At GitHub, our mission has all the time been to innovate forward of the curve and provides builders every little thing they must be happier and extra productive in a world powered by software program. Once we started experimenting with giant language fashions a number of years in the past, it shortly grew to become clear that generative AI represents the way forward for software program improvement. We partnered with OpenAI to create GitHub Copilot, the world’s first at-scale generative AI improvement instrument made with OpenAI’s Codex mannequin, a descendent of GPT-3.
GitHub Copilot began a brand new age of software program improvement as an AI pair programmer that retains builders within the circulate by auto-completing feedback and code. And fewer than two years since its launch, GitHub Copilot is already writing 46% of code and helps builders code as much as 55% sooner.
However AI-powered auto-completion is simply the place to begin. Our R&D staff at GitHub Subsequent has been working to maneuver previous the editor and evolve GitHub Copilot right into a readily accessible AI assistant all through the complete improvement lifecycle. That is GitHub Copilot X—our imaginative and prescient for the way forward for AI-powered software program improvement. We aren’t solely adopting OpenAI’s new GPT-4 mannequin, however are introducing chat and voice for Copilot, and bringing Copilot to drag requests, the command line, and docs to reply questions in your tasks.
With AI out there at each step, we will essentially redefine developer productiveness. We’re decreasing boilerplate and guide duties and making advanced work simpler throughout the developer lifecycle. By doing so, we’re enabling each developer to focus all their creativity on the massive image: constructing the innovation of tomorrow and accelerating human progress, in the present day.
Let’s leap in.
Need to see what’s new? Uncover GitHub Copilot X—our imaginative and prescient for the way forward for AI-powered software program improvement. Study extra >
A brand new AI-powered developer expertise 🧑💻
- A ChatGPT-like expertise in your editor with GitHub Copilot Chat: We’re bringing a chat interface to the editor that’s centered on developer eventualities and natively integrates with VS Code and Visible Studio. This does way over recommend code. GitHub Copilot Chat isn’t just a chat window. It acknowledges what code a developer has typed, what error messages are proven, and it’s deeply embedded into the IDE. A developer can get in-depth evaluation and explanations of what code blocks are meant to do, generate unit assessments, and even get proposed fixes to bugs.
GitHub Copilot Chat builds upon the work that OpenAI and Microsoft have carried out with ChatGPT and the brand new Bing. It’s going to additionally be part of our voice-to-code AI know-how extension we beforehand demoed, which we’re now calling GitHub Copilot Voice, the place builders can verbally give pure language prompts.
- Copilot for Pull Requests: Now you can join a technical preview of the primary AI-generated descriptions for pull requests on GitHub. This new performance is powered by OpenAI’s new GPT-4 mannequin and provides help for AI-powered tags in pull request descriptions by a GitHub app that group admins and particular person repository house owners can set up. These tags are routinely crammed out by GitHub Copilot primarily based on the modified code. Builders can then assessment or modify the advised description.
Enroll your repository within the technical preview >
That is simply step one we’re taking to rethink how pull requests work on GitHub. We’re testing new capabilities internally the place GitHub Copilot will routinely recommend sentences and paragraphs as builders create pull requests by dynamically pulling in details about code modifications.
We’re additionally making ready a brand new function the place GitHub Copilot will routinely warn builders in the event that they’re lacking enough testing for a pull request after which recommend potential assessments that may be edited, accepted, or rejected primarily based on a challenge’s wants.
This enhances our efforts with GitHub Copilot Chat the place builders can ask GitHub Copilot to generate assessments proper from their editor—so, within the occasion a developer could not have enough take a look at protection, GitHub Copilot will alert them as soon as they submit a pull request. It’s going to additionally assist challenge house owners to set insurance policies round testing, whereas supporting builders to satisfy these insurance policies.
- Get AI-generated solutions about documentation: We’re launching GitHub Copilot for Docs, an experimental instrument that makes use of a chat interface to supply customers with AI-generated responses to questions on documentation—together with questions builders have concerning the languages, frameworks, and applied sciences they’re utilizing. We’re beginning with documentation for React, Azure Docs, and MDN, so we will be taught and iterate shortly with the builders and customers of those tasks.
We’re additionally working to carry this performance to any group’s repositories and inside documentation—so any developer can ask questions by way of a ChatGPT-like interface about documentation, idiomatic code, or in-house software program of their group and get instantaneous solutions.
We all know that the advantages of a conversational interface are immense, and we’re working to allow semantic understanding of the whole thing of GitHub throughout private and non-private data bases to raised personalize GitHub Copilot’s solutions for organizations, groups, firms, and particular person builders alike primarily based on their codebase and documentation.
Transferring ahead, we’re exploring one of the best methods to index assets past documentation equivalent to points, pull requests, discussions, and wikis to offer builders every little thing they should reply technical questions.
- Copilot for the command line interface (CLI): Subsequent to the editor and pull request, the terminal is the place the place builders spend essentially the most time. However even essentially the most proficient builders must scroll by many pages to recollect the exact syntax of many instructions. That is why we’re launching GitHub Copilot CLI. It could actually compose instructions and loops, and throw round obscure discover flags to fulfill your question.
Let’s construct from right here 🚀
From studying docs to writing code to submitting pull requests and past, we’re working to personalize GitHub Copilot for each staff, challenge, and repository it’s utilized in, making a radically improved software program improvement lifecycle. Along with Microsoft’s data mannequin, we’ll harness the reservoir of information and insights that exist in each group, to strengthen the connection between all employees and builders, so each thought can go from code to actuality with out friction. On the similar time, we’ll proceed to innovate and replace the center of GitHub Copilot—the AI pair programmer that began all of it.
GitHub Copilot X is on the horizon, and with it a brand new era of extra productive, fulfilled, and completely happy builders who will ship higher software program for everybody. So—let’s construct from right here.
