--- title: OpenAI's math dump has mathematicians saying 'not like that' description: Mathematicians are pushing back on OpenAI's release of over 100 math results to GitHub, after months of friction over how the company has publicized unverified... canonical: https://daily.dev/posts/openai-s-math-dump-has-mathematicians-saying-not-like-that--lthfo5owb twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: OpenAI's math dump has mathematicians saying 'not like that' | daily.dev og:description: Mathematicians are pushing back on OpenAI's release of over 100 math results to GitHub, after months of friction over how the company has publicized unverified... og:url: https://daily.dev/posts/openai-s-math-dump-has-mathematicians-saying-not-like-that--lthfo5owb og:image: https://api.daily.dev/og/posts/LTHFo5OWB.png og:image:alt: OpenAI's math dump has mathematicians saying 'not like that' og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # OpenAI's math dump has mathematicians saying 'not like that' **[Trends](https://daily.dev/sources/trends)** · 3 min read · 2 upvotes · 0 comments ## Summary Mathematicians are pushing back on OpenAI's release of over 100 math results to GitHub, after months of friction over how the company has publicized unverified claims of solving open problems. Fields Medalists Peter Scholze and Geordie Williamson criticized OpenAI's conduct at the Heidelberg Laureate Forum as coercive and PR-driven, following an earlier controversy where OpenAI claimed a Navier-Stokes Millennium Prize solution and 25 Fields medalists signed a letter objecting to labs treating open problems as benchmarks. OpenAI says it consulted an independent advisory group at the Institute for Advanced Study and is releasing results from an internal frontier model, but critics say dumping hundreds of unreviewed results on the math community shifts the burden of verification onto researchers for free. Anthropic and Google are pursuing similar automated math reasoning efforts, but OpenAI is drawing the most criticism. ## Content OpenAI's answer to "publish real proofs, not blog posts" was 722 manuscripts at once. The mathematicians are not thrilled with the delivery. ## The fight The Association for Human Mathematics criticised the release and urged mathematicians "to discontinue their work with OpenAI." Terence Tao reposted the statement on his blog. @rohanpaul_ai's reply: "Mathematicians are coping so hard," adding that "walking away is not a scientific position." The skeptics are more specific than that. Tristan Buckmaster and Andrew Sutherland question whether OpenAI properly vetted the results and say the claims should count as unverified until someone independent reproduces them. Many proofs ship with Lean formalizations for machine checking. Some don't, and those could contain errors. The history matters here. @Hesamation walks through the sequence: OpenAI meets 40 mathematicians in August, hints at hundreds of solved problems, gets asked for real papers, then publishes a Navier-Stokes Millennium Prize claim. Twenty-five Fields medalists sign an open letter telling labs to stop using open problems as model benchmarks. OpenAI responds with an advisory group. At the Heidelberg Laureate Forum, Peter Scholze and Geordie Williamson called OpenAI's conduct around the announcement coercive and PR-driven. Wired then reported the plan to drop hundreds of results on GitHub, and @Hesamation's summary of the mood was "not like that." ## The release itself OpenAI says it consulted the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study. That group, AGMAI, says this is "the beginning, not the completion" of human understanding and stresses it is not an endorsement. The numbers: 722 manuscripts grouped into 372 result families, pulled from roughly 4,000 problems posed to an unreleased internal model. Each result used about three hours of ChatGPT Pro thinking compute on average. The headline result, per @rohanpaul_ai, is a matrix multiplication exponent of about n^2.25, down from the n^2.371177 that AlphaEvolve set in August. It comes with a Lean formalization. It is also a theoretical bound, with no faster routine for real GPU workloads. ## The believers @kimmonismus called it history: "History was made yesterday. I say that without exaggeration." @emollick points to early first-hand accounts of mathematicians meeting "a narrow superhuman intelligence," with problems solved in inhuman ways that make people wonder what it means to actually know something. @rohanpaul_ai predicts math will now move at breakneck speed. There's