What AI is — and is not
The everyday meaning, the older research meaning, and the claims this desk will not make.
Index
Explainer articles
Five pages. The first three are the core path; the last two add the training/inference split and a method for reading claims.
Models versus products
A trained model is not the chat app, the API bill, or the safety filter sitting in front of it.
Risks without hype
Confident errors, privacy, scams, and over-trust. No invented probabilities of doom.
Training and inference
Why “it learned from the internet” is not the same event as “it answered you just now.”
How to read an AI claim
Demos, leaderboards, and press releases — what to ask before you repeat a number.