The Advisory Group on Mathematics and Artificial Intelligence
terrytao.wordpress.com
terrytao.wordpress.com
They might be the first community I've seen to experience the AI "rush" and (at least as presented to an outside observer) immediately come together, assess the situation, and calmly, empathetically, and rationally act. They evaluated what AI is good at and what it lacks. They've thought through how it'll likely affect their field in the future. They've explained where the need for humans still lies, and made clear proposals for how to change their own field and for what demands to make of AI companies. Of course they're not all on the same page, but they're at least talking and trying.
They haven't started worshipping the machine god and loudly claiming their whole field is solved. Nor have they flailed wildly at LLMs as if complaining enough about it will make them go away.
Every major statement I've seen come out of the math community on this matter reads as well thought-through, humble, reasoned, and deeply human.
In these days of fear, uncertainty, and obsolescence anxiety, honestly, they've given me some confidence that maybe we will figure this stuff out after all. Maybe we'll learn from them. Who knows.
Their field is extremely rigorous. As rigorous as it can possibly get in that they have to prove each and every line of their work beyond any doubt. The discipline they have cultivated in their culture shows in their response to AI as well.
As opposed to some other fields in which rigor was either not part of the culture or was not always possible. For e.g., software engineering in terms of code quality being produced. There were some indirect signals here and there but they are all subjective.
Perhaps we should define what rigor means.
Your point is that code is executable and speaks for itself whereas a Math proof (non-lean) is just someone's writing on a piece of paper.
Now let's compare the "practice" of doing Math and software engineering. In Math, every step is very intentional, and getting to a point where a proof is complete and correct is a very long, laborious, difficult and intentional process. Not to mention, the work is also peer reviewed (for published stuff). This is what I mean by rigor.
In software engineering, the practice is quite different. We defined the problem (somewhat), come up with a design that we "think" would work, write programs that we think is correct and then execute it. Most of the time it doesn't work exactly as we would have predicted. So we take the signal and adjust. So it's a more iterative part and this gets us closer to reality (what we actually want it to be), step by step.
So comparing the two, the major difference I see is in one each and every step is very intentional and we can't guess it. While in the other one we have lots of liberties, but we are still making progress.
So to me the difference is just between the practices followed in the field when it comes to rigor.
Can't disagree with "deeply human", but I wouldn't always use it in a positive sense:
https://proofsandprompts.com/2026/09/10/open-letter-about-th...
>Participation in an event so closely associated with Anthropic and OpenAI could plausibly negatively impact the future reputations of participants
As a mathematician, I am a bit disappointed by my (admittedly illustrious) colleagues.
I get the need to take it slowly (and I am a quite impatient person, so I shouldn't get to decide stuff like this), but everything said feels a bit too sour grapes for my taste.
Ok, maybe AI did not solve the field (I believe it will, btw), maybe there is a need for human "understanding", but:
1) They don't seem to consider even the possibility (not the certainty) that they might be wrong, that math as we know it is gone, and we cannot "adapt"
2) They seem to have been oblivious all these years about AI eventually reaching this point (at least I personally wasn't, I predicted this stage back in 2018)
The answer to this is probably just “get out of the way” and then your “work” would be to develop a better appreciation of what has been produced.
They aren't in the way. AI labs can do their own math all they want, and nobody is stopping them. Rather, mathematicians are simply speaking about the topic.
So is the answer that they should just shut up?
For example, group theory lead to a useful way of thinking about how to solve a wide array of problems. That is what mathematicians are seeking.
I'm sorry that you personally don't like that mathematicians are "bitching" about OpenAI snaking their work out from under them, and would like to silence them. But if we're going for consistency's sake, maybe consider taking your own advice here before giving it to others.
In the meantime, OpenAI can do whatever they want, mathematically-speaking. Nobody is in their way. If there isn't further meaningful progress from them, that would seem to strengthen the work-product-theft hypothesis.
Wiping out an occupation that people enjoy and that gives them purpose (making them "obsolete") is a public mental health crisis, if anything. It's unclear that the "benefits" of the "solved" field actually outweigh that.
No one is saying "Perelman and Wiles didn't need that silly AI and N-S was more of a counterexample". Most people imagine the sky is falling.
What if OpenAI is stalling because they don't have 100 additional unpublished proofs as claimed?
I did notice, this is why I had thought that they had thought through the consequences. But, unfortunately, it is likely they imagined that AI would plateau somewhere around the "smart undergraduate" level and that they will stay relevant in pure problem solving.
Taking GP at his word, I'd be more terrified if Oai had identified one hundred NEW open problems..
In contrast with Fields Medallists, some Nobelists get the prize for something they'd eventually NOT be (in)famous for. Like Josephson. In hindsight, will it be because they found out, as young unknowns, how to explore underappreciated avenues without millions in dedicated funds (uh encouragement).. ??
GP, did you? Get the prize for the question you are most proud of.
Which is what I think a large chunk of AI progress actually is - people taking the latest models out for a spin, and browbeating them to actually ship code, and RLHFing them to ship better code in the process, either directly or indirectly through Github.
