AI Art Seething General


That looks like mistaking a plateau (which in some segments of AI is kinda happening) and excessive hype with dying out.

I get that the AI bros have oversold AI solutions and shit often doesn't work as advertised, but what works, works, and there's no going back. While AI degeneracy may be a problem, it's not like the existing models that weren't fed AI-generated inputs will magically disappear.

It's like claiming self-driving cars (promised by Elon to be a thing by 2014) are dying off. No, they aren't, they simply never existed in the first place, and all the research is still being utilized, albeit as not as an impressive functionality, by car makers all over the world.
 
The regurgitation thing, reminds me of the somewhat recent University of Cambridge research, suggesting that the more AI generated images a model is fed during its training, the worst is will be.

sketch-1722540370814.png
"The increasingly distorted images produced by an AI image model that is trained on data generated by a previous version of the model. | M. Boháček & H. Farid/arXiv (CC BY 4.0)"

One of the articles summarizing it, can be found here
 
The regurgitation thing, reminds me of the somewhat recent University of Cambridge research, suggesting that the more AI generated images a model is fed during its training, the worst is will be.

View attachment 6260099
"The increasingly distorted images produced by an AI image model that is trained on data generated by a previous version of the model. | M. Boháček & H. Farid/arXiv (CC BY 4.0)"

One of the articles summarizing it, can be found here
crispy jpegs
 
The regurgitation thing, reminds me of the somewhat recent University of Cambridge research, suggesting that the more AI generated images a model is fed during its training, the worst is will be.

View attachment 6260099
"The increasingly distorted images produced by an AI image model that is trained on data generated by a previous version of the model. | M. Boháček & H. Farid/arXiv (CC BY 4.0)"

One of the articles summarizing it, can be found here
Synthetic data has to be curated, sure, but it's not unusable. There's nothing special about AI generated data that makes it bad to train on, you just cant feed it into itself in an infinite loop and expect good results.
 
The regurgitation thing, reminds me of the somewhat recent University of Cambridge research, suggesting that the more AI generated images a model is fed during its training, the worst is will be.

View attachment 6260099
"The increasingly distorted images produced by an AI image model that is trained on data generated by a previous version of the model. | M. Boháček & H. Farid/arXiv (CC BY 4.0)"

One of the articles summarizing it, can be found here
That reminds me. Some people really thought training ai is like breeding dogs.
6fbhmi58fxed1.jpeg


Ive got some anti ai content. Two of them, in fact

 
The regurgitation thing, reminds me of the somewhat recent University of Cambridge research, suggesting that the more AI generated images a model is fed during its training, the worst is will be.

View attachment 6260099
"The increasingly distorted images produced by an AI image model that is trained on data generated by a previous version of the model. | M. Boháček & H. Farid/arXiv (CC BY 4.0)"

One of the articles summarizing it, can be found here
It's like tumblr/deviantart/twitter draw-slaves training on other draw-slaves until the average image looks like this:
GOPB-zbXYAAPwal.jpeg
 
Wake up, there's a new seething video

And here's some bonus comic seething

This guy's voice is setting my troondar off.
I.. I don't understand those entltled ''artist''. It not like AI art is stealing their ''art'' who still exist uncorrupted. Also Unless the databse or prompt is specifically curated to a specific artist style and even with that in mind the result are kinda mix, AI art is pretty much unique it own way. Those dumbass have just no originality and try to find a scapegoat when the AI begin to look like the most common used style
 
I.. I don't understand those entltled ''artist''. It not like AI art is stealing their ''art'' who still exist uncorrupted. Also Unless the databse or prompt is specifically curated to a specific artist style and even with that in mind the result are kinda mix, AI art is pretty much unique it own way. Those dumbass have just no originality and try to find a scapegoat when the AI begin to look like the most common used style
They've also made two more videos:
Preservetube archive of the channel
 
Large language models have already reach the plateau congratulations you created a random text generator when you train your algorithm to say knowingly on true things such as crime statistics which are a point of fact not a point of conjecture then it is not an AI.
The only thing I writing tools are really good at is correcting my terribly worded sentences into something somewhat resembling what I was trying to say so it's basically a more advanced form of Grammarly.

Also by these people's own logic if they don't lock themselves in a cave and train using nothing besides their own natural instincts they are also stealing other people's art.

Also the United states is rapidly collapsing that's why we doubt that any country is going to be able to function in five years due to changing demographics the destruction of all social safety Nets as well as the break down of civil society maybe Japan.
 
