Generative AI · Established · Intermediate

In-Context Learning

A model's ability to adapt to a new task from information in the prompt alone, without any weight updates.

What In-Context Learning is

In-context learning is what makes few-shot prompting work. The model recognises the pattern in the context and continues it, effectively performing task adaptation at inference time.

How it works

No parameters change. The mechanism is still an active research question, with evidence that attention implements something resembling implicit optimisation over the provided examples.

Why it matters

It is the property that makes one general model usable for thousands of bespoke tasks, and the reason context engineering matters so much.

Common uses

→Few-shot prompting
→Dynamic task definition
→Runtime personalisation

Strengths

✓Instant adaptation
✓No training infrastructure

Watch for

✓Limited by context window
✓Less durable than fine-tuning

Continue exploring

Few-Shot Learning Prompt Engineering Context Window