Few-shot prompting

Including a handful of worked examples in the prompt before the real task, so the model infers the expected format and reasoning style from demonstration rather than instruction alone.

Why exams ask this

Tested against zero-shot as a "when do you pay the extra tokens" tradeoff. Examples steer format and style more reliably than an instruction alone, but each example costs context window budget the model could otherwise spend on the real input.

Relevant to

Related concepts

Resources

No resources linked to this concept yet.