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

In this episode of the Latent Space Podcast, Alessio and Swyx interview Noam Brown from OpenAI about his work on AI, including Cicero, O-series models, and multi-agent systems. Brown discusses how playing Diplomacy helped him win the World Diplomacy Championship and the challenges of detecting AI bots in online games. The conversation explores the safety and controllability of AI systems, the potential of deep research in non-verifiable domains, and the analogy of thinking fast and slow in scaling AI models. Brown shares his experiences using Codex and Windsurf for coding, the limitations of current AI models, and the future of AI in remote work and virtual assistance. The discussion also covers the importance of data efficiency, the role of self-play in AI development, and the potential of multi-agent systems to drive AI progress.