Skip to content
DKKB

Hallucination

Section titled “Hallucination”

Hallucination is model output that presents unsupported or incorrect content as if it were a valid continuation or answer.

The term covers several failure types, including contradiction of supplied evidence and unsupported factual claims.

Hallucination is an evaluation problem, not a single model switch. Grounding, retrieval, constraints, and verification can reduce risk but do not remove it.

Evaluate LLM behavior and hallucination riskMeasure model behavior against representative tasks and verify externally checkable claims instead of treating fluent output as evidence.