OpenAI's latest move in mathematics has triggered an uproar. According to Tom's Hardware, the company last week released 722 manuscripts and proofs, including more than 400 results for formulas that have resisted solution for decades. Researchers now warn that verifying and proving all of this AI-generated work could take years.
Critics are also cautioning mathematicians about sharing their own unpublished research with large language models. The concern is that feeding such material into LLMs may lead to premature or unverified dissemination, and that the field could be overwhelmed by raw outputs rather than supported by careful, community-driven verification.
One professor cited in the report argues that mathematics should focus on building community over raw solutions. The episode highlights a growing tension: AI can generate results at scale, but the discipline's existing validation processes are not built for such volume.