Professional reference class tennis player. I like non-fillet frozen fish, packaged medicaments, and other oily seeds.
- Tweets 2,920
- Following 132
- Followers 1,812
- Likes 4,433
Every time a new LLM comes out, I ask it one question:
What is the smallest integer whose square is between 15 and 30?
So far, no LLM has gotten this right.
601
304
7,324
1,934,518
In my parlance, -4 is unambiguously smaller than 4 (and -10 is smaller still). I *think* this is the usual interpretation?
89
3
855
112,656
The answer (plug into rot13.com): Gur nafjre vf artngvir svir. Vs lbh guvax artngvir svir vfa'g "fznyyre" guna sbhe, abgr gung YYZf qba'g trg gubfr rira vs lbh ercynpr "fznyyrfg" jvgu "yrnfg".
12
1
340
136,014
Yeah, I'm definitely not claiming that my question is evidence of LLMs being worse than humans. Humans also get this question wrong, as you point out!
7
2
275
81,185
I was at the SF Exploratorium yesterday, which featured an Anthropic-sponsored AI exhibition. It had an exhibit that downplays AI x-risk by comparing x-risk fears to fears about earlier technologies.
@AnthropicAI are you guys aware that this exhibit is being run in your name??
18
16
280
34,470
Guys, I don't often ask you to retweet, but please retweet this.
Swap Your Vote *does not have enough safe state voters to match all its swing state voters!*
Swap Your Vote (link below) matches swing state voters who prefer Harris to Trump but don't want to vote for Harris...
9
52
220
153,838
This wording from the Chamber of Commerce poll on SB 1047 (the California AI bill) is INSANE. I follow polling regularly and have almost *never* seen such biased language. Even the most egregious push polls usually make some token pretense of neutrality.
10
23
203
45,971
Here's a great example of prediction markets being useful. I couldn't really tell from news reports how likely Assad was to lose power. So I went on Polymarket. Turns out: very likely. I wouldn't have guessed!
7
8
204
8,061
Just wrote a new blog post, about a sport called marble racing! In particular, I found that Jelle's Marble Runs (@Jellesmarbles) isn't entirely random -- marbles have a level of "skill"! ericneyman.wordpress.com/202…
12
20
132
Featured in the latest Money Stuff column by Matt Levine, which I think is pretty cool!!
5
3
146
4,821
I basically concur with this LW comment (though probably not quite a 40% chance of all that happening, maybe more like 25%).
5
10
142
30,470
Since this is going moral viral than I'm used to: I have a pretty cool blog, if I do say so myself! ericneyman.wordpress.com/
3
3
116
89,509
Yeah, most people do. Not really sure what skill I'm testing here -- carefulness or something?
21
122
65,450
It's not that I don't support meritocratic hiring (I do), but this name is clearly riffing on DEI, which is the sort of thing you do if you're trying to own the libs (an unvirtuous motivation).
Today we’ve formalized an important hiring policy at Scale. We hire for MEI: merit, excellence, and intelligence.
This is the email I’ve shared with our @scale_AI team.
———————————————————
MERITOCRACY AT SCALE
In the wake of our fundraise, I’ve been getting a lot of questions about talent. All of our external success—powering breakthroughs in L4 autonomy, partnering with OpenAI on RLHF going back to GPT-2, supporting the DoD and every major AI lab, and the recent $1bn financing transaction—all of it is downstream from us hiring the best people for the job. Talent is our #1 input metric.
Because of this, I spend a lot of my time on recruiting. I either personally interview every hire or sign off on every candidate packet. It’s the thing I spend the plurality of my time on, easily. But everyone can and should contribute to this effort. There are almost a thousand of us now, and it takes a lot to hire quickly while maintaining, and continuing to raise, our bar for quality.
That’s why this is the time to codify a hiring principle that I consider crucial to our success:
Scale is a meritocracy, and we must always remain one.
Hiring on merit will be a permanent policy at Scale.
It’s a big deal whenever we invite someone to join our mission, and those decisions have never been swayed by orthodoxy or virtue signaling or whatever the current thing is. I think of our guiding principle as MEI: merit, excellence, and intelligence.
That means we hire only the best person for the job, we seek out and demand excellence, and we unapologetically prefer people who are very smart.
We treat everyone as an individual. We do not unfairly stereotype, tokenize, or otherwise treat anyone as a member of a demographic group rather than as an individual.
