Advance Transformations Lead to More Hallucinated Responses
Advance Transformations Lead to More Hallucinated Responses
At a glance
The community member who posted the original post noticed that when using more advanced transformations, the responses they receive are more "halucinated". They guess that this is because they are feeding the transformations to the response generation part as well.
In the comments, another community member suggests that for most cases, they would only keep the transformations for embeddings.
There is no explicitly marked answer in the provided information.
i am asking this as i noticed in the case that i am using some of the more advance transformations the reponses are more halucinated, and my guess is as i am feding them to the reposne generation part as well!
