What is Few-shot Learning?

Machine Learning 4 min read

Definition

Few-shot learning is the capability of large language models to understand and perform new tasks after seeing only a small number of examples (typically 1-5) provided directly in the prompt. This is a emergent property of large models trained on diverse data, enabling rapid task adaptation without fine-tuning.

Comparison with Other Approaches

Method Examples Needed Training Required
Zero-shot 0 None
Few-shot 1-5 None
Fine-tuning 100s-1000s Yes