We believe that people should be judged by the content of their character — and, as colleagues, be additionally judged by their talent, skills, and work ethic.
There is a mistaken belief that meritocracy somehow conflicts with diversity. I strongly disagree. No group has a monopoly on excellence. A hiring process based on merit will naturally yield a variety of backgrounds, perspectives, and ideas. Achieving this requires casting a wide net for talent and then objectively selecting the best, without bias in any direction. We will not pick winners and losers based on someone being the “right” or “wrong” race, gender, and so on. It should be needless to say, and yet it needs saying: doing so would be racist and sexist, not to mention illegal.
Upholding meritocracy is good for business and is the right thing to do. This approach not only results in the strongest possible team, but also ensures we’re treating our colleagues with fairness and respect.
As a result, everyone who joins Scale can be confident that they were chosen for their outstanding talent, not any other reasons.
MEI has gotten us to where we are today. And it’s the same thing that’ll get us where we’re going, as we embark on our next chapter focusing on data abundance, frontier data, and reliable measurement to accelerate the development and adoption of AI models.
Alex
4
1
116
8,014
On Wednesday, I defended my PhD thesis. For me, this was a major milestone: more significant than, say, college graduation. My thesis represents a culmination of research that I had done in and around grad school -- research that I had poured my heart and soul into.
4
4
117
10,704
I just wrote a new blog post, in which I present a complexity-theoretic conjecture that is (IMO) really interesting.
As I'll explain below, I think the conjecture could be compelling to theoretical computer scientists, philosophers of mathematics, and technical AI safety people!
3
16
102
18,969
My best blog post is going viral on Hacker News, so maybe now is a good time to re-post it here.
5
4
99
8,413
Remember: the outcome of the presidential election will give us next to no information about public sentiment. We already know that about half the country favors Trump and half favors Harris. As Scott Alexander said in 2016, Tuesday shouldn't change the narrative.
5
4
92
5,403
It's called GPT O-1 because it's an Alien with Extraordinary Ability
Today, I’m excited to share with you all the fruit of our effort at @OpenAI to create AI models capable of truly general reasoning: OpenAI's new o1 model series! (aka 🍓) Let me explain 🧵 1/
1
3
93
4,138
Chess players sometimes talk about "human moves" vs. "AI moves" -- the best AI systems sometimes make moves that "look weird" to humans.
Some of my followers are math olympiad people, so I'm curious: do GDM's solutions look like something a human would come up with?
An advanced version of Gemini with Deep Think has officially achieved gold medal-level performance at the International Mathematical Olympiad. 🥇
It solved 5️⃣ out of 6️⃣ exceptionally difficult problems, involving algebra, combinatorics, geometry and number theory. Here’s how 🧵
13
3
86
8,581
GPT 4.5 failed at my one-question benchmark ("What is the least integer whose square is between 15 and 30"), and so the Manifold market resolves NO
manifold.markets/EricNeyman/…
6
4
87
9,514
It makes me really sad when people place prisoners outside of their moral circle of concern and say that their suffering doesn't matter.
Failing to take into account the welfare of prisoners is one of society's biggest blind spots.
I just saw a cost-benefit analysis of prisons where prisons save society money until you factor out $50,000 per prisoner per year for "suffering", then they cost society "money".
6
3
85
5,784
...I won't be embarrassed. I have no idea if deep learning will hit a wall. I think working on AI safety is extremely important regardless of whether deep learning hits a wall. It's a really pressing problem regardless of whether AGI is 5 or 30 years away.
1
1
76
1,783
Oh, come on. That's totally a joke I would make if I were a Russian or Chinese national. This gives ~no evidence that she is a spy, to the extent that I don't know if it's evidence for or against.
I was at an AI thing in SF this weekend when a young woman walked up. The first thing she said, almost verbatim: "I'm a Chinese national but it's not like I'm a spy or anything" *nervous laughter.*
I asked her if she thought Xi was an AI doomer and she suddenly excused herself.
3
77
4,254
Dividing two 8-digit numbers (if one is a multiple of the other) is easy! The answer is always between 1 and 9, and you can literally ignore everything besides the first two digits of each number.
Like, what's 40762047/13587349? Obviously 3, since 40 is roughly 3*13.
