OpenAI Partition Principle claim is an academic mess, says mathematician Asaf Karagila
A leading set theorist says OpenAI's preprint on the century-old Partition Principle fails basic standards of academic writing and referencing.

OpenAI’s claims of progress on a century-old mathematical mystery have been met with sharp criticism from one of the world's leading experts on the problem.
The AI company recently released a preprint alongside a vast mountain of computer code written in the Lean programming language, claiming to show new progress on whether the "Partition Principle" implies the "Axiom of Choice". But set theorist Asaf Karagila, who has spent 15 years studying the problem, says the paper is so poorly written and academically sloppy that it deserves a flat rejection.
The dispute highlights a growing frustration among scientists who feel tech companies are dumping poorly verified, machine-generated work onto the academic community and expecting humans to clean up the mess.
What is the Partition Principle?
The debate touches on a fundamental question in mathematical logic. The Partition Principle is a classic, century-old puzzle in set theory. Karagila is so famously associated with the problem that his personal website's "Problems" page promises a bottle of whisky to anyone who finally solves it.
After OpenAI's announcement, his inbox filled with colleagues asking if he was about to mail a bottle of single malt to OpenAI chief executive Sam Altman.
The answer, Karagila wrote on his blog, is a firm no.
Why the OpenAI preprint fails academic standards
While OpenAI's press release framed the paper as a sign of "progress", Karagila described the actual text of the preprint as deeply flawed.
"It sucked," Karagila wrote. "It is unclear, muddled, and has a strange structure."
According to the mathematician, the paper fails basic academic standards on several fronts:
- Strange structure and terminology: The mathematical phrasing is "a bit off," and the preprint includes bizarre, highly specific lemmas that have no logical place in a paper of this nature.
- Lazy referencing: Rather than pointing to established, peer-reviewed literature, the preprint cites unpublished, unrefereed lecture notes—including Karagila's own informal online notes—for basic concepts.
- Unreadable arguments: Usually, a set theorist can skim a paper in their field and quickly grasp the author's strategy. Karagila says the OpenAI paper is so poorly communicated that this is impossible.
"If this was an academic paper submitted to a journal, it should be issued a desk rejection for the quality," Karagila said. "The onus is always on the author to conform and adhere to the communications level."
An AI 'denial of service' on researchers
This sloppy presentation points to a broader, more frustrating trend in the generative AI boom. By dumping incomprehensible, machine-generated papers onto the internet, tech companies are effectively launching a "denial of service" attack on academic time, Karagila argues.
While the media portrays these releases as revolutionary solutions, actual scientists are left with a miserable choice: ignore them, or stop their own active research to painstakingly sift through a machine's messy homework.
There is a political danger to this hype, too. Karagila warns that politicians, funding bodies, and the public—who do not understand how mathematical research actually works—will see these flashy announcements and assume human mathematicians are about to be replaced by bots.
He compares the situation to replacing a chef with automated kitchen tools. A self-dicing knife is a neat gadget, but it is still a long way from preparing a palatable meal.
For Karagila, the episode highlights a clash of cultures. Rather than AI companies doing the hard work to meet the rigorous communication standards of science, they seem to want academics to lower their standards to accommodate AI.
"I am afraid that this just will not do," Karagila said.



























































