Hallucinations refer to the phenomenon where an AI model generates convincing-sounding but factually incorrect information. Since LLMs are not fact databases but calculate probabilities, they can present inventions as true. Countermeasures include retrieval-augmented generation (RAG), source citations, and human oversight ("human in the loop").
In short
What are hallucinations in AI?
Hallucinations are convincing-sounding but false answers from an AI model. Since language models do not store facts but calculate probabilities, they can present inventions as true. Countermeasures include RAG, source citations, and human oversight.
Using AI in your organisation?
We show you which applications are worth it for your organisation.