New Math from OpenAI (thezvi.substack.com)

OpenAI has made a groundbreaking announcement by releasing a set of 722 mathematical results on GitHub, derived from an internal frontier model that tackled approximately 4,000 math problems. This collection notably includes solutions or insights into 90 of the top 500 open problems in mathematics, a feat deemed monumental by experts in the field. Among these results, significant advancements such as a tighter bound related to the Riemann Hypothesis, a reduction in the exponent of matrix multiplication from 2.37 to 2.25, and startling findings on integer multiplication and graph coloring have been highlighted. These achievements have immediate implications for various fields ranging from quantum computing to algorithm development, indicating a potential paradigm shift in how mathematical problems are approached and solved. The implications of this release extend beyond just mathematics; it signals a transformative moment in AI's role in scientific discovery. As the AI-generated proofs are subject to verification, mathematicians are now reflecting on the future of research and problem-solving in the domain. While some scholars express skepticism about the authenticity and significance of AI contributions to math, others argue that this breakthrough marks an epoch in computational mathematics. With many results bordering on some of the most profound problems in the field, October 6, 2026, could be remembered as a pivotal day—one that reshapes our understanding of creativity and intelligence in mathematics, challenging the long-held belief that these pursuits are exclusively human domains.