Event details
217 Comments
- Bing Chat Enterprise is told to be more secure and data will not be used. However the description says: it makes the personal information (user etc) anonymous, and sends its afterwards towards Bing Chat. Meaning data will leak from our tenant to an external server. How can we be sure that the "Bing Chat" service will not use this information for training the next generation GPT?
- JaredAndersen
Microsoft
Bing Chat Enterprise doesn't persist any chat data beyond the duration of your chat session. It's not logged, we don't have eyes on it, and consequently there is no path for it to be used as part of the training set for the underlying LLM. On top of that, during the chat session itself, we anonymize the user account and domain - after determining that they are a valid BCE account, which is what determines that we will handle the chat data differently. - This is an important question.
- Will Bing Chat Enterprise be applied to the Bing Chat button in Edge along with the upcoming Bing Chat button in Windows where the user is logged into both the machine and Edge with their E5 licensed accounts?
- JaredAndersenYes - If you have enabled Bing Chat Enterprise for your tenant and the user has signed in with AAD, then it will be available from bing.com/chat, from the Edge sidebar (clicking the Bing Chat button in the upper right corner) and from Windows Copilot.
- Is there any schedule when M365 Copilot will be available for preview by a broader audience? Will it be available in CDX Demo environments for showcasing the features to customers?
- KevinShermanWe do not have any additional information about when Copilot will be available to a broader audience, inclusive of CDX demo environments. The Early Access Program is our preview; please refer to (https://learn.microsoft.com/en-us/answers/questions/1292974/how-do-i-get-early-access-to-copilot-365) for more detail
- I uploaded a document in onedrive for business and asked bing chat enterprise to extract information from the document while providing the link from onedrive. It answered with information which was not from the document but seemed more like from a template. Like asking for the contact person mentioned in the document and it answered with something like "John Smith, ABC Company, Phone Number 123456789" which was nowhere in the document. When asked from where the information is coming from, it answers with the link from the document. Shouldn't it be able to extract information from documents from Office 365?
- JaredAndersenGood question - Bing Chat Enterprise is able to answer questions about information on the public web. It doesn't have the ability to answer questions about content inside the firewall or in the Microsoft 365 Graph. You would need Microsoft 365 Copilot to do that.
- According to the FAQ entry at https://www.microsoft.com/en-us/edge/bing/chat-enterprise?form=MA13FJ it should be able to "Summarize a work PDFs open in Edge: “Recap the findings of this internal cybersecurity report and the top 3 concerns” and “What are best practices for addressing these kinds of security issues?”"
- I am worried that BCE is "tuned" even more to make it safe, I hope not though. I had one answer that was way off base on creative but then on precise it got it right.
- I had a similar experience just telling it to summarize a PDF from a public web URL. I then asked why did you respond this way as that information isn't found in the PDF. Response of it also checked other sources online. Once I rephrased to specify just the PDF then I got the expected response.
Liliya_BWill BCE usage reporting be available in M365 Admin center?- JaredAndersenGreat question! We don't currently support any form of reporting in the M365 Admin Center, but that is a feature we've heard requests for and we're evaluating.
- Is Microsoft Copilot prone to hallucinations like BCE or are hallucinations not a concern?
- KevinShermanAs you probably know, hallucinations are a bit of a necessary evil of LLMs. The thing that makes them so creative is the same thing that can cause them to get things wrong at times. We're getting much better at reducing the frequency of hallucinations across the board. Additionally, Copilot is less likely to hallucinate than many chat-connected LLMs systems because it grounds prompts before calling the LLM. The grounding of those prompts adds the context and content needed to help the LLM get to a more accurate answer more often. That said, we do know that Copilot will at times be usefully wrong, and have built the UX and UI to help the user recognize this, and work with Copilot to get to a more accurate answer.
- Will it be possible to train Microsoft Copilot on a targeted and private dataset or is that a different AI service Microsoft has or is working on?
