At this year’s Build conference, Microsoft has focused on generative artificial intelligence, announcing various updates to its platforms for creating applications and experiences based on AI: Azure AI Studio and Copilot Studio.
First, a quick overview of these tools. Azure AI Studio is a suite of tools within Microsoft’s Azure OpenAI service that allows users to combine AI models, such as the newly introduced GPT-4 from OpenAI, with their own data to create chat assistants or other applications that “reason” over that data. Meanwhile, Copilot Studio provides tools to integrate Copilot in Microsoft 365, the AI-powered assistant in applications like Excel, Word, PowerPoint, as well as the Edge browser and Windows, with third-party data.
Azure AI Studio, now generally available, will enable developers to create generative AI-powered applications using pay-as-you-go inference APIs. These APIs allow developers to access and fine-tune generative AI models hosted on Azure infrastructure. Microsoft calls this “model as a service” and will launch it initially with models from Nixtla and Core42, with plans to include models from other providers like Cohere, Stability AI, and AI21 Labs in the future.
New preview features in Azure AI Studio will allow customers to train and debug generative applications by comparing different versions, as well as monitor the usage and quality of applications in production. Users will be able to visualize various trends and receive customized alerts.
Azure AI Studio will also integrate with Microsoft Purview (in preview), Microsoft’s service to prevent unauthorized access to data across applications and services. This will enable the detection of potential “data risks” in AI applications, enforce encryption on sensitive data, and control the use of these applications. Additionally, new tools will be launched to prevent AI model “jailbreaks” (methods to disable a model’s safeguards) and detect “hallucinations” (when a model fabricates information without basis).
Regarding Copilot Studio, Microsoft is introducing Copilot agents, described as AI robots that can “independently orchestrate tasks tailored to specific roles and functions.” These agents can manage various business workflows, learn from user feedback, and seek help in situations they cannot handle.
Charles Lamanna, CVP of Business Applications and Platforms at Microsoft, explains in a press release: “Developers assign a defined task to their copilot, equip it with the necessary knowledge and actions, and then Copilot Studio orchestrates dynamic workflows and acts behind the scenes to automate the task.”
Also new in Copilot Studio are extensions and connectors in preview for Copilot in Microsoft 365 and within the Microsoft Teams enterprise collaboration platform. Extensions allow developers to customize AI copilots with instructions, database knowledge, and add-on actions, facilitating tasks like expense report management and employee onboarding. Connectors, on the other hand, enable a copilot to connect with organizational knowledge from various sources.
Lamanna adds: “Extensions expand the actions Microsoft Copilot can perform on behalf of the user, personalize core knowledge with relevant business data, and allow for transfer to other copilots. Copilot connectors include Power Platform connectors, Microsoft Graph connectors, and Power Query connectors, and soon there will be integrations with Microsoft Fabric. This enables copilots to use diverse data sources, including public websites, SharePoint, OneDrive, Dataverse tables, Microsoft Fabric OneLake, and Microsoft Graph, as well as leading third-party applications.”
By: Nestor Castillo, ForAllTechNews Director