Experts Flag Notable Results in OpenAI’s Massive Proof Release
OpenAI released over 700 manuscripts addressing 372 open problems in mathematics, theoretical computer science, physics and related fields. Experts highlighted several results as potentially major, though not all proofs passed Lean verification and some were retracted after errors were found. OpenAI also omitted full prompts and runtime details, but researchers still described the overall batch as remarkable.
OpenAI’s Tuesday release contained over 700 manuscripts on 372 open problems, spanning mathematics, theoretical computer science, physics, and related fields. Experts highlighted two Riemann-zeta results: one establishes a “quasi-Riemann hypothesis,” and another rules out Siegel zeros for the zeta function and many Dirichlet L-functions. Hilbert’s 10th problem also appeared among the featured math topics.
Not every proof passed Lean verification; some were retracted after outside scientists found errors. Complete prompts and runtime information were not provided, and even OpenAI researchers had not read everything. Still, specialists called the overall batch remarkable.
If these results hold up, mathematicians, computer scientists, and physicists could gain new tools and shortcuts for longstanding questions. Research groups may need to adapt to AI-generated proofs by strengthening verification, replication, and review practices. Students and early-career researchers might face a changing landscape of what skills matter. At the same time, retractions and missing methodological details could temper trust, making clear documentation and independent checking more important. The broader public may benefit indirectly through faster scientific progress, but only if experts can reliably separate durable advances from errors.