- You will find that capability in the Azure OpenAI services and associated studio
- But CoPilot will have access to your Sharepoint and OneDrive files, correct?
We are in the "Early Access Program" and are looking forward to evaluate the impact of Microsoft 365 Copilot.
Q1 : Differences between EAP and https://www.microsoft.com/de-ch/microsoft-365/roadmap?featureid=124840
Is the Feature Id 124840 coming in Nov. 2023 independently from the EAP and will be released from Nov. 2023 ?Are the functionalities in the feature 124840 also showcased in the EAP ?
Q2 : Extensibility / Configuration options
- What kind of configuration / extensibility options are available for the Copilot plugins ?
- Will it be possible to create our own Copilot plugins and augment the experience with our own custom data sources ?- Will it be possible to train the LLM on specific datasets ?
- Same for transformers for custom wordings / meanings
- Ben_SummersHi - I can answer Q2. There are two primary ways to extend copilot. First is to build plugins. The second is to build Graph Connectors. Yes, you can create your own Copilot plugins. There are multiple options for creating plugins. You can: Build Teams Message Extensions or Power Platform Connectors (or you can update your existing investments). You can also build/use OpenAI plugins. Fundamentally, a plugin is just a manifest + API, and that API can of course tap into your own custom data sources. You can also create or (or acquire) Microsoft Graph Connectors, which allow you to index additional data sources in the Semantic Index. This will also help Copilot understand things like acronyms used internally, etc. etc. To train an LLM on specific datasets, you'll need to build your own copilot - for example with Azure OpenAI services.
- Ben_SummersHi - I can answer Q2. There are two primary ways to extend copilot. First is to build plugins. The second is to build Graph Connectors. Yes, you can create your own Copilot plugins. There are multiple options for creating plugins. You can: Build Teams Message Extensions or Power Platform Connectors (or you can update your existing investments). You can also build/use OpenAI plugins. Fundamentally, a plugin is just a manifest + API, and that API can of course tap into your own custom data sources. You can also create or (or acquire) Microsoft Graph Connectors, which allow you to index additional data sources in the Semantic Index. This will also help Copilot understand things like acronyms used internally, etc. etc. To train an LLM on specific datasets, you'll need to build your own copilot - for example with Azure OpenAI services.
- When a document or SharePoint page is referenced in Copilot and BCE results, will there be any indication on the document and/or page to show that the data or material has been referenced? E.G. A user asks for more information about a particular topic, and Copilot/BCE pulls results from a SharePoint page, will that show in the usage analytics for that page?
- Ben_Summers
Certainly, if you click the reference link provided by Copilot. Will have to defer to others on the specific topic of Copilot counting as a "user" in this context.
About semantic index:
-Is there a way to determine the applications that the indexer can access?
-I understand that the index will have access to Sharepoint, can you confirm?
-How long does it take from when a new file appears in Sharepoint until it is indexed?
-Is there a way to exclude certain SharePoint sites from indexing without having to modify the permissions of all users or groups?
-Are there limits as to the size of data that can be indexed?
-Will there be tools to monitor this indexing?
-Will the indexer have access to OneDrive?
Thanks!
- Hey Diego, I can help answer some of the rest. Only data, across Microsoft 365 including SharePoint and OneDrive, a user has permission to access is indexed when Semantic Index is enabled. Microsoft 365 Copilot and Semantic Index do not change file sharing permissions or policies in place. Currently there's no indicator for monitoring status of indexing but we're determining what the right approach is.
- Ben_SummersHi Diego. I can answer some of these questions, maybe not all, so here goes. First, Semantic Index has access to SharePoint & OneDrive. Refresh rates will depend on the size & volume of data that must be updated/indexed. I don't have specific performance measurements at this point. Yes, you'll have the ability to exclude SharePoint sites from indexing on the Site settings. That's what I know. We'll have others from the team chime in with the rest of the answers.
