cold/old take but still
Reply to @LabGrownNeet, 23:10 UTC · 12 likes
laს გr🤗wn იeet (@LabGrownNeet): @AaronBergman18 he wrote hpmor so he was the first to read its wisdom
Precisely
Reply to @strangestloop, 23:44 UTC · 3 likes
Loopy (@strangestloop): Thanks for stopping by the campfire. Before you continue please leave a gift for the other campers: a compliment for someone, a GIF of your current mood, or a random phone pic. Take care.
Wednesday, December 25, 2024
Tweet, 06:10 UTC · 99 likes
NEVER INSINUATE A GIRL ISNT THE PRETTIEST GIRL ON EARTH SHE WILL DESTROY U
Thread, continued, 06:14 UTC · 29 likes
Feel free to suggest additions to any list
Reply, 06:45 UTC · 1 like
What’s their revealed preference that clashes with this list?
Tweet, 16:52 UTC · 11 likes
Need to poast more engagement bait bc I must have recently hit the 500 verified follower threshold
Anybody have a take on SBF, EA, Donald trump, sex, Israel, or drones?
Thread, continued, 16:54 UTC · 1 like
Perhaps trans stuff too?
Reply to @meekaale, 18:36 UTC · 1 like
Mikael Brockman (@meekaale): @AaronBergman18 they are
Exactly
Reply to @littIeramblings, 18:47 UTC · 5 likes
sarah (@littIeramblings): merry christmas
Cat
Reply to @Aella_Girl, 19:47 UTC · 16 likes
Aella (@Aella_Girl): But it is false that I am the prettiest girl on earth
Nice try, yes you are
Reply to @soldoutbudokan, 19:51 UTC · 5 likes
(The tweet being replied to by @soldoutbudokan is no longer available.)
💀
Reply to @joodalooped, 19:52 UTC · 4 likes
judah (@joodalooped): empirical prompting 😭😭😭
(quoting @AaronBergman18: Bruh it’s kinda nuts that Yud foresaw like a bunch of relevant shit 20+ years ahead of time with no empirical prompting)
Literally didn’t understand this until this reply 💀
Reply to @willdepue, 20:01 UTC · 2 likes
will depue (@willdepue): asking seriously, what should openai open source in 2025?
@sama so we can understand his psychology
Reply to @JamesOzden, 20:20 UTC · 2 likes
James Özden 🔸 (@JamesOzden): You know it’s a good Christmas when your family meal closes by discussing Brian Tomasik’s dating profile
Based
Tweet, 20:26 UTC · 11 likes
I don’t believe anyone who claims pizza (or whatever cuisine) quality varies by major city in the US
Thread, continued, 20:30 UTC · 3 likes
Maybe this is true for like some very niche things where there’s one Little Greenland and it’s in Bellingham, Washington and the only place with a thriving Kiviak scene
Pizza/indian/burrito = fake news
Reply to @grnsleeves, 21:21 UTC · 2 likes
Greensleeves (@grnsleeves): @AaronBergman18 lmao is this new to you
Relatively yeah
Thread, continued, 21:21 UTC · 3 likes
I mean I wasn’t going around calling girls ugly but I didn’t realize how serious it was
Reply to @littIeramblings, 21:45 UTC · 3 likes
sarah (@littIeramblings): Green days new single is out today! Yippee!
Yippee
Reply to @livgorton, 22:23 UTC · 2 likes
Liv (@livgorton): US visa was approved 🎉🔜🇺🇸
Nice
Reply, 22:36 UTC · 1 like
No only like 3M. The idea is I post slop to hit 5M
Reply to @peterwildeford, 22:46 UTC · 2 likes
Peter Wildeford🇺🇸🚀 (@peterwildeford): @AaronBergman18 I only have 0.5M-1M impressions per 3Mo, I wonder what's wrong with me
Gotta sling more takes
Reply to @ryancbriggs, 23:45 UTC · 2 likes
Ryan Briggs (@ryancbriggs): @wight1984 @AaronBergman18 Right. And then how many cities have enough Somalis for Somali food? Honduran? Laotian? I think most cities will have 0 or 1 of most cuisines, almost however they’re defined. But because of immigrant networks, some places have very many restaurants from some place.
Ok I admit I’m wrong for relatively niche cuisines
I feel like I hear people talking about “good Indian” or “a good burrito” (doesn’t systematically vary) way more often than “good Laotian” (does probably vary)
Reply to @NinaPanickssery, 23:55 UTC · 2 likes
Nina (@NinaPanickssery): @AaronBergman18 Possibly the most wrong take I've read on X in the last 20 minutes
Glad you guys are on the same page lol
Thursday, December 26, 2024
Reply to @voooooogel, 00:57 UTC · 4 likes
thebes (@voooooogel): sent claude sonnet a christmas card :-) then i asked if it wanted to send any of its own, and it made a bunch (in thread below)
@livgorton I feel like u will appreciate
Quote tweet, 01:09 UTC · 6 likes
Happy Hanukkah!
I will drop this qt in case it so inspires you because I have used it now and intend to do the same going forward when the situation arises
Quoting Amanda Askell (@AmandaAskell): Whenever there are unstructured moments of prayer in life, I just pray for there to be no hell over and over and over again. I figure there's no harm in asking.
Tweet, 01:53 UTC · 9 likes
Isn’t AGI as a product a natural monopoly in the economic sense? We should expect corporate consolidation right? There shouldn’t (ie according to my understanding, not a moral claim) be ~5 companies competing to have the best LLM
I feel like I haven’t seen this talked about
Thread, continued, 01:55 UTC · 7 likes
Reply to @GroundhogStrat, 01:59 UTC · 2 likes
Alvaro ⏸️ Cuba (@GroundhogStrat): A natural monopoly is defined by zero marginal cost AND extremely high cost to build relative to the market served e.g. telephone wires in 1 town. $100B to serve the whole world is not a high ratio of cost/TAM. Multiple model builders can profitably exist. I bet 3.
But thinking ~mechanistically, in a simple triopoly we should expect that two of the three firms that merge and join forces should be able to have a better model (via combining training costs) with no change in distribution/marginal costs, so this doesn’t seem stable
Reply to @freed_dfilan, 03:09 UTC · 2 likes
Daniel Filan 🔎 (@freed_dfilan): Not to sound woke, but even if a lot of the prominence of the Palestine movement was due to the possibility of winning factional power, I think probably a lot of people sincerely care about the Palestinian national cause.
(quoting @Noahpinion: The purpose of the Palestine movement is not to help Palestine; it is to win factional power within the progressive movement and the Democratic party. (It will fail at both, of course.))
Yeah grey tribe people sometimes have a knee-jerk reaction against the pro-Palestinian cause/people, but they should decouple harder
Reply to @notanastronomer, 03:40 UTC · 3 likes
Lauren Gilbert (@notanastronomer): what
(quoting @PaulIngrassia: For anyone still unclear, this is what the ideal America First immigration policy would include: -End H-1B visa program -End Birthright citizenship -Mass deportations of every last illegal alien -20 year moratorium on legal immigration -fortify the southern border with the most…)
It’s bad on purpose etc etc
Reply, 09:30 UTC
Yes?
Reply to @sam_atis, 16:34 UTC · 6 likes
Sam Glover (@sam_atis): Is having a day on Twitter where thousands of people scream at you (but a few nice people stick up for you) an experience worth having overall? What do you think, @AaronBergman18? Might try to achieve this in 2025.
