GitHub and OpenAI release Copilot, an artificial intelligence tool to automatically complete code snippets based on the OpenAI Codex neural network model

GitHub, owned by Microsoft, and OpenAI, Inc. have launched a technology preview of a new artificial intelligence tool called Copilot. It became part of the popular source editor Visual Studio Code and is designed to automatically complete code snippets. That is, yes, the developers got into the regiment of intellectual assistants.

According to GitHub, Copilot doesn’t take snippets of source code it’s seen before and almost copied it. Instead, the system analyzes the written code and generates new appropriate code, including individual functions that were previously used. Examples on the project’s website include autowriting code for importing tweets, plotting a scatter plot, or getting a Goodreads rating.

GitHub sees the development as an evolutionary development of the concept of pair programming – the two work together on a project to effectively “catch” each other’s bugs in order to speed up the process. In Copilot, one of the programmers is the virtual assistant.

Copilot is built on a new algorithm called the OpenAI Codex, which CTO Greg Brockman at OpenAI called “a descendant of GPT-3,” a popular neural network model capable of creating text that is sometimes indistinguishable from human-typed text. The system was trained using terabytes of open source code from GitHub, as well as samples of English scripts. And if GPT-3 generates texts in English, then OpenAI Codex generates code. Copilot works best with Python, JavaScript, TypeScript, Ruby and Go, according to GitHub CEO and Xamarin IDE founder Nat Friedman.

In fact, this is the first major product since Microsoft invested $1 billion in developing OpenAI. It will initially be available as a browser and Visual Studio Code plugin, but the OpenAI API will be released in August 2021 so that third-party developers can take advantage of Codex in their applications.

So far, Copilot is in a limited technical beta, but those who wish can register on the project’s website to access it.