anyone who actually knows what they're talking about knows that models learning from their own HUMAN SELECTED, SHARED, AND APPROVED generated content, is completely fine
RETARDS!!!!!!!!!!!!!!!!!!!!!!!
>HUMAN SELECTED, SHARED, AND APPROVED
the cope really is genuine from you disgusting garden gnomes, isn't it? lost all your money on memecoins and now your latest grift is also falling apart. my sides. your only solution is suicide.
anyway, your schizo nonsense is 100% false. everyone knows that all these memeshit algorithms are scraping public websites (automatically) with no filters at all. this is why they're facing copyright and quality problems.
Why do people who know nothing about the subject they're talking about pretend to know what they're talking about?
You literally train the model on whatever art you choose. Yes, there are models that scrape the internet but those are generally trash.
>AI BAD AI BAD MUH IP MUH COPYRIGHT >SAVE MEEEEEE GOVERNMENT HELP HELP DISNEY LOBBY HELP NINTENDO HEEEEELP WE NEED MORE LAWS AND LITIGATION AND PUNITIVITY AND TEAMS OF CORPORATE gnomish SUPER LAWYERS ENFORCING LITERAL THOUGHT CRIME PUNISHMENT >PLEASE PLEASE PLEASE GIVE MICROSOFT A MILITARY ENFORCED EXEMPLARY COMPANY-STATE MONOPOLY ON AI SO MUH HECKIN API DOESNT GET UNETHICALLY SCRAPED
the march of progress will continue and you will be trampled underfoot
>AI 'art' created through referencing AI 'art' is yielding worse and worse results.
Citation needed. I've gone one that suggests the opposite. >StableRep: Synthetic Images from Text-to-Image Models Make Strong Visual Representation Learners >https://arxiv.org/abs/2306.00984 >With solely synthetic images, the representations learned by StableRep surpass the performance of representations learned by SimCLR and CLIP using the same set of text prompts and corresponding real images, on large scale datasets. When we further add language supervision, StableRep trained with 20M synthetic images achieves better accuracy than CLIP trained with 50M real images.
That paper does not mean what you think it means and really, it's trivially obvious. Image generation generates statistically likely images, real images also include statistically unlikely ones. Also, real image-text datasets have very noisy labels, while synthetic images do not. The most you can make of this article is that the distribution of image data generated by image generators now matches the distribution of real images on average, which is a cool way to quantify how good image generation is right now, but that's all.
Not at all, it demonstrates that AI generated image are fine to learn from; though that was already obvious since 99% of public ones are already human curated, and thus they provide some signal for training to learn from.
>HUMAN SELECTED, SHARED, AND APPROVED
the cope really is genuine from you disgusting garden gnomes, isn't it? lost all your money on memecoins and now your latest grift is also falling apart. my sides. your only solution is suicide.
anyway, your schizo nonsense is 100% false. everyone knows that all these memeshit algorithms are scraping public websites (automatically) with no filters at all. this is why they're facing copyright and quality problems.
The AI implosion. In the future we will will have colonies isolated from the rest of us, with no access to AI, to generate more data to be fed into it.
Maybe we already are in one
> The AI implosion. In the future we will will have colonies isolated from the rest of us, with no access to AI, to generate more data to be fed into it. >Maybe we already are in one
That’s basically what BOT is right?
If the data set only includes higher rated ai images then the data set is good, because then the images will be human filtered. That's exactly how AI already works. There is no issue.
SD is only 1B params? kek
we should be autotraining models for any artgays seething on social media who have any clout so people can easily rip their garbage shit off
flush their worthless decades of drawing fetish trash for subhumans down the drain and hopefully make them eclipse trannies in suicide rate.
personally I can't wait for generation Alpha to grow up and start posting about how early AI art was so soulful compared to the too perfect modern AI art we will have then. Maybe they'll even share prompts among themselves to try to capture the characteristic weird-fingers look of early AI art.
So AI art can't replace genuine emotional reactions because... it just cant?
It wouldn't even need to be authentic, if it can recreate an experience to draw influence from it, that would be more than enough in conveying something
I predicted that 6 months ago. I don't know which programs "pull" though, stable diffusion uses existing laion database. The problem is it won't be possible to make newer databases with all generated images in internet.
this is only true for textgen because it's computational prohibitive and the threshold for acceptable SNR in output is very high. for artmorons it's trivial to completely emulate some random twitmorons style with a few hours on a decent gpu
>the programs are now starting to pull from it
What the fuck is this moron talking about? There are people that select AI arts to train a new AI art? I mean, that's possible but why would you. It's sheer retardation and a fantasy from artcels to sprout anything about this technology.
