A new “AI pair programmer” represents a breakthrough in the third revolution in software development: the use of AI in coding.
Powered by a new AI system developed by OpenAI, hopefully GitHub Copilot will soon come to Visual Studio Code and then to the full-fledged Visual Studio IDE.
Right now, in the limited tech preview phase, it’s rough around the edges, but it’s supposedly getting smarter and smarter. Its first implementation comes in a VS Code extension that is now used by a small group of testers. If the tech preview proves successful, there are plans to make it publicly available for VS Code and Visual Studio as a paid product.
« We built GitHub Copilot last year with the incredibly talented team at OpenAI and we’re very excited to show it off today, » said Nat Friedman, CEO of GitHub, on the June 29th Hacker News post. “Hundreds of developers use it in-house every day, and the most common reaction was the exploding emoji. If the technical preview goes well, we plan to scale this up as a paid product at some point in the future. «
As an AI pair programmer, it offers functions and suggestions for code completion similar to IntelliSense / IntelliCode, but goes beyond these Microsoft offerings with Codex, the new AI system from Microsoft partner OpenAI. IntelliCode is powered by a large-scale Transformer model specializing in code usage (GPT-C). OpenAI Codex, on the other hand, has been described as an improved descendant of GPT-3 (Generative Pre-trained Transformer), which can translate natural language into code.
Like IntelliCode, Codex is trained on high quality code repos on GitHub, taking into account the local project context and other factors, to suggest code completion for individual lines or entire functions.
« OpenAI Codex has a broad understanding of how people use code and is significantly better at code generation than GPT-3, in part because it was trained on a data set that contains a much greater concentration of public source code, » Friedman said in a Blog post from June 29th. « GitHub Copilot works with a wide variety of frameworks and languages, but this tech preview works especially well for Python, JavaScript, TypeScript, Ruby, and Go. »
Microsoft and OpenAI entered into a partnership pact as early as 2019, the latest manifestation of which includes new functions for natural language development without code for Microsoft’s Power Apps, which are supported by GPT-3. Now OpenAI Codex is instrumental in the new offering from GitHub, which is owned by Microsoft.
As mentioned earlier, the limited tech preview is missing some features (the VS Code extension shows 325 installs at the time of this writing). « GitHub Copilot doesn’t actually test the proposed code, so the code may not even compile or run, » the FAQ on the project website said. “GitHub Copilot can only save a very limited context, so even individual source files that are longer than a few hundred lines are cut off and only the context immediately preceding is used. And GitHub Copilot can suggest old or outdated uses of libraries and languages for the code anywhere, but you do so at your own risk. «
Additionally, the FAQ states that when trying to understand a developer’s intent, GitHub Copilot is suggesting code that may not always work or make no sense. Therefore, like any code, the suggestions should be carefully tested, reviewed, and reviewed.
But even with these reservations about the technical preview, GitHub hopes for big improvements based on promising initial results. « We recently ran a benchmark on a number of Python functions that had good test coverage in open source repos, » the FAQ says. “We hidden the functional bodies and asked GitHub Copilot to fill them out. The model got this right 43 percent of the time on the first attempt and 57 percent of the time on 10 attempts. And it just keeps getting smarter. «
Over at Hacker News, the inevitable questions about code ownership, licensing, conflicts of interest, and other legal issues popped up. « In general: (1) training ML systems on public data is fair use (2) the output belongs to the operator, just like a compiler, » Friedman replied. « On the education issue in particular, here is OpenAI’s position as submitted to the USPTO. We expect IP and AI to be an interesting political discussion around the world in the years to come, and we look forward to participating! »
The technical preview is limited because the tool requires state-of-the-art AI hardware. Developers who think they have are invited to apply here. “If the technical preview is successful, we plan to create a commercial version of GitHub Copilot in the future. We’d like to use the preview to see how people are using GitHub Copilot and what it takes to run it on a large scale, ”says the FAQ.
OpenAI says it will be releasing Codex through its API later this summer to allow developers to explore many more features of the new AI system that can be built into their own apps.
David Ramel is an editor and writer for Converge360.
A new “AI pair programmer” represents a breakthrough in the third revolution in software development: the use of AI in coding, says GitHub CEO Nat Friedman.
ML.NET – an open source cross-platform machine learning framework for .NET developers – now works on ARM64 and Apple M1 devices, as well as Blazor WebAssembly, with some limitations for both.
Microsoft has released new products to help develop apps for and on Windows 11, including an updated driver kit, a new assessment and deployment kit, and ARM64EC (Emulation Compatible), a new way to create apps for Windows 11 on ARM.
When announcing Windows 11, Microsoft said that a more open Microsoft Store ecosystem will provide more opportunities for developers.
Developers can now set an application-wide default font for WinForms projects with a new API.
Keywords:
