So I previously talked about how code copyrights are nonsense . This is something I still believe in. The thing is, while this argument may seem like I am in favor of "vibe coding", aka using LLMs to code things, I'm actually not. I certainly think coding is not particularly artistic, and is more like puzzle solving; however, I always also vaguely felt that using LLMs to code is still wrong. And now, I think I've figured out why. The core problem is, coding with LLMs is inherently dishonest . Sure, in theory a human nicking code from Github, stackoverflow et al is no different from an LLM "generating" code out of the codebases it has scraped, if one only looks at the end result (and ignores that, since LLMs don't actually "know" or "understand" anything, they're far more likely to randomly insert mediocre or downright buggy code, compared to just...copying code off the web). The problem is, the human nicking the code can choose t...