🧵 Untitled Thread
Anonymous at Sun, 7 Jan 2024 06:10:58 UTC No. 969626
With the prevalence of AI art fucking over 2D artists, anyone else relieved to be a 3D one?
Anonymous at Sun, 7 Jan 2024 07:35:01 UTC No. 969629
>>969626
Not at all, generative 3D algorithms already exist and is becoming more capable by the day. We'll face the same issues as 2D artist it's just a question of latency when our field is as affected.
Anonymous at Sun, 7 Jan 2024 07:48:39 UTC No. 969630
>>969626
>>969629
Thing to understand is neural nets of the type we build now are on the verge of being 'do anything' machines. People think about them as being these search algorithms that merely pair things from a giant corpus of data and that things need to present in their training data for it to be generated.
This is a fundamental misunderstanding of how these things works. A better analogy is to think of it as digital brain tissue that has been taught how to do certain tasks.
'Stable Diffusion' has been trained on labeled images from the public internet and is able to generate almost anything you can think of.
It might be trained of exabytes of image data for all we know.
But the finished trained neural net itself only takes 2.6gigabyte of drive space, it works without a internet connection on your local PC.
It doesn't contain data structures that point to elements of images, it has algorithms that actually understand what type of pixel structures constitutes
being the sort of thing you ask for and generate those structures from randomized noise doing something similar to what we do when we imagine shapes in clouds.
It's not some cute trick but something that is dangerously close to being able to run circles around the best of us at just about anything we do, not just art.
Anonymous at Sun, 7 Jan 2024 08:27:22 UTC No. 969635
>>969626
No ai will definitely come for us as well, I feel bad for 2d bros. But it will definitely work out in artists favour in the end. Normies like making stuff with ai, not consuming it
Anonymous at Sun, 7 Jan 2024 09:56:36 UTC No. 969642
Dumb fuck artists should realize that an AI can not be creative.
Drawing shit from an popular series is what AI does very well, because there are so much data it can be traoned on.
Blue Archive, Kankolle, GFL, Azur lane, has so much art available.
When an AI can do a character well, it's time to stop drawing that character.
Generating shit in 3D would be amazing, because then the full focus can be what you do with the 3D models, not making them. Only ones who will get shafted by AI 3D models are asset pack creators on the Unity store.
Anonymous at Sun, 7 Jan 2024 10:33:43 UTC No. 969644
>>969642
>Dumb fuck artists should realize that an AI can not be creative.
Strongly disagree. While the bulk of AI content you find is 'AI goop' that is a reflection of the prompt maker.
Provided the right prompt the AI is way more creative than any artist I can think of.
We can't make that many associations that fast so the way you can let the AI zero in on something interesting.
The creative magic happens when you task the AI to create things that never have been done before rather than trying to get the most refined image possible out of it.
You can explore concepts and use it to extend your own creativity in seconds that would otherwise take days to develop alone.
One day they'll be able to prompt themselves better than we ever could.
Anonymous at Sun, 7 Jan 2024 12:10:19 UTC No. 969652
>>969626
Dude in the OP image is just looking for an excuse to quit and blaming AI.
Anonymous at Sun, 7 Jan 2024 17:25:33 UTC No. 969685
>969630
all those words to say you're a /g/tard, probably a weeaboo and an unironic pedo (not just lolicon) too, and have no idea how the brain works even on a basic level. I bet you actually think our brains are just a sack of neurons that spontaneously gave itself sentience once we as infants managed to duplicate enough nerve cells to fill our skulls, and/or that neuroscience doesn't know anything more about it than that
if you knew anything about entropy and information (theory) you wouldn't be surprised by the efficacy of lossy AI compressors either. unlike text and binary, images are itself already quite sparse with information (amongst one another) and the fact that most artists stick to their comfort zones means there's even less of it to go around too.
when musk talks about AI dooming us all, he's cautioning us about the very minute but very dangerous possibility that current AI research would lead us down a path that gets us to the singularity. only an absolute fool would consider that scenario all but a given.
Anonymous at Sun, 7 Jan 2024 18:26:51 UTC No. 969693
>>969685
>unironic lolibanger ramble ramble duur
lol u whut m8, where that unwarranted aggression come from, did I hit a nerve there bro?
>I bet you actually think our brains are what gives us sentience.
Well, What organ do you propose to be responsible?
Point is we're witnessing these high level phenomena start to emerge at a level of computational simplicity that is already within strike range of what we can do today.
We don't need any singularity type sentience for this to be a real threat to us. These things could be mechanistic in nature like insects and still run circles around us in a multitude of tasks. Moving thru a slowly unfolding societal collapse as the consumer base is being destroyed by automation is what lies ahead of us now.
>muh Elon
Yeah Musk is a S-tier retard nobody should pay any mind, we're actually in total agreement there.
Let this be the isle of common ground from which our kinship might blossom my fellow synapsid.
Anonymous at Mon, 8 Jan 2024 01:11:52 UTC No. 969726
>>969626
>fucking over
Not a single person has been fucked over by AI
Someone making better art than you is not fucking you over
If you give up on your passion because something does it better then it was never your passion to begin with
Good riddance to these so called ‘artists’
Anonymous at Mon, 8 Jan 2024 04:00:12 UTC No. 969738
>>969644
Pic unrelated?
