Meta introduces CodeCompose, a generative AI tool for coding, similar to GitHub's Copilot. It's currently for internal use only, providing real-time code suggestions for Python and other languages in IDEs like VS Code.
The underlying model is built on Meta's research, with a massive 6.7 billion parameters. CodeCompose leverages surrounding code and comments to offer better suggestions. Thousands of employees already benefit from it, with an acceptance rate of over 20%.
However, concerns arise regarding code-generating AI controversies. Lawsuits accuse companies like Microsoft, GitHub, and OpenAI of copyright infringement due to Copilot's regurgitation of licensed code. Meta hasn't directly addressed this issue.
Security is also a concern, as generative coding tools can introduce vulnerabilities, according to a Stanford study.
To address these concerns, Meta assures that CodeCompose was trained on InCoder, which collected public code with permissive licenses from GitHub and GitLab, along with StackOverflow data. CodeCompose excludes poor coding practices and errors from its training data.
Developers have the choice to follow or disregard suggestions. Security was a "major consideration" during development.
Meta is enthusiastic about CodeCompose's progress and believes that bringing this tool in-house serves their developers best. Exciting times ahead in the world of AI-powered coding!