Post #4118255
2026-07-24 19:44 UTC
@krig@goto.liten.app I've realized that for many¹, these systems really produce something not "just" faster than they could have, but indeed substantially better (on most metrics) - and some are aware of this.
They genuinely can't tell. They don't even edit it because they couldn't improve on it.
They don't want something better from someone else either - because again, they can't tell. And that might then be too difficult to grasp even, so why bother.
¹ if not most, when I have a dark moment
Replies (1)
-
@wwklnd@social.spejset.org 2026-07-26 18:29
@larsmb@mastodon.online @krig@goto.liten.app I'm weirdly torn when it comes to this. I have a feeling that a lot of these models could write better code than I would on a first pass for a lot of things, but on the other hand, whenever I've decided to actually try using a model for something I was working on (like some CircuitPython code for a KMK keyboard) just to see how it would handle it, it has failed spectacularly and rewritten functionality that already existed while also dreaming up fake dependencies.