LLMs
Hallucination
A hallucination is a fluent model output that is unsupported or false in the relevant context.
How it works
Language-model likelihood rewards plausible continuation, not factual verification. Grounding, tools, constraints, and evaluation reduce risk but do not eliminate it.
A concrete example
Illustrative: a model invents a paper title and DOI when the retrieval result contains no such source.
Do not confuse it with Confabulation in humans
model hallucination describes unsupported output, not a claim about human cognition.