What is hallucination?
A confident model statement that is false or made up, such as an invented citation. Spotting them is a core part of evaluation work.
A hallucination is a confident statement from an AI model that is false or made up. Common examples are invented citations, wrong figures, court cases that never happened and functions that do not exist in a library.
Why it happens
Language models generate the most plausible next words, and a plausible-sounding citation is not the same as a real one. Without a way to check, the model has no signal that it is wrong.
How trainers catch them
Fact-checking and evaluation tasks ask you to verify each claim against a reliable source and mark anything unsupported. Experts are hired precisely because a hallucination in a specialist field reads as correct to a layperson. Writing prompts that reliably trigger hallucinations is also valuable work on its own.
Example
Asked for research on sleep and memory, a model cites a 2019 paper with real-sounding authors and a journal name. The evaluator searches for it, finds no such paper, and marks the answer as containing a fabricated citation.
More terms: quality and review
Full glossaryReady to look at roles?
Every listing shows its pay and who it is open to.