Hallucination

My wife and I have a term for a certain kind of person: CBR. Confident But Wrong.

You know the type. Gives you directions with total authority, zero hesitation, and the restaurant they're sending you to closed twelve years ago.

That's AI hallucination.

The model isn't looking things up the way you'd Google it. It's predicting the most confident-sounding response, one piece at a time. When it has the answer, that confidence is earned. When it doesn't, it usually won't stop and say so. It generates something that looks exactly like a right answer. Plausible. Detailed. Specific. Wrong.

That's what makes it sneaky. A hallucination doesn't look like a mistake. It looks like an answer, in the same calm voice as everything true it's ever told you. The fake citation sits right next to the real one wearing the same outfit.

Here's the moment it clicked for me. I asked it for the exact name of a Shopify metafield in my own store. It gave me one that looked perfect. Right format, right wording, the kind of thing I'd have typed myself. Didn't exist. I'd never written it down anywhere, so it just made one up that sounded right. (Lawyers have done the same thing with court cases and gotten fined thousands. Same trick, higher stakes.)

So the fix is simple: verify anything that matters. Names, numbers, quotes, "did you actually do this." If a wrong answer would cost you, check it yourself.

None of this means the tool is broken. It means you treat it like the brilliant, fast, slightly overconfident intern it is. You'd never forward an intern's research to a client without a glance. Same energy here.

CBR. Confident But Wrong. Now you'll see it everywhere, in your AI and in people.