Microsoft's Azure gives a glimpse of GPT-3’s natural language capabilities11/10/2021
Microsoft’s Azure Gives a Glimpse of GPT-3’s Natural Language Capabilities
During Microsoft’s Ignite conference last week, the company announced the launch of Azure OpenAI Service, making the pre-trained neural network available to select clients.
It makes sense that Azure would offer OpenAI’s GPT-3, since Microsoft signed a $1 billion deal with OpenAI to help grow the former’s AI platform. OpenAI eventually gave Microsoft exclusive commercial access to GPT-3, whose model has been trained on trillions of words that can be applied to a variety of use-cases from converting natural language to software code to summarizing large amounts of text and generating answers to questions. During the conference, Microsoft listed specific applications and enterprise uses for the service. For example, a sports franchise’s marketing team could use GPT-3’s capability to generate original content like explaining what’s happening in a game to help the team brainstorm ideas for social media or blog posts and engage with fans more quickly. Azure OpenAI can help companies, say, generate speech for a video game character or create targeted corporate communications materials at speed.
Of course, these new capabilities that Azure’s access to the GPT-3 platform provides aren’t without their drawbacks.
Microsoft recognizes the bias that can come from models pre-trained on vast amounts of internet data, and so has embedded filtering and moderation tools. Users basically feed GPT-3 examples of the kinds of outputs they want it to generate, then filters allow them to tailor the model’s output — adjusting the language to be less formal for video game dialogue, for example, or more promotional for marketing collateral. More importantly, Azure is putting into place tools that will ensure that the AI model isn’t used for harm and meets Microsoft’s protocol for network security, management, privacy, and the like. If all goes as planned, the technology will not generating unwanted or harmful results, said Sarah Bird, Microsoft’s Responsible AI Lead for Azure AI, which can include everything from offensive language to personally identifying information.
Because of the nuances of GPT-3 outcomes, use of Azure OpenAI is initially available by invitation only, and customers who are planning to implement well-defined use cases that incorporate responsible principles and strategies for using the AI technology are prime candidates. But since the partnership allows the company to integrate GPT-3 into its own products, it has the potential to put this powerful AI platform within reach of nontraditional users such as those in healthcare, government, gaming, you name it.
Although the price of the service currently remains unannounced, the future applications for Azure AI are practically endless. Who knows? You might just see natural language processing at work in the most unexpected places — from learning tools and videogames to healthcare.