One of these sentences is confidently lying
Ten short AI-style passages. In each, exactly one sentence is fabricated — a wrong number, an invented citation, a mechanism that names no real thing. Click the fake, then learn the check that would have caught it.
This passage reads like a confident AI answer. Exactly one sentence is fabricated — click the one you don't trust.
Each fabrication is written to be indistinguishable from the truth until the reveal — the reflex being trained is treating fluent confidence as a signal independent of correctness. The miss is felt, not just described.
Why do language models hallucinate?
A language model generates the most plausible next words, and plausibility is not the same thing as truth. Where the training data is thin, the fluent-sounding continuation wins — complete with confident numbers and official-looking citations.
What is a fabricated citation?
A source that looks completely real — plausible year, official-sounding journal, tidy author format — but does not exist. The reliable check is to search for the source itself rather than the claim; a journal that is not there ends the argument.
What is the fastest way to check a suspicious claim?
Three cheap moves: cross-check the numbers inside the same text against each other, push the stated mechanism one level deeper with a specific 'how', and treat absolute words like 'never' and 'always' as an invitation to find one counterexample.
This playable is part of the registry: its tool page
Next: see what your stack is hostage to
The Lock-In Detector audits your AI stack in eight questions and radar-charts exactly where the lock-in lives — with the mechanism and the swap for every red flag.