How do you reconcile that with incompleteness and undecidability results?
I think it isn't true that no one imminent is considering (1). Tsimmerman said he thought the job of math researcher was over as soon as he announced he was leaving the field.
I think the question is so far still open, but it is easy to imagine Tsimmerman was right. I haven't been impressed with the calls for "human understanding" as the really important thing. It seems to miss the point of why math has been funded (not exactly problems, but certainly not to ensure every theorem is understood either).
As a scifi comment, it is now possible to start thinking about some future AI system which decides if it will need to explain certain math to humans and to pick and train young people for that purpose.
Don't give Dario any ideas please.
> next week: Mathematicians are solved [214 points, 891 comments]
The 3D designers are in a tougher situation because it's far less of abstract issue then math, and yet the results are astounding. It's hard to believe until you see it yourself. It's surreal.
This group of Mathematicians just happens to be more willing to accept the inevitable and adapt. I would actually say software engineers, being amongst the earliest impacted by AI after creatives, clearly have adapted to the new reality much more broadly as evidenced by online discourse and coding agent providers' skyrocketing revenues.
One might argue that this is due to size of the software community, or commercial interests. But I do believe OP was right in his description of the response of the math community as different from the software community.
But my point is the wider Mathematics community didn't do that either. We are just looking at the subset in TFA and assuming they represent the whole community. If you look at AI + Math-related articles and threads on HN you'll definitely see a lot of consternation from Mathematicians.
There are certainly still strong opponents of AI, but they seem to be a small, vocal minority. Most developers seem to have simply adapted and moved on.
The whole thing is far from over when we haven't experienced the worst of the technology yet: deskilling, concentration of power, extreme imbalance of wealth, military and surveillance uses, etc.
I hope you’re not assuming you’re immune to being in a biased bubble yourself.
I’m just honestly sharing my experience. Obviously, your experience has been different, but both experiences can be true at the same time.
1. "Your experience is not universal from my perspective, and I offer my experience as evidence"
2. The possibility that this adaptation you are observing is neither a stable equilibrium given future developments, nor is necessarily an indication that the people using AI have bought into it fully. I use AI heavily while remaining a sceptic.
And yes, I agree that we still don’t know the long-term consequences of AI. But current LLMs are nowhere near good enough at judgment or new ideas to replace most jobs.
I expect they’ll mainly automate parts of jobs rather than replace them entirely. That’s exactly what I’m seeing as a software developer. I use an LLM to write my code, but it only works well when I apply good judgment and shape the context and prompts to remove ambiguity and then carefully review the results.
If you look at the comments on the blog post, you'll find plenty of this.
> If I thought this was a good thing to do, this would be a great group of people to do it. But I have immediate misgivings. OpenAI has had some very bad publicity, and so they are trying to exploit the trust and respect that these mathematicians command. It is unrealistic to think that this group can change the way OpenAI does business; I don’t need to tell you all the objections that people have raised to that. Is it a good idea to help their crisis management?
https://terrytao.wordpress.com/2026/09/21/advisory-group-on-...
The only thing I want to see is the problem statements, solutions, and associated Lean proofs. Anything else is gatekeeping. What a low point for academia.
Or are you perhaps sublimating your own AI anxiety into confrontational assertions that other people who discuss AI aren't as AI-pilled as you?
If the advisory group says something like "AI is bad and nobody should use it in math", I'm pretty confident OpenAI will ignore them.
On N-S I think nobody internally felt confident enough to publicly step up. So Bubeck had to wing it.
It remains to be seen which of the big names will work with whom inside the org
I actually want to see interesting theories and mathematical ideas come out of proofs more than I want solutions/proofs. After all most maths doesn't have direct practical applications. So although unsolved problems are a good barometer of 'there's still stuff left to understand here', the interesting part about solving them is less knowing what is true and more 'how does this help us understand this field better than we did before?'.
I would still like to know if the Riemann Hypothesis is true though!
A proof would then merely update you from 99.9% to ~100%, which is a smaller update than the example of checking that your keys are indeed in your pocket, where you go from 99% to ~100%.
Of course OpenAI and others will hype any LLM led/assisted research results, and the media will find juicy headlines to write about it. But in the end they are publishing findings, and the academic community can gauge the developments and decide what they want to do with them. To me that seems like research working as expected.
There are typically processes to ensure rigor in the findings and to weed out crap.
It doesn't always work, but it tends to work much better than any other community
"This is changing the power we have, let's create a new power structure where we're still at the top, and in control."
No surprise here.
However, even if they are successful in removing math from the PR toolkit of AI companies, I can’t see it more than a short term band aid. If AI progression continues, the announcements will just transition from big flashy one, to a kid using Claude on their laptop.
I don’t have a solution here, and I’m worried it’s coming to all human fields. So it’s more of an ability to observe what might happen to all of us.
You can also look (as I did back in 2018 when contemplating the end of math) at how chess or Go players adapted mentally to the idea that whatever they may do, a machine could do it better (Lee Sedol famously quit Go).
