Mastodon Feed
Boosted by baldur@toot.cafe ("Baldur Bjarnason"):
zzt@mas.to ("[object Object]") wrote:
I’ve got very little patience for software engineering “professionals” coming into this thread to pretend that LLMs are good at finding bugs, as if we’re not in the same profession and I haven’t been forced to use the fucking things
all research not funded directly by AI corporations shows they’re fucking terrible at finding bugs and introduce many more than they fix, and that’s in line with personal experience
the machine will constantly gaslight you into thinking your code is riddled with bugs and it’s fixing them, and by pure luck it might even be right! you’re very likely to remember the times it’s right more than the times it absolutely fucks it
it also really, really wants your code to be a formless blob, and the thing’s output will get hilariously manipulative if you push back on it at all, as you would with a person reviewing your code. and people using an LLM to review your code will get real shitty with you too, cause the machine said you’re wrong.