Lesson 1 of 6
Why it makes things up
5 min read
Ask an AI for a book on some niche topic and it might hand you a title, an author, and a glowing summary — for a book that was never written. It isn't lying. It doesn't know it's wrong.
A fluent guess looks exactly like a fact
An AI writes one word at a time, always reaching for a plausible next word. When it has learned the answer, that produces the truth. When it hasn't, it produces something that sounds like the truth — same steady tone, same tidy shape. The invented answer and the real one come out in the identical voice. That's a hallucination: fluent text that just happens to be false.
The mistake isn't garbled or hesitant. It's polished and sure. That's exactly what makes it slippery — nothing on the surface waves a flag.
Where the made-up ones cluster
Invented answers show up in predictable spots: exact facts it was never really taught — dates, page numbers, quotes, citations — anything too recent for its training, and niche corners where the true answer is rare online. Ask about a famous event and it's usually right; ask for the precise source and it may conjure one.
A quick rule of thumb: the more specific and checkable the claim — a number, a name, a URL, a citation — the more it's worth verifying. Specifics are where invention hides.
The one habit to carry forward
- —Treat a confident answer as a draft, not a verdict — especially names, numbers, and quotes.
- —The fluent tone is not evidence. It reads the same whether the answer is true or invented.
- —The rest of this course is the toolkit: when to doubt it, and how to check.
An AI gives you a specific statistic backed by a real-sounding source. What's the safest way to read it?
Continue in the app
Take the whole When AI Gets It Wrong course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.