Friends: Many of you have been asking at both #MSAITour, #CTTT24 and other events exactly HOW does the M365 Copilot work? Like how does it ground in data from our graph, how do we expand the content through graph connectors, where does Responsible AI fall in, etc. I absolutely love the graphic below which can be summarized as: 1. Copilot receives an input prompt from a user in an app, such as Word or PowerPoint 2. Copilot then pre-processes the input prompt through an approach called grounding, which improves the specificity of the prompt, to help you get answers that are relevant and actionable to your specific task. The prompt can include text from input files or other content discovered by Copilot, and Copilot sends this prompt to the LLM for processing. Copilot only accesses data that an individual user has existing access to, based on, for example, existing Microsoft 365 role-based access controls. 3. Copilot takes the response from the LLM and post-processes it. This post-processing includes other grounding calls to Microsoft Graph, Responsible AI checks, security, compliance and privacy reviews, and command generation. 4. Copilot returns the response to the app, where the user can review and assess the response before deciding to publish it. A LOT more details can be found here: aka.ms/LearnHowM365CopilotWo… Let us know what else you'd like to see!

The Old Men Yell At Cloud session was actually super useful at #CTTT24 on all things Copilot🤣 Some 💎 s from @themarkchristie @YannickReekmans @anttipajunen @ThatPlatformGuy and @lauriepottmeyer: -Pick your battles! Don’t try to keep up with EVERY Copilot product and instead choose things you can use NOW with current customers -Build a network of community people who know things different than you and call on them when you need expertise out of your depth. Find them via blogs, and the MS Tech Communities and the Power Platform Community Forum. Secret WhatsApp groups. Speakers at confs WANT you to talk to them and can connect you to people! -Don’t worry about asking “stupid” questions. There is at least one other person who will have the same question! -Cloud is “generally” cheaper than on-prem, it’s also much more secure and reliable. On-prem is also easier to hack. Cold storage is sometimes an exception -(Bing) Copilot creates responses from the AI model (GPTx) and also weaves in responses from the web so you can up-to-date info. ChatGPT also does this now with Bing in the background - GitHub Copilot is the best tool for code generation. Devs who have it for a week refuse to give it up. GH Copilot Enterprise is going -How to roll out M365 Copilot: Guardrail your data with Microsoft Purview, figure out where is your structured data and unstructured data, build an adoption plan via the rings of release strategy with an internal community of champs (Insider program style!). You should set all this up ANYWAY because it will protect your org and help you adopt things faster and better. Have your champs create good prompts for Copilots and send your people to Prompt Training! - For knowledge management, very few people actually do it well. Lots of DevOps usage. One idea is to create markdown files in GitHub and generating webpages a la Patterns & Practices - How to add stuff to your M365 Copilot: graph connectors, Copilot Studio, and plug-ins to M365 Copilot.