Skip to main content
All terms
Prompting

Few-Shot Learning

Giving a model a handful of examples in the prompt to teach a task on the fly.

Definition

Few-shot (or in-context) learning provides several worked examples directly in the prompt so the model infers the pattern and applies it to a new input — no training required. Zero-shot uses none; one-shot uses a single example. It works because large models can generalize from patterns shown in their context.