I think simply never using a project with emoji in the readme is sufficient to capture 90% of slop without removing much useful.
That being said, there is no way to determine AI generated from not, consistently, and over the long term. I worry you’ll have people making accusations and there’s no practical way to disprove them. You’ll also have people working harder to hide AI generated code (maybe that’s good overall if they are actually editing and revising the slop code?).
But if someone uses AI to build a rough draft, then rewrites 80% for clarity, correctness, etc. is that worthy of banning their account? What if someone runs their repo through AI to check for bugs and vulnerabilities?what proportion if content becomes “predominantly”? What about someone using it for translation of the text parts of their repo?
I just think there are very hard problems this policy creates, which take away from other resources to manage.
This is partially why I chose a paid host with resource limits. I have one repo with a large benchmark library that I wouldn’t feel okay putting on a free service, for example.
You can get high accuracy if you don’t care about false negatives and only ban the most egregious shit.
I think simply never using a project with emoji in the readme is sufficient to capture 90% of slop without removing much useful.
That being said, there is no way to determine AI generated from not, consistently, and over the long term. I worry you’ll have people making accusations and there’s no practical way to disprove them. You’ll also have people working harder to hide AI generated code (maybe that’s good overall if they are actually editing and revising the slop code?).
But if someone uses AI to build a rough draft, then rewrites 80% for clarity, correctness, etc. is that worthy of banning their account? What if someone runs their repo through AI to check for bugs and vulnerabilities?what proportion if content becomes “predominantly”? What about someone using it for translation of the text parts of their repo?
I just think there are very hard problems this policy creates, which take away from other resources to manage.
I guess if you can find the worst offenders you can ease server load, like if they make pushes around the clock too fast, that’s one good signal.
Like if you have 2000 pushes in a day (pushes, not commits) and it’s clearly 24/7, that’s probably an autoslopper.
If someone is not doing that, who cares, the “agents directing agents” folks are the ones that cause the most problems.
Hm, I’m a heavy pusher. Tho, I don’t believe I’ve gotten past 200 a day
Git is actually very efficient. You need to be very inefficient with your workflows to cause strain on a git server.
The fact that you can just host a git server for free and most people don’t need to ever worry about limits is a wonder.
This is partially why I chose a paid host with resource limits. I have one repo with a large benchmark library that I wouldn’t feel okay putting on a free service, for example.
There’s been studies about the shape and structure of text. Humans go on tangents AIs don’t