Good morning everybody.
I’m trying to implement a custom knowledge base chat.
I managed using gpt-index, I index all the documents I need and following the user’s question I get it from the index and send chatgpt to respond with this prompt:
Decide whether to answer based on the information below, or your knowledge. Information: {context} answer the following question: {message}
He responds correctly, but when I send him a message just as a complement he is not very “human”
Example:
User: What is Java?
Gpt: Java is a programming language
User: Cool
Gpt: I didn’t find anything referring to legal.
The problem is still legal, I would like it to be more humanized, he could say
Really cool, java is a very good language!.
There is a software that managed to solve this called chatbase
when I send him a “Cool” he gives more humanized responses.
Example of chatbase response:
“Hello! Glad to hear you found it interesting. Can I help you with any questions or additional information?”