Post #4014898
2026-07-22 15:03 UTC
@NullTheFool@mastodon.gamedev.place Also, I'm just going to say that it's incorrect. Humans make different kinds of mistakes, as there is still a certain level of understanding there. Also, we're actually good at intuiting the kinds of mistakes humans make. LLMs don't make mistakes, they extrude the wrong text, it's not a mistake, because they don't have any cognition. And because there is no cognition, the kinds of errors are harder to catch in review, because they happen in ways that are not intuitive to humans.
This is also exactly why machine learning can be a great addition to humans in detecting things, because what either detects is very different.
Replies (1)
-
@sotolf@polymaths.social 2026-07-22 15:15
@ainmosni@social.ainmosni.eu @NullTheFool@mastodon.gamedev.place Having cleaned up some things "created" by coworker with LLMs has shown me that dealing with LLM mistakes usually are way more tiring than fixing errors made by people too. Most of the code looks plausibly good, and the errors are usually way more hidden away than the errors that a normal person would make, they get hidden between layers of abstraction, and it doesn't seem like there are much try at simplifying things, rather the opposite. I at least find myself tire from fixing and debugging those things so much faster than from something where someone at least tried thinking something through.