Terence Tao Leads Association for Human Mathematicians in Joint Boycott of OpenAI

The flashpoint of the controversy was a release. According to QbitAI, OpenAI released a batch of manuscripts claiming to contain solutions to several well-known problems in mathematics, numbering over 700 machine-generated proof documents. The Human Mathematics Association (AHM), led by Fields Medalist Terence Tao, immediately issued a joint statement fully boycotting OpenAI and urging mathematicians to stop collaborating with it.

The tone of the statement was described as extremely severe, and its core allegation is not about whether "AI can do mathematics," but about the manner of this release: no peer review, no readable human logic, and no respect or explanation for prior work. In the words of the report, the statement opens by dredging up old grievances, pointing directly at OpenAI as a company currently mired in litigation over allegations of illegal plagiarism, copyright infringement, and trademark dilution.

Two levels need to be distinguished here: the AHM issuing the boycott statement and OpenAI releasing over 700 documents are confirmed facts; whether "these documents actually solved the problems they claim to solve" has not been independently verified in the available material. The report itself also lists the actual mathematical correctness of the 700-plus documents as an open question.

What shook the mathematics community was a set of repeatedly cited numbers. Public data shows OpenAI used internal models to test approximately 8,000 open mathematics problems, with a hit rate of around 5%; the report also states that solving a high-difficulty problem on average consumes only about 3 hours of GPT-Pro-level compute.

This set of numbers needs to be treated with caution. They come from public data relayed by the report, lacking first-party primary sources, so in terms of evidentiary strength they can only count as source claims, not confirmed facts. But even as claims alone, the signal they convey is clear: if 3 hours of compute can touch a scholar's years or even a lifetime of work, then the cost structure, priorities, and evaluation system of mathematical research will all be repriced.

The report uses "three hours of compute can directly crush a scholar's years or even a lifetime of research accumulation" to describe this impact. This is rhetoric, but the anxiety it points to is real—when the speed of output far exceeds the speed of human digestion, the question shifts from "can it be done" to "what do we do after it's done."