AI glossary
What is an AI hallucination?
A hallucination is when an AI model generates false or made-up information and presents it with full confidence, as if it were a real fact.
It happens because an LLM does not look the answer up in a database: it generates the text that seems most likely. If it does not have the fact, it can still produce something shaped like a correct answer, such as a quote, a date or a reference that does not exist.
It is more common with very specific questions, with recent facts or with topics it saw little text about, and when the prompt pushes it to answer no matter what.
It can be reduced, though not eliminated, by giving the model its sources (RAG), asking it to say when it does not know and checking what matters. In Tokencito, an assistant that makes up answers is one of the systems you get to fix.
An example
Asking a chatbot for three books on a topic with their ISBNs and getting convincing titles that do not exist is a classic hallucination.