A “shot” is a worked example you give the AI inside the prompt before asking it to do the real task. Read a Federal Reserve statement and predict the rate decision — hike, hold, or cut — then reveal the real outcome to check. You are not retraining the model; you are teaching it in the prompt.
Describe the task and ask. The model leans on general knowledge — fast, but can miss domain cues.
Give one input→answer example, then ask about a new case. One example can sharpen the answer.
Give several labeled examples first; the model learns the pattern — often the biggest jump, still no coding.
Keep the real outcome hidden from the model until it answers — the predicted-vs-actual check is the whole point.