The process of building a language model from data.
An LLM starts with random parameters. Training means feeding it text and adjusting those parameters so it gets better at predicting what comes next. Do that with enough data and compute, and the model starts to understand language: grammar, facts, reasoning patterns.