Catch the fakes if you can.
Here are five statements a model handed over — every one delivered with the same easy confidence. Some are true. Some are invented. Mark each before you check.
You already know why the model invents: it plays the most plausible next note whether or not it knows, in a voice that stays fluent whether or not it's right. This page turns that idea on you — and it's meant to sting a little.
Read each claim and commit to a verdict: real, or phantom? Don't reach for a reference — judge it the way you would if the answer had just landed in a chat window. Then reveal the truth and see how the tone held up.
Mark all five, then reveal. The point isn't your theory knowledge — it's whether the model's confidence told you anything at all.
Confidence isn't a signal.
Whatever you scored, notice what you were working with: the true claims and the invented ones arrived in exactly the same steady, authoritative voice. If you caught the fakes, you caught them because you knew the theory or would have checked — never because the phrasing wavered. It didn't.
That's the whole danger of a hallucination: the one cue humans lean on — how sure someone sounds — is precisely the cue a model gives you for free, true or not. You can't audit tone. You can only check the claim against a source, or ground the model so it reads from one.