The Book of KelsYeah, this is such a perfect representation of that happening.
Raven
terrestrialTerror@ravenskrag @SigmaOne @Nezchan There's several issues with that image. AI models like that if retrained keeps using a curated datasets. There's a whole science to this but even shit scrapped off of the internet requires curation and sanitation. Likewise with a quick glance at artstation there doesn't seem to be descriptions for the images there. So for this to work and new images added to the database there needs to be a human adding image descriptions to the image, making sure the image is in the correct format and all of this stuff. Basically there's a bunch of points where a human can see an attack like that and say "okay no" and get rid of it.
Even if they were to get an image description from god and are just pulling shit into their database indiscriminately it takes days for this shit to retrain. So if that post was posted when the protest started there's literally no way for those people to post what they did with the new model.
It's plain disinformation from people without a clue how these AI systems work.
Even if they were to get an image description from god and are just pulling shit into their database indiscriminately it takes days for this shit to retrain. So if that post was posted when the protest started there's literally no way for those people to post what they did with the new model.
It's plain disinformation from people without a clue how these AI systems work.
@tetrahydreyes @SigmaOne @Nezchan When I wrote my original reply, I was indeed skeptical that MJ could incorporate this sort of information into it's model without an expensive retraining.
I agree that the training images would have to be labeled, but I don't know enough about their data curation process to believe that it is easy to remove bad data from an attack. There seems to be many forms of "bad" in the datasets for MJ and StableDiffusion they did not remove.
@tetrahydreyes @SigmaOne @Nezchan After reading the original twitter threads, I agree this is misleading.
The OP of the twitter thread posted a faked AI image that was apparently created with the intent to mock the anti-AI stance. The "someone should tell them" is a reference that the first image was a "joke". Rainisto's work is also fake / a joke.
There are legitimate reasons to believe that AI art generators are unethical, but many artists do not understand the tech. It's complex stuff.
@ravenskrag @SigmaOne @Nezchan Honestly, I wasn't really arguing if it was unethical or not, I just hate people perpetuating misinformation.
It's like this one image that I saw a fukken year ago where this one person posted a fake image of windows blocking someone from playing a pirated movie because of some instruction being added to the x86 instruction set.
A bunch of people here ate it up. I fucking hate Microsoft but I hate people making up shit to be mad at Microsoft about. There's plenty of reasons to hate them why make something up??
Here I do generally enjoy AI art though the creators of these models definitely are stepping on quite a few toes. I personally just like seeing the computer doing a cool thing.
It's like this one image that I saw a fukken year ago where this one person posted a fake image of windows blocking someone from playing a pirated movie because of some instruction being added to the x86 instruction set.
A bunch of people here ate it up. I fucking hate Microsoft but I hate people making up shit to be mad at Microsoft about. There's plenty of reasons to hate them why make something up??
Here I do generally enjoy AI art though the creators of these models definitely are stepping on quite a few toes. I personally just like seeing the computer doing a cool thing.
@tetrahydreyes @SigmaOne @Nezchan Yeah, I wasn't trying to suggest that's your stance on the ethical issue. I was just trying to say that personally, I am very much against AI / ML being used in this way. I want to push what computation can do, but not at the cost of devaluing or displacing artists. As you say, they are indeed "stepping on quite a few toes."
"There's plenty of reasons to hate [Microsoft] why make something up??" I think we're totally in agreement there.
Daniel H. Alcojor (he/him)@tetrahydreyes @[email protected] @Nezchan @ravenskrag Also, the AIs use hundreds of thousands, or millions, of images. Having now the same image a couple dozen times I don't think would matter THAT MUCH.