Glossary
Hallucination
When an LLM generates confident-sounding output that is factually incorrect.
Definition
A hallucination is an output from a large language model that is presented with the same confidence as accurate output but is factually wrong — invented citations, fabricated statistics, non-existent people or events. Hallucinations arise because LLMs predict likely-sounding tokens rather than retrieve verified facts.
Context
Hallucinations are the single most-cited reason enterprises hesitate to deploy generative AI. In practice they are managed — not eliminated — through retrieval-augmented generation (grounding outputs in verified sources), evaluation harnesses, and human review at high-stakes decision points. Modern models hallucinate less than 2022-era models, but rigorous grounding remains essential in enterprise settings.