OpenAI's math breakthroughs fuel growing friction with researchers
OpenAI has announced a string of mathematical breakthroughs over the past year, including a solution to one of the Millennium Prize problems, but the way these results were handled has angered many mathematicians. A new advisory panel meant to fix the relationship is already drawing fresh criticism.
Over the past year, major AI labs including OpenAI and Anthropic have reported striking progress on long-standing mathematical problems. The most notable case is OpenAI's claimed solution to the Navier-Stokes problem, one of the seven Millennium Prize Problems, which had gone unsolved for roughly 90 years and carries a $1 million reward. The company said the breakthrough came from an internal AI model running 10,000 concurrent agents.
While such results might normally be celebrated, they have instead triggered backlash within the mathematics community. Many researchers say OpenAI is charging into problems others have spent careers studying, with little regard for established academic norms. Accusations have surfaced involving the use of other researchers' unpublished work, rushed announcements, and a lack of transparency about training data.
Disputes over sources and ethics
Several mathematicians, including Andreas Thom, have publicly raised concerns that their own interactions with OpenAI's chatbot may have indirectly contributed to the company's results in specialized areas such as non-sofic groups. OpenAI has acknowledged that some of its announced results built heavily on prior work by outside researchers.
A new advisory group
In response to the criticism, OpenAI this week announced an independent panel of elite mathematicians to advise it and other AI companies on how to responsibly handle and release new mathematical findings. But mathematicians who spoke to The Verge, including a panel member, described the rollout as confusing, with little clarity on the group's actual authority or whether OpenAI will heed its advice. The panel's first task is reportedly helping coordinate the release of dozens more unpublished results attributed to an unreleased OpenAI model, a prospect already causing unease among researchers about the potential impact on their field.
