How Does Claude 4 Think? — Sholto Douglas & Trenton Bricken

Highly recommend listening to this episode where Dwarkesh talks with Sholto Douglas and Trenton Bricken from Anthropic about how Claude 4 thinks, RL scaling, and the future of AI agents.

Key Takeaways

• The discussion on mechanistic interpretability and how models develop “circuits” for different capabilities
• Their predictions about autonomous agents - by May 2025, we might see agents capable of complex multi-step tasks like advanced Photoshop work
• Insights into how reinforcement learning is scaling and what that means for model capabilities

YouTube link if you prefer video.