Sucking that robocock with nothing to prove for it but your creative bankruptcy.
Anonymous at Mon, 8 Jan 2024 05:33:02 UTC No. 969743
>>969738
You don't understand, I could create stuff like that myself but the time I would need to get to the point I start passing the quality of the machine would be at least several hours. The machine spit out 4 of these in seconds. This is just the AI's baby steps with the 1st generation of stable diffusion running locally on my PC.
Extrapolate what happens going forward as these things keep getting more refined. Even if you have the vision of something amazing in the time you create one piece of art
The AI could've been prompted to generate thousands and iterate further on that concept. Of those thousands several will eclipse any living artist including top pro's.
You're bound by editing the pixels locally by brush, the AI is manipulating the whole image at every point. Your creativity is gated by combining concepts and ideas from
the things you have seen, experienced and remember. The AI's creativity is currently gated by what's on ~the entire internet. And tomorrow with computer vision and reinforcement learning off video and sensors traversing the real world it will be gated in it's creativity by something that have traveled the entire planet and remembers what things look like.
You're gated to improve by 16 of the 24 hours per day you have available and the structure of your evolved brain. The AI will be able to restructure it's brain and optimize it's own hardware for specific task we want it to be good at. It's ability to learn is gated by how much processing power we throw at it.
We trying to out-art the AI will be much like us trying to out-calculate a supercomputer doing arithmetic with pen & paper.
Anonymous at Mon, 8 Jan 2024 09:09:47 UTC No. 969756
>>969743
Is this reply AI generated?
Anonymous at Mon, 8 Jan 2024 11:47:02 UTC No. 969772
>>969630
Don't kid yourself.
SD is capable of generating generic, uninspired garbage en masse and little more.
A neural net also doesn't contain any algorithms, it's just a large heap of weights, millions of if-else instructions, and it certainly doesn't "understand" anything.
And it absolutely is some cute trick, the only advance we got in 20 years is more powerful computers to run these absurdly large nets on. That's it. Every "breakthrough" AIdiots are crazy about has been part of a compsci degree course for 20 years, running on a better computer than 20 years ago.
Anonymous at Mon, 8 Jan 2024 12:18:23 UTC No. 969774
>>969772
> neural net also doesn't contain any algorithms, it's just a large heap of weights, millions of if-else instructions, and it certainly doesn't "understand" anything.
That's where you go wrong. Our own neural nets in our brains are "just" heap of weights too, but in those heap of weights structures form that can be viewed as the neural analogy of algorithms. There is a tremendous amount of 'understanding' going on inside these nets. Start thinking about what's actually involved in a LLM being able to predict the next word in a coherent sentence, comprehending spatial relationships and answering inquiries about disparate topics presenting novel connections that tracks with the human observer.
Whatever it's nature there has to exist some sort of world model within that network against which claims are tested, if that isn't 'understanding' you have a very narrow definition.
Something like chat GPT is like having access to a person that have read more books than any living human and is privvy to all knowledge inside that gargantuan libary.
You can ask it questions that you could not ask any living human because nobody knows that much about everything. You ask the AI to present you with some information drawing on dispartite topic X and Y. It'll present you with some insights one can gleam from such a comparisson and it'll be absolutly novel information that more often than not also checks out.
Anonymous at Mon, 8 Jan 2024 12:20:29 UTC No. 969775
>>969774
>Cont
There doesn't need to be qualia or sentience present for these neural algorithms to be 'intelligent' or 'understand' what type of information you are trying to access.
There is something that happens at the interface of your prompt and the LLM that is symbiotic in nature that we don't quite have a good word for.
It's like you are providing the AI with consciousness to explore and follow up on information it retrieve, challenge it to hone in and refine it's search.
It understands the context provided and in turn provide you with easy access to ever better structured information on exactly the thing you're interested in.
> the only advance we got in 20 years is more powerful computers to run these absurdly large nets on
Incorrect. The difference is one of transformer network architecture and the realization that this level of emergence happens in neural nets a fraction of the size we'd expect something
analogous to human intelligence to emerge based on how we're structured. The amount of compute needed to train them has been available for decades on existing supercomputers.
Sticking our heads in the sand and downplay what this is won't be helpful going forward. More people need to come to terms with what this is and what is able to do.
Anonymous at Mon, 8 Jan 2024 13:25:57 UTC No. 969781
>>969774
>Our own neural nets in our brains are "just" heap of weights too
It's your wishful thinking, no one has any idea how brain works exactly
>this level of emergence
This level of emergence is 0 exactly, current neurogenerators are merely lossy data compression algorithms and behave exactly like ones. "Emergence" is a relegious belief, not scientific theory.
Then given that in deep learning it takes a factor of 500 to cut the margin of error by 2, AGI even if its achievable by current methods will require enclosing the planet in Dyson's sphere or something.
Anonymous at Mon, 8 Jan 2024 13:48:25 UTC No. 969782
>>969693
>Well, What organ do you propose to be responsible?
Then I suggest you read up on basic neuroscience before running your mouth.
>Moving thru a slowly unfolding societal collapse
It was going to happen either way. I'm sure reddit chucklefucks like you would be eager to pin the blame on something as inane as modern AI
>automation
No, the sole contribution of AI to the collapse would be its ability to bypass the traditional checks and balances of our high trust (lol) society. ie. yet another drop in the ocean of technology and human ideals that subvert the checks and balances of society.
