Hallucination

A hallucination is output from a language model that is fluent and confident but factually wrong, fabricated, or unsupported by the provided sources. Because models generate the statistically likely next token rather than verified facts, they can invent citations, names, or details that never existed.

Definition

Hallucination

Hallucination arises because a language model optimizes for plausible continuations of text, not for truth. When the model lacks reliable knowledge or the prompt invites confident completion, it may produce content that reads correctly but is not grounded in reality — including invented references, quotes, or statistics.

Common mitigations include grounding the model in retrieved sources (RAG), asking it to cite evidence, lowering reliance on parametric memory for factual claims, and using evaluation and self-critique to flag unsupported statements. No single method eliminates hallucination entirely, so verification of high-stakes output remains essential.

Frequently asked questions

What is Hallucination?

A hallucination is output from a language model that is fluent and confident but factually wrong, fabricated, or unsupported by the provided sources. Because models generate the statistically likely next token rather than verified facts, they can invent citations, names, or details that never existed.

Sources

Related terms

← Full AI & prompt engineering glossary

Put Hallucination to work with Prompeteer, the Agentic Contextual AI Platform →