OpenAI releases more than 370 mathematical results, prompting expert concerns

OpenAI released more than 370 mathematical results on Tuesday, covering algebra, theoretical computer science and mathematical logic. The publication prompted concern among mathematicians and leaders of research institutions.

Last month, the company also said it had solved the Navier–Stokes equation, one of the world’s most difficult mathematical problems, which carries a $1 million reward.

Experts call for verifiability

The Institute for Advanced Study in Princeton said it did not support testing the hardest mathematical problems with proprietary models unavailable to the wider mathematics community. The institute said people may be unable to understand, verify or take responsibility for arguments generated by AI.

Tristan Buckmaster, a New York University mathematician, told The New York Times that researchers prompting models to solve equations may have helped the systems reach their results. He said some outputs may involve models completing work begun by other mathematicians.

OpenAI to work with institute

In response to the concerns, OpenAI announced that it would work with the Institute for Advanced Study to give mathematicians a role in deciding how the field should proceed. The company did not say it would stop testing its models on advanced mathematical problems.

The advisory board also called on AI labs to provide “equitable access” to their models for mathematicians worldwide. It said reliance on proprietary internal systems could create a two-tier structure in which AI labs outpace the rest of the field.