Prompt

No jargon, no math

A prompt is the complete text you give the model in a single call, and the model does one thing with it: it continues it. That text usually stacks three layers: a system prompt with the rules of engagement, any context you choose to paste (documents, data, earlier messages), and your actual request. The model has no task menu and no memory of yesterday: it answers from exactly what is in front of it, matching the patterns your text suggests. This is why the same model feels psychic with one prompt and useless with another. Vague input gets a generic answer, because a generic answer is the safest pattern match. Missing context gets invented or ignored. And anything you leave out of the window simply does not exist. Writing a good prompt is less about magic words and more about doing the framing the model cannot do for itself: who is speaking, what material is on the table, what shape the answer must have.