Event banner
Event details
As an Admin, how can we troubleshoot and assist the users when they use the same prompt and receive different response based on their access to the datasets. How can we assist and help them to point out the dataset for which they dont have access that provided them the different response than the other user?
JeremyChapmanMSFT
Microsoft
MicrosoftThe access to data would be only what that user of Copilot has access to. Copilot does not have and cannot grant access to files that the user in that session cannot access.
- Understood. How do we assist in case there is a complaint that the user mentions that he/she has access to same datasets similar to other user. However he/she gets different response from the same prompt?
- I'd explain that Copilot for Microsoft 365 is meant to mimic conversational, human interactions. A human subject matter expert is not going to give an identical word-for-word answer to two different people (unless they are acting like a robot reading from a script or cutting/pasting pre-canned content into the response box, and who wants that?). Also, a person will use different words/language/tone based on who they are talking to - such as a person's role. I'd respond very differently to a person from Finance versus the Sales team. If they want identical responses (similar to a human reading from a script), maybe consider building your own chatbot or custom Copilot?
- JeremyChapmanMSFTThat is the nature of the LLM completion. There will always be some slight variation in response. Also, a response will vary if there is a longer running session with multiple turns (prompts and responses), because that session context is also used for subsequent responses. Two users referencing the same file using the same prompt in new sessions should get somewhat comparable responses, but they won't be identical.
