Hallucination
A confident-sounding output that is false or not supported by the sources the model was given.
What it means
In AI, a hallucination is an answer that sounds right but is invented or unsupported. It can be a made-up case citation, a wrong date, or a figure that appears in no source. Hallucinations are hard to spot because the wording reads like expert writing.
A Stanford study of legal research tools found hallucination rates between 17% and 33% for the products it tested, using tools as they stood in 2024. Rates depend on the task, the sources and the product version. The practical defense is a check step: require a source for each claim, open the source, and treat any claim you cannot trace as unverified. Hallucination rates are useful only when the test set, the date and the product version are stated alongside them.
Continue this reading path
Cite this entry
https://aipoweredresearchtool.com/glossary/hallucination/