prompting

A prompt is simply the text you hand a language model to get it going. The model has already read a huge amount of writing, so it does not need to be retrained for each task; it just continues the text you give it in the most fitting way. Prompting is the craft of choosing those opening words so the continuation is the answer you actually wanted. Ask vaguely and you get a vague reply; ask with a clear goal, the right context, and the shape of answer you expect, and the model usually meets you there.

Mechanically, the model never sees your request as a command. It sees a sequence of tokens and predicts what tokens should come next, one at a time. Your prompt sets the scene that makes a good answer the most likely continuation. That is why the same model can write a poem, fix code, or translate a menu with no change inside it: only the prompt changed. Learning to prompt well is mostly learning to be specific about the task, the audience, and the format, and to give whatever facts the model could not otherwise know.