Suggestion: a more total ban on AI contributions

[daemonspudguy]

AI contributions create innumerable risks to a project's security because it makes it much harder to audit the source code, have massive questions on whether or not they're even legal to redistribute under any license at all, and have severe ethical issues because of both the stolen work that LLMs are inherently built on and the fact that these LLMs are being run in datacenters that largely get built in more impoverished areas because they're less able to fight back against them. Therefore I, as a user but also as someone who cares about this stuff, suggest that a total ban on generative AI in Prism Launcher.

[8ballpizza]

even if AI contributions are banned, people will still use it and just pretend not to. if it is at least somewhat "allowed" then people will be more open about using it and PRs can be properly labelled as AI assisted, allowing for more thorough reviews, etc.

[ptrcnull]

except even with a clear checkbox "did you use AI to write this" some simply won't check it, so reviews should be thorough regardless; meanwhile, a clear no AI policy will make it possible to just ban individuals who try to pass off claudeslop as their own

[lumi-me-not]

In my opinion, it is important to send a message that generative models are not okay and that the project is against them.

Sure, people may try to sneak it in, but that doesn't mean we should just accept it. This technology is very harmful to the world, and the more we make a stand against it, the better. It is better that they have to lie about using it than being able to openly and proudly use it.

Disclosing usage of generative models is just admitting to guilt. Admitting to guilt does not mean you get away with it, that makes no sense.

[solonovamax]

so for this reason alone it should be accepted?

this sounds like a dumb argument to me.

"if you don't accept it they'll do it anyways and lie!!!!!"

then just ban them?

it's like saying

if you don't let people add code that is improperly licensed, then they will do it anyways and just lie.

[TayouVR]

I think we generally agree with you, however banning AI entirely makes it so people just hide what is and isn't AI.

We find it to be better to judge code and contribution by the quality of the contribution. We have had some AI PRs so far, and almost always the PR author can't explain their code or reply to requests for changes. So the AI gets simply filtered out by the "author" being unable to comply with our code quality standards.

On my personal note:

I think the licensing problem is real, but we are so deep that realistically we have no way to ensure the source of any code, even from people who don't use LLMs. A big part of the internet is slop and looking at some random guide or blog post could mean copying code licensed under another license that's been through the slop machine.

Licenses are very important, but I have to be realistic here; us explicitly disallowing slop won't stop it from happening at all.
I also think there are some ethically sourced LLMs, that can even be run locally, minimizing the environmental impact.

I still think LLMs aren't capable of producing actual high quality code, but that aside, blanket banning it is imo not helpful.
Our current AI Policy is fair and allows legitimate use of local and ethical LLMs, acknowledges code ownership and therefore licensing by extension. I don't think there is a need for change here.

if you want to read up on our existing policy, it is here:
https://github.com/PrismLauncher/PrismLauncher/blob/develop/CONTRIBUTING.md#restrictions-on-generative-ai-usage-ai-policy

[daemonspudguy]

There is absolutely no way to have a copyright on AI extruded code in the United States. At all. It is literally not possible per the US court case Naruto v. David Slater et al. and by leaving the door open at all you're just opening yourself up to allowing wholesale unauthorized usage of the project because nobody legally owns it.

[ell1e]

however banning AI entirely makes it so people just hide what is and isn't AI.

Sorry for jumping in, but this confuses me. Wasn't stealing code previously also banned, yet people may have bypassed it? What's your data suggesting that the overall usage of AI will remain high once it's banned?

I wonder, if you trust contributors so little that they won't abide by this at all, perhaps it would be good to screen contributors some more. After all, they could also be sneakily introducing stolen Windows XP leak source code - whether they use an LLM to do that accidentally or on purpose. Lacking trust wouldn't just be limited to secret AI use, or would it?

There are a long row of projecs that seem to be having okay experiences with bans: Asahi Linux, elementaryOS, Forgejo, Gedit, Gentoo, GIMP, GoToSocial, Löve2D, Loupe, NetBSD, postmarketOS, Qemu, RedoxOS, SDL, Servo, stb libraries, Zig.

[ell1e]

ethical LLMs

Have you found a code LLM with properly licensed training data? I would be curious which one, because none that I saw seemed to do that.

[lumi-me-not]

I think we generally agree with you, however banning AI entirely makes it so people just hide what is and isn't AI.

Sure, but I think it's better to ban it and have them lie about it than allow it and send a message that it is okay to use generative models. They should, in my opinion, not be able to proudly use it.