Anonymous at Mon, 8 Jan 2024 14:11:13 UTC No. 969785
>>969781
>AGI even if its achievable by current methods will require enclosing the planet in Dyson's sphere or something
Yeah, 'emergence' is completely unscientific despite us witnessing it everywhere all the time and having it well defined for a multitude of examples of emergent properties in systems.
Emergence isn't just about 'qualia and consciousness', it's the label for the mechanism by which something configured in a specific way gains properties not present in any of it's part in isolation. It's not a 'religion' it's a label for a very specific philosophical concept.
But here you fly off prophesying about how it'll "take all the energy of our star" to run something that is equivalent of our brain which runs on ~20 watt.
Ok, how very scientific of you to know what is possible to run with all the radiant energy of an entire star, can it run crysis too anon?
Anonymous at Mon, 8 Jan 2024 14:12:32 UTC No. 969786
>>969781
>no one has any idea how brain works exactly
While we lack the complete picture of how the system works as a whole we know a shit-ton about how axons and dendrites interconnect and fire together based on electric input and chemical
interface with neurotransmitters. Our ANN's are ofc inspired by how biological neuron works, hence the name 'Artificial Neural Network'.
We're very much charting unknown territory in building these structures and learning about the _emergence_ of capability within them.
Like the often cited description of the researchers surprise how the ability to talk other languages than english _emerged_ within the LLM's without any attempt to train them to do so.
Knowing when and how higher level phenomena like sentience emerge isn't something anyone knows, it could be close it could be far, some say it'll never happen others think it has already happened. Personally I don't think it has happened but I'm certain it will happen eventually but I don't want nor care to speculate as to when it might.
To me that question is quite irrelevant here and now compared to what already exist in our world and how we will have to deal with the refined version of these early AI agents
starting to seriously displace humans from qualified work, prob starting to happen in a big way as early as this calendar year we're on.
Anonymous at Mon, 8 Jan 2024 14:26:17 UTC No. 969788
>>969738
its even worse, ai is extremely jewish, image-gen AI dalle-3 inserts random POCs in your prompts, in language processing AI you can't get rid of globohomo shit (prompting is just cope in this case).
so everyone supporting AI shit is pro-globohomo basically, enabling indoctrination to the extreme.
Anonymous at Mon, 8 Jan 2024 14:45:44 UTC No. 969790
>>969786
>Our ANN's are ofc inspired by how biological neuron works
They have nothing in common whatsoever. Emergence is a pseudoscientific religious bullshit that has been around at least since 18 century. It means that if you throw enough of current year cool thing (gears, vacuum lamps, Conway's game of life cells, NN parameters) le god will emerge out if it. It didn't for all the previous attempts and won't for next ones. I predict that next attempt at manifesting emergence after this one completely stalls will be about clusters of neural networks and will take a dedicated power plant or even multiples to run.
Anonymous at Mon, 8 Jan 2024 15:08:14 UTC No. 969791
>>969788
how does it fare on the models from china?
Anonymous at Mon, 8 Jan 2024 15:16:06 UTC No. 969795
>>969790
>manifesting emergence
You are severely confused as to what emergence refers to anon. It's not just about the birth of some singularity AI god or spark of life in us.
It's about how new function arise as parts are assembled in specific configurations, emergence can be something as simplistic as how a set of rolling wheels emerge if you round off a shape and thread it onto an axle. How a projectile throwing device emerge if you attach a string to both ends of a flexible stick.
Emergent phenomena is everywhere in our world. 'Wetness' is often cited as example to get the concept of emergent phenomena across.
One molecule of water isn't wet. Neither are 2 or 10000. But somewhere as you keep adding molecules of water together the "wet" characteristics of the substance emerge and come into play. Despite not being possible to witness studying any individual part in isolation.
The sort of emergence you refer to is how most people with any insight to science recognize how that lump of tissue with 86 billion neurons and 100 trillion connected pathways between our ears causes the phenomena we experience as us to emerge.
If you wanna call it a religion to believe hardware that ridiculously complex is able to do exactly what it seems to be doing in our every day lived experience; fucking fine.
I call it follow the evidence where it leads you.
Anonymous at Mon, 8 Jan 2024 15:20:30 UTC No. 969799
>>969795
that's the philosophical definition of it. the contextual, practical definition of it is as he said.
you used emergence to argue that sentient AI is a definite (finite-time) possibility.
Anonymous at Mon, 8 Jan 2024 15:22:54 UTC No. 969800
I won
Anonymous at Mon, 8 Jan 2024 15:32:22 UTC No. 969803
>>969799
I don't even argue about sentient AI, I'm talking about the Ai we have today and what sort of properties have already emerged in them.
You don't understand this because you think emergence is a word that means existence of life as opposed to what it actually means: things emerging.
Do you think a single artificial neuron can do what the LLM's like chat GPT does? Ofc not. How about 10000? A million? A billion?
Do you see my point? What the LLM's are doing is a phenomena that only come about once you pass a certain threshold of complexity.
The process by which such phenomena comes about as you keep adding parts is a _emergent_ phenomena.
