The Advisory Group on Mathematics and Artificial Intelligence

I knew mathematicians were a smart bunch, but honestly, the events of the past few weeks have really given me a new level of respect for them.

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.

A comment on the linked post from Burt Totaro:

> 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-...

> We are currently facing the very specific challenge of advising OpenAI on how to coordinate the release of a large number of significant results in mathematics that they report have been produced by their internal model.

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.

I don't really get it, seems like a gate keeping exercise for the academic community.

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.

Academics are the kings of, and have the most to gain from, gatekeeping.

"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.

The purpose is obvious - AI companies are using math to as a PR tool to demonstrate that AI is more capable than humans. This devalues math and mathematicians. There is even a relevant announcement from today, where OpenAI announced they solve 100 open math problems, without even bothering to state what they are.

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.

The composition of the panel might give us some clues as to the nature of OpenAI's unreleased results. Some of Ed Witten's work is quite closely related to the Yang Mills existence and mass gap problem, which is one of the remaining Millennium problems. I went through the list of panel members and asked an LLM to associate each member of the panel with one or two millennium problems. Here is the resulting count:

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.

A mathematician's suggestion:

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.

I noticed Terry Tao is not in this "advisory group" ...

I really wish the models were smart enough to "tail their own logs" and tell us whose work/papers/chats were the catalyst for these math insights. What makes their "internal" model so much better at math? Must be trained on a whole lot of honesty.

discussion of oai announcement: https://news.ycombinator.com/item?id=49790389

Is the advisory group abelian? I'll see myself out...

whats the difference between a mathematician and a pizza? a pizza can feed a family of four

well terry tao no longer has this problem ;)

> Importantly, the group will not be responsible for advising us on how to pace our internal progress on mathematics.

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.

France, US/Italy, France/England, Switzerland/England, US, England, US, US, US.

Incredible international representation...

The way I've understood this whole thing so far, the main reason why the mathematicians are so energized right now is because the AI companies' problem solving spree frames mathematics as a discipline that exists to solve open problems. Mathematicians don't fully share this view even though they value problem solving, and they're afraid that this framing will produce the following chain of thought in the public: Mathematicians get paid to solve problems -> AI solves problems better -> we don't need to pay mathematicians. So they're trying to reframe their work to shift problem solving to the background and keep getting paid. I leave it to the mathematicians to explain what their value is, as they would do a better job of it, and it remains to be seen if this reframing will work.

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.

Okay so more ivory tower slop.

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.

Why don't they just publish the results? What does an advisory group do in this case when none of the members can validate or verify any of the purported solutions?

Related:

Advisory Group on Mathematics and Artificial Intelligence (OpenAI post)

https://news.ycombinator.com/item?id=49790389

Honestly I am not impressed with Tao treating "ai" as a threat to mathematicians. He might as well say math itself, specifically linear algebra and gradient descent are a threat to mathematicians. Maybe think of something original rather than pander the hype of day?

[dead]