flâneur

Begin Proof — Noam Brown - YouTube

youtube.com · 9,286 words · saved by 1 readers

Noam Brown discusses the evolving capabilities of reasoning models in mathematics and the potential for these systems to bridge disparate fields of study. This conversation explores the transition toward superhuman performance, the importance of general-purpose AI development, and how human-AI collaboration may shape the future of mathematical research and discovery. Zeropercent breaks down the fundamental techniques of formal mathematical reasoning. The lesson covers logical operators, quantifiers, set theory, and direct and indirect proof methods using clear, structured examples. Ellie Sleightholm demonstrates how a single, specific instance can invalidate a mathematical statement. By examining a claim regarding odd and even integers, they show the simplicity and impact of disproving a proposition.

Last month, an OpenAI model disproved a long-standing conjecture by Paul Erdős on the planar unit distance problem, producing a construction so intricate that verifying it took a team of external mathematicians a week. In the inaugural episode of Begin Proof, Slater Stich sits down with Noam Brown, who leads work on multi-agent reasoning and test-time compute at OpenAI, and was one of the key people behind o1, OpenAI's first public reasoning model. They talk about what the Erdős result actually means, where the models are superhuman and where they still aren't, and what working…

saved by

related reading