AI glossary
Hallucination
When an AI model states something false or invented — a fact, quote, source or number — as if it were true.
Language models generate plausible text; they do not look facts up unless they are given tools or documents. When they lack the information, they may fill the gap with something that sounds right: a non-existent paper, a wrong date, an invented function in a code library.
To reduce it: give the model the source material (see RAG), ask it to quote or cite, allow it to say “I don’t know”, and verify anything important yourself. Read more in our beginner’s guide to hallucinations and privacy.
Example: You ask for “three studies that prove X” and the model gives three titles with authors and years. One exists; the other two do not. They sound credible because they mix real names with invented titles.
In practice
- Be most wary of exact figures, quotes and references without a link.
- Turn on web search for recent topics.
- For important work, check the fact against the original source before using it.
