The future of programming? GitHub's OpenAI-powered developer tool has arrived
GitHub Copilot for Business, the OpenAI Codex-powered pair programmer extension, is now generally available with an updated version of OpenAI Codex and a new real-time vulnerability filter that catches common security bugs while coding in the editor.
The Microsoft-owned code repository service announced Copilot for Business in November, adding a new teams-based option to the existing Copilot for individuals. The service costs $10 per user per month. Copilot for Business subscription costs $19 per user per month, based on the number of assigned Copilot seats.
Also: Memory safe programming languages are on the rise. Here's how developers should respond
OpenAI's Codex, which powers Copilot code suggestions, translates natural language into code. Copilot can be used with various editors, including Microsoft Visual Studio, Neovim, VS Code, or JetBrains IDEs.
Developer
GitHub says Copilot now has an upgraded Codex model and new capabilities. It has also developed a security vulnerability filter to make Copilot coding suggestions more secure and help developers spot insecure coding pattens on the fly.
The new Codex model -- which Microsoft is offering along with GPT-3.5 and DALL•E 2 to developers through Azure OpenAI Services -- should result in a higher percentage of code being written by Copilot.
GitHub claims that, when Copilot for individuals launched last year, around 27% of developers' code files on average were generated by Copilot. Now, the average for this measurement is 46% across all supported programming languages, while for Java it is 61%.
Copilot's code acceptance rate is lower, but the rate is still on the rise. In June 2022, developers on average accepted 27% of suggestions. This rose to 30% in September and reached 35% in December.
Besides updating Copilot to a newer Codex model, Copilot gained a "paradigm" called Fill-In-the-Middle (FIM), which goes beyond the previous method of only considering the prefix of code to account for known code suffixes and leaves a gap in the middle for Copilot to fill.
"This way, it now has more context about your intended code and how it should align with the rest of your program. FIM in GitHub Copilot consistently produces higher quality code suggestions, and we've developed various strategies to deliver it without any added latency," explains Shuyin Zhao, senior director of product management at GitHub.
GitHub has also updated its VS Code extension with a "lightweight client-side model" that learns about the user's context to reduce the frequency of unwanted suggestions. GitHub claims it resulted in a 4.5% reduction in unwanted suggestions.
Also: Programming languages: Why this old favorite is on the rise again
The new vulnerability filter also uses large language models (LLMs) to "approximate the behavior of static analysis tools". GitHub claims it can block insecure coding patterns in real time and targets very common security issues, such as hardcoded credentials, SQL injections, and path injections.
Copilot for individuals and businesses could help Microsoft bring more users on board to GitHub. GitHub recently reported it had 100 million users -- far more than most measures of the worldwide developer population. GitHub CEO Thomas Dohmke said the developers no longer work for software companies.
"They're an increasingly diverse and global group of people working across industries, tinkering with code, design, and docs in their free time, contributing to open source projects, conducting scientific research, and more," argued Dohmke.
"They're people working around the world to build software for hospitals, filmmaking, NASA, and the PyTorch project, which powers AI and machine learning applications. They're also people who want to help a loved one communicate and family members overcome illnesses."