AI
“On October 6, OpenAI published more than 700 papers presenting mathematical research results produced by one of its internal AI models, sparking outrage among mathematicians." (French original on the slide: "Le 06 Octobre OpenAI a publié plus de 700 articles présentants des résultats de recherche produits par un de leurs modèles d'IA…”
Plain restatementOn 6 October 2026, OpenAI released more than 700 mathematical manuscripts generated by an unreleased internal model, and the release drew strong critical reaction from research mathematicians.
This one is largely correct. On 6 October 2026 OpenAI did publish a very large batch of mathematical manuscripts generated by an unreleased internal model, and the reaction from research mathematicians was strong. The details worth adding: the material went up in a public GitHub repository, not a journal or arXiv, so none of it has been peer reviewed, and only a minority of the headline results have been formally checked so far. The count was 722 manuscripts, but the repository itself groups them into 372 result families, so it is not 700 separate discoveries, and three manuscripts were withdrawn the next day after an error. Nature reported that the release divided mathematicians rather than uniting them in anger, with some celebrating progress on old problems while others objected strongly. The quotes from Fields Medallist Hugo Duminil-Copin in the caption, including crying and saying his field had been annihilated, were reported by Le Monde from a Société mathématique de France seminar on 8 October and check out. The caption's suggestion that OpenAI is working to bury the discipline is an interpretation, not an established fact.
On [drifted from the evidence:] October 6, OpenAI [drifted from the evidence:] published more than 700 [drifted from the evidence:] papers presenting mathematical [drifted from the evidence:] research results produced by [drifted from the evidence:] one of its internal [drifted from the evidence:] AI models, sparking outrage among mathematicians." (French original on the [drifted from the evidence:] slide: "Le 06 Octobre OpenAI a publié plus de 700 articles présentants des résultats de recherche produits par un de leurs modèles d'IA internes, créant ainsi une vague d'indignation chez les mathématiciens")
On 6 [added by the neutral restatement:] October 2026, OpenAI [added by the neutral restatement:] released more than 700 mathematical [added by the neutral restatement:] manuscripts generated by [added by the neutral restatement:] an unreleased internal [added by the neutral restatement:] model, and the [added by the neutral restatement:] release drew strong critical reaction from research mathematicians.
Red-tinted words in the claim drifted from the evidence. Green-tinted words are what a neutral restatement needs.
The trace / claim to source
- The date is correct. OpenAI's own blog post carries the date 6 October 2026.
- The count is correct. The repository README recorded 722 manuscripts at release, which is "more than 700". The current figure is 719 after three withdrawals.
- The attribution is correct. OpenAI states the results were produced by an internal frontier model that has not been released publicly.
- The subject matter is correct. The manuscripts present claimed results on open problems in mathematics and theoretical computer science, across 17 areas by repository count.
- Strong critical reaction from research mathematicians is real and documented by Nature, which described the community as in uproar, and by an open letter from a group calling itself the Association for Human Mathematics urging mathematicians to stop working with OpenAI.
- The caption's Duminil-Copin details check out as reported. Le Monde records the crying and waking-his-daughter remark, the "they annihilated my research field" remark, and the 16 problems of interest to him, all from the Société mathématique de France seminar of 8 October 2026. He also expressed concern about young researchers in a separate RTS interview.
- The claim says "papers", which in a research context implies journal articles or at least arXiv preprints. The evidence is that these are self-published manuscripts in a GitHub repository, released at varying stages of verification, not peer reviewed, with roughly 42 percent of top-line results formalized in Lean as of mid-October. The word "papers" imports a validation step that has not happened.
- 722 manuscripts are not 722 results. The repository itself states the manuscripts are grouped into 372 families, where one family bundles a main result with companion arguments, consequences or alternative proofs. A reader of the claim would reasonably infer several hundred more discoveries than the artifact asserts.
- Three manuscripts were withdrawn on 7 October, one day after release, after a sign error invalidated one argument and the two papers that depended on it. The claim presents the batch as a finished body of results.
- The claim presents the reaction as outrage among mathematicians as a group. Nature reports the drop divided the community, with some mathematicians celebrating the solution of long-standing problems while others objected. Presenting one side of a split reaction as the reaction of the field overstates consensus.
