Generative AI is a big focus for Microsoft at this year's Build conference. And to that end, the tech giant announced a series of updates to Azure AI Studio and Copilot Studio, its platforms for building generative AI-powered apps and experiences.
First, a quick refresher on Azure AI Studio and Copilot Studio. Azure AI Studio is a toolset within Microsoft's Azure OpenAI service that allows customers to combine their own data with AI models, like OpenAI's recently announced GPT-4o, to make inferences about that data. ” You can build chat assistants and other types of apps. Copilot Studio, on the other hand, provides tools to connect Copilot for Microsoft 365 (apps like Excel, Word, PowerPoint, Microsoft's Edge browser, and AI-powered “copilot” in Windows) to third-party data. .
Now generally available, Azure AI Studio will soon allow developers to use pay-as-you-go inference APIs, which allow developers to access and fine-tune generative AI models hosted on Azure infrastructure. , you will be able to build apps that leverage generative AI. Microsoft is calling this the “as-a-service model,” starting with models from Nixtla and Core42, and in the future he expects to see models from additional providers such as Cohere, Stability AI, and AI21 Labs.
Other new Azure AI Studio features in preview allow customers to train and debug by comparing different versions of apps powered by generative AI, and monitor app usage and quality in production. can do. Users can visualize various trends and receive alerts based on custom-defined filters and settings.
Azure AI Studio also prevents unauthorized access to data across apps and services, detects potential “data risks” within AI apps, applies encryption to sensitive data, and protects AI app usage. It's also integrated with Microsoft Purview (in preview), Microsoft's service for managing . The Studio is also shipping new tools to prevent “jailbreaking” of AI models (i.e., workarounds that disable the model's protection features) and to detect hallucinations and cases where the model has completely fabricated facts. I am.
Image credit: Microsoft
On the Copilot Studio side, Microsoft is launching the Copilot agent. The company describes it as an AI bot that “can independently tailor tasks to specific roles and functions.” Copilot agents can leverage memory and contextual knowledge to navigate different types of business workflows, learn from user feedback, and ask for help when they encounter a situation they don't know how to handle.
Charles Lamanna, CVP of Business Applications and Platforms at Microsoft, explains the concept in a press release: “The developer provides the co-pilot with a defined task and equips it with the necessary knowledge and actions. Copilot Studio then coordinates the dynamic workflow and works behind the scenes.” Integrating scenes with… to automate tasks. ”
New extensions and connectors for Copilot Studio are also available in preview versions of Copilot for Microsoft 365 and directly within Teams, Microsoft's enterprise collaboration platform. Extensions allow developers to customize his AI-powered co-pilot with instructions, knowledge from the database, and actions from plugins. For example, you can build a co-pilot to handle tasks like expense reporting and employee onboarding. Connectors, on the other hand, provide a way for developers to “stick” organizational knowledge from a variety of sources to the co-pilot.
“Extensibility expands the actions that Microsoft Copilot can take on your behalf, allowing you to customize the foundation with relevant business data and hand off to other copilots,” Lamanna adds. “And Copilot connectors include Power Platform connectors, Microsoft Graph connectors, Power Query connectors, and Microsoft Fabric integration coming soon. This allows copilots to use a variety of data sources, including public websites, SharePoint, OneDrive, Dataverse tables, Microsoft Fabric OneLake, Microsoft Graph and leading third-party apps.”