🔍 Read the full analysis: From 722 Proofs To What? OpenAI’s AI Mathematics Seeks Direction on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
OpenAI published 722 mathematical manuscripts, organized into 372 families and generated from roughly 4,000 problems by an unnamed model. The work includes claims about major open problems, but outside mathematicians have not confirmed the results; the longer-term question is whether researchers can verify and use them.
OpenAI has published 722 mathematical manuscripts produced by an unnamed, unreleased model, including claims addressing several longstanding problems. The company says the papers were selected from work on about 4,000 problems, but outside mathematicians have not yet confirmed the results, leaving both their correctness and their potential value to the field unsettled.
The manuscripts are arranged into 372 families of related results and cover number theory, geometry, operator algebras, topology, theoretical computer science and mathematical physics. OpenAI says the average result took about three hours of ChatGPT Pro thinking compute. The collection is published under the Apache-2.0 license.
The catalogue includes claimed work on the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, the isomorphism of nonabelian free group factors, and a zero-free region for the Riemann zeta function to the right of Re(s) = 11/12. It also includes a result concerning the Hodge conjecture for CM abelian varieties and work on the Mahler conjectures. These are claims in manuscripts, not independently established breakthroughs.
OpenAI reports that many, but not all, results have Lean formalizations, computer-checkable representations of mathematical proofs. The company’s repository warns that some unformalized results could have issues. It provides ten abridged reasoning summaries, one for each of only ten families, and says it selected the published problems for an “appropriate level of significance.” The source says that selection was made inside OpenAI, not by an independent 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.
What Verification Could Change
The immediate importance is not simply whether a model can produce a large number of mathematical claims. It is whether mathematicians can check, understand and reuse the arguments. A valid proof can settle a question, but its broader influence often comes from techniques that other researchers can apply to new problems.
The Unique Games Conjecture illustrates the possible consequences of verification. The source describes a large body of theoretical computer science built on results that assume the conjecture, including claims about limits on approximation algorithms. A proof or disproof that survives scrutiny could affect that work. But until researchers establish what the manuscript proves and how, its downstream implications remain conditional.
There is also a practical test for AI-assisted research: can the model’s output be converted into mathematics that the community can inspect and build on? A catalogue of impressive-sounding claims does not by itself answer that. The work could produce reusable methods, settle questions without much further influence, or contain errors or mismatches between a claim and the problem mathematicians intended to solve.
mathematics proof verification software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
OpenAI’s Earlier Math Releases
The release follows three other major OpenAI mathematics announcements described in the source. In May, the company’s model produced 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. That process, in which researchers turned machine output into a form they could evaluate, is presented as an example of AI work becoming useful through human scrutiny.
An August release called “Ten Advances” had a more contested result: a claimed counterexample to Connes’s rigidity conjecture was challenged within a day. The critique, as summarized by the source, argued that the constructed groups did not meet a condition required by the conjecture. In September, OpenAI announced a Lean-formalized Navier–Stokes result generated with about 10,000 concurrent agents over 88 hours. The announcement prompted debate over research priorities and how mathematics should use AI; the source describes a declaration signed by 25 Fields Medalists three days later.
The latest catalogue also includes two departures from the stated standard process: the write-up concerning the Riemann zero-free region was edited by humans for readability, and the Hodge result was treated as an exception. The source does not provide enough detail to establish the full role of human input in either case.
As an affiliate, we earn on qualifying purchases.
Claims Await Independent Review
The source material provides no independent verification of the 722 manuscripts as a group and no outside assessment establishing which results are correct. Lean formalization can help check a proof when its formalization is complete and accurately represents the claim, but the source says formalizations are absent for some results. OpenAI itself cautions that some unformalized work may have problems.
It is also not clear how much of the work experts will be able to evaluate from the released material. Only ten abridged reasoning summaries are provided for 372 families, and the source says the Riemann write-up was edited for readability. The amount and effect of human involvement in preparing individual manuscripts remain uncertain.
For each result, researchers will need to determine whether the proof is sound, whether it addresses the stated mathematical problem, and whether it offers ideas others can use. Until that work is done, the catalogue’s claims should not be treated as established solutions or as evidence that the model has made a corresponding number of confirmed discoveries.
mathematical problem solving software
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 review of the manuscripts, including scrutiny of their definitions, proof steps and any Lean formalizations. The source does not give a timetable for reviews or identify a panel tasked with evaluating the full collection. It is therefore unclear when researchers will be able to report reliable conclusions across all 372 families.
Some papers may attract attention sooner because they address problems with substantial existing literatures or have formalized proofs that can be checked. Others may require specialist work to translate long or unfamiliar arguments into forms researchers can assess. Any early confirmation should be tied to the specific manuscript and result reviewed, rather than generalized to the entire catalogue.
The larger measure of the release will be what happens after checking: whether mathematicians extract methods, develop new results or find that particular claims need revision. For now, OpenAI has made a substantial body of machine-generated work public; the mathematical community has yet to establish how much of it stands up.
AI-powered mathematical research tools
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 and drawn from work on roughly 4,000 problems, according to the source material.
Has an outside mathematician confirmed the major claims?
The source says the claims have not yet been confirmed by outside mathematicians. Each result requires independent review; the release alone does not establish that a conjecture has been solved.
Are all the proofs formally verified?
No. The source says Lean formalizations exist for many, but not all, results. OpenAI’s repository warns that some unformalized results could have issues.
Why does the Unique Games result matter if it is verified?
The source says many theoretical computer science results assume the Unique Games Conjecture, including work on the limits of approximation algorithms. A verified result could affect that body of research, though its precise consequences would depend on what the proof establishes.
What happens next?
Mathematicians need to examine individual manuscripts, check their proofs and assess whether the methods are useful. The source gives no review timetable, so when broader conclusions will be available is unknown.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
