For more Microsoft 365 Copilot news, capabilities across more apps, and details on the Microsoft 365 Copilot Early Access Program, check out www.microsoft.com/en-us/microsoft-365/blog/2023/05/09/introducing-the-microsoft-365-copilot-early-access-program-and-new-capabilities-in-copilot/
If our data is not used to train your LLMs, how is it going to get better & evolve over time? Wouldn't it get left behind after some time, as the others are likely using user data to continue to train their models?
I appreciate the approach of "copilot," acting more as a teacher and a tool for inspiration. As a teacher myself, it is lovely to see the methods we use in the classroom being applied here as a teaching method rather than a means to shortcut the work. I have used chat GPT for a while now, and this is far more valuable as an application because it does not give me exactly what I ask for; instead, it encourages me to create it myself. The ability to change the settings of the copilot is also beneficial, as it allows me to learn from its suggestions rather than regurgitate them. Thank you, Copilot team, for making a better AI tool that respects the artists in the world while still providing assets that can help improve talent.
Nah - it'll just make us lazy and dumb(er). The time saved won't be put to better use, it'll just be gaps in someone's day. People are inherently lazy.
Thanks for your comment. We're working on it and recently published details on the Copilot Early Access Program www.microsoft.com/en-us/microsoft-365/blog/2023/05/09/introducing-the-microsoft-365-copilot-early-access-program-and-new-capabilities-in-copilot/
언제 출시될까요? 올해안에 나온다고 소문이 있던데요. 생산성이 얼마나 더 좋아질까요? 회사에서 일하는 방식은 얼마나 달라질까요? When will it be released? I heard it will be released within this year. How much more productivity do you get? How different is the way you work at a company?
Shalom! I am curious about: What about the copyright the books and articles, etc? What about the legal status of the books and the articles used by the LLM? The books that are still under copyright protection require the permission of the author. How Google did obtain the permission from the right holder?
I would like to know why the different email clients best classify emails by conversation by subject (subject), but never by email address (possibly with the possibility of subsequently linking several email addresses to the same person or the same organization ). It would be much more practical to find a previous conversation with a person or from an organization, and simpler than having to go to the search bar, type the name of the person, then copy their email address, then type this address email in the search bar.
Will Microsoft Co Pilot also be integrated into their Office software for macOS? Is Microsoft having discussions with Apple how to integrate their LLMs within macOS and iOS/iPadOS/watchOS etc?
That's not available with Microsoft 365 Copilot. You can check out Azure OpenAI Service for additional customizations with your own in-house developed applications.
@@MSFTMechanics thank you for your reply, we are using AZ OpenAI and also Fine-tuning and user-friendly RAGs as well. Where I can find more detailed about m365 isolated orchestrator?
Thanks for your comment. We've recently broadened access with the Microsoft 365 Copilot Early Access Program. You can find out more here: www.microsoft.com/en-us/microsoft-365/blog/2023/05/09/introducing-the-microsoft-365-copilot-early-access-program-and-new-capabilities-in-copilot/
I would like to know how you can gather enough relevant data, while maintaining all proper user permissions, to generate a learning model that can actually 'learn' given all the constraints to the graph structure of available data sets. It seems to me, this will be limited to connecting existing data points and generating content in basically the same way. While I do see the potential for this to be a powerful tool, I see no other way - it would essentially require all user data to be collected to satisfy the necessary and sufficient conditions for a robust and 'creative' model.
Thanks for your comment. Copilot will be supported initially in English, Spanish, French, German, Italian, Japanese, Portuguese, and Simplified Chinese.
Thanks for your question. Yes, Dynamics 365 will also have Copilot capabilites. You can find out more in this blog: blogs.microsoft.com/blog/2023/03/06/introducing-microsoft-dynamics-365-copilot/
Thanks for your comment. There are lots of built-in intelligent experiences in Power BI powered by GPT. Check out this blog: powerbi.microsoft.com/en-us/blog/enabling-intelligent-experiences-with-power-bi-for-developers-data-scientists-and-data-engineers/
No worries. Power Platform actually started rolling out GPT powered capabilites quite early in 2021. Power Apps Ideas to automatically generate Power Fx is another example using GPT.
Unfortunately I’m unable to find anything in writing from Microsoft stating that that LLMs don’t use our proprietary data for machine learning or that our data is not available to the AI outside our tenant. The assurances in this video are nice but Microsoft themselves have said no such thing in writing. If anyone has seen it, I’d love a link. The AI is promising but not worth it if our trade secrets end up in the public domain.
Thanks for your comment. Here are a few links with Microsoft's commitments in writing: learn.microsoft.com/en-us/microsoft-365-copilot/microsoft-365-copilot-privacy; techcommunity.microsoft.com/t5/microsoft-365-blog/administration-of-microsoft-365-in-the-new-era-of-ai/ba-p/3767079
Thanks for commenting and great question. That is part of the Copilot orchestration for information retrieval using semantic search. There is a great how-it-works video about that here: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-KtsVRCsdvoU.html
@@MSFTMechanics Thank you for responding. One of my hats says informaticist. I also come into contact with eminent researchers who use the answers from large questionnaire surveys to further our understanding of certain critical events. When I point out that any undergraduate student proffering a questionnaire as illogical as the one I have just been asked to complete would receive a C-, at best, I am airily informed that AI would fix the issues that concern me. This is a serious issue that needs to be addressed, I would rather take the specifics off-line, so you can better understand how unrealistic many people's expectations of AI are.
Thanks Robin. You might be interested in retrieval approaches and logic that we covered in a related show: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-tW2EA4aZ_YQ.html and for more details on general semantic search, check out: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-d_6ZNyV1MvA.html
If we have several hundred docs about the same subject (say HR benefits detailing co pay) where data changes over time how does it ‘know’ which is the most accurate? Thanks
Thanks for your comment. This is part of the retrieval process and uses semantic search. There's a great high level overview here: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-KtsVRCsdvoU.html
Thanks for your comment. We haven't done a Mechanics show on Security Copilot yet, but I'd encourage you to check out this one on Threat Intelligence in Microsoft Defender: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-YaZWK-S-lYI.html We explain how you can access both finished and raw intelligence for cybersecurity investigation - and demonstrate how it works.
At the start you said it can gather information from your files and emails. Clearly that isn't done every time you start a chat so you have to be storing something
Thanks for your comment. As demonstrated, each Copilot session or Business Chat conversation starts from scratch and information is retrieved each time on demand.
Thanks for your comment. We actually explain that in the video. Retrieval permissions and access are based on the user using Microsoft 365 Copilot; information is retrieved via the orchestration engine using the Microsoft Graph to append to a prompt as temporary information specific to that conversation, then once the conversation ends, the temporary information is also removed. The process then repeats for any new conversation.
We're very excited about what lies ahead, thank you for sharing. 🚀 This morning we also held a webinar in German on our assessment of Copilot. For those interested: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-4wT5WQWO3ic.html
In practice, copilot will behave like an undergraduate who writes a research paper for you, riddled with errors, and your job will be to carefully sift through and find all of those errors.