GitHub arms Copilot with GPT-4

Copilot is about to become much more chatty

GitHub, the platform for collaborative software development, is enhancing its AI codewriting tool, Copilot, by incorporating OpenAI's GPT large language models, and adding text chat and voice support.

Copilot, which GitHub launched less than two years ago, has already had a significant impact on software development.

GitHub says the tool, built using OpenAI's Codex model, has been involved with writing 46% of the code on Microsoft's cloud repo depot, and has helped developers code up to 55% faster.

Copilot functions as an AI partner programmer by automatically completing code and comments.

The system is now undergoing a significant change under the overarching 'Copilot X' name (which GitHub calls "a vision for the future rather than an available product").

This means GitHub is moving from Codex to the GPT-3.5 Turbo and GPT-4 large language models. OpenAI is shutting down the public API for Codex today, 23rd March.

The move will add new features to Copilot, including carrying on conversations with devs a la ChatGPT, in either Visual Studio Code or Visual Studio.

Using an IDE sidebar, the tool will be able to identify and explain code, propose solutions, and rectify errors.

Copilot Chat entered a technical preview on Wednesday.

"We are bringing a chat interface to the editor that's focused on developer scenarios and natively integrates with VS Code and Visual Studio," the company said.

Copilot now possesses the ability to identify the code that a developer has typed as well as any error messages that appear. The chat functionality is integrated into the IDE.

Copilot Chat provides developers with in-depth analysis and explanations of code blocks' intended functions, generating unit tests, and even suggesting solutions for bugs.

The new features include pull requests, to assist developers by creating AI-generated descriptions.

This new feature utilises the GPT-4 model. It incorporates support for AI-powered tags in pull request descriptions via a GitHub app that individual repository owners and organisation admins can install.

Copilot will automatically fill these tags based on the modified code. Developers can review or alter the suggested description as needed.

GitHub is also developing functionality that will enable Copilot to alert developers automatically if they miss sufficient testing for a pull request.

After generating a warning, Copilot will propose potential tests that can be edited, accepted, or declined based on the project's requirements.

As part of the Copilot X vision, GitHub is also launching Copilot for Docs: an experimental tool using a chat interface to offer users AI-generated answers to questions about documentation, including queries about languages, frameworks and technologies.

GitHub will begin with documentation for React, Azure Docs and MDN.

As a part of the broader Copilot X initiative, GitHub is also introducing Copilot to the command line.

This new feature will help with creating commands and loops and throw around obscure feature flags that are often buried deep within a manual.