Introduction
The future of coding may involve writing far fewer lines of code. GitHub, powered by Microsoft, recently unveiled Copilot, an AI tool driven by GPT-3 that aims to automate code generation. Write a descriptive function name or comments, and Copilot fills in the implementation. This isn't just snippet insertion; it's the creation of novel code tailored to your project's context. Is it revolutionary or just hype? Let's explore.
How GitHub Copilot Works
Copilot works by analyzing your existing code to understand the context. For example, if you're writing a React component and want to create a unit test, simply add a comment describing the test. Copilot will then generate the code, adapting to your component and chosen testing framework. Similarly, tasks like calculating the number of days between two dates, which might usually involve a trip to Google and Stack Overflow, could be automated. Just write the function name, and Copilot implements the function body. If you're not satisfied, you can cycle through alternative suggestions.
Hands-On Experience: Testing Copilot's Capabilities
After installing the Copilot extension in VS Code and authenticating with a GitHub account, the user in the video tested Copilot's capabilities with a vanilla JavaScript project. The first test involved defining a function called intersectionOfArrays. Copilot successfully predicted and implemented the function. However, the initial suggestions weren't ideal, and cycling through alternatives yielded similar results. Opening the Copilot option to synthesize 10 different possible solutions, didn't produce any new or better results, even though a cleaner implementation exists on Stack Overflow.
Next, the video explored building a simple counter app. While Copilot offered suggestions as the state variable was being defined, it sometimes felt overly aggressive with its predictions and even suggested code relevant to React or Vue in a non-React project. A comment like // get button from DOM was autocompleted. When adding an event listener to increment the state on a button click, Copilot surprisingly completed the comment before any code was even written. It set up the event listener correctly, incremented the state, and even updated the UI. However, it also introduced an unexpected modulus three operation, the reasoning for which was unclear.
Adding a button to the HTML with a comment prompted Copilot to autocomplete with the expected ID. In the CSS, a comment describing a red button with a blue background resulted in Copilot implementing the code line by line. However, results varied each time the Copilot was opened, even with the same comment, emphasizing that it's more of a starting point than a code-complete solution.
The Technology Behind Copilot
Copilot is powered by GPT-3, a deep learning model with over 175 billion parameters. GPT-3 is known for its creative abilities based on initial context, such as summarizing complex articles or generating ad copy. It was developed by OpenAI, initially a non-profit founded by figures like Elon Musk and Sam Altman, which later transitioned into a for-profit company. Microsoft invested $1 billion in OpenAI and secured exclusive licensing of GPT-3, positioning them to build unique products leveraging the algorithm.
Microsoft's Embrace and Extend?
Microsoft's increasing investment in developer tools like GitHub repos in VS Code, GitHub Actions, and now Copilot raises questions. While these tools enhance the developer experience, they also collect more data, potentially reinforcing AI algorithms. While this is speculative, the presenter jokes about a hypothetical four-step plan to replace developers with AI: Collect public code, create tools that make developers productive, collect more data as they use it, and then replace the developers with AI. Currently, Copilot is a technical preview but will be developed into a commercial product in the future, similar to AWS's CodeGuru.
Conclusion
GitHub Copilot is a promising AI-powered coding tool that can automate code generation. While it may not produce production-ready code consistently, it can save time, and is the "best AI coding experience" the presenter had. It leverages GPT-3 and a vast dataset of publicly available code. While the future impact of AI on programming remains to be seen, Copilot represents a significant step forward in AI-assisted development.
0 Comments