🔍 Read the full analysis: OpenAI’s AI Mathematics: Can 722 Proofs Lead To A Larger Impact? on ThorstenMeyerAI.com
Get privacy and security gear 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 mathematics manuscripts produced by an unnamed, unreleased model, covering 372 families of results drawn from about 4,000 problems. The company says the claims have not been confirmed by outside mathematicians; whether they lead to reusable ideas depends on independent checking and human understanding.
OpenAI published 722 mathematical manuscripts on Monday, presenting work by an unnamed, unreleased model across 372 families of related results. The collection includes claims about longstanding problems in mathematics, but OpenAI chief executive Sam Altman said the results have not been confirmed by outside mathematicians, leaving their accuracy and potential influence unsettled.
According to OpenAI’s post and the collection’s GitHub repository, the manuscripts span number theory, geometry, operator algebras, topology, theoretical computer science and mathematical physics. They were selected from roughly 4,000 problems posed to the model. OpenAI says the average result used about three hours of ChatGPT Pro thinking compute. The collection is published under the Apache-2.0 license.
The manuscripts include claims about 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. Other claimed results concern the Hodge conjecture for CM abelian varieties and conjectures in convex geometry. These are claims in the released work, not independently established solutions.
OpenAI says some manuscripts have Lean formalizations, but not all. Its repository README warns that “some of the unformalized results could have issues.” The release also contains ten abridged reasoning summaries for 372 result families. OpenAI chose which problems to include based on what it considered an appropriate level of significance; no outside group made that selection. The Riemann write-up was edited by humans for readability, and OpenAI says the Hodge result and Riemann result followed exceptions to its usual procedure.
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.
From Machine Proofs to New Mathematics
The potential impact is not simply whether a proof can settle a famous question. In mathematics, a proof can matter because it gives researchers a method they can adapt, a connection they can extend, or a result that changes what can be shown elsewhere. A correct answer that offers no reusable insight may have less effect on the field than a less sweeping result that opens a new line of work.
The Unique Games Conjecture illustrates why verification could matter beyond mathematics departments. Many results in theoretical computer science are conditional on it, including claims about the limits of approximation algorithms. If the conjecture were proved, researchers would need to assess the consequences for those results. But that chain of implications depends on the proof being correct and on specialists understanding what it establishes.
There is also a practical burden: hundreds of manuscripts require expert attention, and a Lean formalization, where available, checks a formal statement and its proof within a system. It does not by itself establish that the statement matches the intended conjecture or explain why the result is useful. The release’s wider value will depend on independent scrutiny and human digestion, not the count of manuscripts alone.
As an affiliate, we earn on qualifying purchases.
OpenAI’s Recent Math Releases
The release follows several mathematics announcements from OpenAI this year, with mixed reception. In May, the company described work on 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 a counterexample. That process showed one route from machine output to a result the field could evaluate: mathematicians reconstructed and checked the work.
OpenAI’s August collection, called “Ten Advances,” included claims that drew scrutiny. A proposed counterexample to Connes’s rigidity conjecture was challenged on the grounds that the constructed groups did not meet a condition required by the conjecture. The September announcement of a Lean-formalized Navier–Stokes result also prompted debate about AI proof generation, research priorities and whether work on headline problems advances mathematical understanding. The source material says 25 Fields Medalists signed a declaration criticizing the use of famous problems as benchmarks without human understanding; it does not establish that the proof itself was wrong.
Those episodes are relevant, but they do not determine the status of the new manuscripts. Each claim needs to be checked on its own terms. The earlier Erdős work offers an example of human verification; the disputed Connes claim shows why a proposed counterexample can fail even when it appears to address a conjecture.
formal proof verification software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Verification Across 372 Result Families
No outside confirmation is reported for the new collection as a whole, and the source material does not identify independent reviewers who have checked the major claims. It is not clear how many of the 722 manuscripts will prove correct, how many will be formalized, or whether formalized statements precisely capture the conjectures researchers regard as open.
The release also leaves open how much of the model’s reasoning can be reconstructed from the ten abridged summaries, and how much human editing or intervention shaped individual results. OpenAI’s selection process means the collection is not an independently assessed sample of the model’s mathematical work. Claims about exceptionally consequential problems should be treated as claims until specialists publish checks or corrections.
As an affiliate, we earn on qualifying purchases.
Independent Checks Will Set the Course
The next step is for mathematicians to inspect individual manuscripts, test their arguments and compare each stated result with the problem it purports to solve. For results with Lean formalizations, researchers can examine the formal statements and proofs; unformalized work will require conventional expert review. Any errors, qualifications or independent confirmations will need to be reported result by result.
OpenAI has not, in the source material, provided a timetable for outside reviews or named a process for resolving disputed claims. The practical measure of the release will emerge as researchers either extract ideas they can build on, accept a result without finding broader methods, or identify flaws. For now, the manuscripts are a large set of research claims, not a verified set of mathematical breakthroughs.
advanced mathematical physics textbooks
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did OpenAI release?
OpenAI published 722 mathematical manuscripts grouped into 372 families, produced by an unnamed, unreleased model. The work was drawn from roughly 4,000 problems and spans several areas of mathematics and theoretical computer science.
Have mathematicians verified the claimed results?
The source material reports that the claims have not been confirmed by outside mathematicians. Some work has Lean formalizations, while OpenAI warns that unformalized results could have issues.
Did the model solve the Unique Games Conjecture?
The collection includes a manuscript claiming a proof of the Unique Games Conjecture. That claim has not been independently confirmed in the source material, so it should not be described as a settled solution.
Why might these manuscripts matter if they are correct?
Some claims concern problems with consequences for other areas, including theoretical computer science. Their longer-term importance will depend on whether the proofs check out and whether researchers can extract ideas and methods that support further work.
What happens next?
Mathematicians will need to assess the manuscripts individually, checking both the reasoning and the exact statements proved. OpenAI has not announced a timetable for that independent review in the material provided.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
