Hacker News

points by whizzter 2 days ago | 0 comments

I think technically such a model should abide by the licence of the inputs, hence all models would probably be GPL if the FSF had resources to fight it and the legal profession had an inkling of the processes involved.

Or maybe the FSF is just waiting for everyone using AI code to become hooked and then subpoena every Antrophic/OAI/Codex customer and force the most massive source release in history (If Adobe/MS would file lawsuits vs the Craft apps and with their resources create legal precedent it could create chaos, maybe the only part stopping them right now from filing lawsuits is them realizing it could backfire since they're heavy AI users themselves).

[–] origini 2 days ago

But as far as i know GPL does not prevent you from using it for training?

[–] whizzter 2 days ago

No, the question is, if the GPL'ed code/structure is more or less recovered during inference, would a court rule the result to be covered by the GPL?

I know there was some cases in the past where people swapped licences and changed variablenames,etc but the essential code clearly remained.

[–] origini 2 days ago

Yeah for sure. But then the question becomes weird: If its binary equal (like asm) but nothing else, will it hold up?

Im quite aware of software patents, i never had to use some software patent scanner in our compliance CI/CD.

So it might be a real issue no one can or does enforce?

[–] whizzter 2 days ago

A while back I was thinking it wouldn't be enforced, but the entire Craft-clones and REA tools usage explosion kinda hints that disassembling/rebuilding the source is cheap enough that LLM's(ironically) could probably quickly find culprits if you had suspicions about certain apps.

[–] semiquaver 2 days ago

Not a lawyer, are you?

[–] whizzter 2 days ago

Does it matter? Lawyers don't know either since most majorer cases are still in progress.

So far it seems more cases that have been decided in lower instances or has been settled or gone in the direction of the party with most money for lawyers, still even in those cases there seems to be a slight tilt towards rightsholders.

I'm not saying anything categorically, but if some cases go all the way and towards rightsholders, it could become a real shitshow in the software industry since we've adopted AI so heavily and the non-significatng portion of public GPL licenced code could bite people in the ass.

[–] semiquaver 1 day ago

Yeah, it does matter. This:

  > I think technically such a model should abide by the licence of the inputs

is vibes baloney.

[–] whizzter 15 hours ago

Maybe maybe not, something interesting popped up reading the currently top-of-HN article on the GHC policy is that OpenJDK currently forbids AI contributions.

From the page, https://openjdk.org/legal/ai

--------------------------

What are the intellectual-property risks of using generative AI tools?

The Oracle Contributor Agreement (OCA) requires that a contributor own the intellectual property rights in each contribution and be able to grant those rights to Oracle, without restriction. Most generative AI tools, however, are trained on copyrighted and licensed content, and their output can include content that infringes those copyrights and licenses, so contributing such content would violate the OCA. Whether a user of a generative AI tool has IP rights in content generated by the tool is the subject of active litigation.

--------------------------

But they're just Oracle lawyers so it's baloney right?

[–] semiquaver 12 hours ago

You’re confusing two very different things. One is Oracle being careful with the provenance of IP that they let allow into their copyrighted codebase. That’s prudent in the face of real legal uncertainty over the issue. The other is

  > technically such a model should abide by the licence of the inputs, hence all models would probably be GPL

so whether a the weights file of a machine learning model itself can “be” GPL as a result of including GPL content in its training corpus.

That question is a category error because machine learning weights are not copyrightable at all, being mere lists of numbers and facts, the result of a statistical process and not a human creative hand, very much ineligible for copyright protection. They would be protected by trade secret law only, like the coke recipe.

So the GPL, which operates entirely within the domain of copyright law, cannot possibly virally spread from a licensed source code file to a machine learning model which trained on it. That would be a cross species jump. Certainly there could be other legal consequences but in no universe is someone going to jump out and yell gotcha and force OpenAI to personally hand deliver the newly GPLed weights for their frontier model to RMS.

[–] whizzter 2 hours ago

Right, I can agree to a certaint extent that the model itself can be equated to storage.

I think you misjudged my intent with that initial comment, it was mostly that I'd be feeling easier about AI usage if I knew the models WEREN'T trained on GPL'ed code, people producing GPL'ed software has historically been quite profilic in certain areas and the odds of copyrightable parts of GPL'ed code not showing up in AI code managing certain fileformats is definetly nonzero.

Ie, I'd be happier using Claude Code-nogpl, I'd be fine with adding BSD,etc copyright notices if given provenance information from the model, being in the GPL minefield because they don't really seem to track inputs is another.