Microsoft puts AI at the service of no-code development

The Redmond group integrates five versions of GPT-3, an automatic language processing model developed by OpenAI with 175 billion parameters, in Azure.

On the occasion of the Build event to be held from May 24 to 26, Microsoft is announcing the launch of Azure OpenAI Service. Accessible through the Azure Cognitive Service API offering from the US cloud, this new offering provides access to the GPT-3 technology developed by OpenAI. With 175 billion parameters, this giant neural network focused on automatic language processing (NLP) is one of the most advanced deep learning models available today. One of the main goals of the Redmond group is to put this technology at the service of codeless development.

As expected, Microsoft does not provide direct access to GPT-3. This model, although very efficient, is in fact not very operational. Why ? Because it is bulky. Training for a specific use case would consume too much time and machine resources. The learning time is estimated at 34 days on an infrastructure of 1024 GPUs (type NVidia Tesla V100). Result: Microsoft offers versions, lighter and more specialized, all made by OpenIA.

Speech code an application

One of the versions of GPT-3 integrated by Microsoft is Codex. A neural network specially designed to translate voice commands (in English) into application source code. It supports a dozen programming languages. A good way for the American publisher to put a little more stamp on the no-code development segment. “The main benefit is to make programmers’ work more productive,” said Peter Welinder, vice president of products and partnerships for OpenAI at Microsoft.

“Through image recognition and mapping we generate the complete graphical interface of an application”

Microsoft takes the opportunity to announce the launch of GitHub Copilot in final version. Based on Codex, this application is presented as an intelligent programming assistant. In addition to inline voice command translation of code, it is designed to make suggestions for improvement based on analysis of already implemented resources.

In the same vein, Microsoft is further expanding the AI ​​capabilities of its Power App no-code development environment. In 2021, the group unveiled Power App Ideas. Objective: To provide the possibility, again using GPT-3, to create applications via Power Fx, a declarative language derived from the grammar of Excel and therefore known by a large number of users. During Build, Microsoft Power Apps adds Express Design to the building. A brick cut to convert a screen design (in image format, PDF, Figma, etc.) into software. “Through image recognition and mapping, we generate the entire graphical interface: button, window, input field…”, explains Charles Lamanna, corporate vice president of platform and business applications at Microsoft.

More than no code

In addition to no code, the publisher implements GPT-3 technology for other use cases. In addition to Codex, it enriches Azure Cognitive Service with the four other most popular versions of the well-known OpenAI model. Considered the fastest and least resource-intensive, Ada is the one that facilitates the most basic of tasks: analysis, reformatting, or text classification. Then comes Babbage, who is particularly suited to documentary research. Dan Curie, which extends to sentiment analysis, vocal or written interactions, or even summaries generation. Finally, Davinci, the most advanced of the four, helping you solve logical problems, capture the intent of a text, produce creative content, or even manage more complex summary tasks, among other things.

The collaboration between Microsoft and OpenAI is not new. The company from Redmond signed a collaboration with the AI ​​specialist in 2019. This agreement, which generated a $1 billion investment from Microsoft, was initially accomplished through the deployment on the group’s Azure cloud of an intensive computing infrastructure focused on training the startup’s machine learning models. We see the result today.

Leave a Comment