You’re offline. This is a read only version of the page.
31
Need help! This is my first time creating a chatbot on Copilot Studio.
Under Generative AI, I have attached 1 pdf document as the source of data.
Under Topics > Conversational Boosting, there's a "Create Generative Answer" flow that gets triggered when intent of user's message is unknown. Refer to image below. It's supposed to search for data in the PDF and reply the user accordingly. If it fails to provide any output, it'll give a standard apology message that you can see in the image as well.
For some reason, it always fails to give an output (i.e. the answer would be a blank and it'll send the apology message.
I looked at the analytics under "Boost Conversations" and noticed that it fails 50% of the time and the main reason is "no search results" (84% of the failures).
Anyone knows why this happens? How do I fix it?
Thank you in advance!
Yeah this is happening to me also as of now, no matter what I try I get no response from Generative Answers. Raised a support ticket, unsolved. This looks like it's been an issue for months.
I replied to another thread with some debug info if a dev can have a look (cc @HenryJammes ) over here: https://powerusers.microsoft.com/t5/General/No-Generative-AI-answers-from-public-website-URLs/m-p/2775490#M11503
Not sure if this will resolve your issue, but after months with the Microsoft engineers across all support platforms it turns out that MCS uses a 30 second client time out and no retries when use the OOTB solution to hit AOI and custom data, compared to PA's 120 second http action and 8 retries by default. So when we received this information we tried to hit a 3.5-turbo deployment instead and lowering the input query, and voila it started working... Still in converastion with the MS engineers since out customer want to use GPT-4 as we have gone a head and do through power automate, but would prefer to use the OOTB solution in MCS...
@HenryJammes Copilot bot which I have created is hallucinating like hell. I am using SharePoint to upload 5 files. Most of the times it is working like google. The queries which I tried had nothing to do with my data, however, were valid and had answers when you google them. And I got the same results what google, or any other search engine would have given me.
I tried custom instructions as well but no luck.
E.g. query1- How do I get PAN details in India
Bot: please try rephrasing it
query2: Do you know a way of getting my PAN details in India.
Bot: "Replies with complete set of steps"
Hi @Tjan ,
When using custom data and your Azure OpenAi connection, have you experienced that all queries return Blank answers? which automatically triggers the Fallback topic.. Calling AOI through any other method and even with a HTTP inside MSC works, but the built in Generative text always return Blank answers even on Low mode...
Hi Henry,
Extremely sorry for the late reply. Haven't had time to visit the community.
To answer your question: I am using Sharepoint and 'Custom data' as data sources. The latter one being related to: https://learn.microsoft.com/en-us/microsoft-copilot-studio/nlu-generative-answers-custom-data#use-custom-data.
I've gone a head and registered a ticket with MS support to understand what is going on...
The fact that it worked very well at first, then completely stopped working with MSC after the adjusted pointer of AOI to Production really has be wondering... (still working in Playground and Postman...). And now won't even work after re-directing to the original test-setup...
I am experiencing similar issues as well.
For context I am new to copilot, but not to Azure OpenAI.
I have indexed several (94) documents that I am able to retrieve information from fine using openai chat playground, postman, proprietary code calling the API - all return the expected results.
I have also used the same model and deployed to A New Web App and it works flawlessly.
After creating the copilot following this guide my testing method has been to ask a question that references a specific document. Copilot refuses to retrieve information from the document. As suggested in this thread I have lowered the Content Moderation with no success. I have also uploaded the stand alone document following this guide and I do get the expected results.
The document itself is very specific in nature. For an example the document is called "How to get to the store.pdf" In the document is a summary and bullet points:
"To get to the store
1. take 3 steps this way 2. turn left 3. etc etc etc."
Odd behavior is if I ask copilot to tell me "what is in the document How to get to the store.pdf" it will return exactly what I am expecting as a result:
"In the document How to get to the store.pdf it states To get to the store
1.take 3 steps this way 2. turn left 3. etc etc etc"
But if I ask "How do I get to the store?" It thinks, then fails. Again - this query works fine from other interfaces.
Unfortunately, if this is the intended behavior, this makes copilot virtually unusable from a teams perspective. Unless I've missed something, or failed to implement something within copilot itself.
Any advice or tips or guidance would be greatly appreciated.
Thanks,
C
We have been trouble shooting a CopilotStudio Copilot's Conversation boost for a couple of days. It worked flawlessly at first integration with an Azure OpenAi instance and Cognitive search index. Then just stopped working and now only returns blank answers that trigger the fallback "I’m sorry, I’m not sure how to help with that. Can you try rephrasing?" @HenryJammes any related releases or known bugs/limitations here as of this week?
Now we are not able to get it going again, not in the current Subscription resource group setup or any other even public enabled resouces... Both Azure open ai studio playground with same deployment and data model setup works as expected and Postman requests return correct data and citations.. only MSC Copilots that suddently completely switch off... Local uploads in the Copilot still work but is not of interest here.
Hi @Tjan -- curious about the low moderation getting a high moderation. Are you using a public website data source I could use to try and repro? Feel free to IM me details.
My bad. You are right.
#1
Romain The Low-Code...
25
#1
Pablo Roldan
25
#3
stampcoin
10
Share
Report
All responses (
Answers (
