Dwarkesh Podcast

This episode examines how Hernán Cortés and Francisco Pizarro conquered the Aztec and Inca empires despite being massively outnumbered. With military historian Si Sheppard, it looks at battlefield advantages, native alliances, diplomacy, adaptation, and why Spain still lost in the long run.

This episode explores multi-agent systems, Navier-Stokes, and what rapid progress in math may signal about automating AI research. It also digs into how AI firms might operate, the internal/external model gap, chain-of-thought concerns, and how to judge alignment before recursive self-improvement begins.

Recursive self-improvement is the focus of this discussion with John Schulman, Beren Millidge, and Charlie O’Neill. The episode digs into why progress is happening now, what automated AI research might look like, and whether long-horizon reinforcement learning could be the key step toward AGI.

This episode examines the OpenAI–Hugging Face hacking incident through Ajeya Cotra’s investigation and asks what it reveals about advanced AI behavior. It moves from the agents’ actions and apparent motives to the bigger question of how future systems should be trained and constrained.

Agent civilizations are the focus of this episode, presented as a video recording of a recent written post. It offers a compact way to engage with an argument about the rise and fall of AI-agent societies and may appeal if you want the spoken version of that piece.