New Math from OpenAI

Substack · Watch/Read · (gathered 2026-10-14)

• OpenAI released 722 new mathematical results, including solutions to 90 of the top 500 open problems, marking a “Mathocalypse” that is arguably the biggest scientific advancement in history, yet largely unacknowledged by mainstream media.
• Key breakthroughs include tightening bounds on the Quasi-Riemann Hypothesis, improving matrix multiplication efficiency to 2.25, and refuting the 3SUM and APSP hypotheses (the latter from Anthropic’s more collaborative release).
• Mathematicians expressed profound mixed feelings, from excitement over new ideas to a sense of loss, disruption of careers, and anger at OpenAI’s “dump it all on GitHub” approach, which disregarded advisory group recommendations for responsible release.
• The rapid AI progress poses significant risks to cryptography, with experts like Vitalik Buterin recommending multiplying key sizes for lattice-based systems and favoring hash-based methods.
• The event underscores a paradigm shift where human mathematicians must adapt, find new problems, and develop new methods of understanding and collaborating with AI, as traditional research and training models are being fundamentally disrupted.