NYU Professor Accuses OpenAI of Copying Unpublished Math Research Strategy

NYU mathematics professor Tristan Buckmaster, alongside Anthropic mathematician Levent Alpöge, announced three proofs Tuesday, including a preliminary finding on the Navier-Stokes existence and smoothness problem. While significant, this achievement is overshadowed by an accusation that OpenAI attempted to solve the same problem using information obtained from the researchers' work.

Buckmaster expressed regret regarding a secondary aspect of the story. He revealed that during their work, they discovered that information about their progress had been passed to OpenAI. When they inquired, OpenAI claimed a full proof had been achieved, but provided evasive answers regarding the timeline and human involvement, eventually admitting the first prompt was sent only days after their work was seen.

Sebastian Bubeck, who leads OpenAI's mathematical research, refuted these claims as "false and inflammatory." He stated he entered discussions following academic norms and expressed disappointment at the public dispute, noting his commitment to high academic standards.

The controversy centers on the "Navier-Stokes existence and smoothness" problem, one of the Clay Mathematics Institute's Millennium Prize Problems. A solution carries a $1 million bounty. The problem involves understanding the Navier-Stokes equations, which describe fluid dynamics but are poorly understood in theoretical terms.

Buckmaster noted that the specific approach they took was unusual and employed by almost no one else. He found it implausible that OpenAI arrived at the same solution so quickly without prior knowledge of their work, suggesting they adopted the researchers' strategy after seeing their progress.

Although Alpöge is an Anthropic employee, he was not acting on company time. However, Buckmaster alleges that OpenAI viewed Alpöge's affiliation as a problem. He claims Bubeck proposed removing Alpöge's credit as a compromise and threatened Buckmaster's career if he went public, stating, "If you don't want me to be nice, then I don't have to be nice."

Buckmaster also raised concerns about the use of OpenAI's Codex. He utilized Codex extensively and noted that if OpenAI used a model trained on interactions from the Codex interface, they could have potentially regurgitated his work. OpenAI has policies allowing training on these interactions, though users can opt-out.

Buckmaster emphasized he has not seen OpenAI's proof and does not know how their model performed. He decided to speak out because he believes allowing a series of announcements to misrepresent the context would be unacceptable.