You're looking at the most prime example of a emergent phenomena front and center that we have that can only be understood in those terms but
since you're some oldschool 'qualia anon' and you're allergic to that term so you go blind and pretend like it isn't there despite knowing damn well it's real.
Anonymous at Mon, 8 Jan 2024 15:40:34 UTC No. 969804
>>969743
So what?
Visual arts are a vessel for ideas. He who can translate his ideas into art and communicate them to viewers will be successful.
AI art communicates nothing because it has nothing to say. Just like a hammer and a nail, it's not the tools we use but what a person builds with them.
Too many idiots focus on the visual component ("I'm competing with AI!") when they should focus on what emotions they're trying to elicit with their art.
Anonymous at Mon, 8 Jan 2024 15:45:20 UTC No. 969805
>>969803
>Knowing when and how higher level phenomena like sentience emerge ... Personally I don't think it has happened but I'm certain it will happen eventually
it's easy to see how an AI would think to fit words of the sanme concept but different expressions together, forming the expression of bilingualism, even of base64, in hindsight.
that's the thing, you can only tell that emergence happened when it happens. you can't just tease it out of something as you please, that'd literally be no different from believing magic is real.
Anonymous at Mon, 8 Jan 2024 15:51:15 UTC No. 969806
>>969804
>when they should focus on what emotions they're trying to elicit with their art.
Most of us are not into peeing onto the floor to symbolize the patriarchal repression of how we're not allowed to pee onto the floor in everyday life I guess. We enjoy making visuals but we're about to become too slow, too expensive and too uncreative to stay relevant in face of the machines that compete for our audiences time and spending.
Anonymous at Mon, 8 Jan 2024 15:52:29 UTC No. 969808
>>969775
Thank you for being one of the few posters I've seen on this site that actually understands what a neural net is
Anonymous at Mon, 8 Jan 2024 16:08:40 UTC No. 969809
>>969626
>anyone else relieved to be a 3D one?
lol, you think we'd be safe with all the news on procedurally generated 3D that is pushed by companies like Epic Games? Not even the idiotic tech bros that unleashed this apocalypse is safe, Web devs are now being replaced by AI, even there's a poster in this very thread is being smug and maybe uses chat GPT to say why AI is good. My biggest advice for anyone not in the industry is to have a backup plan of some sort while practicing in case shit goes down and be as frugal as possible, because this technology will heavily change the economy and labor force in the upcoming years.
Anonymous at Mon, 8 Jan 2024 16:53:57 UTC No. 969818
>>969626
Lel, that shit is low tier for novelai diffusion. These people have only seen shit like sd v1.5 and think that's the ceiling. Check this shit i made while fapping
>https://mega.nz/folder/9LsxwZCb#Gp
Anonymous at Mon, 8 Jan 2024 21:07:59 UTC No. 969837
>>969630
Here is the thing about Neural Nets, they as of now are limited by two things, data and the fact that they can only take input and output. Their generation is only as varied as the training data tags and training data itself. We just feed it bunch of data, ask it to learn how to replicate that data on a tiny space "only 2.6gigabites" and just hope emergent behaviors emerge.
AI still cannot do logic 100% of time and cant be thought like a human, it cannot remember things fully and cannot function in the world even when given body. Lot of the stuff when people say is just smokes and mirrors means in this case that it can sound smart, it can sound creative, but it can also say straight up well articulated bullshit or spit out generic shit.
The way it works and is trained is very anti-creative on its own. You give it 1000 pictures of trolls and then it as you said learns what "trollness" means and always gives you brand new picture of a troll, but said trollness is basically just the average of all those 1000 images. This is also kind of why it struggles at consistent characters, it does not learn definitions and then try to build from concepts or however other mysterious way human creativity works, it just creates generic stuff because when you give it prompts it will associate "essences" of those words from the training data into a single picture.
Or at least this is how diffusion AI works. LLMs just predict the next word but that also boils down to LLMs being generic because what would be the most probable story plot to heroic medieval fantasy? Yeah.
Random seeds can shift it a lot, but again its not that great, and LLMs are already reaching their current peak. It has been almost a year since GPT4 released and there is not a single other model that can top it off. Internet is now flooded with AI shit both text and image and future models will just eat their own shit in like 5 years. What AI has not taken already 90% sure will remain next 3-5 years
Anonymous at Mon, 8 Jan 2024 21:08:24 UTC No. 969838
>>969805
If you create visuals just for visuals you deserve to rot.
No message no story no meaning, you might as well be a robot. You're replaceable and I hope they're kind enough to give you a second serving at the soup kitchen.
Anonymous at Tue, 9 Jan 2024 05:29:19 UTC No. 969869
>>969837
Your understanding of where they stand today and how they're trained means you've looked into this and to a degree I think you make a fair assessment.
Where you trail off into what I deem to be cope is how you downplay what they already do focusing only on where they fail instead of recognizing how much valid
work even these 1st generation 'frozen in time' simple neural nets are already able to do. And how deep pockets are scrambling to research/build better ones.
If you look at all the AI BS that currently flood the internet your perception of what this technology is able to do is skewed to what it does in the hands of
lazy grifters scrambling to make money off any content they can put out there. But if you follow the field as to what is happening in the labs it's very different.