- Capability extrapolation, in the caption rather than the headline claim: the caption says years of work were reduced to nothing by a single company. What is established is that the manuscripts claim progress on problems researchers were working on, with verification still under way. A claimed and partly unverified result is not the same thing as a closed problem.
- The counts and the problem set description all originate with OpenAI's own repository and blog post. They are reliable for what OpenAI published and for how OpenAI describes it. They are not independent confirmation that the mathematics is correct.
- Imputed motive stated as fact, in the caption: the caption asserts OpenAI appears to be working with the aim of burying the discipline. No evidence supports a statement about the company's intent. OpenAI's own framing is the opposite, and on 15 October it announced a 5 million dollar contribution to a Simons Foundation initiative to help mathematicians work with AI-produced results. Whether that framing is sincere is not something the available evidence settles either way.
- How many of the 372 result families are mathematically correct. Formal Lean verification covers a minority of top-line results and human review of the corpus is ongoing. Further withdrawals are possible.
- Which internal model produced the results. OpenAI describes it only as an unreleased internal frontier model, so the claim cannot be pinned to a named version.
- The caption's specific statement that many young researchers have thesis topics among the published papers. Duminil-Copin's concern for young researchers' careers is documented, but a count or named cases of doctoral theses rendered moot was not found.
- Whether the net balance of reaction in the mathematics community is predominantly critical. Named mathematicians were quoted on both sides, and no survey of the field exists.
OpenAI's own blog post, dated 6 October 2026, states that it is releasing a broad range of new mathematical results produced by an internal frontier model, with Lean proof formalizations and research details on GitHub. The repository README states that the catalogue contained 722 manuscripts organized into 372 families, where a family groups a principal result together with companion arguments, consequences or alternative proofs. The README as it currently reads states 719 manuscripts in 372 families, the count having dropped after three manuscripts were withdrawn on 7 October following a sign error that invalidated one argument and the constructions in two dependent papers. Reporting of the README states the model was posed roughly 4,000 problems and that each result used on average about three hours of ChatGPT Pro thinking compute, with the collection released under Apache 2.0 and at varying stages of verification. Nature reports that the research community is in uproar after OpenAI released a trove of more than 700 mathematical preprints entirely generated by artificial intelligence on 6 October, posted on the software repository GitHub. Nature also reports that although some mathematicians celebrated the solution of long-standing problems, others took to social media to complain about being scooped, and some were incensed at what one mathematician called a "slopocalypse", even if the mathematical content could end up being formally correct. Nature's briefing states that the surprise drop has divided mathematicians. On the Duminil-Copin element, Le Monde reports that the mathematician Hugo Duminil-Copin, Fields Medal 2022, said during a videoconference organised by the Société mathématique de France on Thursday 8 October that he had cried during the night on Tuesday and that it had woken his daughter, and that OpenAI had put online more than 719 manuscripts in 372 families claiming to resolve many still-open mathematical problems, with no fewer than 16 questions of interest to the French researcher concerned. The same report records his remark that his field of research had been annihilated.
Complete reasoning
The reply is formatted for pasting into the thread where the claim is circulating.
Ask this case
Answers come only from the case file above; nothing is added.
Did OpenAI actually publish 700 peer-reviewed papers on October 6?
OpenAI did release more than 700 manuscripts on 6 October, but they were posted to a public GitHub repository, not a journal or arXiv, and have not been peer reviewed. Only a minority of the headline results had been formally verified as of mid-October.
Is 700 the right number of distinct mathematical results?
Not quite. The repository logged 722 manuscripts, but it groups them into 372 result families, where each family bundles a main result with related arguments or alternative proofs, so the actual number of distinct discoveries is lower than 700.
Did all mathematicians react with outrage?
No. Nature reported that the release divided the community rather than uniting it in anger, with some mathematicians celebrating progress on long-standing problems while others objected, with one calling it a slopocalypse.
Are the quotes from Hugo Duminil-Copin accurate?
Yes. Le Monde reported that the Fields Medallist said he cried and woke his daughter, and that his research field had been annihilated, during a Societe mathematique de France seminar on 8 October, and these details check out against that reporting.
Is it true OpenAI is trying to bury mathematics as a discipline?
This is not established. That claim appears only in the caption as an interpretation, not a fact supported by evidence, and OpenAI's own public framing points the other way, including a later announced contribution to a Simons Foundation initiative on AI-assisted mathematics.