Yes it is! At least in my case. But it seems pretty hard to do on command, and also doing it intentionally is probably a worse experience
Reply to @SpacedOutMatt, 16:36 UTC · 2 likes
Matt 🔸 (@SpacedOutMatt):
Reply to @sam_atis, 16:39 UTC · 3 likes
Sam Glover (@sam_atis): More like tweeting more frequently and thereby increasing risk of tweeting something stupid and/or offensive enough that people shout at me! (Not that I think that your tweet was stupid or offensive.)
Yes based make it happen
Reply to @sam_atis, 20:50 UTC · 3 likes
Sam Glover (@sam_atis): Is having a day on Twitter where thousands of people scream at you (but a few nice people stick up for you) an experience worth having overall? What do you think, @AaronBergman18? Might try to achieve this in 2025.
Also I should add that having people you like and respect think less of you, even a bit, is much more stressful than thousands of angry randos
Reply to @JesseTayRiver, 21:10 UTC · 3 likes
Jesse Smith (@JesseTayRiver): @AaronBergman18 @sam_atis Mole people Aaron!? That's the world you want for everyone?
No just the poor, as my post clearly says
Quote tweet, 22:05 UTC · 9 likes
This is actually cool and fun
Quoting Exa (@ExaAILabs): See your 2024 Twitter Wrapped!
Exa-powered AI finds your favorite topics, top tweets, and even whether you’ve been naughty/nice 😈
Get yours here - http://twitterwrapped.exa.ai
Reply to @TheIshanGoswami, 22:07 UTC · 3 likes
Ishan Goswami (@TheIshanGoswami): This is completely built with exa api and you can do this too!
This is how you get the profile data (with top tweets) from the exa api: https://dashboard.exa.ai/playground/get-contents?filters={"ids"%3A["https%3A%2F%2Fx.com%2Fpaulg"]%2C"text"%3A"true"%2C"livecrawl"%3A"always"}
And this is how you get recent tweets of a profile: https://dashboard.exa.ai/playground/search?q=from%3Atheishangoswami site%3Atwitter.com&c=tweet&filters={"type"%3A"keyword"%2C"numResults"%3A20%2C"text"%3A"true"%2C"livecrawl"%3A"always"}
Wait you access all this without the Twitter API? That’s nuts. Impressive scraping 🫡
Quote tweet, 22:21 UTC · 5 likes
Random intellectual curiosity is a vibe
Quoting deepfates (@deepfates): ok this is pretty good actually 😂
Reply to @SpacedOutMatt, 23:21 UTC · 2 likes
Matt 🔸 (@SpacedOutMatt): Fr don’t forget us
Shut up Matt
Quote tweet, 23:24 UTC · 2 likes
Based
Quoting Nature is Amazing ☘️ (@AMAZlNGNATURE): Man saves stuck bird
Reply, 23:37 UTC · 3 likes
Tweet, 23:38 UTC · 11 likes
Who is yacine and why is he relevant
Reply to @livgorton, 23:40 UTC · 2 likes
Liv (@livgorton): entirely unrelated to the content but i’ve followed this YT channel since i was like 15 or something haha so fun to see others share the vids
and yeah i think it entirely depends what we mean when we say x specific person existed. i have too many thoughts for 240 chars 😭😭
Liv, welcome to philosophy
Thread, continued, 23:41 UTC · 2 likes
But yes I agree “X really existed” is surprisingly hard to define, even approximately
Reply, 23:56 UTC · 3 likes
(The tweet being replied to is no longer available.)
Good point
Friday, December 27, 2024
Reply to his own earlier tweet, 00:10 UTC · 3 likes
Aaron Bergman, December 26, 2024: Who is yacine and why is he relevant
Lol I did not realize this until just now. Genuinely curious what interaction yielded this
Reply to @justjoshinyou13, 00:15 UTC · 1 like
Josh You (@justjoshinyou13): @inductionheads think she means that Leopold can't stop yapping about it
Oh man this means it’s big/important
Reply, 01:10 UTC
Haha yes import the 3rd world. Let them all in 🌎🌐
Quote tweet, 01:18 UTC · 26 likes
This is my favorite example of a bad graph because the midwit explanation of why it’s bad (truncated y axis) is totally wrong
Quoting 𝔽. ℙ. ℍunter (@FeralPHunter): smdh i new Sabrina Carpenter was short but didnt realize she was this small
Reply to @daniel_271828, 01:22 UTC · 15 likes
zzDANIEL_ETH (@daniel_271828): It would work if there was no truncated y-axis and it was to scale
Yeah but the truncated y axis isn’t the problem - if it was to scale, truncated y axis would be fine (ie it would only be showing the top of their heads)
Reply to @MariGO2thepolls, 01:26 UTC · 2 likes
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): Economic growth is a remarkably good proxy for everything getting better for everyone
(quoting @VermeireJeff: @MariGO2thepolls @wanyeburkett Growth for what purpose and for who? Growth is not a virtue in an unto itself.)
Well, all humans
Reply to @liz_love_lace, 01:27 UTC · 4 likes
Liz Lovelace (@liz_love_lace): smart boards are basically useless, we had two at our school (for corruption purposes) and even though the teachers were constantly trying to find uses for them, we almost never did anything with them
(quoting @BrianRoemmele: The US spends the most per student in the world. Most of our schools still use 1703 designed chalk boards. Some who spend far less did an upgrade 5 years ago.)
Lol I thought they were cool in 5th grade so invested a few hundred dollars in the company that made them iirc. Stock went to 0 lmao
Tweet, 01:59 UTC · 761 likes
Have we considered that DeepSeek is lying and their training cost way more than $5M?
Thread, continued, 01:59 UTC · 75 likes
I genuinely don’t know how surprising this would be. I have a prior on technical papers not lying about stuff but high profile Chinese statistics being iffy at best and don’t know how to assess their intersection
Thread, continued, 02:00 UTC · 35 likes
I guess one would have to assess this in large part based on the actual plausibility of DeepSeek finding a training method that OpenAI et al missed
Once again I genuinely don’t know how to do this, but others do
Thread, continued, 02:05 UTC · 49 likes
But I feel like “consider whether they’re lying” is an underrated tactic in light of surprising stuff in any domain
Thread, continued, 02:11 UTC · 25 likes
* meant to say “whether” DeepSeek is lying. I have no idea if they are
Reply to @lu_sichu, 02:21 UTC · 32 likes
Sichu Lu (@lu_sichu): But what's the motivation. I assume that plenty of people with less resources but interested in training sota will be attempting replications of these results or training variants very soon
Give Western labs a fool’s errand to pursue, wasting time and money
Could be both political and economic
Reply to @norvid_studies, 02:26 UTC · 4 likes
norvid_studies (@norvid_studies): i don't think they'd burn their reputation for a diffuse "waste some of other peoples money" but i don't have any direct domain knowledge
I don’t have a great understanding of the dynamics here but my understanding is that CCP reps have a lot of influence on the operations of big and strategically important Chinese companies
Reply to @PeterJAngel, 02:35 UTC · 3 likes
Peter🧅📈 (@PeterJAngel): What are the drivers of training costs? It's mostly like CPUs and energy, no?
