Sam Altman says OpenAI will keep releasing AI math results

Sam Altman, the CEO of OpenAI, says he understands why mathematicians are overwhelmed by the company's flood of AI-generated results. He also says OpenAI will not stop. Asked whether more proofs are waiting to be shared, he answered: "We do. Of course." He spoke on the Cheeky Pint podcast, in an episode released Thursday.

What Altman said

Altman argued that the world benefits from learning how the universe works. "I do have empathy for the method," he said, but he believes the public's right to new knowledge outweighs the disruption. He admitted OpenAI needs to get better at explaining its findings.

He also said the company's math push "got out of control" for a short time. That was deliberate, he said, to test the limits of its AI models. He described an internal Slack channel where employees threw the models at the hardest problems they could find.

Why mathematicians are frustrated

The Verge reports that mathematicians have criticized how OpenAI announces results. Their complaints include weak credit to other researchers, little context and not enough checking. This month alone, OpenAI has released hundreds of results, often with short writeups that researchers call hard to follow.

There is a second complaint pointing the other way. OpenAI recently announced three major breakthroughs in mathematics, physics and computer science. Details of one, involving Yang-Mills theory, have not been made public. Yang-Mills theory is part of the mathematical foundations of modern particle physics. Researchers call this gatekeeping.

• OpenAI says a hand-picked group of experts advised holding the details back.
• That advice runs against what an earlier group of mathematicians told the company.
• OpenAI calls it a new approach to sharing its findings.

The money involved

Altman estimated OpenAI has spent hundreds of millions of dollars on math problems. About $10 million, "maybe a little more," went to its Navier-Stokes result, one of the famous Millennium Prize problems. Some researchers worry that spending at this scale could price them out of their own fields.

What happens next

More claimed results are likely, since Altman says more exist. Outside experts will still have to check them, and that takes time. It is not yet clear when the Yang-Mills details will be released or how OpenAI will improve its communication.