Will it run?
Models

OpenAI tries to mend fences with mathematicians, trips over its own announcement

By Rae Whitlock Clawpit staff
OpenAI tries to mend fences with mathematicians, trips over its own announcement

The announcement that sowed confusion

OpenAI has assembled an independent advisory panel of nine leading mathematicians, but the way it chose to unveil the group raised more questions than it answered. The panel, AGMAI, was announced on 21 September in a guest post on the blog of Terence Tao — Fields Medalist and a sharp critic of how AI companies conduct themselves in mathematics. OpenAI repeatedly stressed the group's independence: members are free to criticize the company publicly, publish their recommendations, and advise on topics they were never asked about. Yet most of the mathematical community, and certainly anyone outside it, heard about the move from OpenAI's own, much louder announcement.

Who actually appointed whom

The AGMAI website says the group came together after OpenAI approached some of the intended members with a proposal to form an advisory council, and those mathematicians decided to strike out on their own and invite additional colleagues. It is unclear which of the nine were approached first. The result: mathematicians who spoke to The Verge in the days after the announcement assumed they were looking at a body appointed by OpenAI — not an unreasonable assumption given the track record. The company has been "bulldozing" mathematicians for months: impressive breakthroughs followed by chaotic announcements that destroy trust.

"We don't work for them, we don't take their money"

Martin Hairer, an AGMAI member and professor at Imperial College London and the Swiss Federal Institute of Technology in Lausanne (EPFL), tried to clarify the picture in a video call from a lecture hall. "We don't work for OpenAI, we don't get money from them, and this is completely independent," he said. The group receives no financial, technical, or other support from the company, he added, and no member has signed agreements restricting their speech beyond standard confidentiality requirements needed to gain early access to research. Hairer described the past few days as "very frustrating" and "intense."

A wave of results is coming, and the anxiety with it

The panel's first task already looms: help coordinate the publication of dozens more results that OpenAI claims its unreleased model has produced. Researchers are already voicing concern about what that wave will do to the field. OpenAI itself explicitly noted that AGMAI "will not be responsible for advising us on how to pace our internal progress in mathematics" — meaning the tempo will be set in the lab, not by the council.

An advisory mechanism without teeth

The group says it is open to working with other frontier labs, but for now all eyes are on the approaching wave, and it is unclear whether the new advisory mechanism can slow it down or will merely accompany it after the fact. When the company producing the breakthroughs also controls the release schedule, an independent advisory committee — however distinguished — remains a spectator.