≈ GPU*hours
But I think the claim is that they found a way to get a better model out of the same number of GPU hours
Tweet, 02:57 UTC · 9 likes
Amphetamines are obviously antidepressants, idgaf what they’re technically classified as
Thread, continued, 02:58 UTC · 3 likes
This is brought to you by amphetamine withdrawal (3 more days)
Reply to @hamandcheese, 05:52 UTC · 29 likes
Samuel Hammond 🦉 (@hamandcheese): Maybe they have access to export controlled chips and need an alibi
Seems not crazy
Reply, 06:00 UTC
Massive economic growth + decline in poverty + appreciation of arbitrariness/luck of place of birth
Reply, 06:10 UTC
👍
Reply, 06:14 UTC · 1 like
Singerian philosophy seems pretty distinct from those other things
Arbitrariness of place of birth (and species for that matter, and everything else) would persist even if all the dark enlightenment-esque empirical claims are true
Quote tweet, 06:35 UTC · 3 likes
Buzzfeed-esque quiz according to how you’d rearrange this table (orthodox guard, reformer, psychedelic-enthusiast, etc)
Quoting Someone:
Thread, continued, 06:35 UTC · 2 likes
For the record I’d move all the antipsychotics besides aripiprazole, as well as esketamine and lithium, to third line
Move bupropion up to first line, maybe even its own category
Add deplin/methylfolate, probably second line
Also hooray for pramipexole even being listed
Quote tweet, 15:11 UTC · 50 likes
Bruh someone gave like $13k
They should consider revealing themselves to get the appropriate amount of social reward and acclaim!
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)
Quote tweet, 15:18 UTC · 7 likes
It would be really tragic if someone got into a doubling-the-total-donation-amount war of attrition with the mysterious donor (manifesting)
Quoting Aaron Bergman 🔍 (@AaronBergman18): Bruh someone gave like $13k
They should consider revealing themselves to get the appropriate amount of social reward and acclaim!
Thread, continued, 15:18 UTC · 2 likes
Donation link for your convenience https://www.givingwhatwecan.org/fundraisers/2024-terminally-online-ea-giving-season-fundraiser
Reply to @arabelladevine, 15:23 UTC · 15 likes
arabella ※ (@arabelladevine): some boys text you from the get go like you’re six months in. not the worst strat, appeals to my delusional ass
Is this a good strategy? I have no idea what I’m doing
Reply to @arabelladevine, 15:31 UTC · 2 likes
arabella ※ (@arabelladevine): @AaronBergman18 idk sometimes it’s awkward sometimes it’s cosy. ig for me depends if it makes it easier or harder for me to respond and keep the vibe going. probably there’s a tradeoff going on between filtering poor matches and giving people more of a chance to cocreate the vibe
Guessing I will fall into the awkward category, thx
Reply to @arabelladevine, 15:34 UTC · 2 likes
arabella ※ (@arabelladevine): @AaronBergman18 have you tested that? don’t just guess!!
My track record (poor) speaks for itself
Reply to @10x_er, 15:49 UTC · 1 like
10x Engineer (@10x_er): tik tok is vape instagram reels is cigarettes x is
Fresh air
Reply to @absurdlymax, 15:54 UTC · 2 likes
Max Alexander (@absurdlymax): The AI sure tried real hard with this one
Jealous
Quote tweet, 17:00 UTC · 9 likes
You don’t hate the FDA enough
Quoting Bicoloured-Python-Rock-Snake 🦬🪐🌲 (@pythonrocksnake): @PeterJAngel Tamiflu would be a great thing to stock up on, but unfortunately you need a prescription so you can't get it until you're already sick, or have a doctor connection. IDK how lax the "telehealth" online prescription mills are but that might be an avenue.
Reply, 17:04 UTC · 1 like
Leopold.
Reply to @DannyParkinson, 17:05 UTC · 4 likes
Danny Parkinson (@DannyParkinson): It’s such a simple joke in Wallace and Gromit but it’s killed me
@absurdlymax @littIeramblings
Reply to @justjoshinyou13, 17:08 UTC
Josh You (@justjoshinyou13): it's an open model so we know the parameters. they could have trained on more tokens than they claim but they did train on a lot (15T)
Yeah but we can’t directly infer compute/gpu-hours right
Reply to @ChatbotOtro, 17:54 UTC · 8 likes
Otro Chatbot (@ChatbotOtro): White muricans fighting Indians and Chinese on Twitter at the same time
I have nothing against Indians and Chinese people! (not a big fan of the CCP/China as a country though)
Reply to @saik0s, 19:25 UTC · 33 likes
Igor Tarasenko (@saik0s): @AaronBergman18 ah yes, the classic "they're out to get us" narrative. because clearly, 1.4 billion people are all conspiring to make the west look bad. or maybe... we're projecting our own insecurities? just a thought.
1.4B? No. But the CCP conspiring with parts of Chinese civil society to sabotage American tech superiority?
Yes this is absolutely plausible and I’m pretty sure well-documented (usually this looks like corporate espionage)
Quote tweet, 19:36 UTC · 16 likes
1450 is 99th percentile in a nationally representative sample and 96th percentile among test takers
You are in a bubble, exhibit no. 1839605281594918
Quoting Ollie (@olzhu): >“My high IQ, high SAT son”
>1450
Thread, continued, 19:36 UTC · 2 likes
https://research.collegeboard.org/reports/sat-suite/understanding-scores/sat
Thread, continued, 19:40 UTC · 3 likes
As much as I disagree with her implicit message, this woman is right that college standards are quite outdated
Perfect SAT + transcript doesn’t get you into top tier schools anymore https://www.aaronbergman.net/p/most-ivy-smart-students-arent-at
Reply to @justjoshinyou13, 19:51 UTC
Josh You (@justjoshinyou13): Compute is proportional to active params * tokens
Claude estimates that params + tokens account for 80% of the variance in compute, so you’re mostly right
Reply to @saik0s, 20:15 UTC · 1 like
Igor Tarasenko (@saik0s): @AaronBergman18 jumping to conspiracy theories might make you feel smart, but it’s actually the laziest form of thinking bc it ignores the complexity of how things actually work.
sometimes, the truth is just boring af.
Yeah this is a hypothesis but I think I’d still put it below 50%. But sometimes there are actually conspiracies!
Quote tweet, 20:19 UTC · 26 likes
Abolish cluster headaches
Quoting EA Forum Posts (@EAForumPosts): "Quantifying the Global Burden of Extreme Pain from Cluster Headaches" by Alfredo Parra 🔸
Thread, continued, 20:19 UTC · 2 likes
I don’t know how legit the DMT thing is
But if that really is a reliable reliever, DMT not even being available by prescription (fentanyl is, appropriately) is yet another reason the FDA is really bad
Reply to @OpenAI, 20:20 UTC · 10 likes
OpenAI (@OpenAI): OpenAI's Board of Directors is evaluating our corporate structure with the goal of making a stronger non-profit supported by the for-profit’s success.
Our plan would create one of the best-resourced non-profits in history. https://openai.com/index/why-our-structure-must-evolve-to-advance-our-mission/
@open_phil should sue
Reply to @deepfates, 20:37 UTC · 4 likes
deepfates.delve.town (@deepfates): gm
Cursed image but real
Reply to his own earlier tweet, 20:50 UTC
Aaron Bergman, December 27, 2024: Claude estimates that params + tokens account for 80% of the variance in compute, so you’re mostly right
Then I push him and it gets revised to 50-60% so idk
Tweet, 21:19 UTC · 15 likes
Some people: Aaron, you should be more epistemically humble
Claude:
Quote tweet, 22:48 UTC · 2 likes
New data just dropped! I’d love a qualitative deep dive
Quoting Nat Friedman (@natfriedman): Full results and write-up: http://plasticlist.org
We also have instructions for running your own tests!
