Noam Brown – Agent swarms, alignment, & recursive self-improvement
Summary
Noam Brown discusses OpenAI’s multi-agent systems, highlighting their role in solving a Millennium Prize problem by scaling test-time compute in parallel. He explains how these agents coordinate spontaneously without rigid scaffolding, resembling human collaboration but at vastly higher speeds. The conversation explores the implications for recursive self-improvement, noting that while AI accelerates research, bottlenecks like experimental runtimes prevent an immediate intelligence explosion. Brown also addresses the "jaggedness" of current AI capabilities and the rapid, often surprising pa…
Listen ad-free on Castria