All this complicated math and no one was left sobbing at the kitchen table. This week, OpenAI dumped solutions to 372 of the world’s previously unsolved problems across number theory, algebraic geometry, topology, and other fields most of us have no business trying to understand. The release has mathematicians spiraling as AI models power through problems some have spent their lives working on. In “C+ in Pre-Calc” terms…this may be one of the biggest advancements in mathematics history, and it was announced via a blog post. OpenAI’s 722 pages of findings included advancements in three of the five Millennium Prize Problems, notoriously difficult math problems that come with a $1 million prize if you solve them. The field of mathematics was already riled up about AI. Last month, OpenAI said it solved the Navier-Stokes Equation, one of the Millennium Problems. OpenAI said it wouldn’t claim the money, but that didn’t quell the drama surrounding the development. - Mathematicians are concerned these powerful models have been trained on their research, creating a conundrum over whether the AI is actually thinking through complex problems creatively or just plagiarising at super speed.
- And the AI’s proof used a controversial loophole and an “approach that many experts find unnatural,” according to Scientific American.
Is this all just marketing?AI companies previously used high school math problems (still hard!) as benchmarks for their models, but as their LLMs advanced in sophistication, so have the equations. But it still looks pretty good as a headline when their models can solve a problem that has stumped humans for nearly a century. Your professors are probably mad about this: Mathematicians have warned that the speed racing of research is detrimental to the field—even the independent AI math advisory board OpenAI brought on after the initial controversy has told the company to stop trying to solve math problems. Twenty-five winners of the Fields Medal (the most prestigious award in the discipline) signed a letter on Sept. 25 accusing AI companies of rushing to publish uncredited proofs, warning this could push mathematicians to keep their research private, which would impede progress.—MM | 