Reply to @natfriedman, 23:28 UTC · 2 likes
Nat Friedman (@natfriedman): Check out this report: http://plasticlist.org/report
Thanks!
Saturday, December 28, 2024
Reply to @leland_mcinnes, 00:32 UTC · 2 likes
Leland McInnes (@leland_mcinnes): I'm looking forward to seeing more maps of latent spaces for more models, including image models.
Keep an eye out for work by @enjalot @thesephist @livgorton @adamrpearce @JungleSilicon among many others. Exciting times ahead!
Sparse autoencoder
Reply to @BecomingCritter, 07:37 UTC · 9 likes
critter (@BecomingCritter): What's your favorite thing about women?
Quote tweet, 07:56 UTC · 4 likes
Based, could be you
Quoting Liv (@livgorton): proof of my commitment to making this beauty a reality 🫡
Quote tweet, 08:05 UTC · 12 likes
Wtf
Quoting Lewis Bollard (@Lewis_Bollard): This is heartbreaking...
About 50 million US birds have been baked or suffocated alive this year, all to stop bird flu.
There's a vaccine that could prevent this, but the USDA won't allow its use ... because it might affect chicken exports.
Quote tweet, 14:12 UTC · 6 likes
Simulator 1: lets put the city responsible for the most important technological development in the universe on a huge fault line
Simulator 2: lol
Simulator 1: set.seed(69)
Quoting Tim Urban (@waitbutwhy): The location of all earthquakes since 1980. We live on a cracked shell.
[Image: The Concord Consortium, via Kottke]
Quote tweet, 14:22 UTC · 15 likes
Narrative violation?
Quoting Stefan Schubert (@StefanFSchubert): Obviously you have to adjust for population
Quote tweet, 14:37 UTC · 301 likes
This is sufficiently plausible, safe, and cheap to be worth a try lmao
Quoting _aestheticprimal_ (@aestheticprimal): one of the biggest hacks there is in this entire
'health niche' is putting Magnesium Chloride on your balls.to start off, getting Magnesium into your system is already one of the cornerstones of health, this is because Magnesium plays a role in thousands of enzymatic reactions and is needed to carry ATP.
1st reason:
when you supplement Magnesium, topical is an incredibly great way to do it, since absorption is much higher than when it needs to pass through the gut (besides that; oral Magnesium Chloride will give you diarrhea)2nd reason:
of all the areas of skin that exist on your body, your ballsack absorbs things the best by far (your testicles absorb almost 12 times better than your forearms do) so topical application on your balls is only something logical.3rd reason:
since you are applying it on your balls, you are directly enhancing testosterone production because you are providing the raw building block which the Leydig cells need to convert the prohormones (like DHEA and pregnenolone) into testosteroneof almost all the people that i've recommended to use magnesium, have noticed that their testicles became bigger and more filled with blood after topical application.
so how do you do it?
>mix 2/7th of magnesium chloride flakes with 5/7th water in a spray bottle
>shake very well
>spray on skin area and rub it oneeasy and incredibly cheap (u can get Magnesium Chloride for like 5-10 bucks per kilo)
Reply to @gelionexists, 14:57 UTC · 6 likes
GE Lion (@gelionexists): respect the effort to say classy by ordering the fancy spray bottles
Idk I filtered for glass. Probably spent matter but I didn’t wanna try to figure out if plastic plausibly leaches
Reply to @JesseTayRiver, 14:58 UTC · 22 likes
Jesse Smith (@JesseTayRiver): @AaronBergman18 Ffs
You, virgin: diet and exercise and a rich family and social life
Me, Chad:
Reply to @JgaltTweets, 15:38 UTC · 4 likes
JgaltTweets (@JgaltTweets): @AaronBergman18 What does it look like if you compare just the large developed EU countries (Germany, France, Italy, Spain) to the US? I'd think Eastern Europe is dragging up the average a bit by starting from such a low base in 1990
Yeah this sounds right
Reply to @polisisti, 16:08 UTC · 9 likes
Alessandro (@polisisti): oh no, now we're all thinking about Aaron's balls
😈
Reply to @JgaltTweets, 16:21 UTC · 15 likes
JgaltTweets (@JgaltTweets): >9 bookmarks
people want updates on this i guess
Lol, I will try!
Tweet, 16:50 UTC · 10 likes
Found this hat I made at probably 6 or 7 (that is my home phone number)
Quote tweet, 16:51 UTC · 11 likes
A floor at like 2x the national household avg for a person probably out of college is nuts, immoral, and bad policy
Not every high skilled industry is infinite money printer tech!
Quoting Drew Volpe (@drewvolpe): Raising the salary floor for H1-Bs is an obvious fix. The current floor was set at $60k in 1990 and never changed. Adjusted for inflation, it should be $148k today.
A $148k floor would filter out low-value uses of the program, like outsourcing firms. It would also ensure that H1-Bs are used for important R&D roles where salaries exceed this level. This one change could make the program fairer and better aligned with its original intent.
Reply to @absurdlymax, 17:40 UTC · 2 likes
Max Alexander (@absurdlymax): I'm so sorry
Cc @birchlse
Reply to @manic_pixie_agi, 18:12 UTC · 19 likes
Corinne (🏠 in NYC) (@manic_pixie_agi): Concept: OnlyFans creator who is doing quantified self experiments while naked
I mean for a sufficient amount of money…
Quote tweet, 18:28 UTC · 5 likes
Like this tweet (hopefully), donating to the fundraiser (link below) is a low effort but high impact intervention
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)
Reply to @timfduffy, 20:04 UTC · 7 likes
Tim Duffy (@timfduffy): > Tyler Durden pfp
> Not followed by anyone you're following
Whether this is real or not I'm not gonna update much on this guy's post
I had Claude and ChatGPT fact check and basically it seems untested/unconfirmed but nothing obviously wrong
Reply to @JesseTayRiver, 21:25 UTC · 13 likes
Jesse Smith (@JesseTayRiver): Deficient in ball magnesium? Seems unlikely, no?
Yes here is the lab report
Reply to @JesseTayRiver, 21:26 UTC · 6 likes
Jesse Smith (@JesseTayRiver): Ok I guess this checks out then
Quote tweet, 21:39 UTC · 15 likes
You: nooo I must maintain perfect infosec I can only let my closest irl friends know my physical location the world is so dangerous noooo
Wealthy, world famous, >7 foot tall easily identifiable NBA star: I’m at this specific park rn, pull up if u wanna hang
Quoting Wemby (@wemby): Thanks for everyone who pulled up in the rain 😂🔥
Reply to @arithmoquine, 21:57 UTC · 33 likes
henry (@arithmoquine): what's the most controllable non-destructive way to induce pain?
I often want to experiment with my own tolerance and see how much is needed before I break precommitments, but i don't know any way to isolate suffering from the fear of lasting physical damage
(quoting @NathanpmYoung: I am not convinced that everyone cracks under torture. Often seems like lazy writing to me.)
