Mathematicians marvel, and grumble, at OpenAI’s trove of new results
Mathematicians react with awe and criticism to OpenAI's release of AI-generated solutions to 372 open math and computer science problems.Summary
Two days after OpenAI released AI-generated results on 372 open math and computer science problems, scientists are sifting for breakthroughs. Harvard mathematician Michael Douglas calls it the start of a new era, bigger than last month's Navier-Stokes solution by OpenAI's internal model. However, the quality is mixed: some papers are well-written with improved citations, while others are "AI slop" that is hard to read, though 42% are Lean-verified, ensuring sound logic. Three papers were withdrawn due to errors, raising concerns about the unverified half. Critics, including Andrew Sutherland of MIT, urge OpenAI to also release failed problems for transparency. Some suspect the model generated its own problems. The Association for Human Mathematics condemned the release as a display of power, urging mathematicians to stop working with OpenAI.
(Source：Scientific American)