tqbf wrote a blog post about using LLMs for coding (here https://fly.io/blog/youre-all-nuts/) and in a way I think he's right. not that it's a good tool or that it's "just progress" or that we couldn't possibly do anything about it. but that it's good *enough* for most of the things that programmers do, and that we *won't* do anything about it. programmers aren't a unionized workforce and they have no class consciousness.
this is making me really scared lol
"AI"
@wxcafe I have good news for you: it's still bullshit
Agents don't fix anything and produce more problems than they solve. That is the reason "vibe coding" does not work. After all, if agents were that good, incremental refinement would result in a converging code base... but instead, it diverges
I also can't help but notice that all of the routine tasks they're good at... were already automated to those in the know
This is "AI" hype aimed at self-styled intellectuals and arrogant sr devs
re: "AI"
@jhwgh1968 @wxcafe "I also can't help but notice that all of the routine tasks they're good at... were already automated to those in the know"
This is probably the one thing about this whole charade that infuriates me the most. How everything "AI" is always compared against having basically no tools at your disposal, rather than against the status quo of tooling, *and people actually buying into that narrative*.
re: "AI"
@joepie91 @wxcafe right? Like I had an IDE that could generate boilerplate function templates from inferred use as well as common code patterns and fragments using autofill (from a custom template language) all the way back in 2004!
The marginal benefit of an LLM if it made no mistakes is waaaaay smaller than they think it is. Enough for even a modest number of mistakes to wash it out in terms of execution time