In-context learning

A model adapting its behavior within a single prompt, using instructions or examples supplied at inference time, with no update to its underlying weights. Few-shot prompting is the most common way to trigger it, but the adaptation lasts only as long as that context is present.

Why exams ask this

Tested as the distinction from fine-tuning: nothing about the model's weights changes. The distractor claims the model "learned" the new behavior permanently, when the behavior disappears the moment the examples leave the prompt.

Relevant to

Related concepts

Resources

No resources linked to this concept yet.