GitHub has released a technical preview of Copilot, an AI-powered pair programmer that helps developers write code. Copilot proposes contextual code and functions, and helps solve problems by learning from the code someone writes.
Github Copilot was developed together with OpenAI and is based on the OpenAI Codex. This AI application has knowledge of how humans program code and , according to Github, is better than, say, GPT-3 at code generation because it is specifically trained on a large dataset of public source code.
While writing code, the developer gets suggestions for whole lines of code or even entire functions that Copilot suggests based on the context as you type it, that is, based on the code a developer is working on. Copilot can automatically convert comments into code, complete repetitive code, and test a function as it writes. Users can browse through the suggestions Copilot suggests and choose to accept or ignore them. Suggested code can also be adjusted manually.
Github emphasizes that Copilot is more than an autocomplete, because it understands and learns a lot more from the context. It can tell the difference between, say, a docstring, a comment, a function name, or code itself. In addition, Copilot gets smarter the more used it is because it learns how a developer codes.