@fsf Where can I read about the legal licensing and copyleft issues surrounding generative AI algorithms like LLMs (Large Language Models) like Chat-GPT or Copilot, trained on GPL'd source code?
I wonder if there is a need for a new license that explicitly makes training generative AI on open source code requires the AI model to be open sourced?
Does the FSF have any written opinions or educational materials related to this topic of the relationship between copyleft and generative AI trained on copyleft source code?
@neptune22222 I don’t know any — the EU basically provided a blanket copyright exception for training models as long as the artist did not decline explicitly (maybe that could be a clause for an updated GPL?), I don’t know about specifics in the US or South American, African or Asian states. @fsf
I notice that copyright holders, like Sarah Silverman, Christopher Golden, and Richard Kadrey are suing Meta and OpenAI for copyright infringement, so I hope that copyleft will have its legal battles in a similar vein.
Also, I'm betting that existing GPLv3 licenses don't mention being used as training data in other algorithms. I imagine there could be stronger legal wording in future licenses that specifies consequences for using copyleft software as training data for closed source algorithms.
Defending the copyright system seems at the heart and core of the copyleft free software agenda, and currently Sarah Silverman et al seems to be leading the charge.
@neptune22222@fsf "Defending the copyright system seems at the heart and core of the copyleft free software agenda"—the goal is not to defend the copyright system but to abolish proprietary works.
Copyleft is a means to turn copyright onto itself: prevent people from grabbing power over others.
Without copyright, copyleft would need a way to ensure that the sources are provided with binaries.
But OpenAI is a landgrab: new copyright built from copyleft ⇒ a problem.
I don't think that copyleft actually has any power over proprietary software. I don't see a legal end to users having a choice for proprietary software.
Currently, I see the copyright system as the only necessary legal core of copyleft. I'm open to learning other legal defenses of copyleft that don't depend on copyright at the core.
Is there another way to stop copyleft software from being used to create closed source software?
I imagine legal battles should currently be happening in this space, but I haven't heard of much, which is why I ask.
@neptune22222 Without copyright, copyleft needs other regulation as protection.
There’s the simple way to only allow selling software with full sources.
You’re not allowed to sell electric devices that randomly catch fire, so why should you be allowed to sell software or a device that routinely stops receiving security updates and can’t be fixed if the parent company loses the source?
(That does not originate with me. It’s an old argument I remember.)
@msw Will well-written licenses (and they are), the added restrictions from copyright automatically become part of copyleft enforcement.
GPLv3 was necessary because the technical environment changed and enabled circumvention of copyleft.
Who would have thought that people would actually limit universal computing devices to only run explicitly authorized code? (via allow-list) @neptune22222@fsf
@ArneBab@neptune22222@fsf you'll have to be more specific about what circumvention(s) of copyleft required changes in v3. I'm guessing you mean TiVo-ization?
@msw yes, I mean TiVoization: building a device on which only the one who sold it to you can allow binaries to run on a one-by-one basis.
What do you think why Google and Apple avoid GPLv3 (GNU userland) as if it were poison? If they shipped even one GPLv3 program, they’d have to allow users to put their own keys into the TPM so they could install anything without rooting.
They would lose their power to control the devices against the wishes of the users.
Add comment