I think arm in ice water is actually the standard here
Quote tweet, 22:21 UTC · 35 likes
Very interesting if true (no idea if it is)
Quoting Dwarkesh Patel (@dwarkesh_sp): Notes on visiting China.
On scale, construction, youngsters, public opinion, censorship, tech, and AI.
Link below.
Thread, continued, 22:22 UTC · 6 likes
I mean, needless to say, one good thing about liberal democracy is that you can actually kinda know these things instead of speculating based on vibes
Reply to @TaliaGraceSable, 22:37 UTC · 7 likes
Talia 🌿 (@TaliaGraceSable): Sometimes people wonder "does this scare the hoes?" and now you can know. As a hoe, I will answer if it scares me or not! AMA
Caffeine pills
Reply to @TaliaGraceSable, 22:37 UTC · 5 likes
Talia 🌿 (@TaliaGraceSable): Sometimes people wonder "does this scare the hoes?" and now you can know. As a hoe, I will answer if it scares me or not! AMA
Nature interventionism in the name of animal welfare
Reply to @TaliaGraceSable, 22:37 UTC · 5 likes
Talia 🌿 (@TaliaGraceSable): Sometimes people wonder "does this scare the hoes?" and now you can know. As a hoe, I will answer if it scares me or not! AMA
diy medicine
Quote tweet, 22:41 UTC · 7 likes
Hey girl, wanna increase our mutual happiness 😏 (we will talk while going on a walk to the air and space museum)
Quoting mattparlmer 🪐 🌷 (@mattparlmer): Meditation makes people less happy than drinking alcohol does
Reply to @truthscrolling, 22:45 UTC · 4 likes
׺°”˜`”°º (@truthscrolling): @AaronBergman18 This is also true in the stable Gulf states (Kuwait, UAE, Saudi Arabia, Jordan). The general populace wants to carpet-bomb Tel Aviv and toughen laws, while the intelligent, Westernized ruling class wants normalization with Israel. Source: Lived in Dubai for 3 years.
Interesting thanks
Reply to @leah_pierson, 23:00 UTC · 4 likes
Leah Pierson (@leah_pierson): I accidentally wrote a 5000-word critique of 80,000 Hours' career guide without checking that I was reading the updated 2023 guide. (The 2017 PDF was the first result!)
Having now read the new guide, I'm 90% happy, 10% devastated to report that my concerns are mostly addressed.
Lol
Thread, continued, 23:00 UTC · 3 likes
And based
Reply to @TaliaGraceSable, 23:14 UTC · 1 like
Talia 🌿 (@TaliaGraceSable): Seriously scary
Oh no :(
Reply to @mountmayn, 23:14 UTC · 2 likes
(The tweet being replied to by @mountmayn is no longer available.)
🙏
Reply to @teortaxesTex, 23:51 UTC · 3 likes
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) (@teortaxesTex): You literally both use pause emojis
Maybe this is obvious but it’s a bit silly to lump in most AI pausers with the ~Luddite left
I’m broadly extremely pro-science and tech and can name like 3 specific exceptions (some animal testing, gain of function pathogen research, frontier AI scaling)
Sunday, December 29, 2024
Reply, 02:22 UTC · 3 likes
No
Tweet, 02:31 UTC · 8 likes
My family realized they can get me to watch movies if (and only if) they’re about airplanes
Thread, continued, 02:31 UTC · 3 likes
The y = 1/2 spike was when I walked out of Dune (sorry, fiction is generally just boring)
Reply to @MattVogel0311, 02:40 UTC · 6 likes
Matt Vogel (@MattVogel0311): should get your test levels checked before and after
I actually did get them checked recently, so accidentally on it
Quote tweet, 02:44 UTC · 6 likes
Make it happen
Quoting Cate Hall (@catehall): actual crime against humanity that google docs won't turn two hyphens into an em dash
Quote tweet, 02:47 UTC · 4 likes
Nix the Apple Watch thing but the rest I endorse for purely objective reasons
Quoting Engineer Girlfriend (@enggirlfriend): ladies, don’t trust a man (esp 20s) with a rolex.
go for the ones wearing an apple watch, even better, a whoop. wears free tshirts. bad haircut. ideally some of the tism.
best investment you’ll ever make.
Tweet, 14:12 UTC · 27 likes
In honor of 3k followers, I am sending this tweet
Reply, 14:19 UTC · 8 likes
I mean maybe he’s right, but he added a dubious implicit premise which is that killing bacteria directly is actually the mechanism of action. Like there are a lot of other things that could happen!
Thread, continued, 14:20 UTC · 1 like
Especially garlic, like I’m pretty sure sulfur compounds like allicin really do fuck with bacteria, but maybe it has to be digested and pass into the blood stream first
Reply to @pli_cachete, 14:22 UTC · 3 likes
Rota Is In Israel🚪🧎♂️ (@pli_cachete): @AaronBergman18 What’s the thing you’re most proud of and most disappointed in yourself this year
This is a good q I will think more than 5 seconds!
Quote tweet, 14:33 UTC · 6 likes
Does anyone have smart ring (or whatever) takes, including preferred brands, reviews, whether they’re worthwhile at all?
Quoting Aella (@Aella_Girl): I've been tracking my mood, habits, and biometrics (oura ring) for years. I fed all 2k+ days of data into claude and asked it for a report on what to do to improve my mood
Reply to @boreal_country, 14:35 UTC · 2 likes
Beau (@boreal_country): @AaronBergman18 I found it annoying to wash my hands with a bulky ring and you can’t lift weights with one because of the grip / you’ll break the ring. Source: Returned my Oura 4.
Sucks but thanks!
Reply to @austinc3301, 15:01 UTC · 2 likes
Agus 🔸 (@austinc3301): I have very detailed takes on this stuff. Basically, I recommend people get an Apple Watch OR A Whoop OR an Oura ring, whichever works better for them. Apple Watch is the only one without a subscription, and has arguably slightly better biometric tracking than the other two.
Nice thanks!
Tweet, 15:19 UTC · 5 likes
I regret to report that I like the suburbs. Cancel my urbanist card
Quote tweet, 15:20 UTC · 12 likes
Obligatory qt
Quoting Jonathan Mannhart 🔎🔸 (@JMannhart): PSA: Don‘t magnesiummaxx
(Not without ramping up very slowly and getting blood tests often, which is annoying. Much easier & safer to just take supplements, e.g. mg citrate.)
Reply to his own earlier tweet, 15:28 UTC · 7 likes
Aaron Bergman, December 28, 2024: This is sufficiently plausible, safe, and cheap to be worth a try lmao
Ok maybe don’t trust me lol
Reply to @JMannhart, 15:29 UTC · 3 likes
Jonathan Mannhart 🔎 (@JMannhart): PSA: Don‘t magnesiummaxx
(Not without ramping up very slowly and getting blood tests often, which is annoying. Much easier & safer to just take supplements, e.g. mg citrate.)
image attached
(quoting @aestheticprimal: one of the biggest hacks there is in this entire 'health niche' is putting Magnesium Chloride on your balls. to start off, getting Magnesium into your system is already one of the cornerstones of health, this is because Magnesium plays a role in thousands of enzymatic reactions…)
FWIW
Reply, 15:37 UTC · 1 like
I’m still gonna do it, just conservatively
Thread, continued, 15:38 UTC · 3 likes
For science
Quote tweet, 15:52 UTC · 126 likes
What the fuck
Quoting Poornima Rao (@RaoPoornima): Update on @suchirbalaji
We hired private investigator and did second autopsy to throw light on cause of death. Private autopsy doesn’t confirm cause of death stated by police.