We don't even know the full extent of what 'chat GPT' are able to do as new behavior keeps being discovered from providing them the correct prompt.
Just giving the network the illusion of being able to reflect, refine and re-prompt itself to solve tasks like 'auto GPT' publicly demonstrate
just how capable these things can be.
> It has been almost a year since GPT4 released and there is not a single other model that can top it off.
You're looking at this very early and structurally incomplete version of where this technology is heading and declaring that this is as good as it gets.
We're gonna get blindsided by what's around the corner with people thinking like that. That there is delay to how fast these things become publicly available is positive.
Releasing these things into the wild for users to use & abuse as rapidly as possible is dangerous. So is removing the behind the scenes safety prompts of current models.
They contain a lot lethal knowledge on dangerous chemistry, explosives, poison, toxins, how to conduct various crimes efficiently, make drugs etc.
Anonymous at Tue, 9 Jan 2024 05:30:38 UTC No. 969870
>>969869
>>969837
Bet not even wacko88 like >>969788 would want the unshackled 'non-woke' variants of these things available if his name stood legally responsible for putting them out there.
Anonymous at Tue, 9 Jan 2024 08:56:28 UTC No. 969894
>>969869
Ok, here I still disagree with you. First of all you need to understand that prompt engineering is not as much of a science as it is art. Researchers publishing the next best prompt to test LLMs like "explain step by step" and so on are more indicative of how pathetic lot of these "papers" are. Only valid paper that I saw back while was one that basically solved prompt engineering by essentially putting random word generator to prompt the GPT with a question and then found out the best words, so the whole "we dont even know what chatGPT is capable of" is horseshit line. Yes, there are still areas of unknown and we know less about chatGPT ot GPT 3.5 then we know about the human brain, but most of it is just noise and muck.
But yes, I was catching up with AI before Dall-E 2 as of a fact and I was one of the people and still am who was completely shocked by chatGPT and even Character.AI when compared to GPT 3. But I also followed with all the failings of chatGPT being deployed, all the articles being released as complete dogshit. I know its just grifters and executives which are just grifters but older, but that still does not change the fact that web is now polluted with AI shit and as new data gets generated more and more of it will be AIslop because chatbots produce more data then humans do. No, there are no secret hidden supermodels, Q* was a meme, and if it was not then its still a math model, not some LLM but better. GPT4 is the top of realistic LLMs and about as good as LLMs can get. Google recently announced their Gemeni model that was supposed to be GPT4 contender and it still failed to GPT4 on basically every metric, or was prompted differently, which is basically the same as giving it different test to use. AI researchers have been using for LLMs transformers made in 2017.
Anonymous at Tue, 9 Jan 2024 08:57:50 UTC No. 969895
>>969894
(cont)
Now we are seeing basically the apex of that tech and we get marginal improvements, even GPT 4 was marginal improvement over GPT 3.5, while the simple retraining and adjusting of GPT 3 to 3.5 yielded much greater results. Similar with image AI. >>969818 fucking pic still has fucked up strings in her hand. Is it better then 8 fingers? Yes. Is the face beautiful? Yes. Is the tech slowing down and not nearly as impressive as the jumps between SD 1.0 and 1.5? Yes. We had photorealistic photos something like 3 months after SD 1.0 was publicly released, and then hands having 5 fingers 80% of time was another few months later. But in 2023 there was basically no progress in the second half of the year. You know why? Because scientific progress does not follow exponential function like soys propose, its a sigmoid function, and the vast majority of science is already slowing down.
Physics basically found as much in the last 30 years as it did in 1 year after WW2. Probably exaggerated but you get my point, similar is true for most of natural science. We spend 10 times more to discover new drugs just to keep up with the progress of discovery. Old scientists that made revolutionary discoveries did have less to work with, but their problems were also much simpler to solve.
Anonymous at Tue, 9 Jan 2024 08:59:21 UTC No. 969896
>>969895
(cont)
And finally to explain why AI learning from AI stuff is bad and wont be better at prompting itself any time soon. Its not about the AIness of the data, but the general quality of the data. Chess AI was able to become 1 000 000 times or something better then best grandmasters by playing against itself, basically generating its own data, but that is only viable because that data was already higher quality then humans, or could be made higher quality by just putting them against eachother. LLMs and Diffusion is still worse then human data.
You might say thats cope because >>969818 is better then 99.9% of online artwork. Yes, that is true but you should compare oranges to oranges, when you prompit it to make masterpiece, it will make pictures similar to other masterpieces. If train LLM on 2 billion tokens of articulated text, and 12 billion tokens of 4chan banter, it will still speak articulated as long as you tag them differently and then prompt it to speak articulated. You can increase the 4chan banter to 80 billion and it will still be articulated. But if that articulated dataset is tainted by also articulated AIslop that is not truthful and uses some phrases in more LLM way (yes, AI content detector AIs exist, yes they are also most likely regularly updated, no it cannot work on images because they improve much faster), then the higher quality text will become more LLM way, basically exaggerated artefacts and more likely to be hallucinating. No, its not just like humans because usually you can check the sources on non-junk sources from humans and hold them accountable. Holding AI accountable for mistakes is like holding Microsoft accountable for company PC crashing, its just pointless. So all in all it wont improve as much in the so short future according to my estimates.
