What is hallucination evaluation in large language models?

It measures how often models generate factually incorrect or unsupported content.

Hallucination evaluation quantifies factual inconsistency between model outputs and ground truth or trusted sources. It includes intrinsic hallucinations (contradictions within text) and extrinsic hallucinations (unsupported external claims). Evaluation methods include fact-checking pipelines, retrieval-based verification, and LLM-as-a-judge frameworks with grounding constraints.