Loading…
Glossary term
When a model produces fluent but false or unsupported information.
A hallucination is output that sounds confident and plausible but is false or unsupported, like an invented citation, a made-up statistic or a wrong date. Models generate likely-sounding text, and nothing in that process checks it against reality.
Grounding answers in retrieved sources, asking models to admit uncertainty and verifying claims all help.
It is the main reason AI answers should be checked, especially for medicine, law and finance.