This is a big part of why I think AI has not actually been intelligently designed for use in software. If it had, it would be way more efficient to work with. I do not want to have a conversation with a shitty intern just to fix a bug.

Either I'm accused of not understanding the tool, not trying hard enough to learn it, or of just lying. Every criticism of LLMs has to be prefaced with "of course, there are some useful use cases ..."

There aren't. Full stop.

Even if LLMs did work, there would still be countless reasons not to use it. But it doesn't. And for whatever reasons claiming "this tool doesn't work" is gauche in software. I'm allowed to complain for ethical or moral reasons, but if I say "This tool doesn't work" that's not kosher.

Every time I see someone at work share their "AI" documentation it sucks. It just isn't good. Every piece of AI code I've read is bad.

The only way an LLM creates sufficient output is if your idea of "sufficient" is "insufferably mediocre, lacking context, and wrong".

I’m 120 pages into it, but it feels like im nowhere. The physical footprint is large, the pages are thin, but the font is also larger than normal.

Altogether it confounds my ability to tell how fast I’m reading - how far I’ve come and how far I have to go.

LLMs aren't a gift from the tech gods, they're a mostly bad tool that you're shoving down our throats and claiming you don't have a choice.