Training in one line

Predict the next token, billions of times.

An LLM learns by being shown text with the next word hidden, guessing it, and nudging its parameters when wrong. Repeat across trillions of tokens and skill emerges.