Microsoft AI Studio opens for business, with a nod to safety

Redmond is ready to help folks build AI 'copilots'

Build At Microsoft's Build developer conference on Tuesday, CEO Satya Nadella announced that Redmond's Azure AI Studio is moving to general availability.

"With Azure AI Studio, we provide an end-to-end tooling solution to develop and safeguard the copilot apps you build," he said during the opening keynote presentation. "It's an end-to-end development environment to build, train and fine-tune AI models and do so responsibly."

Redmond is making OpenAI's latest multimodal model, GPT-4o, available in Azure Studio and as an API. The "o" in the model's name means "omni," a reference to its ability to accept different types of input (audio, text, and visual material) and to respond in any of those modes.

The Windows biz also announced Phi-3-vision, a multimodal model family optimized for deployment on personal devices. Phi-3-vision is suited for applications where it can respond to questions about input images or generate charts.

We're also told Phi-3-Silica, the smallest of Microsoft's Phi series of models with 3.3 billion parameters, will be baked into all Copilot+ PCs when they go on sale starting next month. Redmond's Phi-3 models, now generally available, can be found here along with Phi-1 and Phi-2.

Nadella also said that Microsoft is adding the ability to make custom models in Azure AI. "We recognize that there are advanced applications where you need much more customization of these models for very specific use cases," he explained.

Screenshot of Satya Nadella at Microsoft Build 2024 - Click to enlarge

The purpose of all this is to help developers create what Microsoft is calling "copilots" – apps infused with generative AI.

In a blog post, Nabila Babar, principal program manager for Microsoft's AI platform, said that AI Studio makes it easier to set up and manage AI services, and to oversee safety and quality.

"The platform includes Azure AI services like Azure OpenAI Service and Azure AI Search and familiar tooling from Azure Machine Learning, like prompt flow for guided experiences for quick prototyping," said Babar. "It supports code-first SDKs and CLIs, integrated with the Azure Developer (AZD) CLI and AI Toolkit for Visual Studio Code to provide the needed scalability as demand grows."

Azure AI Studio offers more than 1,600 models in its model catalog, from vendors like Meta, Mistral, Microsoft, and OpenAI.

And through a newly announced partnership with HuggingFace, the Azure AI Model Catalog also now includes 20 of the AI startup's models, including Rhea-72B-v0.5 from David Kim and Multiverse-70B from MTSAIR.

The deal makes Azure's AMD-based infrastructure, specifically ND MI300X v5 virtual machines, with AMD GPUs, available to those using Hugging Face Hub.

In addition, Microsoft has devised a "Dev Mode" feature for Hugging Face Spaces, for creating and hosting models, that allows developers to connect Visual Studio Code to their Space and edit code directly, either locally or in the browser. ®