Hi Adam, nice write up. Your examples don’t sound implausible, but I noticed they tend to focus on well-known areas of web development. Video games I can also imagine to some extent, because of all the layers you can put on top of high level game engines.

What about some of the lesser travelled paths? Like systems development - especially systems that are poorly documented. If Chat GPT is trained on what’s known, what happens with little-known areas?

Remind me to ask Chat GPT to write a virtual audio driver for macOS in a year or two. It might be able to find boilerplate based on Apples’s NullAudio code, which is 3,000 lines of necessary no-op. Basically a hello world that doesn’t say anything. I’d be super impressed if Chat GPT can figure out how to turn that 3,000 lines into a working driver based on some functional requirements that you tell it.

Or maybe a bidirectional XPC mechanism that will talk between an application and the sandboxed audio driver. Apple’s sample code and documentation on XPC only supply unidirectional examples. There’s a hint of what you need to do for bidirectional in StackOverflow somewhere, but if the internet is Chat GPT’s classroom, I doubt it can write anything usable.

Edit: these above examples supplement the more detailed but entirely different example in my article on the topic.

In many areas of macOS development, sometimes old examples simply won’t work after the introduction of SIP. Because of this, among various other factors, many internet searches end up with nothing useful. Reading header files and finding private APIs is sometimes the only way - and it still requires a lot of context and original thought.

Maybe if Apple trained Chat GPT it might be a different story, but they’re also pretty secretive themselves, so I don’t see that happening.