0:00 / 3:12Chapter 1 · Text in, numbers out1×CC≡

Transcript · follows along

1. 0:00: A model never sees letters. Before it reads anything, your text is chopped into pieces.
2. 1:10: Common words get one piece. Rare words get split into chunks the model has seen before.
3. 2:10: You pay per token, and the context window is counted in tokens, so it pays to know them.