Altman Says He Gets Why Mathematicians Are Annoyed. OpenAI Is Still Not Slowing Down

Sam Altman has heard the complaints. He's just not changing course.

In an episode of the Cheeky Pint podcast released Thursday, the OpenAI CEO addressed, for one of the first times directly, the frustration mathematicians have voiced over the company's steady drumbeat of AI-generated proofs and results. His position: the disruption is a price worth paying.

"I think, on net, the world deserves to understand how the universe works," Altman said. "But I do have empathy for the method."

That empathy comes after months of what researchers have described to The Verge as a chaotic rollout. OpenAI has been releasing mathematical results at a rapid clip, including hundreds in this month alone, often accompanied by nothing more than brief writeups that mathematicians have called hard to follow and poorly contextualized. The complaints aren't really about whether the results are impressive. Researchers told The Verge the problems are structural: inadequate attribution, little explanation of methodology, and not enough verification before the findings go public. That leaves mathematicians to do the slow, careful work of checking and interpreting OpenAI's output even as the company keeps adding to the pile.

Altman's defense is essentially utilitarian. He argued that the public's right to new scientific knowledge "outweighs" the disruption those releases cause, while conceding that OpenAI needs to get better at how it explains its findings. "I feel unequivocal on this point: Humanity deserves fundamental knowledge, and we should not be sitting on a major fundamental science discovery," he said. "The world deserves that."

A contradiction of its own making

That stance is now colliding with a separate controversy. This week, mathematicians and scientists told The Verge they're frustrated that OpenAI appears to be doing the exact thing Altman says it shouldn't: sitting on major findings. The company recently announced three claimed breakthroughs spanning mathematics, physics, and computer science, but hasn't released full details on one of them, a result tied to Yang-Mills theory, the mathematical framework underpinning modern particle physics.

OpenAI says the delay comes on the advice of a hand-picked group of experts it consulted about how to best release the work. That explanation hasn't landed well, partly because it runs counter to guidance from a previous group of mathematicians OpenAI had consulted earlier. The company frames the shift as part of a broader, evolving approach to how it communicates results. Critics see it as inconsistent: urgency and transparency when it's about volume, caution and gatekeeping when a result might be especially significant.

More proofs than anyone's seen

Perhaps the more striking revelation from the podcast is how much OpenAI hasn't shown anyone yet. Asked whether the company has more mathematical discoveries beyond the hundreds already released, Altman didn't hesitate: "We do. Of course."

He also gave some color on how this all started. Altman acknowledged that OpenAI's push into mathematical problem-solving "did get out of control" for a stretch, but said that was intentional, a way to stress-test how far its models could go. He described an internal Slack channel employees nicknamed "Model Name-Hold My Beer," where staff threw the toughest problems they could find at the models just to see what would happen.

That internal experiment has had a real price tag. Altman estimated OpenAI has spent somewhere in the hundreds of millions of dollars pursuing mathematical problems overall, including roughly $10 million, "maybe a little more," on a single result: a solution tied to the Navier-Stokes equations, one of mathematics' seven legendary Millennium Prize problems.

The money matters as much as the proofs

The spending figures say as much about the current moment as the math itself. They show just how much capital AI labs are now willing to pour into scientific problem-solving, well beyond what academic mathematics departments can typically muster. That imbalance is exactly what worries some researchers, who fear a well-funded AI lab chasing famous open problems could effectively price them out of their own fields, or at least reshape who gets credit and how discoveries get verified and shared.

For now, mathematicians hoping OpenAI might pump the brakes, slow the release cadence, or clean up its communication before dropping more results aren't getting that. Altman's message is that the output will keep coming, the company will try to do better at explaining it, and there's more in the vault than anyone outside OpenAI has seen. Whether that approach builds trust with the research community or deepens the current friction looks like the thing to watch next.