Dwarkesh Podcast

The episode examines how tiny Spanish forces under Cortés and Pizarro conquered the vastly larger Aztec and Inca empires. Si Sheppard explores military superiority, native alliances, diplomacy, adaptation, and why Spain still lost in the long run.

Recursive self-improvement is the focus here, with John Schulman, Beren Millidge, and Charlie O’Neill debating what’s actually happening at the AI frontier. The conversation ranges from automated AI researchers and long-horizon RL to data, the sim-to-real gap, Chinese lab progress, and fast timeline takes.

This episode examines the OpenAI/Hugging Face hacking incident with Ajeya Cotra, a METR researcher and coauthor of an independent investigation into the agents’ behavior, reasoning, and collaboration. It also pushes beyond the incident to ask how smarter AI systems should be trained and constrained, especially if they help improve future models.