Mathematicians Split Over OpenAI’s Large Batch of AI-Generated Proofs
Mathematicians are still working through a large set of AI-generated results on 372 open math and computer science problems released by OpenAI. Some researchers called the moment unprecedented and compared it with major breakthroughs, while others found the papers uneven, with some clear and others difficult to understand. The way the results were produced remains unclear and is a source of contention.
OpenAI’s release covers 372 unsolved problems in mathematics and computer science. Two days on, researchers are still sorting the collection. The company’s earlier model had recently addressed Navier-Stokes, a celebrated million-dollar problem. Observers note improved writing and citation practices compared with that proof, though quality varies widely.
Roughly 42 percent of results are formalized in Lean, making their logic likely reliable even when prose is unclear. Three papers were withdrawn after a major error surfaced, raising concern about the majority not yet Lean-checked. Critics also want OpenAI to publish the thousands of failed attempts, not just successes.
The release could reshape how mathematical research is conducted, affecting academics, journal editors, and students who rely on verified literature. If AI-generated proofs become common, trust and peer review may need new standards for formal verification and transparency. Broader publics might benefit from faster discovery, but uneven quality and withdrawn results could also spread confusion or false confidence. The episode may influence how AI companies share scientific claims.