Lesson 3 of 9
Why does it make things up?
5 min read
Ask for a source and it may hand you a book that doesn't exist — author, title, page, all invented. Why does a helpful tool lie so smoothly?
Built to sound right
Because it predicts likely text, it produces what a good answer would look like — whether or not it's true. When it doesn't know, it doesn't stop; it fills the gap with a plausible-sounding guess. That's a hallucination: fluent, confident, and wrong.
It optimises for plausible, not for true — so gaps get filled with fiction.
Why it won't just say 'I don't know'
Saying 'I'm not sure' is itself a skill the model has to be trained toward, and it's imperfect. Newer models hedge more, but none are immune. The fix is on your side: treat surprising facts, names, numbers, and quotes as claims to check, not answers to trust.
Confidence is free; correctness isn't. Verify the specifics.
Never ship an AI's facts, citations, or figures unchecked — especially names, dates, and quotes, which it invents most fluently.
Staying safe from made-up facts
- —Hallucinations come from predicting plausible text.
- —It fills gaps instead of admitting it doesn't know.
- —Verify specifics — names, numbers, quotes, sources.
Why does AI sometimes state false things so confidently?
Continue in the app
Take the whole AI, Explained course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.