How large language models are trained, explained in plain terms
Large language models learn by predicting the next token across vast amounts of text, then are fine-tuned with human feedback. Here is each stage, drawn from the original papers and NIST guidance.