Microsoft 365 Copilot Early Access Program

Open

Context

I took part in the "Microsoft 365 Copilot Early Access Program" as Quebec champion inside my organization "AM". To maximize the benefits of the program, each business unit has a champion whose the responsabilities were the followings :

• Driving engagement and business transformation by combining technical knowledge with people skills.

• Ensuring that all Copilot EAP Users in their BU have met the required prerequisites.

• Serving as the local Copilot expert and training other Copilot EAP Users in their BU to derive value from Copilot.

• Acting as the first point of contact for Copilot EAP Users in their BU for any questions or issues.

• Identifying and sharing the business value that their BU is gaining from Copilot.

• Speak the local BU language when that is not English.

I took the opportunity to document this new technology and make the information more accessible.

What is Copilot ?

Copilot is a new way of working that combines the power of large language models (LLMs) with your content in the Microsoft 365 apps. It can help you create and improve texts, summarize meetings, transform files into presentations, and much more. Copilot is not just a better way of doing the same things. It’s an entirely new way of working.

Copilot works with the apps you already use, such as Word, Outlook, PowerPoint, Excel and Teams. You can also ask questions and get answers from Copilot using natural language. Copilot can access your data and files, as well as public information from the internet used to train the LLM, to provide you with relevant and personalized suggestions.

Copilot is safe to use with your data and files. They are protected and won't leak outside your organization. Copilot does not store or use ArcelorMittal data to train the models.

Open

AI Copilot as Business Chat (1)

What did I do to help my business unit ?

Creation of a SharePoint Intranet

I have developed a SharePoint Intranet in order to demystify and summarize most ot the info accessible in the following websites :

• Introducing Microsoft 365 Copilot – your copilot for work - The Official Microsoft Blog : remarkable introduction.

• How Microsoft 365 Copilot works : get an inside look at how LLMs work when we use it in Microsoft 365.

Creation of a presentation related to the features (fonctionnal part)

1. The user sends a request to Copilot via Chat.

2. Copilot by anchoring makes calls to Microsoft Graph, which is a gateway to data of all kinds in the Microsoft Cloud. Here, it seeks access to your organizational data.

3. Copilot processes the information obtained, combining it with other types of data to obtain the most accurate and relevant answers.

4. Copilot sends the modified initial information (step 1) to LLM (Large Language Model).

5. LLM sends the response to Copilot.

6. Copilot reprocesses the response it has just received by making additional calls to Microsoft Graph while combining security, compliance and confidentiality checks.

7. Copilot returns an adapted and precise response to the user, along with recommended commands.

Creation of a presentation related to the Architecture (technical part)

1. User sends data (text or files) to Copilot.

2. Anchoring Copilot makes calls to Microsoft Graph, a gateway to data of all kinds in the Microsoft Cloud. Here, it seeks access to your organizational data.

3. Copilot processes the information obtained, combining it with other types of data to obtain the most accurate and relevant answers: the pre-processing stage.

4. Copilot sends the modified message to LLM, which uses the semantic index to interpret user queries and produce sophisticated, meaningful, multilingual responses.

5. LLM sends the response to Copilot.

6. Copilot processes it again (post-processing), i.e. includes additional anchor calls to Microsoft Graph, security, compliance and confidentiality checks.

7. Copilot returns a recommended response and appropriate commands.

8. Copilot processes and orchestrates all this iteratively.

M365 Copilot : the first pillar - Microsoft Graph

Microsoft says: "Microsoft Graph is a gateway that gives you access to the data and information in Microsoft 365. It provides a unified programmability model that lets you access the impressive amount of data available in Microsoft 365, Windows and Enterprise Mobility + Security" (Présentation de Microsoft Graph - Microsoft Graph | Microsoft Learn).

M365 Copilot : the second pillar - Large Language Model (LLM)

An LLM is a transformer that manipulates input data and generates output data in the form of human language. And, An LLM has the capacity to analyze billions of pieces of data with billions of parameters, while recognizing the human language used at the outset.

For more information on what an LLM is : Large Language Models (LLMs): Challenges, Predictions, Tutorial (v7labs.com)

Creation of a presentation related to the data

Copilot or LLM do not use any public instance of OpenAI (GPT).

• The data flow runs within the Microsoft infrastructure, and data at rest is protected by the Microsoft infrastructure ("Microsoft 365 Trust Boundary").

• Data in transit is encrypted during all transfers between architecture components, using various technologies (Transport Layer Security - TLS, Internet Protocol Security (IPsec), BitLocker, per-file encryption): Encryption in the Microsoft cloud | Microsoft Learn

How Copilot protects our data ?

1. Customer content isolation: This is ensured by the main identity provider "Azure Active Directory", which exercises role-based access control.

2. Microsoft physical security: "Microsoft uses rigorous physical security, background filtering and a multi-layer encryption strategy to protect the confidentiality and integrity of customer content" (Microsoft).

3. Encryption of customer content: Microsoft 365 uses service-side technologies that encrypt customer content at rest and in transit, including BitLocker, file-based encryption, TLS (Transport Layer Security) and IP sec (Internet Protocol Security).

4. Customer data control (Compliance): Microsoft comply with widely applicable privacy laws, such as the RGPD, and privacy standards, such as ISO/IEC 27018, ...

Conclusion : remarks

Benefits & challenges only from a user perspective, not from a technical perspective (admin, dev, security, …) : Work Trend Index | Will AI Fix Work? (microsoft.com)