Microsoft loosened access to its Azure OpenAI Service, allowing more developers to tap into the AI service's powerful model for natural language processing in the cloud.
At its Build developer conference this week, Microsoft invited businesses to apply to the limited access preview of its OpenAI service. Until now, the service had been invitation only.
Microsoft based the Azure service on the Generative Pre-trained Transformer (GPT-3) deep learning neural network. The company OpenAI, founded in 2015 by a group of investors including Elon Musk, developed the GPT-3 machine learning model that has over 175 billion parameters. Musk, CEO of electric car maker Tesla, left the company in 2019.
That year, Microsoft invested $1 billion in OpenAI to commercialize GPT-3 and conduct further AI research. The OpenAI model is powerful enough to generate large amounts of relevant machine-generated text from a small amount of input text. OpenAI produces writing that reads as if a human wrote it.
Researchers and academics have used OpenAI through its originating company for years. Nevertheless, the Microsoft announcement is significant, said Andrew Lohn, a senior fellow at the Center for Security and Emerging Technology at Georgetown University.
"It does strike me as an important moment in illustrating the union between Microsoft and OpenAI," Lohn said. "It shows that OpenAI is affiliated with Microsoft in a product development sense."
Andrew LohnSenior fellow, Center for Security and Emerging Technology at Georgetown University
GPT-3 had no match as a trained language model until Meta, Facebook's parent company, released Open Pretrained Transformer (OPT-175B) early this month. The two models are similar in size but differ in how developers, researchers and academics can access them. Meta lets users run the model and its codebase on an on-premises server. OpenAI provides access only through the cloud.
Microsoft's GPT-3 offering is also only cloud-based, and the company vets applicants before providing access. The preview launched at Build includes GPT-3, Codex and embedding models.
Codex translates plain language to code that underlying computers can execute. Embeddings simplify deploying semantic search, a data searching technique that delivers results based on what the AI technology determines is the intent and contextual meaning behind a query.
Developers use the OpenAI Service to provide writing assistance, translate natural language to code, and extract intelligence from massive amounts of data.
National used-car dealer CarMax, which had nearly $19 billion in revenue last year, uses the OpenAI Service to provide 5,000 summaries of 100,000 customer reviews of every make, model and year of vehicles the company sells, according to Microsoft.
Other customers include Farmlands, a New Zealand-based farm supplier. The cooperative uses OpenAI Service to provide call center agents with summaries of previous customer interactions.
Preventing potential GPT-3 abuse
Microsoft has safeguards for preventing abuse of the OpenAI Service. The company will require applicants to specify use cases and follow Microsoft's principles for AI use. Organizations using the service must use people to ensure output accuracy before publication.
In addition, Microsoft has embedded in the service filters that can remove sexual, violent and hateful content. During the preview period, Microsoft plans to add filters and improve the service's anti-abuse capabilities in collaboration with customers.
However, Microsoft wants to avoid becoming the only content gatekeeper, said Sarah Bird, Microsoft's responsible AI lead for Azure.