Microsoft Azure has played a central role in Microsoft’s AI endeavors for several years. Initially, it made Microsoft Research’s deep learning products available as Azure Cognitive Services. Then, Microsoft added tools for developing cloud-hosted machine learning models using Azure. Now, Azure is home to Microsoft’s growing family of Copilots, which utilize Azure OpenAI’s generative AI models and provide customers with access to those models.
To support these tools and allow for customization of cloud service models, Azure needed to provide multiple development environments. As a result, the Azure AI team has been working on Azure AI Studio, a replacement that brings together Azure’s AI development tools. Azure AI Studio is built on responsible AI concepts and supports both pre-defined and custom AI models.
The development of Azure AI Studio represents a fundamental change in how AI models are used. Instead of simply making API calls to individual models, users now create pipelines that combine different aspects of models or chain different models together to create multimodal applications. Tools like LangChain, Semantic Kernel, and Prompt Flow are essential frameworks for controlling and harnessing the output of generative AI models.
Azure AI Studio combines various Azure AI development tools into one environment. In its current public preview, Azure AI Studio is primarily focused on building Copilots, which are generative AI-powered applications. The platform supports mixed-model multimodal tools and the Azure AI SDK. Its goal is to enable users to experiment within the Studio before deploying refined models into production services.
During the public preview of Azure AI Studio, approval from Microsoft is required to use Azure OpenAI models in applications. Users must be working on a project for an approved enterprise customer and collaborate directly with a Microsoft account team. Specific use cases are also required to control access to the service. For example, applications using sensitive data may be limited to internal users on secure internal networks.
Azure AI Studio is a standalone service that can be accessed with an Azure account. It provides a home screen with access to a model catalog, Azure OpenAI service, Cognitive Services APIs, and content safety tools to ensure appropriate training data and prompts. The platform consists of four tabs: Home, Explore, Build, and Manage. The Home tab offers sample projects on GitHub to provide a starting point for building custom code, including tutorials on building Azure AI-powered Copilots and multimodal applications.
To get started with building an AI-powered application in Azure AI Studio, users create an AI-specific resource to manage VMs and services. The platform guides users through an Azure set-up wizard to create this resource and associated AI services. Users can add AI models, such as Azure OpenAI generative AI models, to their Azure AI Studio instance. Models can be selected from a catalog that includes OpenAI, Meta’s Llama, Hugging Face, Nvidia, and Microsoft Research models.
Building an AI-powered application in Azure AI Studio is straightforward. After creating a deployment and selecting a model, users can test prompts and model operations in a playground. Once satisfied, users can modify the model’s behavior by adding data from various sources like files or Azure AI Search indexes. Azure AI Studio also provides cost notifications at each step to help users make informed decisions.
To create complex AI-powered applications, Azure AI Studio includes Prompt Flow, a tool for chaining models, prompts, and APIs. Prompt Flow allows users to manage system-level prompts, user input, and services, creating a flow similar to those built with Semantic Kernel or LangChain. Prompt Flow provides a visual representation of the application’s elements and their connections, allowing users to construct and debug services by linking specific functions. Users can work with Prompt Flow in both Azure AI Studio and Visual Studio Code.
Azure AI Studio is still in preview but offers a unique approach to AI application development. Microsoft’s collection of AI tools demonstrates a commitment to generative AI and incorporates lessons learned in building trustworthy Copilots. This platform promises to provide an efficient path for implementing generative AI applications.
Source link
