Prompt Craft
Steering, in practice
Everything in the previous modules turns into one practical skill: giving the model a context that makes the answer you want the most probable one. Role, context, examples, format and constraints each move the odds.
A prompt is not a command
It is tempting to think of prompting as learning the magic words. It is more useful, and more accurate, to think of it as setting up context, because context is the only thing that determines what becomes likely.
The model is not deciding whether to obey you. It is producing the text that most plausibly follows what it has been given. Every word you add changes what plausibly follows. That is the entire lever, and it is a much bigger one than it sounds.
Notice how much of the request you did not write. The product's own hidden instructions, the conversation so far and any files attached are all in there, all shaping the answer, and some of it may be working against what you just asked for.