In August, OpenAI convened around 40 mathematicians to discuss the implications of AI potentially surpassing human capabilities in mathematics. Attendees reported that OpenAI suggested its models had solved numerous long-standing math problems, but company representatives assured them that solutions would not be released all at once. OpenAI spokesperson Lindsay McCallum stated that the company was unaware of such an assurance. The meeting elicited mixed reactions, with some attendees expressing excitement while others felt apprehensive. Mathematician Bryna Kra from Northwestern University emphasized the need for OpenAI to publish detailed papers rather than brief blog posts, as had been done previously.
According to sources familiar with the plans, OpenAI intends to release hundreds of solutions to unsolved problems on GitHub. McCallum noted that a new internal model had resolved over 100 long-standing open problems, including the Navier-Stokes Millennium Prize problem. The company is seeking to responsibly release these results, guided by advice from the Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study.
However, some mathematicians feel that OpenAI has not learned from past controversies regarding its releases. They argue that the company and its competitor Anthropic are prioritizing their public image over traditional scientific processes for releasing and attributing results. Mathematician Tristan Buckmaster accused OpenAI of preemptively solving a Millennium Prize problem based on rumors of others nearing a solution, which he claims undermined his collaborative work with an Anthropic employee.
In discussions with WIRED, Buckmaster described a negotiation over credit that he felt was mishandled by OpenAI researcher Sébastien Bubeck. Bubeck reportedly suggested that the Anthropic employee should not be included as a co-author, a claim he later denied. Buckmaster expressed concern over the implications of OpenAI's actions on his career.
Nestor Guillen, a visiting math professor at NYU, remarked on the perception of aggressive behavior from AI companies towards mathematicians. McCallum responded to this characterization by stating that the company disagrees with it. Guillen noted that many mathematicians are concerned about the concentration of power within AI companies.
OpenAI formed an advisory group in September to guide its assessment and communication of new results, but many mathematicians believe progress has been insufficient. They are frustrated that results continue to be released via blog posts rather than scientific papers, which complicates verification and often overlooks prior work. Kra criticized the trend of announcing mathematical results through social media, arguing it does not support the academic ecosystem.
In response to the rise of machine-assisted proofs, some mathematicians have developed new tools to help the community navigate AI-generated results. Despite encouragement for OpenAI to utilize these tools, Kra noted that the company has not changed its approach. She also expressed concern that OpenAI's actions do not align with the Leiden declaration, which calls for AI companies to adhere to standards set by mathematicians.
Some OpenAI employees reportedly believe that their technology has rendered traditional mathematics obsolete, although McCallum stated that the company is committed to collaborating with the math community. The initial meeting in August framed the potential for AI to outpace human researchers as a hypothetical scenario, but attendees interpreted it as a cautionary signal.
Bubeck expressed optimism that advanced AI could help mathematicians tackle more ambitious questions and enhance the accessibility of mathematics. Mathematicians acknowledge the need to adapt to these changes but maintain that their field is not finished. Kra highlighted the importance of improved disclosures from AI companies to foster trust and collaboration in the mathematical community. She described the current period as both challenging and exciting.