prompt and completion

The prompt is what you give the model — your question, instruction, or the start of some text. The completion is what it gives back — the continuation it generates. If you type write a haiku about rain, that line is the prompt, and the three-line poem that follows is the completion. Everything you do with an LLM is some version of this pairing.

To the model there is really only one stream of text. It reads your prompt as the context and then keeps predicting tokens to extend it; we simply call the part you wrote the prompt and the part it produced the completion. Because the prompt is the only steering wheel, how you word it strongly shapes the answer — the skill of wording it well is called prompt engineering.