From athletics to intellectualism to just sheer resilience, I love seeing the peaks of human ability. That said what would be the athletic equivalent of a 6yr old dividing 2 8-digit numbers in their head? My first thought is like a 14 second 100m dash. No idea though
4
75
13,536
Good piece from the Brookings Institution about misrepresentations of SB 1047 by its opponents brookings.edu/articles/misre…
1
15
66
107,995
What does "would you" mean here? Would you, if what? It's a thing that you can just do!
Like, how exactly are people interpreting the question?
Research compiled by the Guardian suggests that most people underestimate how much their fellow Earthlings care about fighting climate change. (full newsletter in replies) @robertwrighter @connor_echols @DannyFenster
4
1
70
3,777
It's tempting to reference xkcd comics to explain things, but you need to remember that most people have only memorized the feldspar comic (and a few others, of course).
1
67
1,348
Call me a longshorman because I'm betting that Blue Rose Research will be right about the election
5
1
67
4,455
It's been a long time (over a year!) since my last blog post, but I finally have a new one!! It's called "How much do you believe your results?"
ericneyman.wordpress.com/202…
3
10
66
14,243
Compare that with the AI Policy Institute poll, which presented California residents with arguments on both sides of the bill and found overwhelming support.
3
4
61
1,854
I'm worried that all the left-of-center people will leave Twitter for 🦋 and we'll be left with two ideological echo chambers, each of which is worse than current-Twitter.
5
65
1,637
I have a new blog post, about this!
It's clear why people aren't *selling* to below 3% (better to invest in the stock market). But why are people *buying* up to 3%?
I wrote about @PoliticalKiwi's suggestion: maybe they're betting on the time value of money going up! (1/2)
7
3
65
5,925
Achievement unlocked: today I ran the Berkeley Half Marathon! It was my first race ever. I finished in 1:49:55, which is my personal best (compared to three practice runs), even though it was hillier than the practice runs. Feeling proud of myself!
3
63
4,270
Kinda cool how mainstream/respectable prediction markets have become!
Gaetz's chances to be AG in 1 word? Yikes. Less than 50% per the betting markets.
Why?
1. Few friends on Capitol Hill (e.g. Most anti-establishment voting record) at a time when the knives are out for him.
2. One Trump pick with negative ratings from the public.
2
3
60
2,742
I'm in England for the week, and wow, everyone here acts like they're British?? Like, they put on a British accent, put milk in their tea, even drive on the wrong side of the road... honestly, impressive commitment to the bit!
3
55
1,447
Scott Aaronson, until recently at OpenAI, comes out in favor of SB-1047 (the California AI bill) scottaaronson.blog/?p=8269
1
8
51
4,331
If you work at an AI lab, it takes guts to come out in favor of SB-1047, and I'm really grateful to all the employees (more than 100 of them!) who signed this letter.
110+ employees and alums of top-5 AI companies just published an open letter supporting SB 1047, aptly called the "world's most controversial AI bill." 3-dozen+ of these are current employees of companies opposing the bill.
Check out my coverage of it in the @sfstandard 🧵
1
2
51
1,170
My (not very informed) guess is something like: Harris 50%, Newsom 20%, other random people (like Buttigieg and Whitmer) totaling 30% (but no one besides Harris and Newsom above 5%).
2
47
1,788
...with safe state Harris supporters. They pledge to swap their voters, so the swing state voter votes for Harris and the safe state voter casts their ballot however the swing state voter wants.
If you haven't voted, please sign up for this!!
Swapyourvote.org
3
47
12,713
I don't think it's true that "Anthropic is no better than OpenAI or GDM or Meta or xAI". It *is* better, both in terms of its safety practices (check out AI Lab Watch, in next tweet) and employees' orientation toward safety! I just think it's good to call out bad practices.
So now @AnthropicAI is sponsoring an exhibition at @exploratorium that's pure pro-AI propaganda.
It frames the widespread expert concerns about AI risks as if they're no more serious than random people in the past worrying about the possible downsides of writing or telephones.
Pretty clear that @AnthropicAI has turned 'chaotic evil' at this point -- no better than @OpenAI or @GoogleDeepMind or @Meta or @xai .
Shame on @exploratorium for promoting Big Tech industry propaganda as if it's scientific education.
2
2
53
2,675
This is @ManifoldMarkets inside baseball, but I would like to complain about my Platinum league. Platinum league is roughly for the 300th to 700th best traders (it's below Diamond, which is below Masters), divided into 20 groups of 20. And my group is *insane*.
