Hacker News Reader

Comment by bilekas | Hacker News Reader

Unfortunately as I see it, even if you want to contribute to open source out of a pure passion or enjoyment, they don't respect the licenses that are consumed. And the "training" companies are not being held liable.

Are there any proposals to nail down an open source license which would explicitly exclude use with AI systems and companies?

View on HN

All licenses rely on the power of copyright and what we're still figuring out is whether training is subject to the limitations of copyright or if it's permissible under fair use. If it's found to be fair use in the majority of situations, no license can be constructed that will protect you.

Even if you could construct such a license, it wouldn't be OSI open source because it would discriminate based on field of endeavor.

And it would inevitably catch benevolent behavior that is AI-related in its net. That's because these terms are ill-defined and people use them very sloppily. There is no agreed-upon definition for something like gen AI or even AI.

Even if you license it prohibiting AI use, how would you litigate against such uses? An open source project can't afford the same legal resources that AI firms have access to.

I won't speak for all but companies I've worked for large and small have always respected licenses and were always very careful when choosing open source, but I can't speak for all.

The fact that they could litigate you into oblivion doesn't make it acceptable.

Where is this spirit when AWS takes a FOSS project, puts it in the cloud and monetizes it?

It exists, hence e.g. AGPL.

But for most open source licenses, that example would be within bounds. The grandparent comment objected to not respecting the license.

The AGPL does not prevent offering the software as a service. It's got a reputation as the GPL variant for an open-core business model, but it really isn't that.

Most companies trying to sell open-source software probably lose more business if the software ends up in the Debian/Ubuntu repository (and the packaging/system integration is not completely abysmal) than when some cloud provider starts offering it as a service.

Fairly sure it's the same problem and the main reason stronger licenses are appearing or formerly OSS companies closing down their sources.

you are saying X, but a completely different group of people didn't say Y that other time! I got you!!!!

It’s fair to call out that both aspects are two sides of the same coin. I didn’t try to “get” anyone

um, no it's not. you have fallen into the classic web forum trap of analyzing a heterogenous mix of people with inconsistent views as one entity that should have consistent views

> Unfortunately as I see it, even if you want to contribute to open source out of a pure passion or enjoyment, they don't respect the licenses that are consumed.

Because it is "transformative" and therefore "fair" use.

Fair use is an exception to copyright, but a license agreement can go far beyond copyright protections. There is no fair use exception to breach of contract.

I imagine a license agreement would only apply to using the software, not merely reading the code (which is what AI training claims to do under fair use).

As an analogy, you can’t enforce a “license” that anyone that opens your GitHub repo and looks at any .cpp file owes you $1,000,000.

Running things through lossy compression is transformative?

The quotation marks indicate that _I_ don't think it is. Especially given that modern deep learning is over-paramaterized to the point that it interpolates training examples.

Reply on news.ycombinator.com