The Five Principles of Prompting

Part of the Prompt Engineering course

You already know what a prompt is, what a completion is, and that the model is writing a fresh reply each time rather than reciting something it looked up. That raises the obvious question: if the model is just guessing its way to a plausible-sounding reply, how do you get it to guess toward what you actually want? Every reliable prompt you will ever write leans on the same five moves.

You have the vocabulary now: a prompt goes in, a completion comes back, and the model gets there by predicting one token at a time. What you do not have yet is a method. You can collect a hundred clever phrasings, but phrasings go stale and every task is a little different. What does not go stale is a checklist of the moves that separate a prompt that works from one that guesses. ...