4.13M · Mechanism●●○○○
In-context learning: teaching without training
Put a few worked examples (or a rule / rubric) in the prompt and behavior changes on the spot — no weights updated. The context is a scratchpad the model re-reads from scratch on every call.
You'll get more from this if you've seen2.1Predict the next word2.6From predictor to assistant
1You want ratings in YOUR exact format. No code, no training — just a prompt.
The wall
You have movie reviews and you want each one rated into 1–5 stars — but in YOUR exact house format. You can't retrain the model. Your only tool is the text you send it: the prompt.
the format you want
★★★★☆ — 4/5
Stars, then an em dash, then n/5. That precise shape — nothing else.
weights: unchanged
space play/pause←→ stepR replay
Common questions
What is "In-context learning: teaching without training" about?
Put a few worked examples (or a rule / rubric) in the prompt and behavior changes on the spot — no weights updated. The context is a scratchpad the model re-reads from scratch on every call.
What problem does it solve?
You need the model to tag reviews as 1–5 stars in your exact format, but you can't retrain it. Are you stuck with its defaults?
What will I be able to do after this lesson?
You can explain in-context learning: examples or instructions in the prompt steer behavior without changing weights, and why that powers RAG, tools, and system prompts.
What comes next?
If the prompt can teach it, the prompt is precious — yet the model keeps nothing between calls.
If the prompt can teach it, the prompt is precious — yet the model keeps nothing between calls.
5.1 The model is frozen and stateless