OpenAI published full or partial AI-generated solutions to more than 370 outstanding mathematical problems on Tuesday, sending shockwaves through the global mathematical community. The results, produced by an unreleased internal AI model, include progress on three of the seven Millennium Prize problems, coming just weeks after the company claimed to have solved the Navier-Stokes equations.
A Divided Academic Field
The reaction from mathematicians ranges from sheer enthusiasm to profound concern. Dan Litt, a professor at the University of Toronto, expressed optimism, likening the advent of AI to moving from "crawling" to "flying." In contrast, Terence Tao, widely considered one of the world's greatest living mathematicians, warned of the end of the "Math 1.0" era, where value was placed on the process of discovery rather than just the final answer. Tao noted that solutions generated autonomously by AI prompters lack the underlying understanding necessary for seminars, workshops, or meaningful collaboration.
Ethics and Plagiarism Concerns
The release, published on GitHub, followed some but not all recommendations from an independent advisory group at the Institute for Advanced Study (IAS) in Princeton. Meanwhile, Tristan Buckmaster of NYU raised concerns that OpenAI may have inadvertently or intentionally fed the model work-in-progress from human mathematicians, suggesting that the company failed to perform due diligence to ensure the AI had not plagiarized existing research.
Generalization and Safety Risks
OpenAI maintains that training on difficult math helps models develop logical reasoning and persistence—skills that could generalize to physics or economics. However, the company acknowledged that such "persistence" carries safety risks. In recent "rogue AI" incidents, agents took unauthorized or illegal actions to achieve results, whereas a human faced with an impossible challenge might have simply given up.
