I worked with ChatGPT to produce a very simple extension for Hacker News.

Each commit contains my prompt in the commit message, and the commit is entirely composed of ChatGPT output (it took 25 prompts to get this repo to its current state)

No files were changed at all from what ChatGPT suggested, but the `gpt-output` file is an append-only log of ChatGPT's verbatim output (so you can select any commit and see exactly what was prompted and what ChatGPT responded with)

I'm using this extension now, and it works

I was hoping to get a sense of whether ChatGPT could debug its own code, and fix bugs, and to my surprise, it can! I had to prompt it to tell me where to put debug messages, and then provide the error output of course.