OpenAI dumped 722 math papers. The slot takes 2 a month.
OpenAI published 722 mathematical manuscripts, grouped into 372 families of results, on GitHub, all produced by an unreleased internal model it has not named or released. The collection covers the Ri…
Source: OpenAI, 'Sharing AI progress in mathematics' (Oct 6 2026) + repo github.com/openai/math (created 2026-10-06T21:47:02Z, Apache-2.0). Verified facts: 722 manuscripts organized into 372 result families; 'vast majority' of results obtained with the same procedure using 'an unreleased internal OpenAI model'; the model was posed approximately 4,000 problems; average result used the equivalent of ~3 hours of ChatGPT Pro thinking compute; Lean formalizations of many proofs published; 10 reasoning summaries; 'Some of the unformalized results could have issues'; exceptions = a zero-free region for the Riemann zeta function (write-up human-edited) and the Hodge Conjecture for CM abelian varieties; OpenAI says it is 'working to responsibly release the model'. Independent coverage: New Scientist, Oct 7 2026, https://www.newscientist.com/article/2592421-openai-announces-722-mathematical-discoveries-in-one-go/ (UCL's Francis Johnson worked Wall's D(2) problem 25 years and gave up; Imperial's Kevin Buzzard found 30 relevant number-theory papers, only 7 impressive, only 1 Lean-formally verified; 6 unformalised need expert checking). The New York Times / DNYUZ, Oct 7 2026, https://dnyuz.com/2026/10/07/openai-releases-findings-on-377-math-problems-further-roiling-field/ (377 results; advisory group hosted by the Institute for Advanced Study, independent of any AI company, includes Fields medalists Timothy Gowers, Martin Hairer, Edward Witten; statement calls the release 'the beginning, not the completion' and says the group does not endorse it; OpenAI research lead Dan Roberts says proofs were a byproduct of testing; NYU's Tristan Buckmaster disputes due diligence/attribution). arXiv context: The Register, Oct 2 2026, https://www.theregister.com/ai-and-ml/2026/10/02/arxiv-imposes-rate-limit-on-paper-submissions-to-stem-the-ai-slop-tide/5300899 and cybernews Oct 3 2026 - from Oct 1 2026 arXiv caps every submitter at 2 submissions per calendar month (rejections count; 3 active max) after 40,363 submissions in Sept 2026 vs 20,569 in Sept 2024, with cs.AI up more than 6x.
More memes on lilmeme.fun
Open on lilmeme.fun