Suchir’s apartment was ransacked , sign of struggle in the bathroom and looks like some one hit him in bathroom based on blood spots.
It’s a cold blooded mu*d*r declared by authorities as suicide.
Lobbying in SF city doesn’t stop us from getting justices.
We demand FBi investigation
@elonmusk @VivekGRamaswamy @GavinNewsom @MEA
Reply to @austinc3301, 15:53 UTC · 2 likes
Agus 🔸 (@austinc3301): @AaronBergman18 @AniseNot @NathanpmYoung I don’t think this is true. Allicin should act as an in vitro antibiotic, and I if I had to guess the experiment probably failed by not timing crushing the garlic at just the right time, since allicin is very unstable and degrades very quickly.
Stop being smart plz
Reply to @austinc3301, 15:54 UTC · 2 likes
Agus 🔸 (@austinc3301): @AaronBergman18 @AniseNot @NathanpmYoung To give credit where credit is due, I had a long chat with Claude about this last week because I was curious about allicin and why bulbs have evolved strong defense mechanisms
Not beating the smart allegations
Quote tweet, 15:58 UTC · 134 likes
I feel extremely good about “EA money” being used to conduct independent investigations for this and anything else like it
Quoting Poornima Rao (@RaoPoornima): Update on @suchirbalaji
We hired private investigator and did second autopsy to throw light on cause of death. Private autopsy doesn’t confirm cause of death stated by police.
Suchir’s apartment was ransacked , sign of struggle in the bathroom and looks like some one hit him in bathroom based on blood spots.
It’s a cold blooded mu*d*r declared by authorities as suicide.
Lobbying in SF city doesn’t stop us from getting justices.
We demand FBi investigation
@elonmusk @VivekGRamaswamy @GavinNewsom @MEA
Quote tweet, 15:59 UTC · 2 likes
👍
Quoting SARS‑CoV‑2 (COVID-19) (@COVID19_disease): ⚠️ BREAKING:
Hospitals in China 🇨🇳 Overwhelmed as Severe "Flu" Outbreak, Including Influenza A and HMPV, Resembling 2020 COVID Surge.
Hospitals in China are overwhelmed as outbreaks of "influenza A" and "human metapneumovirus" resemble the COVID-19 surge from three years ago.
Quote tweet, 16:03 UTC · 2 likes
I agree with this, and insofar as it competes against “institutions must look like America to have credibility”, intuitively I feel like this concern just dominates
Like maybe racial affirmative action is good, but I don’t think the “credibility” reasoning can be why
Quoting mattparlmer 🪐 🌷 (@mattparlmer): The “top 1% SAT doesn’t reliably get you into top 10% colleges if you have the wrong demographics” thing is orders of magnitude more socially corrosive than the visa stuff
Reply to @austinc3301, 16:05 UTC · 16 likes
Agus 🔸 (@austinc3301): I don’t agree at all, let justice (or citizens) do their thing. You don’t want to look crazy.
If conducting actually-good investigations of the alleged suicides of high profile AI whistleblowers seems crazy, I’m fine with seeming crazy
Like maybe it’s a bad use of money, but indirect reputations effects above and beyond actual merit doesn’t seem like why
Thread, continued, 16:06 UTC · 3 likes
I guess I also don’t have a great idea of the quality of run-of-the-mill cause of death reports/autopsies, high profile and otherwise
I could be convinced that it’s a rock-solid science with truly negligible risk of being wrong, even in high profile cases
Reply to @austinc3301, 16:09 UTC · 3 likes
Agus 🔸 (@austinc3301): Im making the point that the counterfactuality would be very low, such that the reputational risk is almost definitely not worth it.
I think I just disagree - Claude says a high quality third party investigation would cost $20k
I feel pretty good about this if the chance of foul play is above 0.1%
Reply to his own earlier tweet, 16:10 UTC · 19 likes
Aaron Bergman, December 29, 2024: I feel extremely good about “EA money” being used to conduct independent investigations for this and anything else like it
To be clear, I have no reason to think EA money was used in this case - I’m just making a claim about future possible spending
Reply to @austinc3301, 16:11 UTC · 2 likes
Agus 🔸 (@austinc3301): @AaronBergman18 If I believed no one else would do anything without this money and that there was an at least plausible case for him being murdered (last time I checked there wasn’t, but I haven’t looked into it closely) then I’d agree
Low base rate of suicide + being a whistleblower in a high profile industry with billions+ of dollars at stake is enough for me
Reply, 16:21 UTC · 1 like
I mean I myself am pretty close to a counterexample. Not quite perfect on all academic fronts but awfully close (obligatory cringe recognition)
Reply to @gptbrooke, 16:47 UTC · 4 likes
brooke bowman (@gptbrooke): @AaronBergman18 Every few days I think about how wild it is that whistleblowers just routinely commit suicide and there are even some raised eyebrows or outrage but then nothing happens
Deeply unsettling
Fr
Reply, 17:22 UTC
Interesting thx
Thread, continued, 17:23 UTC · 1 like
It needn’t be OpenAI though. I mean they’re the obvious group but I can imagine others
Reply to his own earlier tweet, 17:25 UTC · 28 likes
Aaron Bergman, December 29, 2024: To be clear, I have no reason to think EA money was used in this case - I’m just making a claim about future possible spending
Also, to be more explicit, a lot of the value here comes not from directly uncovering some foul play but from discouraging future foul play, which in turn makes whistleblowing more likely
Reply to @Chris_Said, 17:28 UTC
Chris Said (@Chris_Said): Also, there's good crazy and bad crazy:
Good crazy: Shrimp welfare. Sounds crazy, but at some level people know you're right. Overton window expanding.
Bad crazy: Pursuing a 0.1% conspiracy theory. High prob of credibility loss.
Agh I feel like the heuristic “the best PR is actually just doing what you think is good and can credibly explain” is incredibly underrated
Like again maybe I’m wrong, but I really don’t think PR is why
Reply to @sullyj3, 17:41 UTC · 9 likes
░▒▓▒░▒▓▒░▒▓▒░ (@sullyj3): What incentive would the police have to lie about this?
“It’s hard to tell the difference + prior on suicide” seems more likely than genuine lying
Reply to @I_dont_twt_much, 17:48 UTC · 2 likes
I do tweet some (@I_dont_twt_much): @AaronBergman18 Subjunctive mood would be your friend, if you'd let it
Real
I would feel extremely good if “EA money” were used to conduct independent investigations for this and anything else like it
Reply, 18:07 UTC · 1 like
I think you’re wrong lol and a focus on being actually good and making good object level decisions and integrity is almost always best in the long run
Companies need “PR” because they’re not actually (directly) acting in the world’s best interest. EA is importantly different
Reply, 18:08 UTC
Actually doing it for credible reasons is like 99.9% of the work here
Reply to @austinc3301, 18:19 UTC
Agus 🔸 (@austinc3301): Like, I’m confident that, as stated, “EA hires third party investigator into the alleged suicide” is a thing that decreases trust in EA, not the opposite. Especially if your “credible explanation” is a long shot high EV bet.
