New Azure OpenAI Service combines access to powerful GPT-3 language models with Azure’s enterprise capabilities

Since OpenAI, an AI research and deployment company, introduced its groundbreaking GPT-3 natural language model platform last year, users have discovered countless things that these AI models can do with their powerful and comprehensive understanding of language.

For instance, a sports franchise that’s developing a new app to engage with fans during games could use the models’ ability to quickly and abstractly summarize information to convert transcripts of live television commentary into game highlights that someone could choose to include within the app.

The marketing team could use GPT-3’s capability to generate original content and its understanding of what’s happening in the game to help the team brainstorm ideas for social media or blog posts and engage with fans more quickly.

At its Ignite conference today, Microsoft announced it will help its customers uncover these kinds of experiences with the new Azure OpenAI Service, which allows access to OpenAI’s API through the Azure platform and will initially be available by invite only. The new Azure Cognitive Service will give customers access to OpenAI’s powerful GPT-3 models, along with security, reliability, compliance, data privacy and other enterprise-grade capabilities that are built into Microsoft Azure.

Microsoft will also offer Azure OpenAI Service customers new tools to help ensure outputs that the model returns are appropriate for their businesses, and it will monitor how people are employing the technology to help ensure it’s being used for its intended purposes.

“We are just in the beginning stages of figuring out what the power and potential of GPT-3 is, which is what makes it so interesting,” said Eric Boyd, Microsoft corporate vice president for Azure AI. “Now we are taking what OpenAI has released and making it available with all the enterprise promises that businesses need to move into production.”

Built by OpenAI, GPT-3 is part of a new class of models that can be customized to handle a wide variety of use cases that require a deep understanding of language, from converting natural language to software code to summarizing large amounts of text and generating answers to questions.

As more people are able to access and use them, the models become even more capable, said OpenAI CEO Sam Altman. He looks forward to the day when you can tell a computer what you want in plain language — even if the request is fuzzy like “find the strategy document I can’t remember the name of but wrote three years ago and has this image in it” — and the software will be able to execute that request.

“GPT-3 has really proven itself as the first powerful, general purpose model for natural language — it’s one model you can use for all these things, which developers love because you can try things very easily,” Altman said. “For a while now, we’ve wanted to figure out a way to scale it as broadly as possible, which is part of the thing that really excites us about the partnership with Microsoft.”

While GPT-3 has been publicly available since last year through an API managed by OpenAI, some potential customers have needed additional layers of security, access management, private networking, data handling protections or scaling capacity — which the Azure OpenAI Service will offer.

The new Azure OpenAI Service will give customers access to OpenAI’s powerful natural language GPT-3 models – with the security, reliability and enterprise capabilities of Microsoft Azure. Azure OpenAI Service can help developers working for a sports franchise create a new app by converting language to software code, then reason over transcripts of live television commentary to offer game summaries for the app and also generate ideas for blog posts and other written content for fans.

Other companies that have already been using the API and want to put those ideas into commercial use will be able to run those solutions on Azure’s global infrastructure to meet their production needs, including critical security, compliance, performance, reliability and scale requirements.

“I believe in people doing what they’re good at,” Altman said. “This allows us to marry all the benefits that Azure customers have come to expect in security and compliance and its massive footprint with all the things that people love about GPT-3.”

Users can also easily teach the models — which have already learned nuances of language from absorbing patterns in billions of pages of publicly available text — to meet specific business needs using their own data. In a process known as “few shot learning,” users only need to show the models a few examples of the kinds of outputs or responses or code they want it to generate.

“It really is a new paradigm where this very large model is now itself the platform. So companies can just use it and give it a couple of examples and get the results they need without needing a whole data science team and thousands of GPUs and all the resources to train the model,” Microsoft’s Boyd said. “I think that’s why we see the huge amount of interest around businesses wanting to use GPT-3 — it’s both very powerful and very simple.”

The potential enterprise uses for GPT-3 range from summarizing common complaints in customer service logs to helping developers code faster without having to stop and search for examples or generating new content as starting points for blog posts, said Dominic Divakaruni, Microsoft group product manager leading Azure OpenAI.

“It helps expedite the process of creative writing, it helps you extract insights from a large amount of text and its code generating capabilities are a great example of the new kinds of business value these models bring,” Divakaruni said. “Customers are learning new things about what it can light up for them each day.”

The Azure OpenAI Service is the latest offering to emerge from a partnership between Microsoft and OpenAI that aims to accelerate breakthroughs in AI, from jointly developing the first supercomputer on Azure to commercializing new AI technologies.

Microsoft, which has a license to the GPT-3 technology that allows the company to integrate it into its own products, is using the Azure OpenAI Service to bring these natural language innovations to customers on a large scale.

Earlier this year, for instance, Microsoft began using GPT-3 in Microsoft Power Apps to help people who have no coding or programming background build apps by translating plain language commands into formulas.

This allows us to marry all the benefits that Azure customers have come to expect in security and compliance and its massive footprint with all the things that people love about GPT-3.

Microsoft subsidiary GitHub and OpenAI also introduced Copilot, a tool that uses a new model based on GPT-3 called Codex, which helps software developers write code more efficiently and avoid repetitive tasks with automatic code completion and suggestions.

Now, the Azure OpenAI Service will offer customers direct access to GPT-3 in a format that is designed to be intuitive enough for developers to use, yet robust enough for machine learning experts to work with the models as they wish.

Because these large language models are trained on vast amounts of internet data, which can include everything from vulgar language to racial stereotypes to personally identifying information, it’s important to give enterprise customers safeguards to help prevent the technology from being used for harmful purposes or generating unwanted results, said Sarah Bird, Microsoft’s responsible AI lead for Azure AI.

That’s why Microsoft initially will make the Azure OpenAI Service available by invitation to customers who are planning to implement well-defined use cases that incorporate responsible principles and strategies for using the AI technology. Collaborations with these early customers will help Microsoft see how its responsible AI safeguards are working in practice and make any needed adjustments, Bird said.

As part of the Azure OpenAI Service, Microsoft will offer new tools to filter and moderate the content of users’ requests and responses to help the models work effectively in each application. Customers will be able to customize those filters according to their business needs, since language that’s appropriate for a video game character may differ from communications aimed at business executives.

Microsoft will also provide safety monitoring and analysis to identify possible cases of abuse or misuse and to help customers ensure their own users are deploying the technology for its intended purposes, Bird said.

Microsoft will also offer customers guidance for using the technology successfully and fairly, such as keeping people in the loop to judge whether the content or code that the model is producing is high quality.

“We expect to learn with our customers, and we expect the responsible AI areas to be places where we learn what things need more polish,” Microsoft’s Boyd said. “This is a really critical area for AI generally and with GPT-3 pushing the boundaries of what’s possible with AI, we need to make sure we’re right there on the forefront to make sure we are using it responsibly.”


Top image: Azure OpenAI Service uses GPT-3 to convert transcripts of live television commentary during a women’s basketball game into short game summaries that the team building an app to engage with fans can pick from.

Jennifer Langston writes about Microsoft research and innovation. Follow her on Twitter.