This is an abomination of a #Rust code base (not sure if you can even call it that).
https://github.com/vjeux/pokemon-showdown-rs
It's pure vibe-coded slop. I've always wondered what it would look like. Now I know, thanks.
Here's the accompanying article by the "author":
The article is quite funny, because you can't really tell if the author is actually serious or just trolling.
@janriemer I wonder how much it cost to run it 24/7 like that
@janriemer Oh no... I hope that doesn't go upstream, my eldest plays that and I can only imagine the places that gets wrong at the high end of competition (where she enjoys playing).
@trini I'm sure this thing won't go anywhere, no worries. 🙂
@janriemer it really seems they went as fast an loose as possible 😂. the code is probably horrible, but the trick with vibe coding to that extreme, is that you don't look at it anymore than a python dev would look at assembly, it's bust not the level of abstraction you work at anymore. But it's only (possibly, not yet proven) sustainable with strong guidelines about code architecture, maintenance, etc, as you would expect a senior engineer to follow, or neither human or AI will understand it.
@janriemer (i didn't open source any vibe coded codebase, in part because i don't think there is anyone wanting that, but i have been experimenting, and you can be a tad more serieous about it than them, and get pretty impressive results)
@tshirtman Yes, if done with proper software engineering practices, this is one of the more useful applications of LLMs in coding, totally agree!
@janriemer there are still problems, as the author points out, in learning to emulate developers, these agents learned lazyness and bad excuses to avoid doing the work, and claude in particular will often claim to be done doing something, when it absolutely is not. I find codex to be more thorough, and to keep doing until a usable solution is produced, but even then, if you don't give it good ways to verify the work is done, it'll often stop short.
@tshirtman Yeah, I don't think this is surprising, given how LLMs are architected.
They are reward-maximization-seekers: anything that looks like it can help them achieve their goal faster, they'll output.
¯\_(ツ)_/¯
LLM Coding Slop
@janriemer
Sure it's not written by an AI?
@musicmatze Oh, you mean the article itself!?
I haven't even thought about that! Could be, who knows...🫠