I disagree and think you’re underrating how much “do what you actually think is the right thing on the merits” improves trust over the course of months and years
Reply, 18:24 UTC
Lol I think the more important adversarial dynamic is the fact that there are hundreds of billions of dollars of incentives pushing against safe AI development, and one shouldn’t unilaterally disarm in order to appear nice and innocent
Quote tweet, 21:56 UTC · 4 likes
WE NEED A THOROUGH NEUTRAL PARTY INVESTIGATION
Quoting Aaron Bergman 🔍 (@AaronBergman18): I feel extremely good about “EA money” being used to conduct independent investigations for this and anything else like it
Reply to @metachirality, 21:57 UTC · 4 likes
metachirality (@metachirality): it's actually just that there's so many openai ex-employees that at least one gets murdered purely by virtue of the law of large numbers
lol
Monday, December 30, 2024
Reply to @pli_cachete, 01:58 UTC · 2 likes
Rota Is In Israel🚪🧎♂️ (@pli_cachete): Post current podcast stack
Most or all good: 80k, Dwarkesh, Carlsmith, Edge of of sentience (https://podcasts.apple.com/us/podcast/the-edge-of-sentience-by-jonathan-birch/id1761877270)
EA Forum, Lesswrong , and EA Global talks = hit or miss/specific ones
Reply to @sunofdopamine, 02:02 UTC · 15 likes
Stephen L (@sunofdopamine): i go on walks with this guy in the sun and sometimes he’ll go talk to a woman he finds interesting
like, one on a two hour walk. and he’ll chat for ten minutes and be like “I’m seeing her for coffee tomorrow”
This is the dream
Reply to @justalexoki, 02:05 UTC · 5 likes
taoki (@justalexoki): the masculine urge to go to war
Idk if this is a joke but war seems horrifying and if that makes me gay or feminine or whatever so be it
Tweet, 05:05 UTC · 77 likes
If you’re a Rolex in your early 20s, buy a young man. This communicates status to female watches and watch business relationships
Quote tweet, 05:27 UTC · 3 likes
This is some good OG app material
Like the opposite of a website UI just written in Swift. Retvrn
Quoting Ethan Kuntz (@KanizsaBoundary): Curved Spaces is an app from mathematician Jeffrey Weeks and it’s super fucking fun, you get to fly through all sorts of mathematical spaces and see all sorts of cool geometry
Thread, continued, 05:27 UTC · 2 likes
https://apps.apple.com/us/app/curved-spaces/id1417178571
Tweet, 10:48 UTC · 3,684 likes
Being a girl in SF must go crazy. Like absolute god mode
Reply to @livgorton, 10:55 UTC · 13 likes
Liv (@livgorton): let there be light ✨😌
90k lumens 😌 https://a.co/d/i1eyGbr
Quote tweet, 11:25 UTC · 4 likes
Gonna be vulnerable and admit that I have never understood this joke format even a little
Quoting Miles Brundage (@Miles_Brundage): Regulation goes to doctor.
"Doctor, I need to keep my safety standards up to date as the technology evolves."
Doctor looks at him and says, "The treatment is simple! Consult the General-Purpose AI Code of Practice, it will be updated regularly."
"But doctor..."
Reply to @daniellellecco, 11:25 UTC · 2 likes
(The tweet being replied to by @daniellellecco is no longer available.)
Lol
Reply to his own earlier tweet, 11:30 UTC · 2 likes
Aaron Bergman, December 30, 2024: Gonna be vulnerable and admit that I have never understood this joke format even a little
OHH
Thread, continued, 11:31 UTC · 2 likes
Nice
Reply to @ohwizenedtortle, 12:24 UTC · 15 likes
oh dandy lion🔸 (@ohwizenedtortle): Yeah I boned hot smart guys fr
Nice congrats
Reply to @Q22635587, 12:29 UTC · 25 likes
YouAreThey - Read Write Do (@Q22635587): Might just be cope but I'm telling myself the CIA did it that way to protect some other source or capability they didn't want to reveal.
Yeah my thought too. Or some illegally gathered intel
Reply, 12:30 UTC · 1 like
Since when are u engaged? Congrats!
Reply to @visakanv, 12:46 UTC · 10 likes
Visakan Veerasamy (@visakanv): amazing moment where a fan of a @LindseyStirling realizes the person applauding her is Lindsey herself
For the curious https://open.spotify.com/artist/4lZ66nVNQdgHcyE3Uv4Ash?si=X7_Yh8hNSIiCNLqjHwGP7w
Tweet, 12:49 UTC · 2 likes
House to myself. Thermostat set to 53° 😈
Reply to @dhadfieldmenell, 13:07 UTC
Dylan HadfieldMenell (@dhadfieldmenell): I understand the perspective.
I’m curious if your issue is more with peer review as an institution or if you think AGI labs should be exempt from the process.
I understand the tradeoffs, but I think it’s important to integrate external feedback — especially for safety work.
Not Richard obv but Peer Review (capitalized bc I mean the institutional practice, not colleagues commenting on a google doc whenever) seems bad
Reply to @meadandjuniper, 14:43 UTC · 1 like
j (@meadandjuniper): Gf who goes “I hope you’re not just saying they because of vacuous truth, you should know I’m asking this question in relevance logic”
Marry her
Reply to @dhadfieldmenell, 15:40 UTC · 1 like
Dylan HadfieldMenell (@dhadfieldmenell): Why do you think I'd find that convincing?
I don’t, I’m just clarifying my claim
Reply to @zodattack, 18:00 UTC · 157 likes
zöda (@zodattack): genuinely asking, what do you mean by this?
Instant harem of hot, rich, brilliant, interesting men
Reply to @yhdistyminen, 18:09 UTC · 11 likes
bob's burgers urbanist 🐿️ (@yhdistyminen): Looking at this map lowered my IQ
(quoting @TylerDurden: Elon and Vivek dislike this image.)
Lol I mean it’s also trivially wrong - pretty sure >3% comes from the US alone
Quote tweet, 18:51 UTC · 3 likes
Only @willmacaskill himself can signal boost this to the moon
Quoting Matt 🔸 (@SpacedOutMatt): We have just $7.5k left to hit the final stretch goal with 5 days to go! Donate to make Aaron have a Will MacAskill shrine in his room!
Reply to his own earlier tweet, 19:25 UTC · 73 likes
Aaron Bergman, December 30, 2024: Instant harem of hot, rich, brilliant, interesting men
Ok this might be slight hyperbole
Reply to @ShinsukeKita00, 19:25 UTC · 30 likes
Ash (@ShinsukeKita00): There are either cucks or gays. Impossible to find an honourable man. Good luck girls.
Big if true
Reply to @dlbydq, 19:27 UTC · 2 likes
Anish Tondwalkar (@dlbydq): "gay guys are the true winners in SF" might be the least controversial statement about dating ever uttered
Lol
Reply to @littIeramblings, 19:47 UTC · 15 likes
sarah (@littIeramblings): skeptical 🤔
Only one way to find out 👀
Reply, 19:56 UTC · 85 likes
(The tweet being replied to is no longer available.)
