A plain-English breakdown of how large language models learn, from training data to token prediction. No PhD required.