2.6 KiB
What's it like to code with GPT-4 and aider?
Below are some chat transcripts showing what it's like to code with the help of GPT-4 using the aider command line tool. In the chats, you'll see a varity of coding tasks like generating new code, editing existing code, debugging, exploring unfamiliar code, etc.
-
Hello World Flask App: Creating a simple Flask app with various endpoints, such as adding two numbers and calculating the Fibonacci sequence.
-
Pong Game with Pygame: Creating a simple Pong game using the Pygame library, with customizations for paddle size and color, and ball speed adjustments.
-
2048 Game Modification: Exploring and modifying an open-source javascript repo for the 2048 game, including adding randomness to the scoring system.
-
Complex Multi-file Change with Debugging: A complex code change involving multiple source files and debugging.
-
Semantic Search & Replace: Updating a collection of function calls, which requires dealing with various formatting and semantic differences in the various function call sites.
-
CSS Exercise: Animation Dropdown Menu: A small CSS exercise involving adding animation to a dropdown menu.
-
Automatically Update Docs: Automatically updating documentation based on the latest version of the main() function.
What's happening in these chats?
To better understand the chat transcripts, it's worth knowing that:
- Each time GPT-4 suggests a code change,
aiderautomatically applies it to the source files. - After applying the edits,
aidercommits them to git with a descriptive commit message. aidernotes each time a source file is added to the chat session. Once added, these files are available for review and editing by GPT-4. The user adds files either via the command line or the in-chat/addcommand. If GPT-4 asks to see specific files,aiderasks the user for permission to add them to the chat.
Transcript formatting
Output from the aider tool is shown in a blockquote like this.
> The user's chat messages are shown on a prompt line like this. They contain they user's change requests, clarifications, etc.
Responses from GPT-4 are in a plain font like this, and often include colorized "edit blocks" that specify edits to the code. Here's a sample edit block that switches from printing "hello" to "goodbye":
hello.py
<<<<<<< ORIGINAL
print("hello")
=======
print("goodbye")
>>>>>>> UPDATED