OpenAI plans to keep publishing AI-generated mathematical results, even as researchers say the pace and presentation of those releases make them difficult to assess. CEO Sam Altman says the public’s interest in new knowledge outweighs the disruption, while acknowledging that the company needs to communicate its findings better.
Why Altman says the work should be public
Speaking on an episode of the Cheeky Pint podcast released Thursday, Altman said he sympathizes with mathematicians trying to make sense of the results. He nevertheless argued that discoveries about how the universe works should not be withheld.
Altman described access to fundamental science as a matter of public interest. In his view, that interest outweighs the difficulties researchers face when a large set of results arrives at once. He also conceded that OpenAI’s communication around its findings needs improvement.
That position leaves two responsibilities in tension: making scientific knowledge available and giving other researchers enough information to evaluate it. A result can be public in the narrow sense that it has been announced, while still being hard for others to verify or use if the explanation and supporting context are limited.
Researchers say the releases are hard to assess
Mathematicians who spoke to The Verge have criticized OpenAI’s announcements for offering inadequate attribution, little context and insufficient verification. They say researchers are left to check the work and interpret what it means, even as more results arrive.
This month alone, OpenAI released hundreds of mathematical results. The accompanying writeups were often brief, and researchers described them as poorly written and hard to follow. The volume and presentation can make it harder to distinguish what has been established from what still needs scrutiny.
That creates a practical burden for the people trying to evaluate the claims. They must work through the material, understand how a result was reached and determine whether it holds up. The source reports both excitement about AI’s potential for discovery and frustration over the way these results have been shared.
Publication and withholding are both under criticism
OpenAI is also facing criticism over findings it has not made public. The company announced three major breakthroughs in mathematics, physics and computer science, but details of one claimed result involving Yang-Mills theory have not been released publicly.
OpenAI says that decision followed recommendations from a hand-picked group of experts it consulted about the release. The approach differs from advice provided by an earlier group of mathematicians consulted by the company. OpenAI describes its process as a new way to communicate findings.
The two complaints point to a difficult question for research announcements: how much information should be shared, and when? Researchers cited in the source criticize both releases that are difficult to check and gatekeeping that leaves important details unavailable. Altman’s argument for openness, meanwhile, raises expectations that claimed discoveries will eventually be accessible to the wider research community.
More results and major costs are in the picture
Altman indicated that the published results are only part of OpenAI’s mathematical work. Asked whether the company had more discoveries beyond the hundreds of proofs already released, he answered, “We do. Of course.” The statement suggests that researchers may continue receiving new results rather than getting a pause to assess the existing ones.
He also said OpenAI’s mathematical work briefly “get out of control,” describing an internal Slack channel called “Model Name-Hold My Beer.” Employees used it to give the models some of the hardest problems they could find, as a way to test their limits.
Altman estimated that OpenAI has spent in the order of hundreds of millions of dollars on mathematical problems. Around $10 million — “maybe a little more” — went toward its Navier-Stokes solution, one of mathematics’ legendary Millennium Prize problems. The spending illustrates the scale of resources the company is directing toward these efforts.
Researchers have raised concerns that such spending could price them out of their own fields. The source does not settle that concern, but it shows why OpenAI’s work is drawing attention beyond the proofs themselves: the pace of AI research, the resources behind it and the way findings are shared all affect how other mathematicians can respond.
For now, Altman’s comments offer little prospect of a slowdown. He says the discoveries should not be kept from the public, and OpenAI has more results to share. Whether its approach can meet that goal while giving researchers the context and verification they say they need remains unresolved.