Fields Medalist Cries: AI Has Destroyed My Research Field

Fields Medalist Cries: AI Has Destroyed My Research Field

Recently, Hugo Duminil-Copin, the 2022 Fields Medalist and French mathematician, lost his composure.

“On Tuesday night, I cried and even woke my daughter up.”

The cause was a recent batch of results released by OpenAI, with more than 700 AI-generated mathematical manuscripts made public. Among them were 16 problems that are precisely the areas Hugo Duminil-Copin has been researching for a long time.

Previously, Hugo Duminil-Copin likened the impact of AI‑generated proofs to a “nuclear explosion.” Every time AI solves a major problem, the mathematical landscape is shaken, and the space left for human exploration shrinks.

OpenAI’s recent batch release of hundreds of results has been described by some mathematicians as a “carpet‑bombing” campaign.

A “Carpet‑Bombing”

On October 6, OpenAI publicly released 722 mathematical manuscripts on GitHub, belonging to 372 result families, each of which may include a core proof, supporting arguments, corollaries, and alternative proofs.

According to the disclosure, this batch of manuscripts contains 377 new proofs. About one‑third of them involve important conjectures that have attracted widespread attention in the mathematics community, another one‑third are of significant value in their respective subfields, and the rest are mainly smaller results or variants of existing proofs.

OpenAI says that the overwhelming majority of these results were generated by the same unreleased internal model. The model began training on August 28, and each result consumed computational effort equivalent, on average, to about three hours of thinking by ChatGPT Pro.

However, after the release, several mathematicians discovered that some manuscripts might not actually prove the claimed conclusions. Even if the proof passes Lean formal verification, it only shows that the internal logic of the formal proposition holds and cannot automatically guarantee that it accurately corresponds to the original argument in the natural‑language manuscript.

Two days after the release, OpenAI retracted three papers that contained errors, leaving 719 papers remaining in the repository.

How should the value of mathematical research be measured?

The value of mathematical research is never limited to the final proof.

In the process of seeking proofs, researchers propose new concepts, create new tools, and discover connections between different fields. A failed attempt may also open a gateway to another problem.

Hugo Duminil-Copin once noted that while tackling a certain probability conjecture he experienced many failures; these failures generated numerous new ideas that were later applied to other research, leading to unforeseen discoveries.

AI providing proofs directly may skip this exploratory process.

The 20th‑century mathematician William Thurston once reflected that after he quickly proved a batch of theorems, the associated research direction gradually lost attention. As researchers moved on, the intuition, experience, and network of communication accumulated in that field also began to dissipate. In his view, mathematical knowledge exists not only in papers but also in the minds of researchers and within the academic community.

Young mathematicians are hit first

In the past, advisors would select a problem for graduate students that was expected to be solvable within a few months. Through repeated attempts, students trained their intuition and gradually learned to tackle more abstract and complex research. Today, such problems used to train young researchers can be solved rapidly by models.

Mathematicians who have secured stable positions can switch directions, but Ph.D. students and postdocs do not have the same amount of time. Their papers, positions, and career prospects often depend on a handful of research problems; once AI completes those problems first, years of effort may lose its value.

Open communication may also be affected. Mathematicians are accustomed to sharing immature ideas at conferences, via email, and in discussions; if AI companies can collect these clues and then focus models on tackling them, researchers might reduce public discourse.

Hugo Duminil‑Copin worries that the culture of free exchange that underpins the development of mathematics is fracturing.

“Alien Mathematics”

Some netizens, while reviewing an AI-generated proof, noticed that the model combined two ideas that had rarely been linked before. The proof is completely correct and withstands scrutiny, but its reasoning feels unfamiliar. Some have dubbed it “Alien Mathematics”.

The netizen believes that similar phenomena may soon appear in physics, chemistry, materials, medicine, and engineering fields.

The way science produces results is changing, and new research pathways may no longer fully follow humanity’s existing habits of thought.

Today may already be the least “alien” moment in science.

Reference link:

https://www.theatlantic.com/technology/2026/10/openai-math-proofs-future/688940/

https://x.com/miniapeur/status/2109237567832949109

https://x.com/haider1/status/2109353337489645622

https://x.com/imjustnewatai/status/2109102057885991250

© THE END 

For reprints, please contact this public account for permission

Submit articles or request coverage: liyazhou@jiqizhixin.com