OpenAI Releases AI Solutions to More Than 370 Math Problems
OpenAI has published full or partial AI-generated solutions to more than 370 mathematical problems, including work on major challenges. The release includes…
OpenAI published full or partial AI-generated solutions to more than 370 mathematical problems on Tuesday, including work on challenges long regarded as major tests for the field. The OpenAI mathematics release contains 722 manuscripts produced with an unreleased internal model, and the scale of the claims has divided researchers.
Some mathematicians see new avenues for research. Others question how the company produced and presented the results, and whether the work can be checked independently. The papers include model reasoning, compute estimates and information about how many problems were attempted.
What OpenAI released
OpenAI said the model took about three hours of computing time per solution on average. Engadget reported that nearly every paper came from a response to a single prompt given to a single AI agent, citing OpenAI’s comments to Scientific American.
The company posted the manuscripts to a GitHub repository and said it followed guidance from its independent Advisory Group on Mathematics and Artificial Intelligence. That guidance included releasing results through traditional academic channels and providing information such as the model’s name, prompts and compute costs.
Some details were not included. Engadget reported that OpenAI did not publish the specific compute times for individual problems or the prompts it used. The company said it would continue exploring community-hosted alternatives and work on citations, mathematical exposition and presentation in future releases.
The results include a claimed solution to the four-dimensional Kakeya conjecture, improvements to computer algorithms and progress toward the Riemann hypothesis. OpenAI also reported progress on three other Millennium Prize problems, but did not say it had fully solved them.
Claims still need scrutiny
The findings follow OpenAI’s earlier claim that an internal model and thousands of AI agents had solved a problem involving the Navier-Stokes equations. That announcement prompted criticism from mathematicians who said they had also been working on the problem with AI tools, including OpenAI’s Codex.
OpenAI denied using those researchers’ work, saying it had not seen it before it was released publicly. The dispute has added to skepticism about the company’s latest claims. The new manuscripts, too, need assessment by mathematicians before their impact can be established.
That verification matters beyond the papers themselves. AI companies have used difficult mathematical problems to demonstrate what their models can do, and researchers have suggested that mathematical training could strengthen skills such as logical reasoning and persistence. Fortune reported that it remains unclear how those capabilities might transfer to areas where answers cannot be checked against a definitive mathematical proof.
There are also concerns about how the results could affect the profession. Dan Litt, a mathematics professor at the University of Toronto, welcomed the work but warned that claims suggesting AI had “solved math” could lead funding organizations to reduce support for research or discourage young mathematicians. “It’s important that society reaffirms support for human mathematical expertise if we want to get anything out of the progress on these problems that AI has made,” he told Fortune.
OpenAI’s release puts a large body of machine-generated mathematics before researchers, but the manuscripts have not yet settled what the results mean for the discipline. The collection includes 722 papers.
Source: bbc.com