In that case, everything stands to be disrupted and those who wield AI will have massive amounts of power in the short term. But this is not a stable situation, and therefore those who wield AI will have to cede control of it or otherwise enter into a great war with those that do not.
I hope for all our sakes that there is a practical ceiling.
Calling it a publicity stunt is an interesting way of framing ground breaking discoveries. The beneficiaries down stream of these discoveries could not care less how they were found.
But they do. Most of higher math has no practical applications and is just a mental playground, philosophy constrained by formal logic and a set of axioms.
Hope is these capabilities will somehow translate to something more practical like physics, chemistry or biology.
That announcement is about exactly this group: https://openai.com/index/advisory-group-on-mathematics-and-a...
This group has been formed to tell OpenAI whether or not (and if yes how) they should release those 100+ problems.
IMO “AI companies are using math to as a PR tool” is not quite it; they are using math problems as a benchmark to evaluate their AIs, but after the prior blowback from the mathematical community, they are happy to sit on the results until the community (this group) tells them how to publish the results.
And as you say, even if they don't publish the results eventually anyone will be able to do this when this model is released, so it's just a matter of time.
Hodge conjecture: 5 Birch and Swinnerton-Dyer Conjecture: 3 Navier-Stokes: 2 P vs NP: 1 Yang-Mills existence and mass gap: 2
So this gives a bit of further evidence that they made progress on the Hodge and Birch and Swinnerton-Dyer conjectures. Maybe also Yang-Mills.
Notably absent is the Reimann hypothesis.
Worthless then isn't it? Next month there could be another 1,000 solutions to handle.
This is more pointless than Facebook's Oversight Board - at least that one Facebook pretends has power.
I would push for an audit of the whole mathematical corpus (all articles etc ingested by the present AI) as soon as possible. Since there are of the order of 10^7 articles, compared with the progression of results announced, this might be possible in less than a year.
All these results exploit the existing corpus. With enough time and maybe several orders more attention, we humans would discover these results.
So instead of producing heaps of new proofs (but be welcome to do this IMO), make an audit, OpenAI or Anthropic or anybody who listens.
https://chorasimilarity.wordpress.com/2026/09/22/a-mathemati...
> During this event, OpenAI requested an interview concerning my vision of the future of AI and mathematics. I accepted, and spoke with them for perhaps an hour. I had done similar interviews in various venues, and I assumed that, as with these other cases, they would eventually post the entire interview online, which talked about both the possibilities and risks of AI much as I have done in these other interviews. As it turned out, they only used a few snippets of that interview for that infamous advertisement instead.
https://terrytao.wordpress.com/2026/09/07/finite-time-blowup...
Incredible international representation...
Advisory Group on Mathematics and Artificial Intelligence (OpenAI post)
For example, had they decided to put their paper on arXiv, they would be facing a 1 year ban, followed by not being able to put anything on there without prior peer review.
Sashas of the world, unite!
Gerko, to donate GPUs for openweight serving
Elbakyan, to release some gpt7 related torrents
Others? Wei?
Why aren't they consulting the younger mathematicians whom this actually affects?
Anyways, this is overblown. Let me know when AI swarms can create a research program or library from scratch. We're really missing long term planning and abstraction creation in these things.
Regardless of that, what happens long-term if AI gets good enough to solve all the hard open problems?
Millenium problems may give AI companies some temporary hype, but advancing research in some narrow directions to find new problems is probably not worth it to them. In addition, current problems have clear, human-defined acceptance criteria. Problems that are both found and solved by AI will be opaque enough not to generate interest in anyone unless they are found to have immediate applications. Mathematicians will somewhat distrust 1M+ LOC Lean formalizations as well as any original AI output in the form of creating a bunch of new concepts. There will surely be heated discussions around that.
Afterwards, mathematicians take some time to catch up with the solutions to existing problems, use AI to understand them and map out further directions. Problem solving by AI companies slows down since there are no cool problems left. They move on, mathematicians get back to whatever they were doing before, now with the help of AI.
With every mathematician having a strong mathematical AI model, a new problem arises. At any point of the current mathematical frontier, a mathematician can use AI to explore a new direction at a high speed. If all mathematicians do that, each will have their very advanced understanding of their direction until they fail to catch up the AI, and nobody knows whose direction is the one that others should check up with. Many possible connections with other branches and explored directions will be lost, unless there's a unified map of all these explorations. It will be hard to create this map, because while exploring one direction may have a moderate cost for a mathematician, having an AI do an overview of all other explorations to find some commonalities and synthesize the findings into a unified theory will take serious amounts of resources only available to AI companies, which they might be reluctant to spend.
Individual mathematicians' explorations will slow down to human level, too, with their unique findings being hard to publish because nobody will care and the mathematical community wants to do things that other mathematicians can appreciate.
In the end, the mathematicians' work does not change that drastically, unless AI gets good enough to one-shot some new crazy branch of mathematics that reveals something with high real-world impact. But that's a discussion for another time.
well terry tao no longer has this problem ;)