🔍 Read the full analysis: After 722 Proofs, OpenAI’s AI Mathematics Still Has A Question To Answer 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 has published 722 mathematical manuscripts, grouped into 372 families and generated from roughly 4,000 problems. The papers include extraordinary claims, but the company and outside mathematicians have not established that the results are correct or that they will produce useful new methods.
OpenAI published 722 mathematical manuscripts on Monday, presenting work from an unnamed, unreleased model that the company says addressed roughly 4,000 problems. The collection includes claims concerning major open questions, but OpenAI has cautioned that the results have not yet been confirmed by outside mathematicians.
The manuscripts are organized into 372 families of related results and span subjects including number theory, geometry, topology, operator algebras, theoretical computer science and mathematical physics. OpenAI says the average result involved about three hours of ChatGPT Pro thinking compute. The papers are published under the Apache-2.0 license, and many—but not all—include Lean formalizations, a way to encode mathematical statements and proofs for computer checking.
Among the headline claims are a proof of the Unique Games Conjecture, a resolution of Hilbert’s tenth problem over the rationals, a result about nonabelian free group factors, a zero-free region for the Riemann zeta function to the right of Re(s) = 11/12, and results concerning the Hodge and Mahler conjectures. These are claims in the released manuscripts, not independently established breakthroughs. OpenAI’s repository warns that some unformalized results could have issues.
The release also includes only 10 abridged reasoning summaries for the 372 families. OpenAI says it selected the problems and filtered for what it considered an appropriate level of significance; people outside the company did not make that selection. The Riemann zero-free-region paper was edited by humans for readability, and the Hodge result and Riemann paper followed procedures that differed from the usual process.
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.
Proofs Need More Than Answers
The central question is not only whether a manuscript’s conclusion is correct. In mathematics, a proof can matter because it gives researchers methods and ideas they can reuse, not just because it settles a named conjecture. If mathematicians can extract such ideas from these papers, the collection could influence work well beyond the original problems. If the proofs are correct but opaque or unproductive, the field may gain answers without gaining much new understanding.
The Unique Games Conjecture illustrates the potential consequences. Many results in theoretical computer science use it to establish limits on approximation algorithms. A valid proof could change how researchers interpret that body of conditional work. But those implications depend on external experts checking precisely what has been proved and whether the proof offers techniques that can be applied elsewhere.
The collection’s size also raises a practical issue: 372 families require human scrutiny. Formal verification can help check that a proof follows from its formalized assumptions, but it does not, by itself, show that the assumptions match the intended conjecture, that every result has been formalized, or that the reasoning is useful to working mathematicians.
Products Worth Considering
Jackery Explorer 2000 v2 Portable Power Station with Two 200W Solar Panels
Samsung 32-inch Odyssey G5 G51F QHD 180Hz Gaming Monitor
Samsung Odyssey OLED G5 27-Inch QHD Gaming Monitor
As an affiliate, we earn on qualifying purchases.
A Mixed Record This Year
This is OpenAI’s fourth major mathematics release this year, according to the source account. In May, its model produced a counterexample to the Erdős unit-distance conjecture. Five mathematicians—Noga Alon, Thomas Bloom, Tim Gowers, Daniel Litt and Will Sawin—then published what they described as a digested, human-verified version. That process turned machine output into a form experts could evaluate and helped establish the result.
OpenAI’s August release, called “Ten Advances,” had a more contested outcome. One claimed counterexample to Connes’s rigidity conjecture was challenged within a day; critics said the constructed groups did not meet the conjecture’s required condition. The September announcement of a Lean-formalized Navier–Stokes blow-up proof also drew attention to questions about how AI mathematics is evaluated. A group of 25 Fields Medalists signed a declaration objecting to benchmark-driven work that, in their view, can sideline human understanding. Their concern was about mathematics’ aims, not a declaration that the proof was wrong.
Those episodes show why outside review matters. A result may be sound but need translation into an understandable argument; it may be disputed because its statement does not match the problem; or it may fail to produce ideas others can use. Verification, interpretation and mathematical value are related but separate tests.
What Independent Review Finds
No outside confirmation of the collection’s headline claims is established in the source material. It is not clear which of the 722 manuscripts have been checked by independent specialists, how many will withstand detailed review, or when assessments will be available. The limited number of abridged summaries also means readers cannot treat those summaries as a substitute for examining the underlying arguments.
It is also unclear how many results have complete Lean formalizations and what level of checking each formalization received. Even where a proof is machine-checkable, experts still need to confirm that the formal statement corresponds to the mathematical question being claimed. The eventual significance of the work—whether it yields reusable techniques, settles questions without broader impact, or contains errors—cannot yet be determined.
Mathematicians Must Test the Papers
The next step is independent mathematical review: specialists will need to check the statements, scrutinize the proofs and identify any ideas that can be separated from the model’s reasoning and developed further. OpenAI has made the manuscripts available, but the source material does not give a timetable for a comprehensive review or name independent teams assigned to each family.
For readers, the clearest measures of progress will be specific papers that receive public verification, corrections or objections, followed by work that uses their methods. Until those assessments emerge, the 722 manuscripts should be treated as a large set of research claims under examination, not as 722 confirmed discoveries.
Key Questions
What did OpenAI release?
OpenAI published 722 mathematical manuscripts, grouped into 372 families and generated from roughly 4,000 problems, according to the release information.
Has the Unique Games Conjecture been proved?
The released collection includes a manuscript claiming a proof, but outside mathematicians have not confirmed the result in the source material.
Does Lean formalization prove that a result is correct?
A Lean formalization can let a computer check that a proof follows within a formal system. It does not automatically establish that the formalized statement matches the intended conjecture, and not every result in the collection is formalized.
Why does it matter whether the proofs produce new methods?
A mathematical result can settle a question while offering few ideas for other problems. Researchers will be looking for reusable techniques and insights, as well as checking whether the claims themselves hold.
What is still unknown?
The extent of independent review, how many claims will be validated, and whether the work will lead to further discoveries are not yet clear.
Source: ThorstenMeyerAI.com
