Noam Brown – Agent swarms, alignment, & recursive self-improvement – Dwarkesh Podcast

Noam Brown – Agent swarms, alignment, & recursive self-improvement – Dwarkesh Podcast — Transcript

Market views

Bullish

OPENAI— OpenAI has achieved breakthrough capabilities with multi-agent systems solving Millennium Prize problems and is experiencing rapid internal acceleration with researchers spending thousands daily on AI coding assistance. The pace of progress is faster than even internal researchers expected, with capabilities advancing beyond 12-month predictions now requiring only 3-month horizons.
AIMODELS— AI reasoning models are demonstrating exponential capability improvements with test-time compute scaling, progressing from grade school math to Millennium Prize problems in just a few years. Multi-agent systems enable parallel scaling of cognitive effort, concentrating thousands of years of human-equivalent thinking into days.
CLOUDGPU— Compute infrastructure is a critical bottleneck for AI progress, with experiments requiring significant GPU resources. Even with brilliant researchers, having 100x less compute would result in substantially less progress, indicating compute remains a key constraint and value driver.
ROBOTICS— AI models are demonstrating immediate capability transfer to robotics applications, with raw models outperforming state-of-the-art robotics-specific models when deployed on mobile manipulators, suggesting rapid advancement in physical AI applications.

Bearish

STARTUPS— While AI helps individual founders, incumbents may gain structural advantages from AI alignment solving organizational misalignment problems. Large organizations with aligned AI agents could operate with startup-level coordination at massive scale, potentially reducing the traditional startup disruption advantage.

Not investment advice.