>the AI is now capable of crawling, training and finetune images without the need of human intervention >might as well produce its output independently without parameters
I can tell you didn't follow the tech progression that much.
Can someone explain why all the discourse surrounding this topic is so vitriolic? Nobody is out there shit-talking lawyers getting replaced by chatbots. My best guess is that since this is the one that creates porn it attracts the terminally online and maladjusted.
For me it is because when it came out I excitedly posted the things I made with it, had a bunch of artists dogpile and send constant hate at me for days afterwards.
I just wanted to share something cool but now I hate all artists.
It's because art in the modern day is based around publishing on twitter and patreon. Hence, these people already have a prominent foothold on popular social media platforms through which they can voice their discontent.
I'm sure other people are upset about things too, but they don't have "twitter clout" so you just don't hear about their complaints.
people who revere people who put images of Jesus in a bottle of piss and live off fetish porn commissiosn which they're so fucking ashamed of they try to change their artstyle so they can't be correlated to their """professsional""" identity have attempted to take the moral highground
best of all the artmorons now have wasted their entire adult life learning a useless skillset and will be starving to death or flipping burgers
none because there aren't, the best "ai detection" most people and scripts do is look for stable diffusion metadata in the pic lel
https://thehive.ai/demos?case_study=ai_generated_image_detection
try it yourself
only good tool that I know of, ignore anything else you find online searching "ai image detection", the rest are garbage
These people think that every time you prompt something in stable diffusion, it searches the internet for pictures and mashes them together. They are literally retarded.
That watermark gets destroyed by pretty much anythink, including upscale. Also it is commonly disabled so SD could shit out images quicker, and because they impace quality.
It will be impossible to use an automated scraper to make an image database for deep learning again. Unless you are willing to invest 100x more into it and run each of the billions of images though another network to detect if it was generated.
AI will use it's own AI images as a data which causes a feedback loop generating more and more retarded images since it won't be able to come up with something new, so its only input will be it's output until it reaches some sort of shizoid singularity.
Humans are not more creative, but humans update their database every second with eyes, feel, smell, sound and taste.
Human stuff serves as those eyes to the AI.
If it keeps referencing itself, it will be more of imagine if human went blind and deaf. He tries to imagine same things he was able to see once before. Every year his imagination will become more and more shizoid, convoluted and detached from reality.
Same with ai
there are six primary reasons to have an image
1. money laundering (modern art)
2. to sound smarter than you are (art in a museum)
3. for assets for another medium (video game, etc)
4. for interacting with other people on imageboards or forums or such
5. for jacking off / because you are addicted to anime and/or supernormal stimulii
6. patent pending
#5 is where all artmorons online got their money while larping #1 wasn't actually just money laundering
anyone who actually knows what they're talking about knows that models learning from their own HUMAN SELECTED, SHARED, AND APPROVED generated content, is completely fine
RETARDS!!!!!!!!!!!!!!!!!!!!!!!
Prepare to apply for a real job.
please write it on your blog instead of here. If anyone gives a shit they can read it there.
>HUMAN SELECTED, SHARED, AND APPROVED
the cope really is genuine from you disgusting garden gnomes, isn't it? lost all your money on memecoins and now your latest grift is also falling apart. my sides. your only solution is suicide.
anyway, your schizo nonsense is 100% false. everyone knows that all these memeshit algorithms are scraping public websites (automatically) with no filters at all. this is why they're facing copyright and quality problems.
>artist cuck copium
all those years down the drain huh
>AI bad
>crypto bad
Why not go for "progressive" bingo and work VR BAD into your rant too
Why do people who know nothing about the subject they're talking about pretend to know what they're talking about?
You literally train the model on whatever art you choose. Yes, there are models that scrape the internet but those are generally trash.
>AI BAD AI BAD MUH IP MUH COPYRIGHT
>SAVE MEEEEEE GOVERNMENT HELP HELP DISNEY LOBBY HELP NINTENDO HEEEEELP WE NEED MORE LAWS AND LITIGATION AND PUNITIVITY AND TEAMS OF CORPORATE gnomish SUPER LAWYERS ENFORCING LITERAL THOUGHT CRIME PUNISHMENT
>PLEASE PLEASE PLEASE GIVE MICROSOFT A MILITARY ENFORCED EXEMPLARY COMPANY-STATE MONOPOLY ON AI SO MUH HECKIN API DOESNT GET UNETHICALLY SCRAPED
the march of progress will continue and you will be trampled underfoot
Great post, love the seethe in the replies.
lmfao what a pathetic post
Ah good so good to see BOT is just as eager to ape out on ai topics as /ic/.
