Is there any wat or best approach of preventing variables in model responses? e.g: [insert dealership business hours]
Hi Ana, welcome to the forum.
One of the best ways would be to provide it the information that you need to it as a part of the initial prompt and then ask it to only use this information to generate the corresponding text. If you are using the API, you would be able to play with temperature setting to ensure that the generation is upto standards while not having any false information.
I am using content management system which is stored as a json object, and using embeddings extract info and attach to initial prompt. however sometimes I get variables
Getting variables after decoding the embeddings into the initial prompt or in the final generation ?
I am getting variables, such as [insert dealership business hours] in final generation.