GitHub Copilot has its quirks

I've been using GitHub Copilot with our production codebase for the last 4 months, and here are some of my thoughts:

The Good:

1. Explains Complex Code: It’s been great at breaking down tricky code snippets or business logic and explaining them properly.

2. Unit Tests: Really good at writing unit tests and quickly generating multiple scenario-based test cases.

3. Code Snippets: It can easily generate useful code snippets for general-purpose use cases.

4. Error Fixes: Copilot is good at explaining errors in code and providing suggestions to fix them.

The Not-So-Good:

1. Context Understanding: It’s hard to explain the context to a GenAI tool, especially when our code is spread across multiple files/repos. It struggles to understand larger projects where changes are required in multiple files.

2. Inaccurate Suggestions: Sometimes it suggests installing npm libraries or using methods from npm packages that don’t exist. This is called Hallucination, where AI-generated code looks convincing but is completely wrong.

3. Complex Code: Occasionally, the code it generates is confusing and complex, making debugging harder. In those moments, I wish I had written the logic myself and let Copilot check for errors or bugs.

Overall, GitHub Copilot has been a useful tool, but it has its quirks. When using large language models, the responsibility always stays with the programmer.

The above is the detailed content of GitHub Copilot has its quirks. For more information, please follow other related articles on the PHP Chinese website!