What is Hallucination?

Full Definition

Hallucination is the phenomenon where a language model produces output that sounds plausible and confident but is factually wrong, internally inconsistent, or entirely made up. Examples include invented citations, fabricated statistics, non-existent people, or incorrect dates. Hallucinations arise because LLMs are trained to predict likely-sounding continuations rather than to retrieve verified facts. Mitigation strategies include Retrieval-Augmented Generation (RAG), which grounds the model in cited source documents; chain-of-thought prompting, which forces explicit reasoning steps; and grounding the model's outputs with tool calls to verified databases. Understanding hallucination is essential for any production AI deployment, particularly in domains like medicine, law, or finance.