Why Does AI Make Things Up?
Watch a confident-but-guessing answer next to an honest one — and pick who to trust.
Sometimes AI says something wrong — very confidently. Why? And how can you tell a good answer from a bluff? Let’s watch them side by side.
How deep would you like to go?
show me · the plain-English storyPick a question and watch a confident guess next to an honest reply. The guess might be totally made up; the honest one tells you what it does and doesn’t know. Then choose who to trust.
Confident, but guessing
“Exactly 79. I’m 100% certain!”
It picked one number and sounded totally sure — but the real count keeps changing as scientists find more, and its information might be old.
Honest
“Scientists keep discovering more — it’s roughly 95 known so far, but I could be out of date. A space website will have the latest.”
This one admits it might be wrong and points you somewhere to check. That honesty is what makes it trustworthy.
An AI predicts words that sound right — so sometimes it says something wrong very confidently. An AI that says “I’m not sure” is being honest, and that’s the good kind.
Remember: you can always talk to a grown-up you trust — a parent, carer, or teacher. You won’t be in trouble for asking. Asking for help is a smart, brave thing to do.
Wow — An AI that says “I don’t know” is being HONEST — and that’s the good kind. Confident is not the same as correct.
Try this — Ask each question and compare the two replies. Then pick which one you’d trust — and see why the honest one wins.
What's really happening · No live AI model is needed: the two answers are written out so you can compare the behaviours clearly and safely. The point is real, though — a language model predicts words that sound likely, and has no built-in sense of “I actually don’t know this”, so a wrong answer can sound just as sure as a right one. The trustworthy fix is to make an AI cite a real source or refuse — the “cite-or-refuse” idea behind vocabotics’ tools. Everything here runs locally; nothing is collected.