My users are now getting info they shouldn’t be and when I click on the investigate button is says “Knowledge base not used”.
How is this possible?! The bot is literally providing old links that are no longer in it’s knowledge base. How is it doing this?!
Hi @Pristine,
“No chunks used” indicates that no knowledge base chunks were appended to the user’s prompt on the backend. The LLM is most likely making up links or retrieving outdated ones from its training data.
Please note that we don’t train LLMs — we only use RAG currently.