Microsoft Expands Azure OpenAI Service with New Features
Launched last year, Azure OpenAI as an addition to the Azure Cognitive Services family was only available on an invitation basis. The Service is a fully managed, enterprise-driven system that provides access to OpenAI's via Azure. Recently, Microsoft has publicized that it's expanding the Azure OpenAI service to interested consumers with a limited access preview. With the release, the Service is also providing more fine-tune capabilities.
Corporate VP of AI platform at Microsoft - Eric Boyd, said they wish to help people move from the experimentation phase to leveraging this tech as a central part of their venture to develop.
Fully Managed Models
The models powering Azure OpenAI, including GPT-3, can be personalized to execute tasks, like translating natural language into code to generate answers to questions. The model has been publicly available since 2020 via OpenAI's API. However, the Service adds corporate-personalized layers to the models that API doesn't, including private networking, access management, and greater scaling capacity.
While people could previously fine-tune the Azure OpenAI Service models using their own data example, the fine-tuning potentials are expanding with the service updates. Beyond improved fine-tuning and new models, Azure OpenAI Service now provides access to Codex, which can produce code given a natural language prompt.
Codex can be leveraged for programming tasks, including transpilation, code explanation, and refactoring code in languages spanning Go, TypeScript, Python, etc.
According to Boyd, Azure OpenAI Service absorption has been robust, with unnamed Fortune 500 companies leveraging the Service. When asked about wider availability, Boyd added that the Service will remain limited in access for the predictable future. Companies that plan to implement well-defined use cases can only sign up for the Service, and they require to explain that they're going to leverage the models reliably.