Hallucinated nonsense is easily prevented by restricting output to verbatim content from a pre-vetted knowledge base. Think fuzzy search for people who prefer t

Hallucinated nonsense is easily prevented by restricting output to verbatim content from a pre-vetted knowledge base. Think fuzzy search for people who prefer to ask questions instead of listing keywords. I doubt this is what they've done here, but not all AI chatbots have to be thin wrappers around a large language model.