Microsoft has introduced a groundbreaking solution called Microsoft 365 Copilot that uses powerful large language models (LLMs) to transform how we work. By connecting with our organizational data, Microsoft 365 Copilot provides a range of helpful features in different Microsoft 365 applications. Let's explore how it works, with a focus on understanding large language models, ensuring privacy, and uncovering the functionalities of Microsoft 365 Copilot.
Understanding Large Language Models (LLMs)
Large language models, like those used in Microsoft 365 Copilot, gain their knowledge by being trained on a massive amount of public data. They learn language, context, and meaning from books, articles, and websites. These models are designed to understand and respond to natural language prompts, which can be statements or questions. When we ask a question or make a statement, the LLM generates a response based on its training and understanding of the context. By giving more details in our prompts, we can provide the LLM with additional information to generate more accurate responses. It's important to note that LLMs only remember the ongoing conversation and don't use the information exchanged to improve their training.
Customizing Prompts for Enhanced Knowledge
To better understand how prompts can be customized, let's look at an example using Microsoft Bing Chat's public service, which is unrelated to organizational data. If we ask a random question like, "What color shirt am I wearing today?" the LLM will intelligently respond but won't be able to see us answer the question accurately.
However, if we include descriptions of our outfits in the prompt, the LLM will use that context to generate a more appropriate response.
By providing context-rich prompts, we can continue asking related questions, building on the conversation and enabling the LLM to understand the context better. It's important to remember that LLMs don't retain information from previous conversations, ensuring privacy and data protection.
What is Microsoft 365 Copilot?
Microsoft 365 Copilot is a new feature that allows you to use natural language prompts and commands to interact with your data and apps in Microsoft 365. Microsoft 365 Copilot uses large language models (LLMs), which are advanced AI systems that can generate natural and coherent text based on your input. Microsoft 365 Copilot also uses your data in the Microsoft Graph, which is a network of your calendar, emails, chats, documents, meetings, and more. By combining the power of LLMs and Microsoft Graph, Microsoft 365 Copilot can turn your words into the most powerful productivity tool on the planet.
Some of the benefits of Microsoft 365 Copilot are:
It can unleash your creativity by helping you write, edit, summarize, and create content in apps like Word, Excel, PowerPoint, etc.
It can unlock your productivity by helping you find, launch, and work across apps like Outlook, Teams, etc.
It can uplevel your skills by helping you learn new things, improve your writing, and get feedback.
Microsoft 365 Copilot vs ChatGPT
Microsoft 365 copilot is private because it uses your data in the Microsoft Graph, which is a network of your data in Microsoft 365 such as your calendar, emails, chats, documents, meetings, and more. It can access your data and personalize your experience, but it does not use your data to train AI models or share it with third parties.
On the other hand, ChatGPT is a general-purpose LLM that uses data from the internet to generate text based on your input. It does not have access to your data or use third-party services. ChatGPT also uses your data to improve the performance of the AI models for better results. It may not have up-to-date or accurate information and may produce errors or inappropriate responses.
About Copilot Content?
The responses generated by our AI system are not 100% accurate. While working to make them better, it's important for users to use their own judgment when reviewing the output before sharing it with others. Microsoft 365 Copilot capabilities provide drafts and summaries to the users, but it's necessary for users to review and verify the content instead of relying solely on automation.
The AI models are constantly improving their algorithms to address issues like misinformation, content blocking, data safety, and preventing the promotion of harmful or discriminatory content.
It's important to note that Microsoft does not claim ownership of the content created by Copilot capabilities. It is possible that similar responses may be generated for similar prompts or queries from different users or customers.
Working of Microsoft 365 Copilot:
Microsoft 365 Copilot is an intelligent system that enhances the functionality of Microsoft 365 apps and provides users with helpful features and prompts. It works by combining various technologies and models developed by Microsoft to ensure secure access, usage, and management of business data.
Semantic Index for Copilot: It is available to Microsoft 365 E3 and E5 customers, and utilizes multiple Language Learning Models (LLMs) on top of Microsoft Graph. It interprets user queries and generates sophisticated, meaningful, and multilingual responses to enhance productivity. This index allows quick search through billions of mathematical representations (vectors) to provide users with the most relevant and actionable information within their organization.
Microsoft Graph: This is a network of your data in Microsoft 365, such as your calendar, emails, chats, documents, meetings, and more. Microsoft 365 Copilot uses the Microsoft Graph to access your data and personalize your experience. Microsoft 365 Copilot also uses Semantic Index for Copilot, a new capability that maps your user and company data in a sophisticated way2.
Microsoft 365 apps: These are the apps that you use every day in Microsoft 365, such as Word, Excel, PowerPoint, Outlook, Teams, etc. Microsoft 365 Copilot is integrated into these apps and works alongside you to help you with various tasks. You can interact with Microsoft 365 Copilot using text or voice input in these apps.
Business Chat: This is a new experience that works across all your data and apps in Microsoft 365. You can give natural language prompts to Business Chat and get relevant information, actions, or suggestions. Business Chat is powered by Bing Chat, a conversational AI platform that leverages Microsoft’s expertise in natural language processing, machine learning, and knowledge graphs.
All these components are part of the Copilot System, an underlying AI stack that powers user experiences across Bing chat, Copilot in Microsoft 365 apps, and cross-app intelligence. It consists of the foundation LLM, AI platform, skills repository, and runtime, working together to deliver intelligent features and capabilities.