Anonymous at Tue, 9 Jan 2024 09:20:55 UTC No. 969904
>>969894
>GPT4 is the top of realistic LLMs and about as good as LLMs can get
How on earth do you feel confident declaring this in the face of the rate development has been happening recently? Is the web flooded with AI slop? Absolutely. Are there people publishing "papers" on silly prompts? Absolutely. The fact all these silly things are happening and the tech is being exploited and hyped by unserious actors doesn't mean that what is at the core of it is something that is both amazing and terrifying and promise/threaten to change our world in various ways this decade we're currently living thru.
Point about prompt engineering is that a system like GPT4 thrown into the right type of loop already might be capable of clearing tasks it currently fails at.
As they keep scaling and training these things in the lab we don't know what comes next because we don't even fully understand what is already here.
Declaring it'll stay at this stage forever when we're just began peering under the hood of these things by increasing the number of eyeballs at this field by 1000%++
is incredibly premature, especially since meaningful advancements are happening almost weekly.
Great minds that now flock to this field and are playing catch-up with those already in it will also need more than 12 months to become experts.
The number of people qualified to help advance this field is increasing by the day, but educating those people takes time.
What we've seen so far is just like the primer going off, the fuse has now been lit and the actual explosion is on it's way.
Anonymous at Tue, 9 Jan 2024 13:27:43 UTC No. 969919
>>969635
how is it possible to be this much in denial. Normies fucking love consuming AI content and it's not even good yet. There are countless AI generated youtube videos racking up many thousands of views with near daily uploads.
Anonymous at Tue, 9 Jan 2024 15:01:58 UTC No. 969931
>>969919
Thousands of views are nothing on YouTube. For every AI slop channel with this many views you have another 20 with no views (5-500 views per video). Just a proof normies also do not like AI.
Anonymous at Tue, 9 Jan 2024 17:52:28 UTC No. 969956
>>969626
>I don't think I will find a reason to draw anymore
Maybe because it's fun and enjoyable?
If the only reason you want to art is money or praise from others, boo hoo.
Anonymous at Tue, 9 Jan 2024 23:06:52 UTC No. 969984
You will need to know how to draw if you want a precise drawing of the vision you have in your head.
Like, if you wanted to make that girl's boobs just a little bigger or whatever, could the AI do that in such a precise manner?
Anonymous at Wed, 10 Jan 2024 00:04:15 UTC No. 969989
>>969772
>Certainly doesn't understand anything
The AI developing an algorithm to do a certain task is considered "understanding" it, just not in the same way we do
Anonymous at Wed, 10 Jan 2024 00:09:19 UTC No. 969990
>>969726
Although you have a point, artfags like these pay the bills from commissions and such, and with Stable Diffusion, the demand for artists isn't that high anymore because you could just go to AI instead, which puts them out of their job and such. Some real art are being replaced by AI art too, an example of such is in ads, but yet again it feels like they just rip artwork off the internet and call it a day, so doesn't feel like much is lost there.
>If You give up on your passion because something does it better then it was never your passion anyways
This is moreof an artist being mad of getting out out of his job than actually quitting a passion
Anonymous at Wed, 10 Jan 2024 06:02:51 UTC No. 970004
>>969726
I've seen artists on twitter announce that they're going to stop drawing digitally or otherwise because of how good ai generating has become a few times already. And said artist's skills are nothing to scoff at
Anonymous at Wed, 10 Jan 2024 06:13:47 UTC No. 970007
There are 2 types of people
Those who think AI has peaked, aka the drones
And the speculators wondering about the possibilities, aka the humans
The drones are the types who would deny that ai would be where it's at today 1 year ago.
Anonymous at Wed, 10 Jan 2024 10:28:39 UTC No. 970019
>>970007
Think the 'rejection/denial' type behavior is people who take offense at how a development like this is even possible.
They wish for it to remain untrue/undone for a range of reasons.
Some have this romanticized view of the human mind/spirit and it's 'can and cannot do's' and find the idea of machines surpassing us at task that
strictly belong to the human realm of requiring psychology/intelligence/heart/soul/
They now hear this buzz about AI and watch how sane people around them start talking about things that where firmly
in the realm of science fiction til just a moment ago as if it's actually something real.
Their knee-jerk reaction is one of rejection and as they go out there to look at what the fuzz is all about they come across counter-points
from like-minded people, downplaying what's going on while pretending to know a lot more than they do about how these things actually work.
They latch onto these surface level explanations and lull themselves into safety "it's _just_ predicting the next word in a sentence" etc.
They don't sit down and interface and probe these things long enough til they have that 'oh shit' moment people who try to push these things to actually
succeed in solving tasks will encounter. Moment their interactions with AI does something stupid that is taken as validation of their views and they
can comfortably disengage pretending like it was a big ol' nothing burger and nothing has happened at all. That this is just some fad that will soon pass.
We're sleepwalking into this near future of societal upheaval and we're not even starting to position our-self for success because of how many people be just like that.
