Prompting & In-Context Learning

zero-shot prompting

Zero-shot prompting means you just ask, with no worked examples attached. You write something like translate this sentence into French, or is this review positive or negative, and let the model answer straight away. It is the most natural way to talk to a modern assistant, and for common, well-understood tasks it often works on the first try because the model has seen thousands of similar requests during training.

The word zero refers to the number of demonstrations in the prompt: none. The model leans entirely on what it already learned, plus the wording of your instruction. This is fast and cheap, but it is also where clarity matters most: with no examples to imitate, the model fills any gap with its best guess about what you meant. When a zero-shot answer is wrong or oddly formatted, the usual fixes are to sharpen the instruction or to switch to few-shot prompting and show it one or two examples.

Also called
zero-shot