New Microsoft 365 Copilot Tuning | Create fine-tuned models to write like you do
Fine-tuning adds new skills to foundational models, simulating experience in the tasks you teach the model to do. This complements Retrieval Augmented Generation, which in real-time uses search to find related information, then add that to your prompts for context. Fine-tuning helps ensure that responses meet your quality expectations for specific repeatable tasks, without needing to be prompting expert. It’s great for drafting complex legal agreements, writing technical documentation, authoring medical papers, and more — using detailed, often lengthy precedent files along with what you teach the model. Using Copilot Studio, anyone can create and deploy these fine-tuned models to use with agents without data science or coding expertise. There, you can teach models using data labeling, ground them in your organization’s content — while keeping the information in-place and maintaining data security and access policies. The information contained in the task-specific models that you create stay private to your team and organization. Task-specific models and related information are only accessible to the people and departments you specify — and information is not merged into shared large language models or used for model training. Jeremy Chapman, Director on the Microsoft 365 product team, shows how this simple, zero-code approach helps the agents you build write and reason like your experts — delivering high-quality, detailed responses. Keep information permissions as-is. Use your organization’s knowledge and sharing controls. See how Copilot Tuning works. Guide Copilot with labeled examples. Copilot learns to reason and write like you are your expert team. Check it out. Build Copilot agents powered by your fine-tuned models. Automate work with your tone, structure, and standards. Take a look at Copilot Chat. QUICK LINKS: 00:00 — Fine-tune Copilot 01:21 — Tailor Copilot for specialized tasks 05:12 — How it works 05:57 — Create a task-specific model 07:43 — Data labeling 08:59 — Build agents that use your fine-tuned model 11:42 — Wrap up Link References Check out https://aka.ms/FineTuningCopilot Unfamiliar with Microsoft Mechanics? As Microsoft’s official video series for IT, you can watch and share valuable content and demos of current and upcoming tech from the people who build it at Microsoft. Subscribe to our YouTube: https://www.youtube.com/c/MicrosoftMechanicsSeries Talk with other IT Pros, join us on the Microsoft Tech Community: https://techcommunity.microsoft.com/t5/microsoft-mechanics-blog/bg-p/MicrosoftMechanicsBlog Watch or listen from anywhere, subscribe to our podcast: https://microsoftmechanics.libsyn.com/podcast Keep getting this insider knowledge, join us on social: Follow us on Twitter: https://twitter.com/MSFTMechanics Share knowledge on LinkedIn: https://www.linkedin.com/company/microsoft-mechanics/ Enjoy us on Instagram: https://www.instagram.com/msftmechanics/ Loosen up with us on TikTok: https://www.tiktok.com/@msftmechanics Video Transcript: -You can now teach or fine-tune your Microsoft 365 Copilot experience by creating your own task-specific fine-tune models that channel your expertise and experience to carry out specialized jobs and tasks accurately and on your behalf. In fact, from Copilot Studio, anyone can use this zero-code approach to teaching Copilot’s underlying model the skills from your organization to produce more usable, high-quality responses that can be as detailed as they need to be, even hundreds of pages long to get the job done. And the model remains exclusive to your organization and only the people and departments you specify. -If you compare this to the traditional way of doing this until now, this level of customization would require data science, machine learning, and coding skills. So this process is a lot simpler. And unlike existing approaches where, as a data scientist, you may be copying data into locations that may not be aware of your protections and access controls, this is enterprise-grade by design. You just focus on the outcome that you want to achieve. And because your data stays in place, your existing data access and protection policies are respected by default. Let me show you the power of this in action by comparing the results of an agent that’s calling a fine-tuned, task-specific model of Copilot versus one that’s just calling the original underlying Copilot model. So both agents are configured to author loan agreement documents. On the left is our agent using the task-specific model, on the right is our SharePoint-based agent using a general model. -Now, both agents are focused on the same exact underlying knowledge. It’s all in a SharePoint location, as you can see here with this precedent file set. And both user prompts are identical with example reference files and the client term sheets containing new information. In fact, this is a precedent file that I’ll use. It’s a long and detailed document with 14 pages and more than 5,000 words. The term sheet is quite a bit shorter as you can see here, but it’s still long and detailed with information about the loan amounts, all the details, and if I scroll all the way down to the bottom, you’ll see signatory information for both parties. -So let’s go back to our side-by-side view and run them. So, I’ll start with the general model agent on the right. And it starts to generate its response. And I’ll let this one respond for a moment until it completes. There we go. And now I’ll move over to the agent on the left. It immediately informs me that it’ll receive an email once it’s finished. Now, this is going to be a longer-form document, so we’ll fast forward in time to see each completed response. -So, starting with the general model, I’ve copied it into a Word document, and the output is solid. You’ll see that the two parties are correct, the loan structure, all the amounts are also correct from the term sheet, but it has a few tells. It’s missing a lot of specificity and nuance that a member of our legal team would typically include in all of the terms. It’s also very summarized and not how our firm would draft an agreement like this. When I scroll down to the bottom, the signatories and addresses are captured correctly and match the term sheet. That said, though, it’s just four pages long and has around 800 words, versus more than 5,000 words in our precedent document. So it kind of follows the 80–20 rule where a good portion of the response could maybe work with some edits, but it’s not reflecting how my firm thinks and how it writes when authoring legal documents like this one. -So let’s go ahead and look at the results of a fine-tuned, task-specific agent. So immediately, you can see this document is verbose. It’s 14 pages long with more than 5,300 words. The word count doesn’t always equate to quality, so let’s look at the document itself. Now, as I scroll down, you’ll see that this agent has been taught our