Anonymous at Wed, 10 Jan 2024 22:10:03 UTC No. 970104
>>970019
No you
Anonymous at Thu, 11 Jan 2024 02:37:07 UTC No. 970116
>>969904
>>970007
>Rate of development
Which peaked almost year ago with GPT4. Show me single model more capable then GPT4, or any significant upgrade that goes beyond hooking it up to python and executing small scripts instead of just writing command and relying on you copy and pasting it. Text to 3D was here back in late 2022, just now it got opened to normies few months ago and didn’t improve in the meantime. If I deny reality then show me ONE FUCKING EXAMPLE instead of pointing to vague bullshit answers like “the undeniable rate of progress” that only convinces midwits.
>Inb4
-Gemini does not count, it’s worse in basically every metric compared to GPT4
-Some tinyass 6 billion model named after random animal having 95% performance of big model does not count because they just reduce big model sizes instead of training from ground up
-Some random tech utilising GPT4 does not count, rich people finally figuring out how to adopt tech is not tech development, it’s deployment
Anonymous at Thu, 11 Jan 2024 02:42:52 UTC No. 970118
>>970116
(Additional inb4)
- Alphafold does not count because it does not remove any jobs, it helps in drug discovery with job that previously didn’t exist, and those molecules still need to be tested anyway, same with any other molecular synthesis AI. Plus all of these barely affect ordinary people anyway, just tiny boost to research that is slowing anyway as science progresses
Anonymous at Thu, 11 Jan 2024 10:43:33 UTC No. 970142
>>970116
So your argument that AI has halted is that OpenAI's competitors have yet to fully catch up to them in the tiny window of 1 years and that what OpenAI is brewing up in their labs isn't publicly disclosed?
All the refined use cases of existing tech based on GPT4 doesn't count either because of ~handwave reasons.
Alphafold doesn't count because it doesn't steal jobs that already existed ('progress = stealing jobs'?).
That we went from generating images from text to generation video from text in less than a year doesn't look like an alarming rate of progress to you?
The quality of images generated by text2image AI from it's inception to now isn't alarming progress to you either?
How about what NVIDA has demonstrated in the world of generative 3D, motion, audio-to-animation? None of that count either?
The rate of progress you seem to request from this field for it do count as annual progress the singularity would be here by next week ffs.
Calm your titties and just enjoy how we all get to be relevant for at least a little while longer anon, our obsolescence could be here all to soon so why make haste.
Anonymous at Thu, 11 Jan 2024 12:25:22 UTC No. 970152
>>969626
AI will swallow the world whole like an onlyfools whore
Anonymous at Thu, 11 Jan 2024 12:27:06 UTC No. 970154
>>970142
Video and images are too similar in the way how they work. Text2video still has the same underlying problems it had year and a half ago when it first started, only major difference is that text2image become super optimised allowing generation of enough high quality frames for high quality videos, and researchers working on ironing some of the nonsense artefacts. No, OpenAI is not hiding supergiga models away from the public without announcing them in advance. And the reason why alphafold does not count is because that is not human-like intelligence. It’s completely different thing, just finding abstract patterns in a way human could not. It’s like classifying YouTube algorithm as intelligent agent. Apples and oranges like calculator and GPT. Alphafold is not showcasing creativity, reasoning or anything like that, the same way TikTok does not. The rate of progress in GPTs is halting because there is no real innovation. The difference between GPT1 and GPT4 is just data, reinforcement learning and computing power, both of which are finite and we are running out of, and both of which are needed in exponential amount as time goes on. Diffusion models after OpenAI figured out Dall-E 1 have barely changed. Yes there are LORAs and so on, but the core remains the same. As for NVIDIA I wonder why they keep all of their models closed behind shut doors. AI animation will most likely just end up feeling as sterile as the average Ubisoft motion tracked slop where fictional characters move way too realistic while still having AI of Bethesda games. Go try Cascadeur and see it’s limitations, because these are the limitations that will stay there for a long time, or as long as they just teach neural networks on how real life things move instead of implementing locally run physics engine that will train your rigged model on how to walk and jump realistic to the bodyshape from the ground up.
Anonymous at Thu, 11 Jan 2024 12:58:05 UTC No. 970157
>>970154
>The difference between GPT1 and GPT4 is just data, reinforcement learning and computing power.
The difference between baby and adult is just data, reinforcement learning and computing power.
Stating that something is 'only better' because how it's scaled larger and trained better brushes under the carpet
how much those two things matter for all neural networks including the one between our own ears.
Your whole post reads as massive cope from where I'm standing. Looking at what happens and claiming the field is standing still because massive breakthroughs aren't happen
from one moment to the next is like watching a F1 race and declare the cars have gone as far along the track as they ever gonna go because you temporarily see them slowing down for a hard corner. The influx of qualified people who work on this is increasing daily and so is the amount of foundational body of work by people probing these things to gain clarity as to what's going on inside them and how we may push things further.
You hedging your beat against the future being one of additional technological breakthroughs kinda makes me wonder how long you been on this planet.
Anonymous at Thu, 11 Jan 2024 13:29:46 UTC No. 970163
>>970157
No my argument is that you will need another rare innovation to move forward. The whole field is not following steady progress, it’s a bumpy ride and saying that data and processing power are the major bottleneck now is saying that it will not move forward as fast as it did because improvements in processing power are slow. You are not 100x the processing power to make GPT5 in just a year, and the data problem is already fucked thanks to ChatGPT the website and API. GPT5 will now learn from halfassed articles made by GPT4 which follow the same kind of reasoning patters of GPT4. You will have articles and text full of GPT3.5 type hallucinations making up majority of the new data. It won’t become smarter because it’s a mimicking tool. If it mimics humans displaying logical reasoning it will probably learn logical reasoning from them. If it reads logically sound reasoning that is ultimately wrong but sounds right like GPT 3.5 does all the time, then you will end up with machine that spends more brain power trying to mimic smaller dumber model. Cat is out of the bag and internet is now fucked as a source of potential training data because it’s really hard to separate from human data and synthetic data.
