Skip to content
MEVZULARI
MEVZU N°124ISTANBUL
Glossary · Intermediate · 2020

In-context Learning

The model's ability to learn a new task purely from information in the prompt, without parameter updates.

EN — English term
In-context Learning
TR — Turkish term
Bağlam-İçi Öğrenme

In-context learning is an LLM's ability to perform a new task purely from instructions and examples in the Prompt, without any updates to the model's weights. GPT-3 (2020) made it clear that this capability emerges sharply at large scale, and the modern LLM ecosystem was built around it. Zero-shot and Few-shot are the two main modes of the paradigm. What mechanism produces it — meta-learning, an implicit form of gradient descent, or something else — is still active research, and one of the most-studied questions in Mechanistic Interpretability.