Techniques & Methods

One-Shot / Few-Shot

One-shot and few-shot learning refer to using one or a handful of examples to guide model behavior—either in training (meta-learning) or at inference time (in-context learning). LLMs demonstrate remarkable few-shot capability: include 2-3 examples in the prompt and performance on new tasks improves significantly.

Few-shot prompting is now a standard technique in LLM deployment. It bridges the gap between zero-shot (no examples) and fine-tuning (many examples), offering a practical middle ground requiring minimal data preparation.

Authority Links

Few-Shot Learning — Wikipedia

How few-shot learning works in large language models.

GPT-3 Few-Shot — arXiv

Original GPT-3 paper demonstrating few-shot learning capabilities.

Related Terms

Techniques & Methods

One-Shot Learning

Model's ability to learn and make accurate predictions from only a single example.

Techniques & Methods

Few-Shot Learning

Model's ability to generalize from only a handful of labeled examples.

Techniques & Methods

Zero-Shot Learning

Model's ability to correctly perform tasks it was not explicitly trained for.

Techniques & Methods

Prompt

Text input provided to an AI model to guide the content and format of its response.

One-Shot Learning Offline Reinforcement Learning