In-Context Learning
Domain: LLM Concepts
🎯 Motivation
The model learns from information provided in the prompt itself - examples, instructions, context - without updating its weights. It’s temporary, just for this conversation.
📋 Concrete Example
Give the model 3 examples of translation, then ask it to translate a new sentence. It learned the pattern from the examples in context - but this learning vanishes when the conversation ends.
🔗 Analogy
Learning rules for a board game by watching one round - you understand how to play from watching (in-context), but you haven’t memorized the rulebook (no weight update). Next game, you need to watch again.