Hallucination

A model producing confident text that is not true. It is the default behaviour of a text predictor, and the reason a model's answer must be checked against a source.

• See also: Retrieval augmented generation
• See also: Evaluation
• Used in: LLMs

Back to the glossary