PodcastIntel
Sign in Get Started Free
Latent Space: The AI Engineer Podcast
Latent Space: The AI Engineer Podcast

Scaling Test Time Compute to Multi-Agent Civilizations — Noam Brown, OpenAI

Jun 19, 2025 · 1h 17m
AI Summary
  • Noam Brown's work on solving Poker and Diplomacy demonstrates test-time compute scaling with multi-agent reasoning; System 1/2 analogy oversimplifies—reasoning models work differently than fast/slow thinking in humans
  • Reinforcement fine-tuning and long-term model adaptability outperform over-reliance on scaffolds and routers; fragility in tool use remains challenge for production agent systems
  • Multi-agent intelligence hypothesis suggests emergent civilization-like behavior; implicit world models, memory reuse, and PR review processes needed for better AI developer collaboration

Guests on This Episode

NB
Noam Brown
1 podcast appearance

More from Latent Space: The AI Engineer Podcast

View all episodes →

Get AI Summaries for Every New Episode

Subscribe to Latent Space: The AI Engineer Podcast and get AI summaries, guest tracking, and email digests delivered automatically.

Sign Up Free →