Chapter C · 8 pages · ~115 min

What is a large language model?

Build a practical mental model of how large language models learn patterns, generate text one token at a time, gain useful capabilities, and sometimes produce convincing mistakes.

Last reviewed 2026-08-08.

What you will be able to do

  • Explain next-token prediction without implying that a model literally thinks or retrieves exact training records
  • Trace how pretraining turns examples into reusable statistical patterns stored in model weights
  • Describe generation using tokens, probability distributions, temperature, and context
  • Recognize useful capabilities, in-context learning, hallucinations, and important operating limits
  • Select and use a model by balancing quality, speed, cost, privacy, context, and evaluation evidence

Lessons in this topic

  1. 01The next-token machine13m
  2. 02How pretraining builds patterns14m
  3. 03How generation unfolds15m
  4. 04Capabilities in context14m
  5. 05When fluency fails14m
  6. 06Worked trace: from prompt to answer16m
  7. 07Choosing and using models15m
  8. 08Mastery check: connect the pieces14m

Sources