OpenAI Claims Hundreds More Math Discoveries

OpenAI just dropped a data bomb on mathematicians: 377 newly solved problems across fields from algebra to topology, all credited to an advanced, unreleased AI model. The release follows the company's headline-grabbing claim that it cracked the Navier-Stokes equation, one of math's million-dollar "Millennium Problems," and again raises the question of whether the system is discovering ideas or simply finishing what human researchers started, per the New York Times. OpenAI has released summaries of just 10 of the solutions, which took an average of three hours of computing time.

An independent advisory board hosted by the Institute for Advanced Study, which does not support the testing of AI models on advanced math problems, has urged OpenAI to fully share prompts and step-by-step reasoning, calling Tuesday's dump "the beginning, not the completion" of understanding. Even once more details are released, "the full impact of the results will likely take some time to be felt as mathematicians assess and digest them," per the Verge. Some mathematicians worry the company hasn't done enough vetting and that their own work may be feeding the AI. OpenAI counters that testing on cutting-edge problems is necessary to improve its tools.