Audio Flamingo 3 - Fully Open Large Audio Language Models
Audio Flamingo 3 - Fully Open Large Audio Language Models
We present Audio Flamingo 3 (AF3), a fully open state-of-the-art (SOTA) large audio-language model that advances reasoning and understanding across speech, sound, and music. AF3 introduces: (i) AF-Whisper, a unified audio encoder trained using a novel strategy for joint representation learning acros...
You're viewing a single thread.
> fully open
> looks inside
> not open source
To save everyone a click: It's a non-commercial license (with a very rude yoink clause, if anyone is foolish enough to build something on it.)
By the by, there's a good chance that AI models are not copyrightable under US law; making the license moot in the US. In other regions, such as the EU, it likely holds.
3.3 Use Limitation. The Work and any derivative works thereof only may be used or intended for use non-commercially. Notwithstanding the foregoing, NVIDIA Corporation and its affiliates may use the Work and any derivative works commercially. As used herein, “non-commercially” means for non-commercial research and educational purposes only.
i know that AI output is not copyrighteable, because it wasn't made by a human.
however the model itself is a product of a shit ton of work. and I doubt any court will claim them non copyrighteable.
That's not how US copyright works.
My idea would be to slightly modify / fine-tune a model and then redistribute that modified version. And claim the same Fair Use, AI companies use to take people's copyrighted work. Either that makes it Fair Use as well, or the "no originality required" collapses or the entire business model.
People do that all the time.
I don't see how that would be fair use or what the argument is supposed to be.
Let me warn you that Lemmy is full of disinformation on copyright. If you picked the idea up here, then it probably is absolutely bonkers.
In any case, fair use is a US thing. In the EU, it would still be yoink.
I think I used a bit too much sarcasm. I wanted to take a spin on the idea how the AI industry simultaneously uses copyright, and finds ways to "circumvent" the traditional copyright that was written before we had large language models. An AI is neither a telephone-book, nor should every transformative work be Fair Use, no questions asked. And this isn't really settled as of now. We likely need some more court cases and maybe a few new laws. But you're right, law is complicated, there is a lot of nuance to it and it depends on jurisdiction.
There is a lot of disinformation being spread on copyright because major rights-holders hope to gain a lot of money for nothing.
US fair use has always worked like this. Other countries without fair use had to make laws to enable AI training. I know about Japan and the EU.
It is precisely because of these new laws that AI training in the EU is possible at all (eg by Mistral AI or by various universities/research institutions). But because of lobbying by rights-holders, this is quite limited. It's not possible to train AIs in the EU that are as capable as those from the US, where Fair Use comes directly from the constitution and can't be easily lobbied aside by monied interests.
Hmmh. It's a bit complicated. "Fair Use" is a concept in Common law countries, but lots of European countries do it a bit differently. We here in Germany need specific limitations and exceptions from copyright. And we have some for art and science, education, personal use and citations and so on. But things like electronic data transfer, internet culture and more recently text- and datamining needed to be added on top. And even datamining was very specific and didn't fit AI in it's current form. And we don't have something like Fair Use to base it upon.
From my perspective, I'm still not entirely convinced Fair Use is a good fit, though. For one it doesn't properly deal with the difference of doing something commercially and for research or personal use, and I believe some nuance would help here, big rich companies could afford to pay something. And as AI is disruptive, it has some effect on the original work and balancing that is somehow part of Fair Use. And then the same copyright concept has higher standards for example in music production and sampling things from other songs that are recognizable in the resulting work. And I don't think we have a clear way how something like that translates to text and AI. And it can reproduce paragraphs, or paint a recognizable Mickey Mouse and in some way it's in there in the model and leads to other issues. And then all the lines are blurry and it still needs a massive amount of lawsuits to settle how much sounding like Scarlett Johansson is too much sounding like her... I'd say even the US might need more clarity on a lot of legal questions and it's not just handled by Fair Use as is... But yeah, "transformative" is somewhat at the core of it. I can also read books, learn something and apply the knowledge from it. Or combine things together and create something new/transformative.
Of course, it would be better if governments would pass sensible laws on AI training. These lawsuits are a complete waste. But you can see the problem in Europe. The copyright industry has too much power. You don't get good laws. (In fairness, Japan did pretty well.)
For one it doesn’t properly deal with the difference of doing something commercially and for research or personal use, and I believe some nuance would help here,
That needs to be considered in fair use, but I don't see what difference it would make here.
big rich companies could afford to pay something.
That's a line by the copyright lobbyists. But economics doesn't work like that.
In a competitive market, producers must pass on costs. EG coffee and cocoa beans have become more expensive on world markets in the last year, so now coffee and chocolate are more expensive in stores.
AI is quite competitive. If AI firms are forced to pay license fees, then AI subscriptions will become more expensive for consumers. The money goes straight from everyone to rights-holders; a few people at the top.
Sure. I mean we're a bit different at both sides of the Atlantic. Europe regulates a lot more. We're not supposed to be ripped off by big companies, they're not supposed to invade our privacy, pollute the environment unregulated... Whether we succeed at that is a different story. But I believe that's the general idea behind social democracy and the European spirit. We value our freedom from being used and that's also why we don't have a two weeks notice and we do have regulated working hours and a lot of rules and bureaucracy. The US is more freedom to do something. Opportunity. And in my eyes that's the reason why it's the US with a lot of tech giants and AI companies. That just fosters growth. Of course it also includes negative effects on society and the people. But I don't think "right" and "wrong" are fitting categories here. It's a different approach and everything has consequences. We try to balance more, and Europe is more balanced than the US. But that comes at a cost.
That's a line by the copyright lobbyists [...]
Well, I don't think there is a lot of good things about copyright to begin with. Humanity would be better off if information were to be free and everyone had access to everything, could learn, remix and use and create what they like.
I think of copyright more as an necessary evil. But somehow we needed Terry Pratchett to be able to make a living by writing novels. My favorite computer magazine needs to pay their employees. A music band can focus on a new album once they get paid for that... So I don't think we need copyright in specific. But we need some way so people write books, music etc... Hollywood also did some nice movies and tv shows and they cost a lot of money.
I don't have an issue with AI users paying more. Why should we subsidise them, and force the supply chain to do work for a set price? That's not how other businesses work. The chocolate manufacturer isn't the only one making profit, but an entire chain from farmer to the supermarket gets to take part in earning money, which culminates in one product. I don't see why it has to be handled differently for AI.
And what I like about the approach in Europe is that there is some nuance to it. I mean I don't agree 100% but at least they incentivise companies to be a bit more transparent, and they try to differentiate between research to the benefit of everyone and for-profit interest. And they try to tackle bad use-cases and I think that's something society will appreciate once the entire internet is full of slop and misinformation by bad actors. Though, I don't think we have good laws for that as of now.