OpenAI drops nearly 400 AI-generated math results, leaving researchers years of verification work

OpenAI has released nearly 400 AI-generated mathematical results spread across more than 700 manuscripts, a deluge that researchers told The Verge could take years to properly understand and verify. The work spans fields from combinatorics and number theory to topology and mathematical physics, and the collection is so sprawling that OpenAI published its own guide for navigating the associated GitHub repository.

A release too large to read

According to The Verge, which spoke with more than three dozen mathematicians, reactions ranged from "staggering" and "unprecedented" to "pure insanity" — awe mixed with deep anxiety about what the flood means for the discipline. Several researchers said that simply working through the roughly 40-page table of contents and abstracts took the better part of an hour. "Just going over the entire list of abstracts is overwhelming," Álvaro Lozano-Robledo of the University of Connecticut told the outlet.

The manuscripts cover combinatorics, several branches of geometry, number theory, theoretical computer science, algebra, topology, probability, statistical mechanics and mathematical physics. As one mathematician put it to The Verge, if the AIs disappeared tomorrow like aliens leaving Earth, the field would spend the next ten years studying what they left behind.

Formal verification lags behind

Scattered through the collection are formalizations in Lean, the proof assistant that allows results to be checked computationally, and which helped researchers assess OpenAI's earlier mathematical claims. But coverage is incomplete. OpenAI acknowledged on GitHub that the manuscripts are at different stages of verification, and said that only 300 top-line results out of 719 manuscripts — roughly 42 percent — had been formalized, with more to be added over time.

Mathematicians complained to The Verge about the shortfall, noting that even where Lean code exists, confirming that the formalization actually proves the stated claim is itself slow work. Several who examined the papers said the quality of the formalizations was inconsistent, and that the verified statements did not always map neatly onto the claims in the accompanying manuscripts.

Kevin Buzzard of Imperial College London, working in algebraic number theory, said he found only around six theorems in his area that immediately stood out, and few of those appeared to be formally verified. Left choosing between reading possibly incorrect material, waiting for others to read it, or waiting for formalization, he could not yet say whether the results are even correct.

Slopocalypse fears, partially allayed

Ahead of the release, researchers had taken to calling the expected flood the "slopocalypse," a nod to OpenAI's earlier mathematical write-ups, which drew criticism for sloppy presentation and poor or nonexistent attribution. The Verge reports that early impressions suggest more care went into this batch, though mathematicians admitted the bar set by previous publications was low.

Better is not the same as adequate. Brendan Hassett of Brown University said the write-up of the problem he knows best made little sense after a quick read, and that he would not spend more time on it had a person written it — before noting there were still hundreds of other preprints. The Verge points out his comment came before OpenAI retracted three papers.

Other researchers flagged papers that appear to cover ground already trodden by others, though they were wary of saying so publicly before proper review. Some noted unusual brevity, with results normally developed over hundreds of pages compressed into a few dozen or fewer. Nalini Joshi of the University of Sydney said some papers she examined have short bibliographies, and given past criticism of OpenAI's crediting practices, she is wary the attribution may be missing the complete story.

Impressive work underneath

Despite the presentation problems, The Verge reports an overarching consensus that the release contains genuinely impressive mathematics of very high caliber. The tension is stark: the same volume that overwhelms the field's ability to review it is also producing results researchers describe as serious. One researcher warned the situation could cause a huge collapse in academic culture and hollow out mathematics faculties, and accused AI labs of ignoring the social consequences.

Why it matters

The release demonstrates that AI systems can now generate frontier mathematics at a speed and breadth no human team can match, and beyond even OpenAI's own capacity to internally scrutinize. That shifts the burden of verification onto a community with limited review bandwidth, straining norms of attribution, credit and peer assessment. Many researchers fear OpenAI will not wait for the field to catch up before releasing more. If that pace continues, mathematics may need new workflows — prioritized formalization, distributed review, and conventions for citing machine-generated proofs. How the discipline adapts will likely become a template for other fields confronting a flood of AI-generated research.