Hallucination
When a language model produces output that sounds confident but is factually wrong or fabricated.
Definition
When a language model produces output that sounds confident but is factually wrong or fabricated. Hallucinations happen most when the model is asked about things outside its training, with no grounding documents, or under vague instructions. Designing systems that prevent or catch them is core operational work.
Example
Asked for a customer's order history with no database access, a model may invent plausible-looking order IDs. The fix is grounding: give it tool access to the real order system instead of letting it guess.
See it in context Learn how Hallucination fits into the bigger picture of how software actually works.
Read the Guide →