Decoders
Plain English for the concepts that keep coming back. Written once, revised when the world moves, dated with the last review.
Context window
How much the model can hold in mind at once.
Reviewed 2026-09-20
Agents
A model given tools, a goal, and permission to take steps on its own.
Reviewed 2026-09-20
Evals
The tests that say whether the thing actually works for your job.
Reviewed 2026-09-20
Retrieval-augmented generation (RAG)
Look the answer up first, then let the model write it.
Reviewed 2026-09-20
Hallucination
Confident, fluent, wrong — and structurally unavoidable.
Reviewed 2026-09-20
Open weights
The model file itself, downloadable and runnable on your own hardware.
Reviewed 2026-09-20
Inference cost
What each answer costs, and why the pilot was cheap and the rollout is not.
Reviewed 2026-09-20
Sovereign AI capability
A contested phrase covering everything from a local data centre to a national model.
Reviewed 2026-09-20
The EU AI Act risk tiers
Four tiers, and the one that matters to government is high-risk.
Reviewed 2026-09-20
Shadow AI
Staff are already using it, and a ban mostly moves it onto their phones.
Reviewed 2026-09-20