The most prestigious law school admissions discussion board in the world.
|
 |
    16 posters qualify.
You have 0 points. Copying this list does not send funds. Paste it on Disperse, then connect your wallet there. |
Leading academic mathematicians now worried about AI putting them out of work
Poast new message in this thread
Date: May 10th, 2026 12:58 PM Author: azure thriller hall rigor https://x.com/wtgowers/status/2052830948685676605
AI can now generate results in a couple hours that are basically equivalent to a chapter in a math phd thesis. Pretty 180. Hopefully these people can learn to do something actually productive for humanity instead of getting paid to solve puzzles only they care about. SWE and math being wrecked by AI should free up a lot of human capital. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878547) |
Date: May 10th, 2026 12:58 PM Author: Rebellious Roast Beef Mother xo's own Terry Tao didn't seem so worried when he appeared on Dwarkesh (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878549) |
 |
Date: May 10th, 2026 1:27 PM Author: aromatic submissive field haunted graveyard Yeah because he's already part of the priesthood that needs to be kept around to "interpret" ai results (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878571) |
|
Date: May 10th, 2026 1:28 PM Author: controlling a virtual character on a computer underrated (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878572) |
|
Date: May 10th, 2026 1:33 PM Author: talking corner (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878578) |
|
Date: May 10th, 2026 1:38 PM Author: azure thriller hall rigor and he also thinks that will always be the case, when 4 years ago these models couldn't reliably solve grade school level problems. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878587) |
|
Date: May 10th, 2026 1:50 PM Author: Rebellious Roast Beef Mother Hallucinating is part of the reason that LLMs work at all, they have to be non deterministic is my basic understanding. Therefore there will almost certainly need to be a human in the loop for truly important problems. With that being said, Erdos problems appear to not be that meaningful for real world applications so maybe he will get automated by clankers. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878606) |
|
Date: May 10th, 2026 1:51 PM Author: controlling a virtual character on a computer cr (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878609) |
|
Date: May 10th, 2026 2:50 PM Author: aromatic submissive field haunted graveyard Humans confabulate shit all the time too and it often gets reified as knowledge for long periods of time. Look at literally the entire history of human knowledge. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878670) |
|
Date: May 11th, 2026 7:47 PM Author: ferengi (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880821) |
|
Date: May 10th, 2026 4:01 PM Author: azure thriller hall rigor there's nothing inherent about the model architectures that means they will always bullshit. the sampling aspect is beside the point. they are being trained to minimize prediction error on arbitrary human text. at the limit, this means no hallucinations. this isn't practical in reality because the models can't memorize everything, but different types of post-training will allow the models to recognize potentially incorrect tokens and self-correct. they can already do this reasonably well, which is people can now vibe code reasonably large SWE projects. model reliability in verifiable domains is growing quite rapidly. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878796) |
|
Date: May 10th, 2026 4:08 PM Author: controlling a virtual character on a computer LLMs have no empirical model of the world. They cannot check hypotheses. All they can do is create them. They need a human to check them
The reason why they are able to self-check hypotheses in math and coding is because you can literally run checks of those within digital engines, on a computer
You can't do that for irl. LLMs will always be confined to digital computer environments (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878810) |
|
Date: May 10th, 2026 4:48 PM Author: aromatic submissive field haunted graveyard Math isn't empirical though
edit: I guess you actually already addressed this to an extent in your poast re: your "math and coding" example, but this thread is about it's use in math which is purely abstract, which is what humans traditionally sucked at except for the smartest ones and also the perfect use-case for ai. In math and theoretical physics ai can traverse vast theoretical spaces in short periods of time and find new structure and rule out others which vastly speeds up discovery in those areas. We need humans to prompt and curate sure, but AI is going to be hugely important to these fields. I think its a mistake to think the only use for AI is completely decommissioning meat machines. Although it likely ultimately will retire many of them. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878839) |
|
Date: May 11th, 2026 3:53 PM Author: ruby bull headed cuckold set multimodal LLMs clearly learn predictive models of the world. There is a reason why generative models can create realistic lighting and scattering effects in images and videos. Predicting or generating the next token requires the model to construct a way to synthesize reality. As the models create ever more realistic output, they can do RL inside these models and create thought patterns that work increasingly well to deal with real life problems. Think of something like MuZero but for arbitrary domains. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880358) |
|
Date: May 11th, 2026 4:00 PM Author: controlling a virtual character on a computer No, they are only capable of predictions within their training data. They cannot generalize, period. LLMs are incapable of generalization as a basic property of how they work
I can't tell if people who claim they can generalize have become so entranced by LLMs that they genuine believe it, or they're just lying to hype up the technology, or a blend of both
Legit general AI is coming. People will build virtual RL environments with real world physics and throw massive amounts of money and compute at them. They will find clever ways to impart some base level of virtual environment RL training to robot brains and so the robots have enough training scaffolding that they can gather their own data irl and learn effectively from that, slowly but surely