Large language models have already reach the plateau congratulations you created a random text generator when you train your algorithm to say knowingly on true things such as crime statistics which are a point of fact not a point of conjecture then it is not an AI.
The only thing I writing tools are really good at is correcting my terribly worded sentences into something somewhat resembling what I was trying to say so it's basically a more advanced form of Grammarly.
I disagree. I use chatGPT daily at my job, with the web search functionality it's basically just a better form of google/stackoverflow. Sure it gets things wrong and it's cucked by microsoft but it doesn't make it not useful.
And it's only a matter of time until FOSS models catch up.
 
I disagree. I use chatGPT daily at my job, with the web search functionality it's basically just a better form of google/stackoverflow. Sure it gets things wrong and it's cucked by microsoft but it doesn't make it not useful.
And it's only a matter of time until FOSS models catch up.
call me when it can replicate its own code
 
  • Optimistic
Reactions: Ether Being
That reminds me. Some people really thought training ai is like breeding dogs.
View attachment 6266024

Ive got some anti ai content. Two of them, in fact
View attachment 6266032
View attachment 6266037
First one I kind of get, but with the second one, I have to say that there is no distinction between the Cal Arts style that has propagated throughout mainstream animation and AI LORAs. They both produce the same shit over and over, but AI can do it faster. The "Capitalism will make animation stagnant" argument has been around since the time of Richard Williams, but quality stuff like Miyazaki or Disney prints money and is relatively creative enough.
 
First one I kind of get, but with the second one, I have to say that there is no distinction between the Cal Arts style that has propagated throughout mainstream animation and AI LORAs. They both produce the same shit over and over, but AI can do it faster. The "Capitalism will make animation stagnant" argument has been around since the time of Richard Williams, but quality stuff like Miyazaki or Disney prints money and is relatively creative enough.

don't most communist countries hate art for the potential of being rebellious? Capitalism may suck but it suck less when your life's put on the line for making fun of their country's government.
 
These artists all realize that there is nothing special about their work, no "divine spark", no special sauce that makes it distinct. That's why they freak out. If AI art was always clearly AI art and inferior, there would be nothing to worry about and nobody would talk about it, and that is the core of the matter - everything else being equal, on observation alone (and please read this sentence carefully) there is simply no difference between a given piece of human made art and a lucky draw by the AI. Don't take my word for it, just listen to the artists. They seem to think so, otherwise why would they freak out? As this technology will get better, you'll see more of this in the future in other fields. I think it hits artists especially hard since many of them go through their life being told that they have a gift, that they are special. Being shown by a computer program of all things that in the grand scheme of things, their "gift" can be statistically approximated like everything else probably is a particularily bitter pill to swallow for them.

There was a new model released to the general public on friday (FLUX by black forest labs) that comes pretty damn close to Dall-E 3 but has open weights, meaning everyone can run it. Not everyone will because it has rather high requirements, but it for example doesn't really always fuck up hands anymore. (you can scroll that thread for more examples, it was also stated that it will get all the fancy tools like control net, so more professional work will be possible with it) I want to remind people here that stable diffusion, especially the early versions, were first and foremost made to be efficient models that can run on average hardware, not the best models possible.

Saying AI tech has "peaked" is such a weird statement to make at this point in time, especially considering that new developments outpace old ones on approximately a three month scale right now. What is currently going on in this space feels like Moore's law on steroids. I went through that time of computer evolution in the 80s and 90s and it really feels exactly the same. I can only imagine such statements come from ignorance of the developments, an incredibly short attention span or wishful thinking. Computers also got a lot of hate before everyone owned one so even that aspect of this entire thing is not really original.

Also even if LLMs wouldn't change from today on (and they will), their accessibility and existence has already changed the world forever. Never before could a computer read/hear/see/process/create/understand/explain the world around it and human language in a natural way and this has been one of the biggest goals of of computer science since the first conception of what we understand as modern computing. Just because you cannot think of an application, don't assume nobody can. Granted, they might not be as efficent and useful as we need them to be for some tasks, but so weren't the early computers. They still eventually changed the world forever.
 
don't most communist countries hate art for the potential of being rebellious? Capitalism may suck but it suck less when your life's put on the line for making fun of their country's government.
No, that's why every commie country from the USSR to modern China and North Korea have state sponsored art of all sorts and why Western commies dream of getting gubmint money for shitting out their Calarts tier scribbles.
 
Back