Here is a step-by-step explanation of how Microsoft 365 Copilot works:
STEP 1: User prompt from Microsoft 365 app (Excel, Word, PowerPoint, OneDrive, OneNote, etc.) which is then sent to Microsoft Copilot.
STEP 2: Copilot will then access the Microsoft Graph and Semantic Index for Pre-processing through a grounding approach. This step improves the specificity of the prompt by making it more relevant and actionable to the user's specific task. To achieve this, Copilot accesses Microsoft Graph, which contains the user's organization's data. The data used by Copilot is limited to what the user already has access to through their existing Microsoft 365 role-based access controls.
STEP 3: Now, Copilot will send the modified prompt to the LLM (Large Language Model). In this step, Copilot retrieves relevant information from various sources, including the user's organization's data and knowledge base articles. This information is combined and used as input to a Language Learning Model (LLM), enhancing the prompt.
STEP 4: After that, Copilot will receive back the response from LLM. The response generated by the LLM is then post-processed.
STEP 5: Now, again Copilot will access the Microsoft Graph and Semantic Index for Post-processing. The post-processing stage includes additional grounding calls to Microsoft Graph, responsible AI checks, security, compliance, and privacy reviews, as well as command generation. These steps ensure that the generated response aligns with the user's context, complies with AI ethics and privacy guidelines, and meets security standards.
STEP 6: At last, Copilot will send the response and app command back to the Microsoft 365 App where the user can interact with the recommended response and take appropriate actions.
Copilot continuously iterates and orchestrates these sophisticated services to deliver results that are relevant to the user's business. Contextual relevance is achieved by leveraging the user's organization's data throughout the process.
Empowering Workflows with Microsoft 365 Copilot
Microsoft 365 Copilot is a helpful tool that works seamlessly with different Microsoft 365 apps to make work easier and more efficient. Here's how it can empower your workflows:
In Word, Copilot can help you create new documents like business proposals. It uses information from your existing files to assist you in writing.
In Outlook, Copilot can compose email replies for you. It looks at the content you select and generates a response, saving you time and effort.
With Copilot in PowerPoint, you can turn your written content into attractive presentations with just one click. This improves communication and engagement.
In Teams, Copilot can generate summaries of your meetings, highlighting the important follow-up actions that were discussed.
When using Business Chat in Microsoft Teams, Copilot can gather information from different sources and bring it all together, making it easy for you to catch up on anything you might have missed.
These features demonstrate how Copilot can enhance your productivity and collaboration by integrating with Microsoft 365 apps.
Privacy and Data Protection Measures
The permissions model in Microsoft 365 helps prevent unintentional data leaks between users, groups, and tenants. Microsoft 365 Copilot only presents data that each individual has access to, using the same controls for data access used in other Microsoft 365 services. The Semantic Index respects user identity-based access boundaries, ensuring that the grounding process only accesses content authorized for the current user.
If you have data protected using Microsoft Purview Information Protection (MPIP or AIP) labeling, that protection will still be in place. Although Copilot-generated content doesn't currently inherit MPIP labels, Copilot does cite the original source, which retains the label.
Microsoft 365 Copilot takes strict measures to protect them, some of them are:
Logical isolation: Customer content within each Microsoft 365 tenant is logically separated using Azure Active Directory authorization and role-based access control.
Physical security: We employ stringent physical security measures, background screening, and encryption strategies to maintain the confidentiality and integrity of customer content.
Encryption: Microsoft 365 uses various encryption technologies to encrypt customer content at rest and in transit, such as BitLocker, per-file encryption, Transport Layer Security (TLS), and Internet Protocol Security (IPsec). More details about encryption in Microsoft 365 can be found in the Encryption in the Microsoft Cloud documentation.
Privacy compliance: Microsoft is committed to complying with applicable privacy laws like the GDPR and privacy standards like ISO/IEC 27018, which is the first international code of practice for cloud privacy.
When Microsoft 365 Copilot is Available?
If you are interested in using Microsoft 365 Copilot, here is how you can access it:
You can join the Microsoft 365 Copilot Early Access Program, an invitation-only, paid preview that will roll out to an initial wave of 600 customers worldwide. This program will allow you to test Copilot and provide feedback to Microsoft. You can apply for the program by filling out the form.
If you are selected for the program, you will need to follow these steps to use Copilot:
Open the Microsoft 365 application that needs help, such as Word, Excel, or PowerPoint.
From the top menu bar of the app, tap the Help button.
In the pop-up menu, select the “Use Copilot” option.
You can also get ready for Copilot by enabling Semantic Index for Copilot, a new capability in Microsoft 365 E3 and E5 that maps your user and company data. This will help Copilot understand your intent and context better and provide you with more relevant and personalized responses. You can enable Semantic Index for Copilot in the Microsoft 365 admin center.
You can expect the official release of Microsoft 365 Copilot later this year. Microsoft 365 Copilot will be compatible with existing Microsoft 365 subscriptions that include the Microsoft 365 apps, such as Word, Excel, PowerPoint, Outlook, Teams, etc. You will be able to interact with Microsoft 365 Copilot using text or voice input in these apps.
Conclusion
Microsoft 365 Copilot revolutionizes work processes by leveraging large language models and seamlessly integrating with Microsoft 365 applications. With a clear understanding of LLMs, privacy considerations, and the functionalities of Microsoft 365 Copilot, users can unlock its transformative capabilities while ensuring the privacy and security of their data.