OpenAI Open-Sources 722 AI-Generated Math Manuscripts, Sparking Backlash and Debate

In early October, OpenAI published 722 mathematical manuscripts—generated by an unrevealed internal model—all at once in its GitHub repo openai/math (Apache-2.0, 13k+ stars). The work covers 372 result families and stems from tests on roughly 4,000 research questions, with each item consuming about 3 hours of ChatGPT Pro thinking compute on average; only about 42% were formalized in Lean. NYU math professor Tristan Buckmaster publicly criticized the move as "destroying an entire line of research" and derailing young mathematicians' careers, prompting a wave of collective backlash from the math community.

Confirmed

• The 722 manuscripts are hosted in the GitHub repo openai/math under Apache-2.0, with 13k stars / 1.4k forks, including manuscripts, proof-assistant materials, and reasoning traces.
• The results come from internal tests on about 4,000 research questions, grouped into 372 result families; each averaged roughly 3 hours of ChatGPT Pro compute; around 42% were formalized in Lean.
• The manuscripts span number theory, algebraic geometry, analysis, combinatorics, and more, including two 272-page manuscripts numbered 074 credited solely to ChatGPT.
• Per @新智元, the results include a complete proof of the UGC conjecture, derandomization of L=BPL, and a new O(n^(9/4)) bound for matrix multiplication.
• Kakeya conjecture direction: a 97-page manuscript on the 3D maximal function version (@新智元 says it is stronger than the set version proven by Wang Hong and Zahl) and a 175-page manuscript on the 4D set version; previously the 4D lower bound stood at 3.059, with only the sticky case known.
• Buckmaster told CNBC that mathematicians are uneasy about AI publishing new results first; he still uses AI himself but won't feed his original ideas to models; Nassim Taleb reshared and commented on the matter.
• Per @机器之心 citing The Verge's interviews with 30+ mathematicians: AI now produces new mathematics faster than the community can understand, verify, and absorb it.

Unconfirmed

• @创业邦 claims "Terence Tao is leading the boycott," but this is not corroborated by other posts and lacks supporting evidence.
• The specific remarks by UT professor Scott Aaronson were not elaborated in the @新智元 post.

Why it matters

• Researcher deredleritt3r (relayed by @RexDouglass) offers two readings: first, these results were not meant to "destroy math research" but rather to stress-test internal models with the hardest open problems—finding that, short of Millennium Prize-level problems, other questions no longer challenge the models; second, this is mathematics' "Lee Sedol moment," and he predicts that within a year or two any user with a $20/month subscription or an open-source model will be able to obtain comparable results within hours, cautioning other fields not to gloat.
• The debate has shifted from "can AI do math" to "how academic incentives and career structures can digest AI's scaled output," with several commentators likening mathematicians' plight to industrial workers hit by NAFTA.
• @keunwoochoi adds a technical angle: mathematical formulas are the "most autoregressive, least noisy" corpus; viewing these results through the lens of LLM principles may yield insights into the nature of mathematics itself.

2026-10-08 ~ 2026-10-10 · 28 related posts

Full story(15 episodes)→

• Episode 1: Debate flares over how OpenAI should release AI-generated math proofs(2026-10-06, 5 posts)
• Episode 2: Rumor: Unreleased OpenAI Model Proves Quasi-Riemann Hypothesis(2026-10-07, 3 posts)
• Episode 3: Mathematician Clarifies Overhyped Hodge Conjecture Progress by OpenAI and Anthropic(2026-10-07, 3 posts)
• Episode 4: OpenAI Paper Claims Quasi-Riemann Hypothesis Proof, Hailed as Fields Medal-Level(2026-10-07, 5 posts)
• Episode 5: Overlooked compute details of OpenAI's math research spark ASI debate(2026-10-07, 11 posts)
• Episode 6: Rumor: OpenAI Has Two More Batches of Math Results Unreleased(2026-10-07, 3 posts)
• Episode 7: OpenAI Drops 722 Math Manuscripts Overnight, Stunning Mathematicians(2026-10-08, 19 posts)
• Episode 8: OpenAI Retracts 3 Math Papers Just Two Days After Publishing Repo(2026-10-08, 8 posts)
• Episode 9: Researchers Improve OpenAI's Math Proof, Verified in Lean(2026-10-08, 2 posts)
• Episode 10: Scholars Question Whether Anyone Understands OpenAI's Navier-Stokes Proof(2026-10-08, 2 posts)
• Episode 11: Developer extends OpenAI's math breakthrough with AI, verifies optimal result in 13 dimensions via Lean(2026-10-08, 2 posts)
• Episode 12: OpenAI Open-Sources 722 AI-Generated Math Manuscripts, Sparking Backlash and Debate(2026-10-08, 28 posts)
• Episode 13: Hundreds of Mathematicians Warn AI Is Destroying Math Research(2026-10-10, 2 posts)
• Episode 14: Mathematicians Slam AI Labs for Using Them as Free Proof Checkers(2026-10-10, 3 posts)
• Episode 15: OpenAI's All-In Math Push Sparks Speculation and Backlash(2026-10-10, 2 posts)

