What is Few-shot Learning?
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 |