Loading page…
The Institute for Advanced Study-hosted group can advise on review, standards and release of mathematical findings, but its remit does not include pacing OpenAI’s internal work.
Listen to this story
OpenAI has created a nine-member mathematics advisory group hosted by the Institute for Advanced Study to scrutinize the significance, disclosure, coordination, and academic standards surrounding AI-generated mathematical results. The panel can publish advice, disagree publicly, provide unsolicited input, and change its membership, but cannot influence the speed of OpenAI’s internal math research and will not be paid by the company.
The panel’s mandate covers evaluating and releasing results, not slowing or directing OpenAI’s internal mathematics progress.
OpenAI says an internal model trained from August 28 solved more than 100 open problems; those claims remain company assertions.
The group includes nine advisers, including Timothy Gowers, Edward Witten, Martin Hairer, and Melanie Matchett Wood.
OpenAI has formed an independent mathematics advisory group to advise on how emerging AI-generated results are reviewed and communicated. The group can weigh in on disclosure and research standards, but not on how quickly OpenAI advances its internal mathematics work.
The Advisory Group on Mathematics and Artificial Intelligence is hosted at the Institute for Advanced Study. OpenAI says it will advise on assessing the significance of new results, communicating them, coordinating their release, and meeting academic and professional research standards. Its remit also includes advice on how OpenAI tools can support mathematical research and learning.
OpenAI says members may offer advice the company did not request, comment publicly on its impact on mathematics, and publish their advice. The company also says members will not be paid by OpenAI and that the group can change its own membership. Those provisions give the panel a route to disagree in public, though its role is advisory.
OpenAI framed the group as a response to rapid internal progress. It says a model whose training began on August 28 has resolved more than 100 long-standing open problems across most areas of mathematics. Those are company claims about an internal model, not findings independently established by this announcement.
Among the work OpenAI cites is its claimed resolution of the Navier–Stokes Millennium Prize problem, one of the Clay Mathematics Institute’s seven Millennium Prize Problems. Earlier this month, OpenAI said an internal system produced a proposed proof involving finite-time singularity formation and formally checked it in Lean, a proof-verification system. OpenAI says it does not intend to claim the prize.
The group gives mathematicians a defined role in how OpenAI assesses and presents mathematical results. Its influence will be clearest when its advice on a consequential finding is made public—or differs from OpenAI’s preferred plan for sharing it.
Editorial analysis
The important test is not whether an advisory group exists, but whether it becomes part of the process before a high-profile result is presented as a breakthrough. OpenAI has given the group room to offer unsolicited and public advice, an unusual feature that could make disagreement visible. Yet the company has explicitly excluded its internal mathematics pace from the mandate. The next meaningful signal will be how OpenAI handles a major result when the group’s view on review, credit or timing differs from the company’s preferred approach.
Loading discussion...
Make your case
A view to debatePublic advice may shape AI math releases without governing the underlying research pace.
Explain what would make the panel’s advice matter.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.