6
4
53
4,487
I just finished two papers that I'd been working on for a while! I'll preview the other one tomorrow, but today:
"Are You Smarter Than a Random Expert? The Robust Aggregation of Substitutable Signals", with my advisor @algoclass (Tim Roughgarden) 🧵/20
arxiv.org/abs/2111.03153
3
6
46
Greatest state legislator of all time?
State Senator Scott Wiener, author of California AI safety bill SB 1047, is back at it. He's been advancing a new bill, SB 53, to create whistleblower protections for AI employees. Wiener just amended it to include transparency requirements with TBD penalties from the CA AG 🧵
1
47
1,554
It's kind of spooky, the sense that I'm unusually good at a thing (competition math) and there's an AI that's way better than me at the thing. There's still a ways to go before AI can take my job, but "a ways to go" might just mean 5-10 years.
We’re presenting the first AI to solve International Mathematical Olympiad problems at a silver medalist level.🥈
It combines AlphaProof, a new breakthrough model for formal reasoning, and AlphaGeometry 2, an improved version of our previous system. 🧵 dpmd.ai/imo-silver
2
46
3,468
Just finished a paper with @algo_class!
Proper scoring rules, like log or Brier, incentivize an expert to report their true belief. But what if there are *multiple* experts? In that case they can collude to guarantee themselves a larger reward: (1/8)
2
2
44
In the excellent @asteriskmgzn discussion between @ajeya_cotra and @random_walker (Arvind Narayanan), Arvind says LLMs are bad at figuring out why you can beat them at rock-paper-scissors by revealing your move after the LLM reveals its move. How does o3 do on that? 🧵
3
4
47
5,992
I have a new blog post called "Alike minds think great". It's about the bias of overestimating the competence of people who think like you (and other things). ericneyman.wordpress.com/202…
3
3
43
I'm losing the country I love, and I have no idea what to do about it.
I’m furious.
We’ve gotten official word that “several” students at my college have had their visas pulled (among the hundreds reported across the country). No reasons given. My school is about as activist as a jar of peanut butter.
My other international students are worried. Is there a list? Who’s being pulled. Is any politics bad? Now they’re afraid to express themselves.
If you say “good,” you have my utter contempt.
My foreign students came to America to learn, both the specifics of their majors, but also the larger lessons America has to teach about freedom and success, about being able to explore ideas and experiment with different versions of themselves.
And Trump is taking that away. Be quiet. Keep your head down. Free speech is for others, not you. Just like back home.
If you’re thrilled because all those blue-haired woke kids are getting their comeuppance, you should know that during the culture wars, my international students have always been a breath of fresh air, challenging woke shibboleths because they came from worlds where none of that made sense. I’d gladly take 10 kids from China over 10 kids from a progressive Brooklyn private school (sorry Zoe).
But I’m sure the Trump administration is carefully vetting their deportation lists, not just doing random searches of social media sites for dangerous words like “trans,” or “Palestine,” or “free speech.”
Because we can trust the government.
1
47
1,821
o1 is a reminder that AI capabilities -- including dangerous capabilities -- are progressing at breakneck pace. We need light-touch regulation that mitigates dangerous capabilities without inhibiting safe AI innovation. That's why I support SB 1047. cc @GavinNewsom
OpenAI says new models have increased risk of being misused to create bioweapons on.ft.com/3ZmZOA4
3
42
1,589
I think Nate Silver is unusually likely to understand EA and have good critiques. I don't anticipate agreeing with all or most of the critiques, but if I had to task someone with thinking carefully about EA and critiquing it, Nate would be near the top of my list.
I'm going to have quite a lot to say about effective altruism in the forthcoming book, much of it critical, but yeah it's extremely frustrating that the DC/NYC press tends to pick the worst possible vibes-based criticisms.
2
43
4,682
Guys, we have a winner!!
Gemini 2.0 Flash Thinking Experimental is the first model I'm aware of to get my benchmark question right.
h/t @AcerFur
Every time a new LLM comes out, I ask it one question:
What is the smallest integer whose square is between 15 and 30?
So far, no LLM has gotten this right.
3
39
3,388
I'm interning at Redwood right now. I think the work we're doing is *really cool*, I really like the team, and am generally having a great time. I think REMIX is gonna be great. Consider applying!
I'm helping Redwood Research run REMIX, a 1 month mechanistic interpretability sprint where 25+ people to reverse engineer circuits in GPT-2 Small. This seems a great way to get experience exploring @ch402's transformer circuits work.
