All terms

Hallucination

When an AI model states something false or fabricated as if it were fact.

Catching hallucinations is a core part of evaluation work: does the model cite a paper that doesn't exist, invent a legal precedent, or state a wrong number with total confidence? Flagging hallucinations accurately is exactly why domain expertise matters - a non-expert often can't tell a confident wrong answer from a correct one.

Example

An AI cites "Smith et al., 2019" as the source for a claim, but no such paper exists - a domain expert catches it because a non-expert would have no reason to check.

Common questions

The model generates text based on patterns, not a live fact-check against a source of truth - that's exactly the gap human evaluators are hired to catch.

Related terms