When AI solves problems no researcher could crack for 30 years
OpenAI's AI models have proved three major unsolved problems in maths, computer science, and physics — and it means AI is no longer just a tool for routine work.
- 3
- major unsolved problems solved in one announcement
- 30 years
- how long the maths conjecture was open
- $1M
- prize for each Millennium Prize Problem
- $852B
- OpenAI's valuation
Last Monday, OpenAI announced that its AI models had solved three major unsolved problems — in mathematics, computer science, and physics. These are problems that some of the world's best researchers had worked on for decades without success. AI is now doing original science, not just helping with tasks humans already knew how to do.
In maths, the AI proved a 30-year-old conjecture about how waves behave, and it did this for every possible dimension at once. In computer science, it disproved a famous assumption that hundreds of researchers had relied on for their own work. In physics, it produced the first rigorous proof for a kind of quantum theory that is part of one of the seven Millennium Prize Problems — math problems so hard that any proof wins $1 million. Expert groups are still checking all three results, so none are fully confirmed yet.
OpenAI is careful about what it claims. It calls the AI's outputs 'raw research data.' Human expert groups are now the ones deciding what the results mean and how to publish them. No academic credit goes to OpenAI. The AI produced the raw material; humans are still the interpreters. That matters: this is not AI replacing scientists, it is AI producing work that scientists must verify and understand.
The real point for builders and businesses is not about science prizes. If AI can crack problems that stumped entire fields for 30 years, it will also start solving problems that your company has quietly accepted as too hard or too expensive. Right now, most businesses use AI to speed up things they already knew how to do. The cost of difficult thinking is about to fall.
Start looking at the problems you have quietly shelved — the ones that seemed to need a rare expert, a six-figure consultant, or just luck. If AI can prove a 30-year-old theorem, it can probably also find the stubborn bug, draft the argument you thought required a specialist, or model the scenario your team never had time to run. The real shift is in what becomes affordable, not just what becomes possible.