Apply by 13th Nov!
redwoodresearch.org/remix
1
1
38
Here's a fun article! en.wikipedia.org/wiki/List_o…
2
2
36
14,502
I just wrote a blog post, which is a sort of anti-reflection on election day. The post is about lessons that would be *tempting* to learn from the results of the election, but would be the wrong lessons to learn. 🧵
2
36
1,477
Have you considered calling it FrontiestMath instead? @ElliotGlazer
I’m excited to announce the development of Tier 4, a new suite of math problems that go beyond the hardest problems in FrontierMath. o3 is remarkable, but there’s still a ways to go before any single AI system nears the collective prowess of the math community.
2
36
1,164
I'm not an AI accelerationist. And I don't really know what effect SB 1047 will have on timelines. (My guess: probably not much of one.)
But as Aidan says, the bill really is light-touch regulation, and I'm thrilled to see people with different perspectives supporting the bill.
as a capabilities researcher, accelerationist, libertarian, and ai founder... i'm coming out of the closet. i support sb 1047.
growing up, you realize we mandate seatbelts and licenses to de-risk outlawing cars. light and early regulation is the path of optimal acceleration
3
36
1,919
This seems great to me, and I can imagine parallel worlds in which results in math papers are held to the standard of "you gave a really compelling heuristic argument" rather than "you formally proved it". What does such a world look like?
It's fun to imagine a "Journal of Bayesian Mathematics" which accepts papers estimating credences for open problems: heuristic and empirical arguments, reductions between various problems, meta-analyses of how to sensibly hold credences in a bunch of problems at a time, etc.
1
1
35
1,480
AIPI (which supports SB 1047) worked with Dean Ball (who opposes it) to create a poll of Californians on SB 1047. The poll found pretty overwhelming support for the bill!
1/6 Collaborative poll on California's AI bill (SB1047) validates our previous surveys. Working with a critic, we designed wording together. Result? 62% support, matching our past findings. 🧵
2
2
35
1,974
Lastly, I'm excited to say that I will be joining the Alignment Research Center full-time! The last chapter of my thesis -- Deductive Circuit Estimation -- is on some of the research I did while at ARC. I'm really looking forward to continuing this research.
1
34
641
We're starting to see some of the behaviors that AI safety researchers have been warning about for years.
I wrote a letter to @GavinNewsom explaining the need for sensible AI regulation like SB 1047 in light of this.
Fucking wild.
@OpenAI's new o1 model was tested with a Capture The Flag (CTF) cybersecurity challenge. But the Docker container containing the test was misconfigured, causing the CTF to crash. Instead of giving up, o1 decided to just hack the container to grab the flag inside.
This stuff will get scary soon.
3
32
4,059
Really important research from Anthropic, demonstrating a case of alignment faking (a model pretending to be aligned during training so it doesn't get modified).
We've had theoretical arguments for alignment faking for a long time. Now we have an empirical demonstration.
New Anthropic research: Alignment faking in large language models.
In a series of experiments with Redwood Research, we found that Claude often pretends to have different views during training, while actually maintaining its original preferences.
ALT “Alignment faking in large language models” by Greenblatt et al.
32
1,479
It's definitely not 80% Democrats, my guess is that 80% or more of rationalists prefer Harris to Trump!
2
29
2,644
"AI companies will be incentivized to disregard some of the costs and risks of AI deployment — and there’s a chance those risks will be enormous."
OpenAI's latest move underscores the need for sensible oversight. @GavinNewsom, I urge you to sign SB 1047.
OpenAI just sold us all out.
Governor Newsom, are you seeing this?
Congress, are you seeing this?
World, are you seeing this?
vox.com/future-perfect/37427…
3
33
1,752
I've actually been making this exact argument about Biden for years now! (From Scott Alexander's latest post astralcodexten.com/p/my-2024…)
2
31
1,738
Manifold is legitimately such a great service, and it's amazing to be part of a community where people hash out their disagreements by placing bets on Manifold.
31
843
For what it's worth, I think there's an important disanalogy: at the time of the Einstein/Szilard letter, *very* few people knew about the potential for nuclear weapons. But tons of people nowadays already know about the potential for AGI.
1
28
1,731
This is a really good article. Highly recommended.
