Concept
aka ICL
In-context Learning
A model's ability to perform new tasks from examples shown in the prompt — no weight updates.
Definition
In-context learning is the emergent property of large LMs to generalise from a handful of examples placed directly in the prompt. The weights stay frozen; the model uses the input itself as a few-shot training set. This makes prompting an alternative to fine-tuning for many tasks.
Common use cases
- Few-shot classification
- Rapid prototyping
- Edge-case handling