OpenAI Announces Independent Mathematics Advisory Group

OpenAI announced an independent advisory group of mathematicians on September 21. It arrives ten days after 25 Fields Medallists signed a declaration attacking the race among AI companies to solve mathematical problems. In the same post the company said a new internal model it began training on August 28 has resolved more than 100 long-standing open problems. OpenAI confirmed in numbers exactly the pace the criticism was aimed at, while handing the question of how to communicate the results to people outside the company.

The group is hosted at the Institute for Advanced Study and has nine initial members. They are François Charles (ENS-PSL), Camillo De Lellis (IAS, GSSI), Timothy Gowers (Collège de France, Cambridge), Martin Hairer (EPFL, Imperial College London), Nikhil Srivastava (Berkeley, Simons Institute), Ulrike Tillmann (Oxford, INI), Ravi Vakil (Stanford), Edward Witten (IAS) and Melanie Matchett Wood (Harvard). The nine seats are spread across universities and institutes, and not one of them belongs to OpenAI.

The post states that the group operates independently from OpenAI. It can offer advice that has not been requested, comment on OpenAI's impact on mathematics, and make its advice public. Members are not paid by OpenAI, and the group can change its own membership. The company wrote that "its value depends on its members being able to exercise their own judgement and challenge ours."

A sentence in the same paragraph draws the boundary. The post states that the group will not be responsible for advising OpenAI on how to pace its internal progress in mathematics. What is open to advice is how results are assessed and how they are communicated, not how quickly they are produced.

The numbers the company placed in the same post show why that line was drawn. OpenAI began training a new internal model on August 28, and said the model has resolved the Navier–Stokes Millennium Prize problem and, beyond that, more than 100 long-standing open problems across most areas of mathematics. The pace of that progress surprised the mathematicians inside OpenAI, the company said, which is why it has been discussing internally how best to inform the community so the field can prepare and adapt.

The immediate trigger for the group was a declaration published on September 11 under the title A Severe Misalignment of AI in Mathematics, whose 25 initial signatories are all Fields Medallists. Terence Tao, who received the Fields Medal in 2006, wrote in the post carrying the declaration that it was unfortunate there had not been time for a more consultative process as with the Leiden declaration, but that the urgency of the situation meant a statement had to be released sooner rather than later. The declaration has also been posted on a separate web page, which is collecting further signatures.

The declaration's argument gathers into a single sentence about misaligned goals. The signatories wrote that over the last few months the mathematical capabilities of large language models have improved dramatically, to the point of solving major outstanding problems in many fields, but that the push by AI companies to use mathematical problems as a benchmark is detrimental to mathematics as a science and to its community. "The goals of the AI companies and the goals of the mathematical community are severely misaligned," they wrote.

The harm the declaration names is not abstract. "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 of others," the signatories wrote, adding that, as in all creative professions, this raises severe attribution and plagiarism questions. The same passage says that without the willing mathematicians who take care of their development and integration into the mathematical canon, AI-conceived ideas would never become fully alive and the human transmission chain between mathematicians would be lost.

The declaration starts from the position that solving problems is a tool rather than the goal. The signatories wrote that famous problems have long served as landmarks and lighthouses against which improved understanding is measured, and that solving one has been a sign of new insights and interesting methods. What followed was a long and arduous process of talks, discussions and simplifications leading to a textbook presentation a student can read, and their warning is that mass-producing true and false statements at an ever faster pace could destroy that fertile ground instead of feeding it.

Friction between the two sides kept building before and after the declaration. According to press reports, Tristan Buckmaster, a professor at New York University, said OpenAI pressured him not to credit a collaborator who solved an important mathematical problem because that collaborator works for Anthropic. The same reporting said OpenAI withdrew its sponsorship of a mathematics event at Caltech after criticism from researchers there, and that the company's proof remains unverified.

METAL checked both lists and found one name in common. Martin Hairer, who received the Fields Medal in 2014, is one of the 25 initial signatories of the September 11 declaration and one of the nine initial members of the advisory group OpenAI announced ten days later. A person who signed the declaration taking a seat on the group means it was not assembled by routing around its critics.

METAL reported on the declaration in which 25 Fields Medallists drew a line for AI companies. The question left open then was whether an AI company would build a channel to answer that demand, and the answer, ten days later, is this advisory group. OpenAI wrote that working with the group is a first step and that difficult questions lie ahead about how AI can support mathematical understanding and how the benefits of these capabilities can reach the wider community.

The announcement also went out through the company's official account the same day. OpenAI wrote that the group will advise on how it assesses and communicates new mathematical results, upholds academic and professional standards, and builds tools that support mathematical research and learning, adding that it wants mathematicians at the centre of shaping how AI supports mathematical understanding. When METAL checked, that post had 870,000 views.

Seen with a reporter's eye, the structure of this announcement is familiar. Assessment, communication and academic standards are opened to the outside; speed and training stay inside. The unpaid terms, the clause letting the group change its own membership and the promise that it may publish its advice will only be priced on the day the group first says something in public that the company did not. More than 100 problems have already been resolved, and when and in what form those results will be released has not been settled. What the mathematicians asked for was not a halt but time to understand, and who sets the length of that time is not written into this announcement.