AI Is Now Producing New Mathematics. Deep Expertise Is Becoming a Career Edge.

OpenAI has published a broad range of new mathematical results produced by an internal frontier model the company has not released to the public. The collection — 722 manuscripts grouped into 372 result families — appeared in a public GitHub repository, drawn from an evaluation of around 4,000 open problems at an average compute cost of roughly three hours of ChatGPT Pro thinking per result.

Many of the proofs are formalised in Lean, a programming language that lets a computer verify a proof's logic, and OpenAI consulted the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study on how to release the results. The company itself adds two caveats: nothing has been peer reviewed yet, and results without a Lean formalisation could contain issues.

The claimed results span number theory, geometry, complexity theory and mathematical physics — progress on problems that mathematicians have worked on for decades.

For students, the career signal is easy to miss. The people who can judge, verify and extend work like this are mathematicians and domain experts, not prompt engineers.

Takeaway

Some of the most interesting AI careers may belong to people who understand something other than AI extremely well — mathematics, biology, physics, finance. Domain depth is what lets you tell whether an AI system's impressive output is actually correct.

Related: Knowing when AI is wrong is becoming a critical skill.