Jargon:Hallucination
1 min read
Say this at the party
“When an AI states something false with total confidence, like a made-up fact, quote or source.”
In plain English
Language models are built to produce likely-sounding text. Most of the time, likely and true line up. Sometimes they don't, and the AI fills a gap with something that sounds right but isn't.
It doesn't know it's doing it. A made-up court case or book title comes out in exactly the same confident tone as a real one.
Newer models do it less, especially when they can search the web and show their sources. But it hasn't gone away.
Where it came from
Researchers used "hallucination" for translation and image systems that invented details long before chatbots. ChatGPT made it everyday language, and Cambridge Dictionary picked "hallucinate" as its word of the year for 2023. In 2025, OpenAI researchers argued it happens partly because training rewards a confident guess over "I don't know".
See it in action
A lawyer asks an AI for cases that support their argument and gets six, with names, dates and quotes. Two don't exist. Courts around the world have now dealt with hundreds of filings like this.
What people get wrong
- It isn't lying. The AI has no intent. It's guessing, and it doesn't flag the guess.
- A link or citation isn't proof. Open the source and check it says what the AI claims.
Where Ugenie fits
That's one reason for the AI Council: Ugenie asks several AIs at once and flags where they disagree.
Get Ugenie →Go deeper
- PaperKalai et al., OpenAI, 2025Why Language Models HallucinateWhy training pushes AI to guess instead of saying "I don't know".
- PaperJi et al., 2022Survey of Hallucination in Natural Language GenerationThe big overview, from before ChatGPT.
- GuideDamien CharlotinAI Hallucination Cases DatabaseA running list of court cases with made-up AI citations.
Where it comes up
Related jargon
Drafted with AI, checked by the AI Council (Claude, ChatGPT and Gemini) and signed off by Phin.