11–20 of 361 posts
In my opinion LLMs are intellectual property theft. Just as if I started distributing copies of books. This substantially reduces the incentive for the creation of new IP. All written text, art work, etc needs to come imbued with a GPL style license: if you train your model on this, your weights and training code must be published.
yes, this is indirectly hinting that during training the GPL tainted code touches every single floating point value in a model making it derivative work - even the tokenizer isn't immune to this.
I know for a fact that all SOTA models have linux source code in them, intentionally or not which means that they should follow the GPL license terms and open-source part of the models which have created derivative works out of it. yes, this is indirectly hinting that during training the GPL tainted code touches every single floating point value in a model making it derivative work - even the tokenizer isn't immune t…
A tokenizer's set of tokens isn't copyrightable in the first place, so it can't really be a derivative work of anything.
In my opinion LLMs are intellectual property theft. Just as if I started distributing copies of books. This substantially reduces the incentive for the creation of new IP. All written text, art work, etc needs to come imbued with a GPL style license: if you train your model on this, your weights and training code must be published.
The problem here is enforcement.
It's well known that AI companies simply pirated content in order to train their models. No amount of license really helps in that scenario.
In my opinion LLMs are intellectual property theft. Just as if I started distributing copies of books. This substantially reduces the incentive for the creation of new IP. All written text, art work, etc needs to come imbued with a GPL style license: if you train your model on this, your weights and training code must be published.
>This substantially reduces the incentive for the creation of new IP. Not all, but some kind of IP. Some of those that is created for sake of creating it and nothing else.
It's come up quite often even before AI when people released things under significantly looser licenses than they really intended and imagined them being used.
In my opinion LLMs are intellectual property theft. Just as if I started distributing copies of books. This substantially reduces the incentive for the creation of new IP. All written text, art work, etc needs to come imbued with a GPL style license: if you train your model on this, your weights and training code must be published.
> if you train your model on this, your weights and training code must be published. This feels like the simplest & best single regulation that can be applied in this industry.
I can see this being extremely limiting in training data, as only "compatible" licensed data would be possible to package together to train each model.
Earlier quoted context omitted.
> if you train your model on this, your weights and training code must be published. This feels like the simplest & best single regulation that can be applied in this industry.
It'd substantially reduce the incentive for the training of new models.
for the burglar
Earlier quoted context omitted.
> if you train your model on this, your weights and training code must be published. This feels like the simplest & best single regulation that can be applied in this industry.
It'd substantially reduce the incentive for the training of new models.
Facebook was busted torrenting all sorts of things in violation of laws/regulations that would lead to my internet being cut off by my ISP. They did it at scale and faced no consequences. Scraping sites, taking down public libraries, torrenting, they just do whatever they want with impunity. You should be angry!
Earlier quoted context omitted.
> if you train your model on this, your weights and training code must be published. This feels like the simplest & best single regulation that can be applied in this industry.
I feel to be consistent the output of that model will also be under that same open license. I can see this being extremely limiting in training data, as only "compatible" licensed data would be possible to package together to train each model.
That's part of the point.
In my opinion LLMs are intellectual property theft. Just as if I started distributing copies of books. This substantially reduces the incentive for the creation of new IP. All written text, art work, etc needs to come imbued with a GPL style license: if you train your model on this, your weights and training code must be published.
Does anyone know of active work happening on such a license?