Decoder. plain-English AI glossary

In-context learning

● Core

Also called ICL

The model picking up a new skill on the fly from whatever you put in the prompt — without any retraining.

Think of it like

A quick learner who adapts to a new task just by reading the brief, without going back to school.

Example

Give it a few English-to-French translations, then a new English sentence — it translates, even though you never fine-tuned it on French.

How it actually works

The model adjusts its behavior based on the patterns in its current context — no gradient updates, no parameter changes. This emergent ability is one of the most surprising properties of large models, and it's why prompt design has such outsized impact.

For product teams

The reason prompt design matters so much — the model adapts on the fly from whatever you feed it.

For engineers

Emergent task adaptation from prompt content without gradient updates; mechanism debated.

Few-shot — Few-shot learning is the most common form.
Why context engineering matters so much.
Inference — Happens inside the context window at inference time.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open Decoder Add to Chrome