0:00 Microsoft 365 Copilot is
0:02 powered by what we call
0:04 the “Copilot System”,
0:05 a sophisticated processing
0:07 and orchestration engine.
0:09 It harnesses the power of
0:10 three foundational technologies:
0:12 The Microsoft 365 apps,
0:14 that's Word, Excel, PowerPoint,
0:16 Outlook, Teams, and more.
0:18 The Microsoft Graph,
0:20 that's all your content
0:21 and context, your emails, files,
0:23 meetings, chats, and calendar.
0:25 And a Large Language Model,
0:27 or LLM, a creative engine
0:29 capable of parsing and producing
0:30 human-readable text.
0:32 All accessible
0:33 through natural language.
0:35 It starts with a prompt
0:36 from you in an app.
0:37 Copilot pre-processes the prompt
0:40 through an approach
0:40 called “grounding”.
0:42 Put simply, grounding improves
0:44 the quality of the prompt,
0:45 so you get answers that
0:46 are relevant and actionable.
0:49 One of the most important
0:50 parts of grounding
0:51 is making a call to the
0:52 Microsoft Graph to retrieve your
0:54 business content and context.
0:57 Copilot combines
0:58 this user data from the graph
0:59 with other inputs
1:00 to improve the prompt.
1:02 It then sends
1:03 that modified prompt to the LLM.
1:06 You may be familiar
1:08 with what an LLM can do
1:09 based on your experience
1:10 with ChatGPT or Bing Chat.
1:13 Copilot takes the response
1:14 from the LLM
1:16 and post-processes it.
1:18 This post-processing
1:19 includes additional
1:20 grounding calls to the graph,
1:22 responsible AI checks,
1:24 security, compliance,
1:25 and privacy reviews
1:26 and command generation.
1:29 Finally, Copilot
1:31 sends a response to the user
1:33 and commands back to the apps.
1:36 Copilot iteratively processes
1:38 and orchestrates
1:39 these sophisticated services
1:41 to produce a result
1:43 that feels like magic.
