Lezing Nebius: Reinforcement Learning in Modern LLMs

Gratis

---

How do LLMs move from “predicting the next token” to actually solving problems? A big part of the answer is Reinforcement Learning. We’ll start with a quick recap of RL fundamentals, then explore how these ideas play out in modern LLM systems. Along the way, we'll share our hands-on experience with agentic RL, verification, the real-world challenges we’ve hit, and the infrastructure we built to solve them. To wrap things up, we’ll walk through a practical example of training a model using our own RL platform.