Very good quality bait, nice post
mad cow disease / scrapie
>feeding animals to animals
what are we going to call the results of using bots to feed bots?
>AI 'art' created through referencing AI 'art' is yielding worse and worse results.
Citation needed. I've gone one that suggests the opposite.
>StableRep: Synthetic Images from Text-to-Image Models Make Strong Visual Representation Learners
>https://arxiv.org/abs/2306.00984
>With solely synthetic images, the representations learned by StableRep surpass the performance of representations learned by SimCLR and CLIP using the same set of text prompts and corresponding real images, on large scale datasets. When we further add language supervision, StableRep trained with 20M synthetic images achieves better accuracy than CLIP trained with 50M real images.
That paper does not mean what you think it means and really, it's trivially obvious. Image generation generates statistically likely images, real images also include statistically unlikely ones. Also, real image-text datasets have very noisy labels, while synthetic images do not. The most you can make of this article is that the distribution of image data generated by image generators now matches the distribution of real images on average, which is a cool way to quantify how good image generation is right now, but that's all.
Not at all, it demonstrates that AI generated image are fine to learn from; though that was already obvious since 99% of public ones are already human curated, and thus they provide some signal for training to learn from.
> and the citation is nonsense
not coping well, are you? will you be livestreaming your suicide too?
it's okay to admit when you don't understand it
>I'm not coping! I'm not coping! You are!
The AI implosion. In the future we will will have colonies isolated from the rest of us, with no access to AI, to generate more data to be fed into it.
Maybe we already are in one
Why don't you gays write a science fiction book. Maybe the AI can help you with it.
>Maybe the AI can help you with it.
It probably would, if AI couldn't handle processing more then few paragraphs at once.
> The AI implosion. In the future we will will have colonies isolated from the rest of us, with no access to AI, to generate more data to be fed into it.
>Maybe we already are in one
That’s basically what BOT is right?
It's not "the" AI implosion, but an AI implosion
Duh. What do you think Earth is? In other to farm joy you need a world full of suffering.
loser BOT coomers think ai is the greatest thing ever, they're too stupid to know it's a dumb parlor trick. to apes technically looks like magic
If the data set only includes higher rated ai images then the data set is good, because then the images will be human filtered. That's exactly how AI already works. There is no issue.
Made up story by someone who has no idea how models are even trained.
this, seething art trannies are mad and trying to cope since they may have to find a real job
the copium continues
>1B image model already created this much seethe.
Can't wait for a good one.
SD is only 1B params? kek
we should be autotraining models for any artgays seething on social media who have any clout so people can easily rip their garbage shit off
flush their worthless decades of drawing fetish trash for subhumans down the drain and hopefully make them eclipse trannies in suicide rate.
Not a transformer either. would be much better at copying an artists art style on lower sampling data otherwise.
>most of the data sets are up to just 2021
"Artists" are the most retarded people on this planet.
I hope they all starve to death.
personally I can't wait for generation Alpha to grow up and start posting about how early AI art was so soulful compared to the too perfect modern AI art we will have then. Maybe they'll even share prompts among themselves to try to capture the characteristic weird-fingers look of early AI art.
>personally I can't wait
You don't actually have to wait though:
I don't disagree with the video. Quirks/Artifacts have their value.
that's pretty interesting actually
I'm not sure I agree completely, but he has a lot of good points
So AI art can't replace genuine emotional reactions because... it just cant?
It wouldn't even need to be authentic, if it can recreate an experience to draw influence from it, that would be more than enough in conveying something
that settles it
the literally who cartoon avatar twat has spoken
I predicted that 6 months ago. I don't know which programs "pull" though, stable diffusion uses existing laion database. The problem is it won't be possible to make newer databases with all generated images in internet.
This has been discussed ever since day one. For the AIs, history ended in 2021 and we can only drip-feed them anything more recent
this is only true for textgen because it's computational prohibitive and the threshold for acceptable SNR in output is very high. for artmorons it's trivial to completely emulate some random twitmorons style with a few hours on a decent gpu
commissions for porn drying up I see
>the programs are now starting to pull from it
What the fuck is this moron talking about? There are people that select AI arts to train a new AI art? I mean, that's possible but why would you. It's sheer retardation and a fantasy from artcels to sprout anything about this technology.
>There are people that select AI arts to train a new AI art?
Yes, the people called automated web crawlers
>the AI is now capable of crawling, training and finetune images without the need of human intervention
>might as well produce its output independently without parameters
I can tell you didn't follow the tech progression that much.
