Professor Buckmaster announced the possible use of his developments by OpenAI

New York University mathematics professor Tristan Buckmaster said he could not rule out the use of his work by OpenAI to solve one of the seven most important mathematical problems.

According to Buckmaster, he and his colleague Levent Alpege have made progress on a number of fluid dynamics problems, including an as-yet unpublished result for a modified version of the Navier-Stokes equations that describe the flow of liquids and gases. One of the millennium problems is associated with these equations, for the solution of which a million-dollar prize is provided. They used several AI models in their work, including Claude from Anthropic and Codex from OpenAI.

Last week, Buckmaster reached out to OpenAI to clear up rumors that Anthropic had allegedly solved an important math problem. A few days later, he was informed that OpenAI's internal model had proven this behavior for a version of the equations where the fluid was subject to an external force.

“I asked when the first prompt was sent. OpenAI did not answer this question directly for some time. In the end, they agreed that this happened within the last few days, after information about our work reached OpenAI,” Buckmaster said in a statement.

The mathematician asked OpenAI if the model had learned from or had access to his Codex sessions.

“I was told that the model does not access user data. I asked again, this time about training, and did not receive an answer,” the statement says.

According to Buckmaster, OpenAI offered him two options: either he first publishes his result on Euler's incompressible equation, and the next day OpenAI separately announces its solution to the Navier-Stokes problem, or Buckmaster himself writes an article about OpenAI's result, publicly admitting that their internal model obtained it. At the same time, the company insisted that Levent Alpege be excluded from the list of authors, explaining this by the fact that he works at Anthropic. Buckmaster allegedly refused both options.