[domepro]

Yeah, I'm really wondering if I'm dumb or something with this whole AI train, but my job as a developer that maintains existing (sometimes quite old) systems is much more about determining what actually needs to be done, gathering lost / not lost knowledge about the existing state of things in systems and figuring out where to change the code for maximum impact and minimal side effects. Lots of talking to customers/non-technical people as well to figure out if existing features can cover something they need and they just don't know about it and I happen to know about it etc.

I've been watching the AI hype (again) and it's completely going over my head, I really really don't get it - I doubt {x}GPT would be able to analyze hundreds of thousands of project code across a big system (multiple languages, multiple services, interconnectedness of everything) and tell me what to do, even in the case I would want to paste in everything proprietary to a 3rd party service which I really don't.

Maybe it just works for a subset of programming, but then again I don't really see how it differs to reading docs for generating boilerplate or whatever it's useful for in greenfield projects. I was also not really impressed by it generating improvements to code when you ask it the right questions since that's the hard part of the job, not typing out the code.

Maybe I'm just a dinosaur and dumb.

[tzekid]

Don't downplay yourself like that. A 25k token GPT model is a step in that direction but we're 3 - 15 years from your vision of valuable.

Currently, the Language Models (GPT or GithubCopilot) are mostly good as ... copilots. You bounce off ideas off them or use them to kickstart something you want to do. It's great for junior devs (me!) but it still cannot do big context or cognition (what you describe), that part will be done by humans for quite a while.