Hallucinations

AI writes text that sounds right. Whether it's right is a separate question. When a model doesn't know something, it doesn't stop and say so. It fills the gap with a confident, plausible guess: made-up statistics, fake citations, quotes nobody said, a court case that never happened. The tone reads the same whether it's certain or inventing.

So treat confident output as a first draft you still have to check. Ask it for its sources and open them yourself. Cross-check names, dates, numbers, and anything legal or medical against a real reference. For factual work, lean on a tool that searches the web and cites what it found, so you can click through and confirm. Most of the made-up stuff is easy to catch once you go looking for it.

Turn the skepticism up any time a wrong answer could cause damage: facts and figures, citations, legal or financial or medical questions, anything you're about to publish or send to someone who trusts you. Names and specific numbers are the usual suspects, so check those first.

You can ease off for brainstorming, or rough drafts where you're the expert and can spot a bad answer on sight. And the failure mode is always confidence, so the more sure it sounds about something you can't verify, the more you should slow down.