Are We Misreading the AI Exponential? Julian Schrittwieser on Move 37 & Scaling RL (Anthropic)

In this episode of The MAD Podcast, Matt Turck interviews Julian Schrittwieser, a prominent AI researcher from Anthropic, about the exponential progress of AI. They discuss the common misunderstanding of AI's trajectory, comparing it to the early stages of COVID, and Julian's predictions for AI capabilities in 2026 and 2027, including AI agents working autonomously for extended periods and matching or outperforming human experts in various occupations. The conversation covers the importance of task length as a metric, the role of benchmarks like GDP-Val, and the potential for AI to make novel scientific discoveries, possibly even winning a Nobel Prize. They also delve into the evolution of AI through AlphaGo, AlphaZero, MuZero, and the intersection of reinforcement learning (RL) and AI agents, discussing the challenges and benefits of combining pre-training with RL. Finally, they address the impact of AI on jobs, the importance of safety and alignment, and the need for political and economic solutions to distribute the benefits of increased productivity.