AI
Hallucination
A hallucination is generated content that is plausible and confidently stated but factually incorrect or unsupported by any source, arising from a model's objective of producing likely text rather than verified text.
Grounding output in retrieved sources, requiring citations, and constraining responses to supplied context materially reduce the rate without eliminating it.
For any application where accuracy matters, the design assumption should be that some output will be wrong, and the system should make that detectable.