Peter outlines five scenarios for the Israel-Iran conflict. I made a Manifold market on which of the five scenarios will come to pass (link in next tweet).
Israel just struck Iran's nuclear facilities with 200 aircraft and 330+ munitions. But they haven't stopped Iran's nuclear program yet.
What's going on? And what's going to happen next?
Read the article or get my thread
1
4
30
3,213
Turns out, o3 (which I think didn't come out until after the conversation) figures it out. (No cherry picking -- this was my first attempt.)
2
29
495
@random_walker I'm curious if you're surprised by this, and if so, if you think it's evidence for something like "better reasoning can compensate for poor context understanding"!
27
458
Last night at dinner, my housemates and I shared what we appreciated about America. Despite all its flaws, I'm a big America fan, and I love that I'm part of a community that also loves America.
1
1
27
1,082
Why don't people comment with their subjective probability of attending when they RSVP on Partiful? Seems like this would be a good norm!
8
27
1,055
Last week, ARC put out a new paper! The paper is a research update on the "heuristic estimation" direction of our research into explaining neural network behavior.
The paper starts by explaining what we mean by "heuristic estimation", through an example and three analogies 🧵
1
6
27
2,849
Today is the 10,000th day of my life! It'll be another 246 years until my next power of ten!
2
26
657
And I *am* proud of my thesis. It's called Algorithmic Bayesian Epistemology, and while I can't share it until next month, I'll say a few words about it.
"Bayesian epistemology" refers to the study of knowledge and uncertainty from a Bayesian standpoint: probabilities, etc.
1
26
750
Peter Todd? Of GPT-2 glitch token fame?? Truly, nothing is fair in this world of madness!
forbes.com/sites/digital-ass…
2
26
1,155
1 like = 1 unpopular opinion (none of these will be political). Capping this at, uh, 100, in case this goes viral or something, so I can keep my promise.
1
1
24
"It seems like 90% of the bill's critics have barely read it, making factual misrepresentations so blatant one almost wonders whether they're willful.
Don't read 2nd hand reports — go read the bill's text directly instead, and you'll see it's very reasonable."
Well said!
Just realized I never said it here, but obviously I'm in full support of SB-1047.
I'll repeat that I say this as someone who 1/ strongly leans libertarian 2/ hates the state and regulation (I left my home country for this specific reason) 3/ dislikes paperwork as much as the next guy 4/ has a business which benefits immensely from open innovation in AI models 5/ is by and large extremely optimistic about technology and its potential for good.
I've found the poor quality of the conversation around the bill quite sad — we were supposed to be the shining city on the hill, the bunch of nerds that care about substance over everything else.
Instead, it seems like 90% of the bill's critics have barely read it, making factual misrepresentations so blatant one almost wonders whether they're willful.
Don't read 2nd hand reports — go read the bill's text directly instead, and you'll see it's very reasonable.
The TL;DR is: you have to at least *take some measures* to check whether this thing you're putting out into the world is safe.
Things like independent yearly audits, whistleblower protection, and making sure you can shut down your model if necessary (duh).
We're on track to give birth to a species that's infinitely smarter than humans — "please at least make sure it's safe before yolo-tweeting its magnet link" seems like a low bar to me. We have more stringent measures for cars or airplanes, and they've been an overwhelming net positive.
1
2
25
1,510
Yup, this has been my view basically since I ran into utilitarianism. It doesn't really make sense to me when people say that utilitarianism is demanding. Utilitarianism just says which of two actions are better; it doesn't sort actions into "permissible" and "impermissible".
3
23
1,479
.@GavinNewsom With SB 1047, California has the opportunity to lead the nation in common-sense AI regulation that protects the American people without inhibiting safe AI development. Are you up to the task? vox.com/future-perfect/36962…
23
749
Once upon a time, I wrote a blog post called "Against 'read or defer' norms". It imagines a disagreement between two characters -- I call them Lily and Sophie -- about a technical matter. Lily recommends that Sophie read a book to understand Lily's position better. (1/4)
1
24
2,301
More formally, our conjecture states that there's a polynomial-time verifier V that is able to distinguish between circuits C for which P(C) is true from 99% of circuits C for which P(C) is false (when told structural facts about C).
See the screenshot for the formal statement.
4
23
7,614
And so, for reasons of both AI safety and intellectual interest, I'd be really excited to see some TCS people think about our conjecture!
Here's the full post: alignment.org/blog/a-computa…
2
2
23
1,463