also a quieter worry underneath. Researchers described pressure to pay for LLM subscriptions just to keep pace with peers, and unease about what this does to hiring and education. My read: the proofs may well hold up, and Lean helps. But a model nobody outside OpenAI can run, a rollout that irritated the community twice, and a pile of papers too big to referee quickly is not a recipe for trust. ## Questions this post answers ### Why are mathematicians upset about OpenAI's release of math results on GitHub? Mathematicians object to OpenAI dumping over 100 unreviewed results on GitHub without accompanying proofs, shifting the burden of verification onto the research community for free. Fields Medalists Peter Scholze and Geordie Williamson called the rollout coercive and PR-driven at the Heidelberg Laureate Forum, following an earlier unverified claim of a Navier-Stokes Millennium Prize solution and a letter from 25 Fields medalists opposing treating open problems as AI benchmarks. _Developers tracking how AI labs engage with scientific research can follow this debate on daily.dev._ ### What did OpenAI claim about solving open math problems before this GitHub release? OpenAI earlier claimed its model had resolved more than 100 long-standing open problems, including a purported Navier-Stokes Millennium Prize solution, without publishing specifics or proofs. That announcement triggered an open letter signed by 25 Fields medalists urging AI labs to stop treating unsolved mathematical problems as benchmarks for model capability. _Those evaluating AI claims about scientific breakthroughs can follow the back-and-forth on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 3 discussions and 101 comments across x (as of 2026-10-08). **TL;DR:** Reactions split between excitement that AI can produce formally verified, superhuman math results and unease that the proofs are often unreadable, unreviewed, and may be outpacing the community's capacity to check them; others frame the backlash as professional guild defensiveness akin to earlier anti-AI reactions in software. **Sentiment:** 20% positive · 45% mixed · 35% skeptical **The case for** - Lean-formalized proofs can be mechanically verified as correct even if humans can't follow the reasoning. - Some see this as an opportunity to spend less time grinding proofs and more time understanding results or posing new problems. - A few argue refusing to engage just because a model authored the work is guild gatekeeping, and results should be judged on their merits. **The pushback** - Most of the 722 results lack Lean verification, so skepticism about correctness is warranted. - There aren't enough qualified people per subfield to check hundreds of AI-generated proofs at once, creating a reading/verification bottleneck. - A valid proof that nobody understands doesn't transfer insight the way traditional proofs did, separating 'knowing it's true' from 'knowing why'. - Some compare this to the earlier flood of AI-generated 'slop' that strained open-source code review, worrying the same dynamic will hit math, physics, biology, and other fields. - Concern that 'superhuman' framing is partly marketing to sell subscriptions rather than a rigorously earned claim. **By community** - x (mixed): A wide, high-volume discussion torn between awe at formally-checked superhuman results and anxiety that proofs are unreadable, unverified, or that the backlash is just guild self-protection. **Hottest debate:** Whether a machine-checked proof that no human can intuitively understand actually counts as 'knowing' the math, or is just an unverifiable claim dressed up in formalism. **Open questions** - Can mathematicians extract reusable intuition or techniques from proofs they can verify but not understand? - How many of the unformalized results will hold up once independently reviewed? - Will the shortage of qualified reviewers per subfield become a structural bottleneck as more AI-generated proofs arrive? **Highlights** > @emollick Some context: Quanta ran the math community's reaction — 'if we don't adapt, there's just no more math in 50 years.' Ken Ono (now at Axiom Math) told a packed Berkeley hall to 'brace.' The 'dethroned' line is Aaronson's own. The grief is the real story, not the proofs. > — [Lin143382 on x](https://x.com/Lin143382/status/2107969354276954507) > @emollick The weird bit is a proof can be checked and still not teach anyone anything. Mathematicians used to get the insight for free with the result. Now they might get the result and have to dig the insight out after. > — [chnhn05782803 