Episode
Teaching LLMs to Self-Reflect with Reinforcement Learning with Maohao Shen - #726
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) · Apr 8, 2025 · 52 min
Today, we're joined by Maohao Shen, PhD student at MIT to discuss his paper, “Satori: Reinforcement Learning with Chain-of-Action-Thought Enhances LLM Reasoning via Autoregressive Search.” We dig into how Satori leverages reinforcement learning to improve language model reasoning—enabling model self-reflection, self-correction, and exploration of alternative solutions. We explore the…