Using the Azure Open AI connector for Conversation boosting.
We sometimes get different answers for the same question, I think it might be because it uses the context of the previous questions. Is there a way to limit the conversation to the latest question or can we give the use an option not to include previous conversation as context?
Second issue the citations sometimes do not contain the content of the reply. I also played around with the .net implementation, the citations were inaccurate, not sure if they use the same in the background. I will raise it on the azure sdk git register too.
1. It's the nature of GPT models that their answers can be different. It's not possible to alter the context sent. You need to design the solution, so that it takes into account that this is the way that this technology works. :-)
2. The citations, at least for the text content, contain only the reference to the text which was used, but it's based on the index search. Sometimes it's quite quirky to find the information that was used to generate the answer.
All in all - let's wait for it to be more mature!
In case of any other questions, let me know. If the answer helped you, mark it, so that others can benefit from it.
Thank you for your reply! To ensure a great experience for everyone, your content is awaiting approval by our Community Managers. Please check back later.