GitHub Copilot for Business, OpenAI’s Codex-powered pair programming extension, is now available via a Codex update.
The Microsoft-owned code repository service announced Copilot for Business in November. The GitHub Copilot for Individuals service costs $10 per user per month. Subscription to Copilot for Business costs $19 per user per month, depending on the number of Copilot “seats” assigned.
The OpenAI Codex, which powers Copilot’s code suggestions, translates natural language into code. Copilot can be used with various editors (Integrated Development Environment – IDE), including Microsoft Visual Studio, Neovim, VS Code or JetBrains IDEs.
Copilot’s code acceptance rate is low, but rising
Copilot now has a new Codex template and new features. For example, there is a security vulnerability filter to make Copilot’s programming suggestions safer and help developers spot insecure development patterns when writing code.
The new Codex model, which Microsoft is offering with GPT-3.5 and DALL-E 2 to developers through Azure OpenAI Services, is expected to increase the percentage of code written by Copilot.
GitHub says that when Copilot launched for consumers last year, about 27% of developer code files on average were generated by Copilot. Today, the average for this metric is 46% for all supported programming languages, while for Java it is 61%.
Copilot’s code acceptance rate is lower, but the rate is still rising. In June 2022, developers accepted an average of 27% of suggestions. This rate rose to 30% in September and reached 35% in December.
Fill-In-the-Middle, the flagship feature
Copilot also gained a “paradigm” called Fill-In-the-Middle (FIM), which goes beyond the previous method of only considering the code prefix. Now it is possible for Copilot to fill fields in the middle of the code. “That way, it now has more context about your code and how it should align with the rest of your program. The FIM in GitHub Copilot consistently produces higher quality code suggestions, and we’ve developed various strategies to deliver them without additional latency,” says Shuyin Zhao, Director of Product Management at GitHub.
GitHub also updated its VS Code extension with a “lightweight client-side model” that learns user context to reduce the frequency of unwanted suggestions. GitHub claims this resulted in a 4.5% reduction in unwanted suggestions.
The new vulnerability filter also uses large language models (LLMs) to “approach the behavior of static analysis tools”. GitHub claims it can block insecure programming models in real time and targets very common security issues, such as hard-coded credentials, SQL injections, and path injections.
Developers coming out of software companies
Copilot for home and business could help Microsoft get more users onboarded to GitHub. GitHub recently announced that it has 100 million users, well above most measures of the world’s developer population. GitHub CEO Thomas Dohmke also said that developers no longer just work for software companies.
“It’s an increasingly diverse and global group of people working in different industries, tinkering with code, design and documents in their spare time, contributing to open source projects, conducting research scientists, and more,” Dohmke explained. “These are people who work around the world to create software for hospitals, cinema, NASA and the PyTorch project, which powers AI and machine learning applications. They are also people who want to help a loved one to communicate and family members to overcome illnesses.”
Source: “ZDNet.com”