RL for LLMs

Reinforcement learning (RL) is how today’s chat and reasoning models are trained after pre-training: the model writes its own answers, gets a score, and makes high-scoring answers more likely. Each page pairs a real-world example with an LLM example. You only need basic probability.