The Quality of GitHub Copilot Generated Code: A Literature Review

GitHub Copilot is an AI-pair programmer tool powered by Codex, a GPT-۳based model trained on billions of lines of source code from public repositories. When provided with a description of the desired functionality, Copilot can generate the corresponding code in several programming languages. In this paper, we review previous empirical studies that evaluated the code generation capabilities of Copilot and present their results in three categories: code correctness, efficiency, and security. It is evident that the performance of Copilot depends on the difficulty level of the problem at hand. It is also a common finding that prompt engineering techniques can significantly enhanceits performance. Therefore, future works should evaluate Copilot using a varied range of problems and employ different prompt engineering techniques to utilize this tool to its highest potential.

کلیدواژه ها:

GitHub Copilot ، Codex ، AI ، Software Development