that the non-profit LAION is going as far as intimidating the creators of the images they used.
They don't publish the image, yes, and that is also their reasoning. However, in my opinion, the intention behind all of this seems obvious. They circumvent copyright claims by being a non-profit and not publishing, but with the clear intent that the image will be used to train some system further down the road. LAION appears to be a key player in how these text-to-image models dodge the copyright bullet.
Intimidating? They mention that if you sue them on copyright ground when they do nothing copyright related, they are able to claim damages. Which seems pretty fair, as the claims by the photographers are clearly in bad faith.
Links are still mostly legal in Germany, if the link is to something which is not, it’s a different situation and different from "hey, I own the copyright of my images, don’t link to them!"
We all know the purpose for which the images linked in the dataset will be used. "We are a non-profit organization and provide only a link" is akin to taking people for fools. Why can't these systems be trained exclusively on images for which people have given consent or for which money has been paid?
> They don't publish the image, yes, and that is also their reasoning.
No, their reasoning is that they only keep a link to the image so there’s nothing to remove.
A more interesting angle would be copyrightable ALT tags in the form of poems or other creative content. But, the cat is already out of the bag as it’s probably easy enough to strip poetry or other copyrightable out of alt tags with the technology we’ve got at this point.
> No, their reasoning is that they only keep a link to the image so there’s nothing to remove.
I'm well aware of it, and that's also what I wanted to say. However, as I mentioned, the dataset with the links (to copyrighted material) is provided for machine learning purposes. Saying "We only provide links" is a lazy excuse from a guy with a smirking face.
that the non-profit LAION is going as far as intimidating the creators of the images they used.
They don't publish the image, yes, and that is also their reasoning. However, in my opinion, the intention behind all of this seems obvious. They circumvent copyright claims by being a non-profit and not publishing, but with the clear intent that the image will be used to train some system further down the road. LAION appears to be a key player in how these text-to-image models dodge the copyright bullet.