OpenAI Says One Model Resolved 100 Open Problems. The Mathematicians It Recruited Get the Announcement, Not the Pace

OpenAI announced on 21 September that it has formed an independent Advisory Group on Mathematics and Artificial Intelligence, made up of nine mathematicians and hosted at the Institute for Advanced Study in Princeton. In the same post it explained why the group was needed.

On August 28, we began training a new internal model. In addition to resolving the Navier–Stokes Millennium Prize problem, this model has now resolved more than 100 long-standing open problems across most areas of mathematics.

Nothing accompanies that sentence. No list of problems, no proofs, no indication of which areas of mathematics are involved. The group's own site says its present task is "advising OpenAI on how to coordinate the release of a large number of significant results in mathematics." The results sit inside OpenAI, the company has told the world how many there are, and the first job of the mathematicians it has just recruited is to work out how to let them out.

What the group is, and what it is not

The nine are François Charles, Camillo De Lellis, Timothy Gowers, Martin Hairer, Nikhil Srivastava, Ulrike Tillmann, Ravi Vakil, Edward Witten and Melanie Matchett Wood. Among them are three Fields medallists, and in De Lellis, one of the leading figures in the fluid-dynamics literature the Navier–Stokes claim lands in.

Its independence is real as far as it goes. Members take no payment, the group says it "operates independently of any AI company," it can comment on OpenAI's impact on mathematics without being asked, and it publishes what it recommends. It also states plainly that "we do not have decision making power at any AI company, and the responsibility for the decisions made by any company will rest with that company." The enforcement mechanism is nine reputations and a public website.

It answers the second objection, not the first

On 11 September, twenty-five Fields medallists published a declaration titled "A Severe Misalignment of AI in Mathematics." Its opening claim is structural:

The goals of the AI companies and the goals of the mathematical community are severely misaligned.

Its specific complaint is procedural:

Often these solutions are announced in a rush, leaving no time for a proper writeup, the isolation of new methods and ideas, and citing relevant previous work.

The advisory group is a precise answer to the second sentence and no answer at all to the first. Rushed announcements are exactly what a body advising on "the responsible presentation and release of mathematical results" is built to slow down. The misalignment the declaration described is a claim about what the two sides are for — insight on one side, solved problems as a demonstrable capability on the other — and OpenAI has carved that question out by name. Pace is the one topic the group is told it is not responsible for.

That carve-out is narrower than it first looks, and more revealing. OpenAI has conceded the announcement and kept the research. A group that can tell the company how to publish a hundred results cannot tell it to produce ninety.

One signatory of the declaration, Martin Hairer, has joined the group. Terence Tao, who published the declaration and signed it, hosted the group's own announcement as a guest post on his blog with an editor's note establishing that it was not his text.

The dispute nobody is advising on

The Navier–Stokes announcement on 8 September did not only draw an objection about pace. Axios reported that the NYU mathematician Tristan Buckmaster, who had been working on related fluid-dynamics problems with Anthropic researcher Levent Alpöge, raised questions about how OpenAI came to the problem, whether private material from their use of Codex fed into the company's solution, and an alleged push to drop Alpöge from authorship because of his employer. Those are allegations, not findings.

They are also outside the group's charter. "Responsible presentation and release" governs what a result looks like on the way out. It does not reach how a lab decides which problem to attack, what it may learn from the research its own tools sit inside, or who gets named when a result arrives. That is the harder conflict, it is the one that produced a public accusation rather than a polite letter, and the new body has not been asked to touch it.

What would show it is working

The test is short and near. The group publishes its recommendations. If the first hundred results arrive as a list with proofs, prior work cited and human contributors credited, the arrangement did something. If they arrive the way the count did — a number, an unnamed internal model, and a date — then nine mathematicians were the announcement.

Sources: OpenAI: Advisory Group on Mathematics and Artificial Intelligence · Advisory Group on Mathematics and Artificial Intelligence (agmai.org) · Announcing the Advisory Group on Mathematics and Artificial Intelligence — guest post, What's new · TechCrunch: OpenAI forms math advisory group as its AI resolves more than 100 open problems · A Severe Misalignment of AI in Mathematics — What's new · OpenAI: On the Navier–Stokes Millennium Prize Problem · Axios: OpenAI's historic math solution overshadowed by credit controversy · Quanta Magazine: AI Has Solved One of Math's $1 Million Millennium Prize Problems · Scientific American: OpenAI claims blockbuster math breakthrough amid swirl of controversy