Few-Shot
Providing a small number of input-output examples in the prompt so the model can imitate the pattern.
In plain language
Few-shot performance depends on example quality, not count. Examples should cover edge cases, formatting details and confusable categories, and match the real input distribution. Common failures include all-positive easy examples that break on negatives, or inconsistent formats that teach contradictory patterns. Too many examples consume context and cost; two to five high-quality examples are usually enough.
Example
For support intent classification, include edge cases such as “change address after shipping,” not just refund and logistics, or the model will force them into existing categories.
Recommended tools
Related terms
In-Context Learning
A model learning a task temporarily from instructions and examples in the prompt without updating its weights.
Prompt Engineering
Designing instructions, examples and constraints so a model reliably produces the desired output.
Zero-Shot
A model performing a task without any examples of that task in the prompt.
Agent Memory
A mechanism for an agent to store and retrieve user preferences, past tasks and key facts across sessions.