>believes twitter crayonbrain logic
Can someone explain why all the discourse surrounding this topic is so vitriolic? Nobody is out there shit-talking lawyers getting replaced by chatbots. My best guess is that since this is the one that creates porn it attracts the terminally online and maladjusted.
For me it is because when it came out I excitedly posted the things I made with it, had a bunch of artists dogpile and send constant hate at me for days afterwards.
I just wanted to share something cool but now I hate all artists.
They fucking started it.
Just imagine that face is the last thing you ever see
>he says with a traced picture
lmfao. the jokes make themselves at this point with these mentally handicapped gays
The barrier to entry for law in market is higher than art. Several things contribute to this.
Laws constantly update.
Context window for modern day LLM is not large enough to be practical for law.
Market disruption in art is here and now.
I'm friends with several lawyers. Their main barrier to entry is that they're fucking lawyers. They're even more powerful of a union than doctors.
Drawing is literally the only thing theses people have, even if every lawyer is replaced today most lawyers would have another job by tomorrow.
If it makes you feel any better their parents will still be cleaning up bedrooms that look like this soon.
It's because art in the modern day is based around publishing on twitter and patreon. Hence, these people already have a prominent foothold on popular social media platforms through which they can voice their discontent.
I'm sure other people are upset about things too, but they don't have "twitter clout" so you just don't hear about their complaints.
people who revere people who put images of Jesus in a bottle of piss and live off fetish porn commissiosn which they're so fucking ashamed of they try to change their artstyle so they can't be correlated to their """professsional""" identity have attempted to take the moral highground
best of all the artmorons now have wasted their entire adult life learning a useless skillset and will be starving to death or flipping burgers
>source: my ass
Maybe if twitter chud "artists" actually did their commissions instead of complaining that drawings of morons look too white they wouldn't be replaced
yeah, it's called inbreeding.
>So, apparently
Truth. I believe.
Kessler syndrome, many such cases
I would simply shoot down the debris or use machines to push it out of orbit
there are scripts to detect AI-generated art that work like 99% of the time
AI training on itself is a non-issue even without human curation
> work like 99% of the time
Any links?
none because there aren't, the best "ai detection" most people and scripts do is look for stable diffusion metadata in the pic lel
https://thehive.ai/demos?case_study=ai_generated_image_detection
try it yourself
only good tool that I know of, ignore anything else you find online searching "ai image detection", the rest are garbage
The amount of high-quality fappable AI furry porn is just too much. I can't possibly coom to it all. I've tried.
I think that SD/midjourney adds a steganographic watermark to it's output
These people think that every time you prompt something in stable diffusion, it searches the internet for pictures and mashes them together. They are literally retarded.
That watermark gets destroyed by pretty much anythink, including upscale. Also it is commonly disabled so SD could shit out images quicker, and because they impace quality.
lol this is simply not true
i can't believe humanity has declined so much to the point that people will believe what [random retard on twitter] says
It will be impossible to use an automated scraper to make an image database for deep learning again. Unless you are willing to invest 100x more into it and run each of the billions of images though another network to detect if it was generated.
Why go on Twitter and lie like that?
Inbred AI.
AI will use it's own AI images as a data which causes a feedback loop generating more and more retarded images since it won't be able to come up with something new, so its only input will be it's output until it reaches some sort of shizoid singularity.
Humans are not more creative, but humans update their database every second with eyes, feel, smell, sound and taste.
Human stuff serves as those eyes to the AI.
If it keeps referencing itself, it will be more of imagine if human went blind and deaf. He tries to imagine same things he was able to see once before. Every year his imagination will become more and more shizoid, convoluted and detached from reality.
Same with ai
eat shit moron, LAION-6B + boorudumps and finetunes is more than we will ever need for anything.
>Artmorons still seething
based, wish you could get aids with your unemployment you fucking moron of a """profession""""
There is a ton of low quality AI art out there but that doesn't mean there arent talented people that will be able to do great things with it
there are six primary reasons to have an image
1. money laundering (modern art)
2. to sound smarter than you are (art in a museum)
3. for assets for another medium (video game, etc)
4. for interacting with other people on imageboards or forums or such
5. for jacking off / because you are addicted to anime and/or supernormal stimulii
6. patent pending
#5 is where all artmorons online got their money while larping #1 wasn't actually just money laundering
synth data has been shown to be even more effective for training so this take is meaningless
makes sense, less subhuman artgays in synth data
>ITT: seething art morons telling other people to seethe despite seething for over a year now about AI
the approach of open assistant is the future. let autists or experts hand curate answers or questions and feed those into the database
Why do trans people love AI so much? 90% of the people I see talking about and defending it are trans
>no talent scaly gay overdosing on copium
Love to see it