PeterI worked quite a bit with ChatGPT last week and one thing that stands out to me is LLMs will always write more lines of code rather than fewer. they love adding another type check, another try/except block, another condition to the YAML file. just in case! "your code is almost perfect, but consider adding this one thing..." it reminds me of someone who pointed out that a line of code isn't an asset, it's a liability because it's another thing to maintain.
Brian Vargas@peter@thepit.social I got in an argument with the work PR bot the other day because it was insisting I add null checks to handle a condition that wouldn't ever happen. And when I pointed that out, it moved the goal posts to "defensive coding" in case it might happen someday. And when I finally told it was wasting my time because it was a dumb bot – no wait, I wasting my OWN time arguing with a dumb bot – its response was blocked by "security policy".
It was really hard to not think it was sulking.
Peter@Ardvaark@mastodon.world lmao PR bot???? fuuuck
Brian Vargas@peter@thepit.social I’m still reserving judgement. It frequently makes good suggestions, and sometimes catches real bugs. But it just as frequently makes inane suggestions to make massive refactoring on otherwise simple changes, and sometimes is just straight wrong.
It’s not clear yet whether the time spent sifting through the slop is worth the things it gets right. I’m leaning towards “not worth it” right now, but that could change if it gets better.