Comment by jandrese

I think some of it is just needing to be a counter balance to the unending hype train from the AI companies that have an existential need for AI to be the next big thing. That and seeing these amazing demos and press releases, then trying it out yourself and finding that it's not nearly as capable as advertised and many of those demos are bullshit.

Even the "programmers are now obsolete" predictions have really not panned out as more people discover that while AI is an absolute boss at solving college homework level problems, it entire falls apart completely or requires a ton of babysitting to work on problems larger than homework exercises.

It's certainly useful at times, but you always have to keep problems about homework sized and generally focused on a single thing that has been done many times before or you'll have a bad time.

Good uses of AI: Asking it to implement a GStream Webp to backing store module in a GTK project.

Bad use of AI: Asking it to implement a libnftables interface for adding and removing firewall rules. The AIs completely hallucinate the entire API and output useless garbage. Plus the API is high level so even when you try to input it into the context the AI engine gets very lost. You can fight with trying to get it to understand, but in the end it's faster to just write the code yourself.