[tazjin]

My current take is that if you start an open-source project now, you should go full AGPL (or similar copyleft license), and require a CLA for contributors.

If your thing ends up actually good you now have a defence against exploitation, and a way to generate income reliably (by selling the code under a different license). afaik, organisations like the FSF even endorse this.

[heavyset_go]

AGPL is my first choice of license, but its efficacy does not necessarily come from its teeth, but from the aversion legal departments have towards the license. It's similar to how the GPL used to be, or still is, treated. Along with compatibility with other AGPL projects, that's the reason I use the license.

There are situations that the AGPL does not cover that could be considered leeching from the commons.

I think we need stronger licensing, and binding contracts that forfeit code recipients' right to fair use in order to hinder LLM laundering, along with development platforms that leverage both to limit exploitation of the commons.

[matt-p]

I agree, I'm quite curious on what feelings are about still putting it in a public GitHub repo?

AI models will train on your codebase, unethical actors will still take it and not pay. Others can give the .zip to Claude and ask it to reimplement it in a way that isn't license infringement. I think it really turns open source upside down. Is this a risk worth taking or best to just make getting the source something that's a .zip on a website which the models realistically won't train on.

[TeMPOraL]

Or maybe ask yourself why are you doing open source in the first place?

AI training on your code is success if you care about your code being genuinely helpful to others. It's a problem only if you're trying to make money or personal reputation, and abusing open source as a vector for it.