being in programming circles, I often hear about how good LLMs are at programming. hell, I've even seen it leak into non-developer circles.
I won't lie, their code is not... the worst I've ever seen -- I've seen people do much worse, LLM code can be surprisingly good... their code can also be terrible slop, though! and so, a person has to watch over the process and review the code it produces -- and therein lies my issue.
I went into programming because I like writing code! I like solving problems and I love trying to come up with concise-yet-readable ways to express these solutions!
I DON'T love reviewing code -- it's a process that's typically boring, yet also requires sustained attention, lest you miss some kind of subtle error between files. usually, you can at least put some trust in the code, knowing that another person -- the one who wrote it -- had also put some thought into it, and that you can always ask that person for clarification, in case of doubts. that's not the case with LLMs.
yet the future that AI maximalists propose is one where the machine does the fun part (programming), and I not only have to do the boring part (reviewing), but do it extra carefully. once again we've arrived at the machine that does your art so you can do your chores, and to quote something I'm SURE you've heard at least once by now:
I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes.
I'm sure someone would be quick to point out that I can also have LLMs do code review -- and sure, I absolutely can! having seen how github's copilot performs at my job, though, I'm not sure I'd be able to trust them. too often does copilot miss anything but the most obvious issues, while occasionally also hallucinating nonsense elsewhere.
I'm not sure if I'm building towards any important point here, I think I just wanted to vent my frustrations a little. I can see AI being a tool to leverage. I've used copilot once or twice before, to take care of tedious, boring bits like boilerplate, or troublesome conversions, it works quite well! I would hate it if it became the main way of working with code, like so many AI maximalists propose, though.
⊹˚ * ⊹ ˚ ⊹
thanks for reading!! :3