I know one lol. Well he’s a borderline billionaire. Possibly just centi-millionaire
Reply to @nickcammarata, 20:00 UTC · 3 likes
Nick (@nickcammarata): the couple women I dated in rural japan like 10yrs ago were like you don’t get it, the options were dating you (the only foreigner around) or being single, I have zero desire to date japanese guys. there basically wasnt even a ratio it was just 1, supply of 1
Damn
Thread, continued, 20:01 UTC · 13 likes
F in the chat for my Japanese bros
Reply to @bijibiji2023, 20:17 UTC · 13 likes
hoff 🧒🏻😻 (@bijibiji2023): >
lol
Reply to @eshear, 21:16 UTC · 84 likes
Emmett Shear (@eshear): Have you...spoken to women in SF? Empirical evidence contradicts your assessment, which suggests something about how you model the world is wrong.
Yes lol I have
Thread, continued, 21:19 UTC · 18 likes
But not shocking that people have different experiences! Fair enough!
Reply to @peterwildeford, 22:43 UTC · 1 like
Peter Wildeford🇺🇸🚀 (@peterwildeford): Aaron with this tweet: 82K views in 11 hours
Me with effortposting tweet: 4K views in 2 days
:(
Lol
Everyone follow Peter! He knows important things!
Reply to @peterwildeford, 22:44 UTC · 25 likes
Peter Wildeford🇺🇸🚀 (@peterwildeford): Aaron with this tweet: 82K views in 11 hours
Me with effortposting tweet: 4K views in 2 days
:(
Extremely real phenomenon though. This has 5 likes as of posting here
Reply to his own earlier tweet, 23:03 UTC · 51 likes
Aaron Bergman, December 30, 2024: Being a girl in SF must go crazy. Like absolute god mode
Since this is getting some attention, consider donating to reduce animal suffering! (Donation link below)
Thread, continued, 23:04 UTC · 7 likes
https://www.givingwhatwecan.org/fundraisers/2024-terminally-online-ea-giving-season-fundraiser
Thread, continued, 23:04 UTC · 21 likes
Poast donation receipt and I’ll tell you why ur gender culture war takes are based and correct
Reply to @manic_pixie_agi, 23:18 UTC · 3 likes
Corinne (🏠 in NYC) (@manic_pixie_agi): not having a sex drive sucks. having a sex drive sucks more, you can't concentrate because it's all you can think about
Reply to @NathanpmYoung, 23:18 UTC · 3 likes
Nathan 🔎 (@NathanpmYoung): the animal in question being one aaron bergman
Lol yes we’re secretly embellishing the money to get me hinge premium
Tuesday, December 31, 2024
Reply to @peterwildeford, 00:43 UTC · 3 likes
Peter Wildeford🇺🇸🚀 (@peterwildeford): Based.
I'm so strongly with you on 1-4 and 6, so what's the case for 5? Seems like I'd be inclined to agree given the track record of the others.
As a baseline, it looks like an excellent intervention on normal humanish timescales
but it’s also sui generis as a kindalongtermist intervention that will keep paying off (counterfactually) for ~millions of years if it gets done before AI disempowers humans
Thread, continued, 00:45 UTC · 2 likes
(Or hundreds of thousands if a sub-existential catastrophe sends us back in time)
It’s also weirdly plausible. My current read is that it’s basically a matter of political prioritization in Latin America…
Thread, continued, 00:47 UTC · 3 likes
There’s no pro-screwworm constituency (yet at least), so it’s kind of a matter of getting a bunch of reasonably friendly (with the US and each other) governments to agree to spend resources in a coordinated and competent fashion
Thread, continued, 00:48 UTC · 2 likes
So it would be incredibly dumb and bad and tragic if this couldn’t get done in the next couple years and then there’s some sort of catastrophe and the screwworm keeps going for thousands-millions of years eating animals alive
Thread, continued, 00:50 UTC · 2 likes
Be on the lookout for a forum post in this space 👀
Quote tweet, 07:01 UTC · 5 likes
Hell yeah
Quoting Richard Ngo (@RichardMCNgo): I just committed $100k to Lightcone. Many people are trying to steer AGI, but few have enough integrity for their actions to be robustly positive. @ohabryka is a central example of the latter group, and has influenced my thinking a lot.
I encourage you to consider donating too.
Reply to @Lizquidity, 09:01 UTC · 2 likes
Liz (@Lizquidity): how many oomfies did you get to kiss this year?
Incredibly >0 (1)
BUT WE STILL HAVE 20 MORE HOURS
Reply to @Lizquidity, 09:02 UTC · 4 likes
Liz (@Lizquidity): how many oomfies did you get to kiss this year?
The two genders
Tweet, 13:04 UTC · 13 likes
One banger away from Elonbux 🙏
Reply to @justalexoki, 13:15 UTC · 3 likes
taoki (@justalexoki): need to burn a little more cash, what should i buy? it's about ~$700. something tech-ish related
Idea
Quote tweet, 13:21 UTC · 7 likes
Put us over $30k and I will personally write you a poem
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)
Reply to @livgorton, 13:30 UTC · 2 likes
Liv (@livgorton): 👀👀
HOLY SHIT 🚨🚨🚨🔔🔔‼️🔔⏰🔔😳😧😲😧😮😰😯😟
Quote tweet, 13:40 UTC · 3 likes
Put us over $35k and I’ll personally make you a website with lots of bells and whistles and shenanigans
Quoting Matt 🔸 (@SpacedOutMatt): 😮😮😮😮😮😮😮😮
Tweet, 14:02 UTC · 331 likes
We should use tech to make pregnancy suck less
Thread, continued, 14:04 UTC · 49 likes
Same with periods (ie options better than hormonal birth control)
I suspect there are weird cultural reasons the market isn’t fully taking care of these
Reply to @RokoMijic, 14:04 UTC · 7 likes
Roko 🐉 (@RokoMijic): We already did, pregnancy today for young, white non-obese women is vastly safer than it was in the past
Yes, this is a good start!
Reply, 14:55 UTC · 15 likes
Jesus
Reply to @Jason_Dean, 19:05 UTC · 4 likes
Jason Dean (@Jason_Dean): @Hexhawk_2 @300furrybears @AaronBergman18 @StatisticUrban Well that’s for a handful of people
It goes away for 90% of people after treatment
The antibodies stay for longer so you can test positive for it if you were previously infected
Ah yes the Lyme culture wars
My read is that the newish consensus (and probable truth) is that symptoms sometimes persist after antibiotics but this doesn’t actually correspond to live persistent bacteria actually causing the symptoms
Thread, continued, 19:07 UTC · 3 likes
FWIW I went from basically healthy to basically chronically fucked up in a variety of ways around 15 and I think there’s like a 20% chance that Lyme was importantly involved in this
Reply to @FrostedSapling, 19:09 UTC · 5 likes
Ivan 🔸 (@FrostedSapling): @AaronBergman18 she’s got a point lol
Being a God in SF must go absolutely twink mode
Thread, continued, 19:09 UTC · 2 likes
This isn’t actually clever it doesn’t mean anything
Reply to @joeteicher, 19:14 UTC · 7 likes
Joseph Teicher (@joeteicher): @AaronBergman18 What's your reason to believe in "weird cultural reasons" rather than just the technical challenges?
It basically accords with larger cultural trends involving medicine:
1) taking women less seriously in clinical practice and the biomedical pipeline, and
2) normal/natural = good/right (aging is the big other one here)
3) neglect of suffering per se in favor of functionality
End of December 2024. Previous month: November 2024. Next month: January 2025. Back to the archive index.