But LLMs cannot generalize. LLM technology is limited to the world of language. Coding, writing, anything digital bc everything digital is code based. Nothing beyond that irl (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880386) |
|
Date: May 11th, 2026 6:41 PM Author: ruby bull headed cuckold set What precisely do you mean when you say they can’t generalize? Every time you give them a novel coding problem they are reusing learned patterns and algorithms to generalize. There has to be generalization involved or they wouldn’t be useable for anything and could only repeat exactly what they saw in the training data. I assume you mean something like out of distribution generalization. They can’t do that well, although in practice interpolation in a high dimensional space necessarily means extrapolation, so there is no clean break there. All you can say is the generalization is imperfect (although rapidly getting better and conceivably solved by models that can use more inference compute at run time) (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880685) |
|
Date: May 11th, 2026 7:24 PM Author: controlling a virtual character on a computer i meant out of distribution generalization, yes, sorry
the issue is that language does not capture anything close to Reality (the physical world). it seems to me like people are getting way too caught up into thinking there actually is some kind of overarching Platonic World Of Forms that sits above Physical Reality that LLMs are tapping into and operating on. but this isn't true. language is just language. it's not the real world. LLMs are only ever going to be able to "model" and extrapolate within this limited Language-World, because that is what they are trained on. they have no mechanism to interact directly with Reality
like i said above, this only applies to LLMs. other forms of AI that *can* interact directly with Reality will be developed. but this distinction matters imo because LLMs are being dishonestly hyped up as the end all be all of AI, and purported to be able to do things they cannot actually do (or will ever be able to do). this dishonesty can potentially become very dangerous if people in power decide to cynically (or even just ignorantly) appeal to LLM outputs for their decision-making and moral judgments (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880777) |
|
Date: May 11th, 2026 7:29 PM Author: aromatic submissive field haunted graveyard “language does not capture anything close to Reality.”
That is way too strong. Language is not reality, but language is not arbitrary noise either. It is a high-bandwidth social compression layer over reality. Models in theoretical physics aren't reality per se either (general relativity, quantum theory), but they are still immensely predictive representations however lossy, actually the most accurate we have. If representations are nothing then I guess might as well throw out all of human knowledge. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880787) |
|
Date: May 11th, 2026 7:36 PM Author: controlling a virtual character on a computer "it's like, a compression layer over reality, more or less, you know?" isn't anything close to Reality though. "kinda sorta in the ballpark" is not even close to Good Enough
the trajectory for LLMs seems obvious. they will keep getting better and better (at increasingly inefficient compute ratios) at Language-World tasks. but they're never doing anything outside of that. we need AI that can train directly on Reality for general intelligence (which depends on world-modeling of Reality) (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880790) |
|
Date: May 11th, 2026 7:58 PM Author: aromatic submissive field haunted graveyard Sure I think if the target is general embodied cognition I think the system needs direct contact, modeling to "reality".
But I still think the phrase “compression layer over reality” should not be interpreted as “kinda sorta in the ballpark.” a theory in a physics textbook isn't reality in itself either, it isn't identical to reality but it still is encoding real constraints and makes a real diff.
I get your point, that llmss can model the world only through the representational channels they are trained on, but what they are trained on still constains information about the real world even if lossy (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880853) |
|
Date: May 11th, 2026 7:56 PM Author: ferengi I definitely agree that a relationship with the real world – either real or simulated, IMO – is important for AGI (not sure if that's the same as general AI?) but my understanding is that these world-interactive machines are still built on LLMs, just with sensing layers on top. There's not a totally different type of system that I'm aware of. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880846) |
|
Date: May 11th, 2026 7:25 PM Author: aromatic submissive field haunted graveyard This is wildly false. Not defensible. Any model that can handle a sentence, code snippet, analogy, instruction, or combination of concepts it has never seen verbatim is exhibiting some form of generalization. LLMs plainly do this. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880779) |
|
Date: May 11th, 2026 7:59 PM Author: ultramarine passionate liquid oxygen theatre yes but at what cost? (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880857) |
|
Date: August 1st, 2026 1:04 PM Author: Metaphysics is Gay that's not the limit at all imo iirc afaict (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#50039051) |
|
Date: May 10th, 2026 1:25 PM Author: talking corner Lol fuck each and every one of them. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878567) |
|
Date: May 10th, 2026 1:31 PM Author: Cream Concupiscible Public Bath lol @ ugly nerds seeing their one asset (being 'smart') completely devalued overnight. being smart is worthless now. they have nothing. now they're just ugly *and* useless. (this applies to xo, inasmuch as anyone was ever smart here). (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878577) |
|
Date: May 10th, 2026 1:35 PM Author: seedy underhanded university useless brakes and it appears good looking hot people can now be good looking, hot AND smart if they so require (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878582) |
|
Date: May 10th, 2026 1:47 PM Author: filthy crawly antidepressant drug boiling water And soon ugly people will become hot thanks to radical new biology unlocked by AI (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878599) |
|
Date: May 10th, 2026 3:44 PM Author: Bearded marvelous space Or, in the alternative, turned into Soylent Green. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878761) |
|
