Microsoft expands AI reach with Azure AI studio, genAI tools
Microsoft's new platform is a studio for building, exploring and testing models. The tech giant is also incorporating Dall-E 3, GPT-3.5 Turbo and GPT-4 Turbo within Azure AI.
Microsoft on Wednesday took significant steps to expand its AI ecosystem with a host of new generative and traditional AI capabilities and updates to its Azure AI line of products.
The tech giant unveiled the new products and updates at Microsoft Ignite 2023. The conference, held this week in Seattle, focuses on enterprise technology and cloud computing.
The new updates and capabilities are part of Microsoft's concerted campaign to support the use and development of generative AI systems and advance in the AI race with other companies.
The slew of AI product releases is also the fruit of the cloud provider's exclusive relationship with AI vendor and LLM provider OpenAI. Microsoft recently invested $13 billion into OpenAI.
New Azure AI platform
A key new product is Azure AI Studio, now available in preview.
The platform is a single platform for building, testing, exploring and deploying AI systems using AI tools and machine learning models.
With Azure AI Studio, developers can build new generative AI applications, including applications with Microsoft's generative AI-based assistant, Copilot, using customizable tooling and models.
Users can access data sources such as Microsoft Fabric, OneLake and Azure AI Search. They can select models from a catalog of open source models, evaluate model responses, identify fine-tune opportunities and scale proof of concepts into production, according to Microsoft.
The new AI studio is a logical progression for Microsoft as a full AI stack vendor, according to Futurum Group analyst Mark Beccue.
It gives them a potentially powerful play given how much they have learned in managing AI models.
Mark BeccueAnalyst, Futurum Group
"It gives them a potentially powerful play, given how much they have learned in managing AI models," Beccue said.
While AI Studio is comparable with other out-of-the-box generative AI systems such as Google's Gen App Builder, Microsoft's experience managing OpenAI models gives it a potentially powerful play, Beccue said.
Moreover, Microsoft also has experience testing Copilot with customers in preview.
The Copilot capabilities in Azure AI Studio are similar to what Microsoft already does in its other enterprise applications of Copilot. Developer can issue a command in English while writing a larger piece of code, said Gartner Research analyst Sid Nag.
"It's yet another functionality to help developers write software without having to do a lot of actual code writing," Nag said.
New multimodal capabilities
Besides the new studio platform, Microsoft introduced multimodal AI capabilities now available in Azure OpenAI Service. The new platform enables organizations to build generative AI systems with images, text and video.
The new features include the addition of image-generating model Dalle-3, large language models (LLMs) GPT 3.5 Turbo, GPT-4 Turbo and GPT-4 updates. Bing Search and Advanced Data Analytics Plugins have also been added.
"Our promise to you is simple," Microsoft CEO Satya Nadella said during a keynote at Ignite on Wednesday. "As OpenAI innovates, we will deliver all of that innovation as part of Azure AI."
With GPT-3.5 Turbo and GPT-4 Turbo, users can extend their prompt length. GPT-3.5 Turbo is generally available now, while GPT-4 Turbo and Dall-E 3 are in preview, with general availability of GPT-4 Turbo expected in December.
Azure OpenAI Service also has GPT-4 updates to let enterprises fine tune and customize AI models for their needs.
The benefit of having Dall-E 3, GPT 3.5 Turbo and other capabilities under Azure OpenAI Service is that it brings all these different features under one place, Nag said.
"If I'm an enterprise building content, I don't want to go look at these different transformer search engines individually," he said.
Azure OpenAI Service users now also have access to Bing Search.
Other updates in Azure OpenAI Service include a new video-to-text summary feature that enables users to generate text summaries from videos.
Users can also transform video content into searchable content using LLMs with Efficient Video Content Search. This lets users locate specific topics or videos inside a video.
Microsoft introduces the new AI Studio and other products Wednesday at Ignite 2023.
Responsible AI and infringement
Microsoft also addressed responsible AI concerns by expanding access to the tech giant's Copilot Copyright Commitment (CCC) to customers using Azure OpenAI Service.
The option includes new documentation from Microsoft to help customers mitigate the risk of infringing on content copyrights.
Microsoft, OpenAI and others are involved in AI lawsuits accusing the tech companies of infringing on copyrighted works.
While Microsoft did not clarify what its new documentation will say, customers that comply with it will get the most advantage of CCC, Beccue said.
However, he said how much of a peace of mind customers gain with this type of documentation depends on the use case.
"Text generation could be an issue if you are developing marketing or advertising text," he continued. "The image gen issue is already well established. Personally, I wouldn't trust or ask any gen AI to write marketing or advertising copy."
Azure OpenAI service customers can also use Azure AI Content Safety, now generally available, as an API feature. The new service helps organizations detect harmful content.
Nvidia builds on Azure
Meanwhile, through an expanded partnership with AI chipmaker and software vendor Nvidia, Microsoft's new NCCv5 virtual machine series for Azure will feature Nvidia H100 NVL GPUs. Next year, Microsoft said it will also add the Nvidia H200 Tensor Core GPU to its Azure fleet.
Nvidia has also built an AI foundry service for Microsoft Azure so that enterprises can connect their custom model with Microsoft's cloud services.
The foundry service is made up of a collection of Nvidia AI foundation models, Nvidia Nemo and Nvidia DGX Cloud supercomputing services. It help enterprises create custom generative AI models.
The service addresses a need for domain-specific language models in the generative AI market, Nag said.
Recent systems that have addressed this requirement include BloombergGPT.
"Generative AI is the single most significant platform transition in computing history," Nvidia CEO Jensen Huang said while appearing on stage with Nadella at Ignite.
"We're going to do for people who want to build their own proprietary large language models, what TSMC [Taiwan Semiconductor Manufacturing Company] does for us," Huang said referring to the world's largest AI chip provider.
Systems like these do not require limitless amounts of data such as the GPT LLMs but need computing capability with the latest silicon technology.
Getty Images is among the customers using AI Foundry to build custom models, according to Nvidia.
Other AI updates and releases
Microsoft also rolled out other new features for Azure AI. They include:
Capabilities to help developers summarize and translate language apps. They include a task-optimization summarization tool powered by LLMs, including GPT-4 and GPT-3.5 Turbo.
A machine translation model that uses customers' data to align translations to business context.
A text-to-speech avatar to generate a realistic copy of a person speaking on the input text and video data of a real person talking. Both preview and custom avatars are available in preview.
Updated features within Azure AI Vision include liveness functionality and vision SDK to help prevent facial recognition spoofing and add liveness and face recognition to mobile applications.
Vector search and semantic search are now generally available in Azure AI Search. Vector search allows for faster responses to user queries of LLMs. Semantic ranking prioritizes the most relevant search results from Microsoft Bing.
Azure Machine Learning includes updates in Model Catalog. New models include Code Llama, Stable Diffusion and OpenAI's Contrastive Language-Image Pretraining. The model catalog now hosts Nvidia's new family of Nemotron-3 8B foundation models that support enterprise chat and Q&A applications and have multilingual capabilities with proficiency in more than 50 languages. OneLake is now available in preview to create a transition between Microsoft Fabric and Azure Machine Learning.
A model-as-a-service coming soon in preview. It will enable developers to integrate open source models such as Llama 2 from Meta and Jais from G42.
Esther Ajao is a TechTarget Editorial news writer covering artificial intelligence software and systems.
