Artificial Intelligence Resolves Mysteries that have puzzled mathematicians since 1946 — and they’re Shocked!

Over the past few months, OpenAI, Anthropic, and Epoch AI have solved various longstanding math problems such as the Unit-Distance Conjecture (since 1946), multiple Erdos problems, and even managed to solve FrontierMath (twice), which was explicitly crafted to be insoluble by AI models through brute force. Meanwhile, more than 3,000 mathematicians have signed the Leiden Declaration, raising concern over the future of the field.

1What Was Actually Solved

OpenAI refuted the Unit-Distance Conjecture, which had baffled mathematicians for decades. A group of researchers from Carnegie Mellon used SAT solvers and AI-generated code to solve a Ramsey theory problem. Epoch AI solved FrontierMath (twice), which consists of a collection of olympiad/research level math problems explicitly crafted to be unsolvable by AI through brute force.

2The Community Reacts: From Optimism to Alarm

AI is like “a diligent and tireless graduate student” says Terence Tao, who believes we’re entering a “golden age” of cooperation between humans and AI. On the other hand, Timothy Gowers argues that “the mathematical culture is being eroded” because “there will be nobody around who knows how to appreciate the beauty of the proof once a machine has generated it”. None of the six Millennium Prize Problems ($1M prize for each) have been solved by AI thus far.

Math has become the first “pure” discipline where AI is consistently surpassing the top human performers in their own domain. This will fundamentally alter the way mathematicians work, but also change our understanding of what constitutes mathematical understanding and the need for the human role in mathematics.