Date: May 11th, 2026 8:10 PM Author: aromatic submissive field haunted graveyard chad von neumanns (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880877) |
|
Date: May 11th, 2026 7:29 PM Author: Soul-stirring becky good thing law relies on artificial bullshit not actual smartness (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880786) |
|
Date: May 11th, 2026 8:00 PM Author: ultramarine passionate liquid oxygen theatre AI can't take a good depo (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880859) |
|
Date: May 10th, 2026 1:41 PM Author: Cream Concupiscible Public Bath (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49878590) |
|
Date: May 11th, 2026 3:47 PM Author: ruby bull headed cuckold set (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880334) |
|
Date: May 11th, 2026 7:22 PM Author: Diverse point generalized bond (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880771) |
|
Date: May 11th, 2026 7:42 PM Author: ultramarine passionate liquid oxygen theatre Good luck learning statistics from an AI (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880801) |
|
Date: May 11th, 2026 7:45 PM Author: adventurous offensive menage "mathematics departments, who owe a duty of care to their students"
LOL, is this rube not familiar with the modern university system? (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880809) |
|
Date: May 11th, 2026 8:01 PM Author: Bipolar Prole Knife No one will have any kind of job in 10 years. It’s communism. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49880860) |
|
Date: May 20th, 2026 7:52 PM Author: azure thriller hall rigor And now it solves the first truly major unsolved problem. Value of human intelligence is heading straight to zero. 180^180
(https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49892236) |
|
Date: May 20th, 2026 8:52 PM Author: controlling a virtual character on a computer but there already weren't any mathematicians doing anything useful or holding real jobs (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49892322) |
|
Date: July 10th, 2026 11:52 PM Author: ultramarine passionate liquid oxygen theatre (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49992364) |
|
Date: July 10th, 2026 11:48 PM Author: azure thriller hall rigor GPT 5.6 just solved an extremely well known problem in graph theory that had stumped them for a long time.
https://x.com/__eknight__/status/2075643450196971805
The sentiment in math research circles is becoming increasingly negative given even current abilities and the extreme rate of progress. LJL. We don’t need smart people to spend decades working on “theoretical” math. The sooner we can automate this, the better society will be (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49992358) |
|
Date: July 10th, 2026 11:53 PM Author: ultramarine passionate liquid oxygen theatre I wonder why mathematicians aren't cerebrating these intellectual achievements in their chosen field (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49992366) |
|
Date: July 10th, 2026 11:57 PM Author: aromatic submissive field haunted graveyard If a human did it without AI they would celebrate. But since its mostly just researchers from BigAI companies using internal models to generate proofs as part of research pipelines to prove AI is 180, they are butthurt about it and not celebrating (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49992371) |
|
Date: July 11th, 2026 12:05 AM Author: azure thriller hall rigor cr. They even make up reasons to be angry about it. “openai trained on all of our papers!” Uh, yes, you made them public. Not to mention a lot of the training involves synthetic data anymore. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49992383) |
|
Date: July 11th, 2026 12:14 AM Author: aromatic submissive field haunted graveyard yeah I think its going to be especially funny when the big labs are eventually directly wired in to continuously train on all the latest embargoed data and newest research papers from all academic publishers and are just extrapolating theories that are decades ahead of institutions. they'll probably have to guardrail against it for alignment and institutional legibility purposes. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49992388) |
|
Date: July 11th, 2026 1:22 AM Author: Diverse point generalized bond wtf the prompt is written so meanly (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49992454) |
|
Date: July 11th, 2026 1:27 AM Author: aromatic submissive field haunted graveyard Do not return merely because current approaches fail or agents report theorem-strength gaps. Continue launching new rounds, reopening blocked approaches only when there is a genuinely new mechanism, and searching for fresh formulations. Return only when a complete affirmative proof has been found and survives adversarial audit. Do not return a reduction, partial result, isolated missing lemma, “best effort” summary, or explanation of why the problem is difficult. Spend at least 8 hours on this before even thinking of returning or giving up. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49992460) |
|
Date: July 11th, 2026 1:33 AM Author: ruby bull headed cuckold set LLMs are often lazy as fuck with hard problems and need to be forced to grind (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#49992464) |
|
Date: July 18th, 2026 7:14 PM Author: ruby bull headed cuckold set Got every question right on this year’s International math olympiad. Also recently resolved a major problem in statistics. It’s over for humans.
(https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#50008906) |
|
Date: July 29th, 2026 4:02 PM Author: ruby bull headed cuckold set This stuff is delicious. “oh no, i am going to have to kill myself because a machine is better at deducing things from abstract, human created bullshit only a few people care about.” these fags need to get a grip:
https://kirwinhampshire.substack.com/p/the-dark-night-of-mathematics (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#50032480) |
|
Date: August 1st, 2026 1:02 PM Author: azure thriller hall rigor https://x.com/nabeelqu/status/2083543826057048373
OpenAI proves 10 problems that are fields medal worthy. We went from struggling with grade school level math problems to this in about 4-5 years and vast amounts of more compute are coming online every year.
AI skeptics crying, losing hope. (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#50039045) |
|
Date: August 12th, 2026 2:02 PM Author: ultramarine passionate liquid oxygen theatre https://x.com/i/status/2087527300975714499 (https://xoxohth.com/thread.php?thread_id=5865573&forum_id=2#50063732) |