Glossary
AI hallucination
When a large language model confidently generates false information — usually plausible-sounding but factually wrong.
Hallucinations occur because LLMs predict likely text rather than retrieve verified facts. The model has no internal concept of "I don't know." Common in: citations (fabricated paper titles), legal cases (made-up precedents), historical details, current events past the training cutoff. Mitigations: retrieval-augmented generation (the model cites sources it actually searched), explicit "I don't know" training, and external verification. For news consumers, the rule is: never treat an LLM answer as primary source. Always demand the citation and verify it independently. See also: primary source, fact-checking.
Related terms