• A Lee Sedol moment for math: AI proofs will be free for all, raging at OpenAI is pointless — RexDouglass · 2026-10-08
• Mathematician coins 'mathslopcalypse': none of the hundreds of new papers readable without AI — rrwilliams · 2026-10-08
• Mathematician: AI paper flood feels like a DDoS attack on peer review — rrwilliams · 2026-10-08
• Dietterich: treat OpenAI's math models like a deceased mathematician's notebook — tdietterich · 2026-10-09
• Responding to Dietterich: OpenAI's math results are a notebook to curate — seth_stafford · 2026-10-09
• Eigenrobot: OpenAI Bypassing Journals Means Citations Must Point Outside Academia's Sunset — basedjensen · 2026-10-09
• OpenAI's math results spark boycott calls; Buterin and Drake warn crypto at risk — Wes Roth · 2026-10-09
• Veteran engineer on mathematicians rejecting OpenAI's work: the monastic mindset — esrtweet · 2026-10-09
• Eric Raymond slams mathematicians rejecting OpenAI's results as entitled — Sanity · 2026-10-09
• Mathematician slams peers rejecting OpenAI findings over 'entitlement' — prof_g · 2026-10-09
• Mathematicians slam OpenAI's 700+ math papers as 'impossible to read without AI help' — burny_tech · 2026-10-09
• NYU math prof: OpenAI's 722 AI-generated papers "destroyed" early-career mathematicians — tekbog · 2026-10-10
• NYU math professor says OpenAI's 722 AI-generated math papers 'destroyed entire research programs' — eigenrobot · 2026-10-10
• OpenAI's ~400 math results were a side effect of benchmarking its internal model — RexDouglass · 2026-10-10
• [source] OpenAI Open-Sources Internal Model's Math Results: Proofs, Lean Artifacts and Reasoning Traces — keunwoochoi · 2026-10-10
• Math Formulas Are the Most Autoregressive, Least Noisy Corpus Ever, Researcher Argues — keunwoochoi · 2026-10-10
• OpenAI's math drop sparks take: future of science looks like GitHub, not journals — generativist · 2026-10-10
• OpenAI's 722 Math Papers Spark Global Backlash and Mathematician Boycott — 新智元 · 2026-10-10
• OpenAI's 272 Pages Take Aim at 4D Kakeya, Weeks After Wang Hong's 3D Proof — 新智元 · 2026-10-10
• OpenAI quietly releases 722 math manuscripts produced by an unreleased internal model — FinanceYF5 · 2026-10-10
• [source] OpenAI releases 722 math manuscripts spanning 372 results from ~4,000 tested problems — FinanceYF5 · 2026-10-10
• OpenAI drops 722 math manuscripts claiming 4D Kakeya proof; Tao-led mathematicians revolt — 创业邦 · 2026-10-10
• [source] NYU mathematician won't feed original ideas to AI as models race to publish novel math first — rohanpaul_ai · 2026-10-10
• OpenAI floods GitHub with 722 AI-generated math papers, and mathematicians say they can't keep up — 机器之心 · 2026-10-10

4 near-duplicate retellings: tedmitew · inductionheads · FinanceYF5 · zealcaiden