firm-specific patterns and the clauses that we use in existing case files. It is structured and worded things just like the precedent document. It’s reasoning and writing with more precision, like an experienced member of our firm would. And while as with any other AI-generated document, I still need to check it for accuracy, it really captures that extra detail and polish to save us time and effort. So model fine-tuning is a powerful way to tailor state-of-the-art large language models that are used behind Copilot to your specific needs. -And as you saw, it also can significantly improve the handling of specialized tasks. So let me explain how fine-tuning works in this case. Unlike Retrieval Augmented Generation, it doesn’t rely on search and orchestration processes that run external to the large language model. The additional knowledge added as part of the fine-tuning process is a protected container of information that attaches the large language models training set to teach it effectively a new skill. Now, it’s never merged into the LLM or used for future model training, and is temporarily attached to the LLM when it’s needed. Again, the skill and knowledge that it contains is exclusive to you and the people or groups that you’ve shared it with, so it can’t be accessed without the right permissions. -Next, let me show you what it takes to create and fine-tune your own task-specific model. I’m in Microsoft Copilot Studio, which you can reach from your browser by navigating to copilotstudio.microsoft.com. I’m on the task-specific model page and I want to customize a model to generate partner agreements. So I’ll paste in a corresponding name. Then I’ll paste in a description. Then as the task type, I’ll select a customization recipe that reflects what I want it to do. And my options here include expert Q&A, document generation, and document summarization, with more task types coming over time. From there, I can provide additional instructions to tailor the fine-tuning recipe, like how the model should use original files, for example, to inform the structure, formatting, company-specific clauses, and other areas important to your model, like we saw before. -Next, I can define my own knowledge sources. Now, these can use information from SharePoint sites and folders, and soon, you’ll be able to add information external to Microsoft 365 using Microsoft Graph connectors. In this case, I’ll define a SharePoint source. Then browse the sites that I have access to. I’ll choose this folder inside the Agreements library. And from there, I can even drill into specific folders for the precise information that I want to use to teach the model, which I’ll do here with the Agreements folder. -For permissions, this process aligns to the enterprise-grade controls that you already have in your organization backed by your Microsoft Entra account. Now, the next step is to process the data you selected for training or what’s known as data labeling. So here, you’ll be presented with data labeling tasks in small, iterative batches. They’re kind of like questionnaires for you to complete, where the fine-tuning process will generate documents and request assessment of them for clarity, completeness, accuracy, and professionalism. This process requires subject matter expertise to open these documents and rate the quality of the generative output for each. I’m just going to show one question here, but you’d repeat this process for every batch. And once all batches are labeled, I can start model training. Now, this will take some time to process, so I’ll fast forward a little in time. -Now with everything finished, I can publish the model to my Microsoft 365 tenant. And it will be available to anyone we’ve shared it with, like our audit team from before, to build new agents. And the process I just showed is called supervised learning, where the model is trained on label data. And soon, you’ll also have the option to use reinforcement learning to enhance the agent’s reasoning capabilities. Now let me show you how to build an agent from Copilot Chat that can leverage our new task-specific model for partner agreement generation. So I’m going to select Create agent. And for the purpose, I have a new option here to build a task-specific agent. Next, I can choose from the existing task-specific models. So I’m going to choose the one that we just created for new partner agreements. There we go. And with any agent, I just need to give it a name. Now I’ll paste in a description for people on the team to know its purpose and what it can do. -And next, I can specify additional instructions as guidelines to provide more context to the agent, as I’m doing here to ensure the structure aligns with our organizational standards. Because this is a very specific agent to write partner agreements, I’ll just specify one starter prompt with details for referencing a precedent source document to start with and a term sheet to get specific new information from, kind of like we saw before. Now, the preview on the right looks good, and I can create the agent right from here. For sharing, permissions also need to align with whoever my task-specific model was shared with, which, as you’ll remember, again, was our audit team. In this case, for my own validation, I’ll select only you so that I can test it before sharing it out with other auditors on my team. -So let’s go ahead and test it out. So I’m going to use the starter prompt. Then I’ll replace the variable file names here. I’ll use the forward slash reference, starting with the precedent file. Now I’ll look for the term sheet file. There it is. From there I can submit my prompt. This is going to take a moment for the response. You can see the structure with sections based on our task-specific files used with the fine-tuning. It tells me that it’ll send me a Word document and email once it’s finished again. In fact, if I fast forward in time a little, I’ll move over to Outlook. And this is the file the agent sent me with links to the new agreement draft. So I’ll open it using Word in the browser. There’s my agreement. And you’ll see it follows exactly how we wrote the precedent agreement. As I scroll through the document, I can see all the structure and phrasing aligned with how we write these types of agreements. In fact, this Representations and Warranties section is word for word direct from our standard terms that our firm always incorporates. And that’s it. My agent is now backed with my task-specific, fine-tuned knowledge, and it’s ready to go and I’m ready to share it with my team. -So those are just a few examples of how fine-tuning in Microsoft 365 Copilot can give you on-demand expertise, and task-specific models respond more accurately using your specified voice and process so that you and your team can get more done. -To find out more, check out aka.ms/FineTuningCopilot, and keep watching Microsoft Mechanics for the latest tech updates, subscribe to our channel, and thanks for watching.
