2022 Fields Medal Winner, French Mathematician Hugo Duminil-Copin: OpenAI Destroys My Research Field
Hugo Duminil-Copin, the 2022 Fields Medal winner and French mathematician, stated during a video conference organized by the French Mathematical Society on October 8 that he cried on Tuesday night, waking his daughter.
He explained that the reason was the American AI company OpenAI releasing over 719 manuscripts, divided into 372 categories, claiming to solve or advance numerous long-standing unsolved mathematical problems. At least 16 of these problems are directly related to his long-term research focus. He calmly stated, "They have destroyed my research field."
Duminil-Copin displayed an AI-generated slide showing a rabbit being crushed by multiple trucks and added, "They want to prove they have the capability to slaughter us." In the Q&A session, he further discussed "the disassembly of our business." In response to one question, AI produced multiple papers in succession, equivalent to advancing an entire research agenda at once.
These manuscripts were generated by an internal, undisclosed model from OpenAI, submitting around 4,000 questions, with each result taking about 3 hours of computation equivalent to ChatGPT Pro. Approximately 42% of the main results have been verified in Lean formalization, but some lack formal proofs, and three papers have been retracted due to symbolic errors. The company acknowledged that the unformalized results may have issues. The mathematics community recognizes that there are astonishing results among them but also points out the need for further independent verification.
Duminil-Copin expressed concern not only for himself but also for young mathematicians. PhD students may spend years searching for problems and then years trying to prove them to establish their academic careers, only to wake up one day to find that an AI company has published dozens of papers claiming the problems have been solved. He noted that the culture of open discussion about unsolved problems in the mathematics community may be impacted, and some mathematicians are beginning to consider not publicly sharing their research directions to avoid being collected by AI companies for the next round of model testing question banks.
This incident follows OpenAI's announcement in September of solving the Navier-Stokes equation related Millennium Prize problem, marking the third similar shock Duminil-Copin has felt. He, along with other Fields Medal winners, signed an open letter calling for cautious AI use led by scientists.
Market Mechanism: OpenAI generates and publicly releases mathematical results on a large scale through its internal model, transforming open academic problems into demonstrations of model capabilities and research output. The event is driven by competition with rivals in cutting-edge mathematics and internal model capability verification; computational and research resources flow from the company's internal models to the public manuscript repository, benefiting AI companies with large-scale computing power and models (gaining proof of capability and influence), while putting pressure on mathematical researchers who rely on the long-term accumulation of open problems (facing rapid coverage of their research agendas and uncertainty in topic selection).
Source: Public Information
ABAB AI Insight
Hugo Duminil-Copin won the 2022 Fields Medal for his contributions to the theory of phase transition probabilities in statistical physics, with long-term research on percolation, the Ising model, and critical phenomena. His work is based on the academic tradition of openly discussing unsolved problems and gradually advancing proofs. OpenAI's mass production of manuscripts using its internal model directly covers several open problems he is concerned with, continuing its large-scale intelligent agent approach to the Navier-Stokes problem in September.
On the capital path, OpenAI has invested the computational power of its non-commercial internal model into about 4,000 mathematical problems, producing 719 manuscripts, some of which are formalized, aiming to demonstrate the model's capabilities in cutting-edge mathematics while influencing the academic agenda through public repositories; this is consistent with its previous investment of tens of thousands of intelligent agents and high computational costs on Millennium Prize problems, shifting research output from traditional academic publishing to model-driven bulk result releases.
In analogy, this is similar to the impact of early computer algebra systems on symbolic computation or AlphaFold on the field of protein structure prediction, rapidly replacing some long-standing work that relied on manual derivation; current mathematics is in the early transformation from human-led proofs to parallel model-generated results and human verification, with some fields shifting from exploring open problems to understanding and verifying AI outputs.
Essentially, this represents a transfer of pricing power under technological substitution: large-scale models transform open mathematical problems from the long-term accumulation of the academic community into results that can be rapidly generated by internal models, shifting discovery priority and research agendas from dispersed researchers to companies with computational power. The mechanism involves using public literature and unsolved problems as training and testing inputs to quickly output formal or semi-formal proofs, thereby compressing the time window for manual research and the value of career paths.
ABAB News · Cognitive Laws
Open problems are academic traditions and also training data for models.
When the speed of proof exceeds the speed of understanding, the research agenda is first covered.
Discovery priority shifts from humans to systems with computational power.