Unanswered

AI-Guided Chatbot without OpenAI Data Connection

(0) Share

Report

Posted on by BennyS

Hi community!

I have currently a small assignment to create a Virtual Agent Pilot. I want to create a chatbot for FAQs which seems to be a standard idea for this.

However, creating hundreds of topics and hardcoding them doesnt seem like a good solution.

The generative answers solutions for Power Virtual Agents seems promising, however, data is leaving the tenant which does represent a data risk.

What would your recommendations be to solve that approach? Is there any "non-OpenAI" solution to combine text understanding in the chatbot?

Thank you!

Categories:

General topics

I have the same question (0)

All responses (12)

Answers (0)

Sort by

MattJimison 579 Most Valuable Professional on at

Like (0)

Report
Copy link

Link copied!
Hi, @BennyS .

Fortunately, there are a lot of options to keep your data in your tenant, as the OpenAI-based features in PVA utilize Azure OpenAI, which is different than OpenAI, and your data isn't used for training models, making it a much more secure option.

Check out this page for more details on how you can leverage your company's SharePoint Online and/or Azure data with the Generative AI features:
https://learn.microsoft.com/en-us/power-virtual-agents/nlu-boost-node

This includes the following internal sources:

Azure OpenAI on your data

SharePoint

OneDrive for Business

Custom data (internal or external): Supply your own content from any source, for example, from a Power Automate Flow, a Skill, or other source.

Find this post helpful? Please mark it as the solution and/or provide kudos so that it will help others in the future.

Cheers,
Matt

Was this reply helpful? Yes No
BennyS 8 on at

Like (0)

Report
Copy link

Link copied!

Thank you Matt!

The most concerning part for me is the "abuse monitoring", that MSFT employees are able to access prompts and responses at every time. In another forum, someone told me that access to the data has to be approved by the business every time. Is this also applying to the OpenAI for VA?

In Short:
In Microsoft OpenAI Services (e.g. generated responses in Virtual Agents), do MSFT employees need the business approval when conducting abuse monitoring deep-dives?

Was this reply helpful? Yes No
Jameslew_MSFT Microsoft Employee on at

Like (0)

Report
Copy link

Link copied!

Thanks BennyS for your question; given the nature of it, I'd like to clarify further.

By "the business", do you mean the customer whose application was related to the "abuse", or do you mean Microsoft's "business" to approve such a thing? Can you be more specific?

In general, for Power Platform and Azure services, no individual operator has standing access to production customer data; multiple gates exist which prevent this; specific hardware, requirement for access to be approved by a designated team member, and membership in specific security groups, along with demonstrated need for a specific case. For an abuse monitoring type of issue there would be additional constraints applied.

Was this reply helpful? Yes No
MattJimison 579 Most Valuable Professional on at

Like (0)

Report
Copy link

Link copied!

Hi, @BennyS .

This is a great resource to help answer your questions:

https://learn.microsoft.com/en-us/legal/cognitive-services/openai/data-privacy

That includes a link to the below form where you can request that your data not be monitored at all:
https://customervoice.microsoft.com/Pages/ResponsePage.aspx?id=v4j5cvGGr0GRqy180BHbR7en2Ais5pxKtso_Pz4b1_xURE01NDY1OUhBRzQ3MkQxMUhZSE1ZUlJKTiQlQCN0PWcu

Find this post helpful? Please mark it as the solution and/or provide kudos so that it will help others in the future.

Cheers,
Matt

Was this reply helpful? Yes No
BennyS27 40 on at

Like (0)

Report
Copy link

Link copied!

Hi @Jameslew_MSFT
Thank you for your fast response!

With the business I am referring to us - so the customers using the capabilities. Is our approval required for MSFT to get access for abuse monitoring? I read this in another thread and I would like this approach, but somehow I think that this doesnt make sense.

I guess major concern that there IS an access to our data for MSFT which we cannot restrict/monitor/avoid due to the abuse monitoring scheme.

Was this reply helpful? Yes No
BennyS27 40 on at

Like (0)

Report
Copy link

Link copied!

Another question in the same context.
The GPT Functionalities are expected to be rolled-out in Europa on Monday - the functionality still is not visible in our environment, nor can I enable it in the "settings" and there is still the "preview" stuff listed.

Does this functionality require any special license? Which licenses are required for this functionality?
How long does it take to be rolled out to our tenant?

Thanks!

Was this reply helpful? Yes No
HenryJammes Microsoft Employee on at

Like (0)

Report
Copy link

Link copied!

Not sure where that information comes from, but Generative AI features in PVA, like Generative Answers, Plugin Actions, and Copilot are not expected on Monday to become available in more regions outside of the US. There is active work to make this happen as soon as possible though, so stay tuned.

Was this reply helpful? Yes No
BennyS27 40 on at

Like (0)

Report
Copy link

Link copied!

Hi @HenryJammes
thank you for your answer!

I found the release plan here: https://releaseplans.microsoft.com/en-us?app=Power+Virtual+Agents&status=new it is stating that it has been rolled out in october for general availability.
Deep-diving in the release plans leads to this article:
Use internal SharePoint content as a source for AI-created generative answers | Microsoft Learn
This article is also stating "Geographic Areas" including Europe.

The development timeplan in this article General availability deployment - Power Platform | Microsoft Learn indicates 2nd October for Germany etc and 16th October for Europe.

Did I get the information here wrong?

Edit: Our goal is to use an internal sharepoint information via generative answers for a smoother chatbot building experience

Was this reply helpful? Yes No
HenryJammes Microsoft Employee on at

Like (0)

Report
Copy link

Link copied!

Hi @BennyS27,

Thanks for raising this, we clearly need to update our release plans to reflect the latest dates. It's not far from now, but it isn't October.

Regarding the "General availability deployment" this is about activating the "2023 Release Wave 2" in your environment in case you haven't done so already. What this effectively does is enabling the "early access features". Generative answers isn't one of them.
You can get see list of early access features here: 2023 release wave 2 features available for early access | Microsoft Learn.

Henry

Was this reply helpful? Yes No
BennyS27 40 on at

Like (0)

Report
Copy link

Link copied!

Thank you for your response again and sorry, I guess I am getting confused now.

The functionality has been in preview (US only) before. This is exactly what we are currently seeing in the VA Portal.
According to my understanding, "General availability" is the next release phase after "public preview". Consequently, the documentation reveals that this functionality has taken this step towards GA and is officially in this release wave as general availability.

We already have this wave enabled in our environment but this did not change anything. Could it be that Microsoft is rolling out this release in waves across tenants and this can take some time still because we're not one of the first to get it?

Quote: "In 2023 release wave 2, we're expanding the breadth of sources that your bots can use to generate answers and including internal SharePoint sites as a source."
So this all does not sound like early access, more like a generally coming and also maybe already available feature.

I'm currently just hoping that it will take maximum 2-3 weeks from now on to be officially available, because we have this pilot planned internally to evaluate the capabilities in our CoE.

Was this reply helpful? Yes No