🔍 Read the full analysis: OpenAI’s 722 Proofs Put AI Mathematics At A Crossroads Of Questions on ThorstenMeyerAI.com
Get the latest gadgets delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
TL;DR
OpenAI published 722 mathematical manuscripts produced by an unnamed, unreleased model, covering 372 families of results selected from about 4,000 problems. The results include claims about major open problems, but outside mathematicians have not confirmed them, and their value will depend on verification and whether researchers can use the methods.
OpenAI published 722 mathematical manuscripts on Monday, presenting results generated by an unnamed model that the company has not released. The papers cover 372 families of related results, including claims concerning major open problems, but OpenAI and the source material say the work has not yet been confirmed by outside mathematicians.
The manuscripts span number theory, geometry, operator algebras, topology, theoretical computer science and mathematical physics. OpenAI says they came from roughly 4,000 problems posed to the model, then filtered by the company for what it considered an appropriate level of significance. The average result used about three hours of ChatGPT Pro thinking compute. OpenAI published the work under the Apache-2.0 license.
The catalogue includes claims involving the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, the isomorphism of nonabelian free group factors, a zero-free region for the Riemann zeta function to the right of Re(s) = 11/12, the Hodge conjecture for CM abelian varieties, and conjectures in convex geometry. These are claims in manuscripts, not independently established solutions. The source account says the Riemann and Hodge write-ups were exceptions to the usual process; humans edited the Riemann paper for readability.
Many results have Lean formalizations, a form of computer-checkable proof, but not all do. OpenAI’s repository README cautions that “some of the unformalized results could have issues.” The company supplied ten abridged reasoning summaries for 372 families, leaving most results without that short-form explanation. The selection and significance screening were also conducted by OpenAI, rather than by an independent mathematical panel.
722 proofs, one question: will any of OpenAI’s AI mathematics actually lead anywhere?
An unreleased, unnamed model produced claimed proofs of results that would each define a career. Sam Altman calls them “claims not yet confirmed by outside mathematicians.” The real question isn’t whether it’s impressive. It’s whether answers nobody understands become discoveries anyone can build on.
Same day: Alon, Bloom, Gowers, Litt, Sawin post a digested, human-verified version. The model for success.
Connes rigidity counterexample challenged within a day — constructed groups fail the required condition. Three rival machine “counterexamples” from different labs now circulate.
~10,000 agents, 88 hours, est. ~$22M at retail. Priority dispute; 25 Fields Medalists sign “A Severe Misalignment” — not saying it’s wrong, saying it’s not understood.
Altman now hedges at announcement — a shift from September. Verification has barely started.
Humans extract the technique, write it up, build on it. This is where downstream discovery comes from.
The question is answered; nobody learns anything reusable. Closes a door without opening a field.
The proof breaks, or proves a statement that doesn’t match the conjecture as mathematicians mean it.
The Unique Games Conjecture is the clearest case. Results like the optimality of Goemans–Williamson for Max-Cut are proved assuming UGC. A correct proof converts them all — no understanding required. A zero-free strip for zeta works the same way for prime-distribution results. Free group factors, Kadison, Mahler would redirect whole programmes — but how depends on the method, which means digestion.
Technology. A Navier–Stokes blow-up proof doesn’t change how anyone designs aircraft; engineering turbulence models never depended on the answer. Near-term consequences are mathematical, not industrial. “AI will cure cancer next” skips several steps.
“Verification abundance, adjudication scarcity” — making proof-checking cheap doesn’t reduce the burden of deciding what’s true and what matters. 722 manuscripts land on a review system built for a trickle, filtered by a selection nobody outside OpenAI made.
Humans re-deriving results, like Alon–Gowers et al. in May
Other people’s work building on these manuscripts
How many unformalized results survive expert checking
Do the Lean statements match the real conjectures?
Do any survive peer review?
Some of it, yes — where a literature is waiting (UGC), a correct proof pays off immediately; where a proof carries a new technique humans digest, it can open a field. Most of it, probably not on its own: at 722 manuscripts with 10 reasoning summaries, the Four Colour pattern is the likely default unless mathematicians are funded and given time. And some will be wrong — OpenAI says so itself. It’s an industry pattern, not one company’s: the forced-Euler result came from an Anthropic researcher, and rival machine-generated Connes “counterexamples” circulate from different labs. The proofs arrived this week. The discoveries, if they come, will arrive at the speed of human understanding.
Verification Will Shape Their Value
The release raises two separate questions: whether each proof is correct, and whether it gives mathematicians ideas they can use. A formalized proof can help establish that a specific argument follows within a formal system, but it does not by itself show that the result is important, that the proof matches the intended conjecture, or that its method will advance other work. Unformalized manuscripts require additional scrutiny.
