You're viewing part of a thread.
The present models are, is the point i think the other person is making.
I haven't looked into it too much, because I don't bother to use these things myself. But if I remember, there are some systems that are open-source, can be run locally, and then a person could train those systems on only public domain and freely licensed works. That is the kind I'm talking about, so bringing up the systems I'm not talking about is just a strawman.
I haven’t looked into it too much
The closest thing to what you're discussing is what's built over at db0
However what you're hypothesizing cannot exist, due to the nature of a diffusion model. The source that the images are generated from, all of that was stolen in 2021. It was illegally scraped.
AIHorde is moving in the right direction, but we need a system where the people who's source work is being scraped need to either consent to the use of their artistic labor, through copy left practices. Or they deserve to be compensated through royalties.
In latestage capitalism the only power is that which lies within capitol. That has been stolen by the large AI companies. What you're proposing through your hypotheticals is closing the barn door after the horses have got out.
That is the kind I’m talking about, so bringing up the systems I’m not talking about is just a strawman.
I'm actively refuting what you're proposing because not only does it not exist, but it's currently impossible, due to the nature of the tech it's based on. This is not a strawman argument, this is a refutation of a naive take that does nothing to push forward the tech in a direction which you yourself would like it to go.
Sorry to be such a wet blanket. If you'd like to know more about Diffusion models and what they stole here's a great place to start. https://systemicalternatives.org/2023/06/06/the-controversy-of-ai-art-theft/
If you want to work towards what your hypotheticals state you could consider contributing to aihorde.net.
My response again is from the viewpoint of someone who is constantly refuting "Why don't we just...." naval gazing and how trolls and other bad faith actors hide behind that type of language.
When you say, "However what you’re hypothesizing cannot exist, due to the nature of a diffusion model. The source that the images are generated from, all of that was stolen in 2021. It was illegally scraped.", are you referring to Laion-5B?
And others yes. It really depends on where you look and who you ask. I think Laion was the beginning of the end but some people say it was Google in 2010 or so
https://www.adweek.com/lostremote/google-set-to-open-image-recognition-technology-via-apis/
Some go even earlier. But I think wholesale theft started around 2020-2021 at scale. Arxiv below for reference.
https://arxiv.org/abs/2210.08402
Also I really do apologize for my tone earlier. I've given all of the beneficial doubt I can really.
Yeah, I'm sorry but none of this seems to add up. I read the book you linked, the entire thing, and I didn't really learn anything new from it. In my own searches I've found a number of competing image generation systems, most of them just use Stable Diffusion - including Horde, which makes sense as it is open-source and readily available. But others have their own solutions, and some even use their own datasets. I've also done other searches and found a variety of instructions on how people can put together their own datasets. Between all of the options, it seems pretty clear to me that even now it is entirely possible for people to use these AI tools without ever needing to come into any contact with "stolen" (ie., copyright infringed - it's not theft) artwork.
But I'm not going to link to any of these things because I don't want to endorse or promote anything without taking sufficient time to be really sure that they are in fact ethical, which honestly I'm probably not going to do at least for the time being, since I have no need to use any of these anyway. But yeah, ethical ai art is entirely possible, no copyright ever needs to be infringed, and this is something anyone can do today if they're inclined to make the searches and assemble free source images themselves.
I understand what you're saying the point to me that sticks out was the usage of "hypothetical", which to me would mean things that don't exist, which wouldn't be a strawman as i understand it. With that being said I'm genuinely not trying to be an ass, just explaining the way I perceived the comments, I hope it's not needed to say, but I meant no offense in my reply of your comment.
No offense taken. It is still a strawman though, because regardless of the realness of what I'm talking about, they're still referring to something different and then attacking that as if it is the thing I'm talking about.
There is no strawman I'm attacking. I'm addressing you directly. I'm not reforming or re-addressing your argument, I'm refuting it.
Your premise is flawed and as you state yourself it comes from a place of naivety as you don't understand the tech. Which is fine, we don't have to understand everything, but don't cry foul when your premise is corrected in a public forum. I replied up the chain with what you could do and projects that are close to what you propose. I also gave you resources that you can go through to learn more if you have the interest.
But it doesn't change the fact that maybe the greatest theft of artistic labor in our lifetimes has already happened and there's very little independent artists can do about it.
You're correct. Trolls love to hide in grey area naval gazing questions. My kneejerk reaction is to directly refute and shut it down. I definitely err on the side of asshole now though.