Truth meter
Seven AI answers arrive in the same confident voice at the same high confidence — but only some are grounded in a real, checkable source; the rest are fluent fabrications. Judge each one, then see the trap: the average confidence of the true answers and the made-up ones is nearly identical. Tone is not evidence of truth. Runs entirely in your browser (0 uploads, works offline).
How to use it
- Read each question and its confidently-worded answer.
- Mark each Grounded or Fabricated on your best read.
- Score it — and see the confidence gap between the two piles, which is essentially zero.
What this clears up (the fundamentals)
- Confidence is generated, not earned — a model sounds equally sure when it's right and when it's inventing.
- Tone carries no signal — the true and fabricated answers here are delivered at nearly identical confidence, so fluency can't separate them.
- A source is the only reliable tell — grounded claims cite something checkable; fabrications say "no source" behind the same smooth wording.
- Verify specifics — precise numbers, names, and citations are exactly where confident fabrication hides.
Where it's used
An AI-literacy drill — the fastest way to internalise that confident output must be verified, not trusted. It's a miatz build-lab concept playable; learn to build it yourself.
FAQ
Why can't I tell truth from fabrication by tone?
Because a language model writes both in the same fluent, assured style. Confidence is a property of the writing, not of whether the claim is true, so tone gives you nothing to go on.
What is the "confidence gap"?
The average stated confidence of the grounded answers minus that of the fabricated ones. Here it's near zero by design — proof that how sure an answer sounds tells you nothing about whether it's real.
What's the reliable signal then?
Whether the claim carries a checkable source. Grounded answers point to something you can verify; fabrications don't, no matter how confident they sound.
Is anything uploaded?
No. The cards and scoring run entirely in your browser — nothing is transmitted or stored. Turn off your Wi-Fi and it still works.
Limits
A teaching deck of illustrative answers, not a live model — real fabrications are open-ended. The reflex it trains (distrust tone, verify sources) transfers directly.
Related
Part of the Demystify Playgrounds. Explore the rest from the Playgrounds home.
Bookmark this page (Ctrl+D, or ⌘D on Mac) — it works offline the next time you need it.