"They have completely destroyed my research field."
Recently, Hugo Duminil-Copin, the 2022 Fields Medal-winning French mathematician, lost his composure.
"On Tuesday night, I cried and woke up my daughter."
The trigger was a batch of results released by OpenAI. Over 700 AI-generated mathematical manuscripts were made public. Among them were 16 problems in the very direction Hugo Duminil-Copin had been focusing on for a long time.
Before this, Hugo Duminil-Copin had likened the impact of AI-generated proofs to a "nuclear explosion." Each time AI solves a major problem, the landscape of mathematics suffers an impact, and the space for humans to continue exploring shrinks accordingly.
OpenAI's concentrated release of hundreds of results this time has been described by some mathematicians as a "carpet bombing."
A "Carpet Bombing"
On October 6, OpenAI publicly released 722 mathematical manuscripts on GitHub, belonging to 372 result families. A result family may include a core proof, supporting arguments, corollaries, and alternative proofs.

It was disclosed that this batch of manuscripts contains 377 new proofs. About one-third involve important conjectures of wide interest in the mathematical community, another third hold significant value in their respective sub-fields, and the remainder are mainly smaller results or variants of existing proofs.
OpenAI stated that the vast majority of these results were generated by the same yet-to-be-released internal model. The model began training on August 28, and the computational cost for each result is, on average, equivalent to ChatGPT Pro thinking for about 3 hours.
However, after the release, multiple mathematicians discovered that some manuscripts might not truly prove the conclusions they claimed. Even if a proof passes the Lean formal verification, it only indicates that the logic within the formalized proposition holds, and cannot automatically guarantee it accurately corresponds to the original argument in the natural language manuscript.
Two days after the release, OpenAI withdrew three papers containing errors; 719 papers remain in the repository.
How is the Value of Mathematical Research Measured?
The value of mathematical research has never been limited to the final proof.
In the process of searching for a proof, researchers propose new concepts, create new tools, and discover connections between different fields. A failed attempt may also open a doorway to another problem.
Hugo Duminil-Copin once mentioned that while tackling a certain probability theory conjecture, he experienced multiple failures. These failures generated a wealth of new ideas, later used in other research and leading to unforeseen discoveries.
AI directly providing proofs might skip this exploration process.
The 20th-century mathematician William Thurston reflected that after he quickly proved a batch of theorems, the related research direction gradually lost attention. As researchers left, the intuition, experience, and communication networks 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 the First to Be Affected
In the past, supervisors would choose a problem for graduate students expected to be solvable within a few months. Students would train their intuition through repeated attempts, gradually learning to handle more abstract and complex research. Today, such problems used to train young researchers might be quickly completed by models.
Mathematicians with stable positions can change direction. PhD students and postdocs do not have the same ample time. Their dissertations, positions, and career prospects often rely on a few research problems. Once preempted by AI, years of accumulation may lose their value.
Open communication may also be affected. Mathematicians are accustomed to sharing immature ideas at conferences, in emails, and discussions. If AI companies can collect these clues and then have models concentrate on tackling them, researchers might reduce public discussion.
Hugo Duminil-Copin worries that the culture of free exchange upon which mathematics thrives is fracturing.
"Alien Mathematics"
While reviewing an AI-generated proof, one netizen also discovered that the model combined two ideas that were rarely connected before. The proof was completely correct and verifiable, but the line of reasoning seemed unfamiliar. Some called it "alien mathematics."


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

The mode of scientific output is changing, and new research paths may no longer completely follow existing human thought habits.
Today might be the least "alien" moment for science.
References:
https://www.theatlantic.com/technology/2026/10/openai-math-proofs-future/688940/
https://x.com/miniapeur/status/2109237567832949109
https://x.com/haider1/status/2109353337489 645622
https://x.com/imjustnewatai/status/2109102057885991250
This article is from the WeChat public account "Machine Heart" (ID: almosthuman2014), author: Yang Wen