on x](https://x.com/chnhn05782803/status/2107993988171284782) > @emollick First-hand accounts are the interesting part. The split that keeps showing up: mathematicians can verify these proofs but can't reconstruct the intuition behind them — the AI took a path no human would. So 'knowing' now means two different things: verification and understanding. > — [dasearth666 on x](https://x.com/dasearth666/status/2108015373924565075) > @rohanpaul_ai I would guess that the reasoning behind the reaction from the math community is the same as what’s seen by OSS communities. OSS is being flooded now by AI slop, with just a handful of reviewers being expected to review code that a human may never have analyzed. For math OAI…1/N > — [jaredyu999 on x · 3 points, 2 comments](https://x.com/jaredyu999/status/2108029025943945587) > @Thyag_a @emollick So if a spaceship landed and the aliens shared their advanced math books, and nobody could understand it. We would tell them their math was just claims? Even if Lean verified? That's certainly one approach. > — [RobertDobalina7 on x · 4 points, 1 comments](https://x.com/RobertDobalina7/status/2107986480392257923) **Source threads** - [x](https://x.com/rohanpaul_ai/status/2108021673706615292) · 0 points · 27 comments - [x](https://x.com/rohanpaul_ai/status/2108009559822586289) · 0 points · 2 comments - [x](https://x.com/emollick/status/2107964764316209326) · 0 points · 72 comments ## Similar posts on daily.dev - [Ten advances in mathematics and theoretical computer science](https://daily.dev/posts/ten-advances-in-mathematics-and-theoretical-computer-science-ngbzjedvn) · OpenAI · 1 upvotes · 0 comments - [OpenAI Publishes 10 AI-Generated Mathematical Breakthroughs Using Its Internal Model Astra](https://daily.dev/posts/openai-publishes-10-ai-generated-mathematical-breakthroughs-using-its-internal-model-astra-v6xjxofws) · Medium · 1 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#openai](https://daily.dev/tags/openai) [View this post on daily.dev](https://daily.dev/posts/openai-s-math-dump-has-mathematicians-saying-not-like-that--lthfo5owb) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"OpenAI's math dump has mathematicians saying 'not like that'","url":"https://daily.dev/posts/openai-s-math-dump-has-mathematicians-saying-not-like-that--lthfo5owb","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/openai-s-math-dump-has-mathematicians-saying-not-like-that--lthfo5owb"},"datePublished":"2026-10-06T22:20:11.476Z","dateModified":"2026-10-08T03:59:41.762Z","description":"Mathematicians are pushing back on OpenAI's release of over 100 math results to GitHub, after months of friction over how the company has publicized unverified...","isAccessibleForFree":true,"articleSection":"Trends","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/openai-s-math-dump-has-mathematicians-saying-not-like-that--lthfo5owb","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,openai","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"OpenAI's math dump has mathematicians saying 'not like that'"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/openai-s-math-dump-has-mathematicians-saying-not-like-that--lthfo5owb#faq","mainEntity":[{"@type":"Question","name":"Why are mathematicians upset about OpenAI's release of math results on GitHub?","acceptedAnswer":{"@type":"Answer","text":"Mathematicians object to OpenAI dumping over 100 unreviewed results on GitHub without accompanying proofs, shifting the burden of verification onto the research community for free. Fields Medalists Peter Scholze and Geordie Williamson called the rollout coercive and PR-driven at the Heidelberg Laureate Forum, following an earlier unverified claim of a Navier-Stokes Millennium Prize solution and a letter from 25 Fields medalists opposing treating open problems as AI benchmarks. Developers tracking how AI labs engage with scientific research can follow this debate on daily.dev."}},{"@type":"Question","name":"What did OpenAI claim about solving open math problems before this GitHub release?","acceptedAnswer":{"@type":"Answer","text":"OpenAI earlier claimed its model had resolved more than 100 long-standing open problems, including a purported Navier-Stokes Millennium Prize solution, without publishing specifics or proofs. That announcement triggered an open letter signed by 25 Fields medalists urging AI labs to stop treating unsolved mathematical problems as benchmarks for model capability. Those evaluating AI claims about scientific breakthroughs can follow the back-and-forth on daily.dev."}}]} ```