It's impossible, I'll never understand why anti ai repos draw such a hard line instead of saying hey we won't even look at 1000+ LOC PRs or design reworks with no discussion.

Both of those are pretty reasonable policies that will kill most of the fly by AI requests.

Yes, and I do agree with this.

If it’s not possible to know if a contribution is AI and it’s a good contribution, it doesn’t matter much. And if it’s bad there’s no point in discussing if it is AI.

Prob banning ppl from running agents that are automatically pushing to github and spamming all day

If you’ve seen a typical GenAI pull request, it’s blatantly obvious. If it has been curated well, sure it’s probably not feasible to know, but “slop” is easy to tell, and the target users have very slop contributions.

Well yes it is obivous if it is vibecoded usually, but if you use it as a tool you write most of it by hand or have it help out with certain functions etc. I hope you get my point! :D

Presumably this is why it says “substantially generated by generative AI”, a diplomatic way of saying “GenAI slop”. GenAI isn’t necessarily completely off the table, but a very common usage scenario is. Admittedly seems subjective, but it provides a criteria to save maintainers from having to review unreasonable contributions without them having to identify some more concrete specific issue that is likely there, but it’s going to be hard to find.

Also, unless it is very digestable, then I wouldn’t even trust myself to be vigilant enough to catch the real issue even if I looked. It does a good job of looking credible like it should be right, even when it isn’t.

“How would they know if my contribution is actually compatible with their open source license? I could just contribute code from an incompatible license, or proprietary codebase, without telling anyone.” This was supposedly an actual argument against FOSS licenses back in the days. Yeah, you can’t be 100% sure, but: You made your intent clear. And that’s honestly what most people care about. Then all you need to do is do some rudimentary checks for quality and maybe do a quick vibe-check (heh) on the person contributing.

Okey thank you. My question was an actual question, if there was like a “tell” or something! :D

There’s dozens.

Large commits, over documentation, and not being able to answer questions about the OR are good starts usually.

Those aren’t really tells for someone just doing “tweaks” like they said. Or someone using a little “AI” like google search. Etc.

Ofc there will be obvious cases but in general there’s no way to detect how code has been generated - especially since there’s no such thing as “AI” in the first place.

My honest answer?

If someone is putting in the effort to tailor their LLM to produce results like that, I’m actually fine with.

I have yet to see an LLM thats able to do that by default. The only way you get there is have a human so close in the loop, theyre still doing the thinking work, but just not the physical typing. At least as I’ve seen, and I have less of a problem there.

The shelved things I have there are environmental, political, human factors, and stolen IP. Those are still fundamental blockers to me being okay with AI. But from a technical output, I dont have a problem with that limit

Exactly. Even if it isn’t immediately detectable, once it is you’ve run afoul of the rules twice, so you can get extra punished.

Thats interesting history, thanks for explaining!

They won’t…

Knowing piefed they were using AI, got called out, and are now “banning” it for the optics.

It would be worse if they actually believed it, but they probably don’t.

They won’t. These rules are about making sure the author reviews. It’s pretty easy to tell when Claude makes a PR, but it’s impossible to tell you used Qwen, fixed the duplicate code, removed useless verbose comments, and moved things to the right place. Reviewers are perfectly happy with the later.