Mathematical proofs often matter for the techniques they introduce, not only for the question they settle. The source account contrasts OpenAI’s earlier Erdős unit-distance result, which mathematicians converted into a human-readable, verified account, with the possibility of a correct but hard-to-use proof. In that view, the practical test for this catalogue is whether researchers can digest the arguments and build on them—not simply how many papers the model produced.
Some claims could have wide implications if verified. The Unique Games Conjecture, for example, underpins conditional results in theoretical computer science about the limits of approximation algorithms. A proof could change the status of work that assumes the conjecture. But the release does not establish that the claim is correct, and the consequences cannot be assessed until specialists examine the argument.
As an affiliate, we earn on qualifying purchases.
Earlier Releases Show Mixed Results
This is described in the source material as OpenAI’s fourth major mathematics release of the year. In May, the company reported that its model had found a counterexample to the Erdős unit-distance conjecture. Five mathematicians—Noga Alon, Thomas Bloom, Tim Gowers, Daniel Litt and Will Sawin—then posted a human-verified account of the result. That process offered a way for the mathematical community to inspect and assess machine-generated work.
OpenAI’s August release, called “Ten Advances,” had a more contested result: a claimed counterexample to Connes’s rigidity conjecture was challenged because critics said the constructed groups did not meet the conjecture’s required conditions. In September, OpenAI announced a Lean-formalized Navier–Stokes result produced by about 10,000 concurrent agents over 88 hours. That announcement prompted a separate dispute about priority and the purpose of AI mathematics. The history does not determine whether the new papers are right, but it shows why publication counts and headline claims are not substitutes for external review.
The source account also describes a debate among mathematicians about whether using famous open problems as AI benchmarks serves the discipline. The concern is not necessarily that machine-assisted proofs are invalid; it is that results may be delivered in forms that mathematicians cannot readily understand or reuse. The current collection makes that distinction especially relevant because it includes hundreds of papers but only ten abridged reasoning summaries.
mathematical proof verification software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Hundreds of Claims Await Review
No outside verification of the new catalogue is reported in the source material. It is not yet clear which of the 372 result families will withstand specialist review, whether the formalized and unformalized papers will fare differently, or whether any claimed proof establishes precisely the statement mathematicians consider open. The source account does not provide independent assessments of the headline results.
It is also unclear how OpenAI ranked the results beyond saying that it filtered roughly 4,000 problems for an appropriate level of significance. The company has not supplied an abridged reasoning summary for every family, and the unnamed model has not been released. Those limits make it difficult for readers and researchers to reproduce the process or assess how representative the published papers are of the model’s work.
Even results that are eventually verified may have different consequences. Some could yield useful techniques; others may settle a question without producing methods that transfer to other problems. The catalogue’s eventual mathematical value cannot be inferred from the number of manuscripts or the prominence of the conjectures they address.
formal proof checkers for mathematicians
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Mathematicians Must Test the Papers
The next step is independent examination of the individual manuscripts, including checks of formalized proofs and close reading of arguments that are not formalized. Researchers will need to establish whether each result is correct and whether it addresses the relevant conjecture in the form claimed. The source material gives no timetable for that work or for OpenAI to publish additional reviews.
For the most consequential claims, the key milestone will be a clear account that specialists can verify and explain. The May Erdős episode offers one example of that process, but it does not guarantee similar outcomes across this much larger collection. Until reviews appear, the 722 papers should be treated as a set of AI-generated mathematical claims under assessment, not as 722 confirmed discoveries.
AI-generated mathematical manuscripts
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did OpenAI publish?
OpenAI published 722 mathematical manuscripts grouped into 372 families of related results. The company says they were generated by an unnamed model using roughly 4,000 problems as inputs.
Have mathematicians confirmed the claimed proofs?
Not according to the source material. The claims have not yet been confirmed by outside mathematicians. OpenAI’s repository also warns that some unformalized results could have issues.
What are some of the major problems mentioned?
The manuscripts include claims about the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, the Hodge conjecture for CM abelian varieties, and a zero-free region for the Riemann zeta function. These remain claims until specialists verify them.
Does formalization prove a result is important?
No. A Lean formalization can help check whether a proof follows within a formal system, but it does not establish the result’s broader significance or whether its methods will be useful to other researchers.
What happens next?
Mathematicians will need to examine the papers, verify their arguments and assess whether the results can be understood and used. No review schedule or timetable for further OpenAI material is specified in the source account.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
