Whats best way to get confidence score of a response ?
I have PDF data inside a vector database.
now i ask it questions.
how can I know the response is not hallucinating and AI model is confidence thats its giving a response
1 Like
Not really possible, I believe.
Best to warn users to check data, that it might not be correct, etc.
This will change as the models continue to improve, but for the moment, there’s nothing like this in place.
does Langchain or some vector DB offer this service ?