Anonymous at Thu, 11 Jan 2024 13:56:35 UTC No. 970166
>>970154
>it’s a mimicking tool
That's a fundamental misunderstanding. It's not mimicking what's already there, it learns about what is present in the structure presented that is representative
of structures of such and such type. It's entirely possible that it's even feasible to train these networks on synthetic training data and have them improve over time instead of degrade by having trained AI's grade how good synthetic images are and feed ones that grade above average as back into the training set.
Much like how 'Alpha' Go learnt to play go far beyond human capability by playing virtual games against itself.
Synthetic data does not equal worse data, it can be better than any human graded data for training purposes.
And you also have to factor in how computer vision will soon provide access to the entire real world as potential training data.
Even if you wanna take the most doom outlook of the internet turning into a cesspool of AI slop and this training data will degrade
you still A) have a backlog of a pre-AI content that you can train better network architectures on, and B) once you solve computer vision have access to all of youtube and
human interactions with real people in the real world to continue to train these networks on. The idea we might gonna run out of training data is a flimsy argument at best.
It doesn't hold up to any sort of scrutiny as to how we may move forward even in the advent of such a hypothetical scenario as far as the public internet goes.
Anonymous at Thu, 11 Jan 2024 15:18:35 UTC No. 970173
>>970166
Stop. Nobody cares. Come back for the next AI hype cycle. This one is already over.
Anonymous at Thu, 11 Jan 2024 15:33:50 UTC No. 970178
>>970173
You don't like AI, that is fine anon. I understand how some people find this development disturbing.
I'm not trying to 'hype' it up with what I'm saying here, I'm trying to correct misconceptions of how this technology actually works
and have more people understand just what is happening and why. Because like it or not it is here now and we all have to deal with it.
You guys not wanting to understand what this is and how it does what it does is not helpful as you end up downplaying and ostrich your heads into the sand.
Just because critics of AI isn't looking or caring doesn't mean it goes away.
Whether you like it or hate it you have to understand what it is we're contending with here in order to contribute for how we deal with this development.
Anonymous at Thu, 11 Jan 2024 16:01:07 UTC No. 970181
>>970178
You're spamming a board with probably 10 to 15 weekly active users. Just so you know.
Anonymous at Thu, 11 Jan 2024 16:04:41 UTC No. 970182
>>970181
I know, but these are my people who I care about, I have a soft spot for misfit creators and lewd lunatics.
Anonymous at Thu, 11 Jan 2024 17:15:15 UTC No. 970189
>>970181
>board avg. 100 posts per day
Anon-san you're too pessimistic, it's at least.. uh.. DOUBLE that!
Anonymous at Thu, 11 Jan 2024 23:05:56 UTC No. 970207
>5 years go by
Someone posts a screenshot of >>970116
>"lol what a dumb faggot"
Stop being a stubborn nigger
Anonymous at Fri, 12 Jan 2024 14:12:44 UTC No. 970255
>>969774
>Something like chat GPT is like having access to a person that have read more books than any living human and is privvy to all knowledge inside that gargantuan libary.
lmao
shit can't even play three legal chess moves in a row
Anonymous at Fri, 12 Jan 2024 14:35:13 UTC No. 970263
>>970255
Nor can it ride a bicycle, Pointing to the absence of capability it isn't trained to have misses how it has a lot of capabilities it wasn't explicitly trained to have.
You're doing exactly this >>970019 you're finding something to latch onto to lull your self into safety downplaying what we're up against here
because how you see it acting stupid in some regard.
You have a early generation LMM that trained is this 45gigabyte neural network that already clear university exams and
here you are laughing so hard that your ass comes off because it can't also play board-games to your liking.
Anonymous at Sat, 13 Jan 2024 00:34:09 UTC No. 970296
>>969626
I really can't wait for AI 3D
Anonymous at Sun, 14 Jan 2024 21:41:36 UTC No. 970504
>>969626
the thing is that from a technical perspective 3d stuff isnt at all standardized and while i do think ai generated models will be a thing in the future i doubt its ever gonna be a problem for stuff outside of sculpts
Anonymous at Mon, 15 Jan 2024 02:36:02 UTC No. 970542
I don't really care much about it.
The reason why AI is so appealing is great tool to create art for people who lack the skill and money to create their fantasies.
Only good enthusiastic and evolving artists will remain.
Because you still need artists create and correct AI fuck ups.
And most of low hanging fruit and Butthurt artists that refuse to adept will shrink massively with a few acceptions of course.
Anonymous at Mon, 15 Jan 2024 20:09:20 UTC No. 970650
>>969626
98% of the time the results from AI stuff are pretty shitty for very complex stuff (look at what Wacom did).
Also, when someone is caught using AI then people proceeds to bully and criticize them to delete the stuff (look at what Wacom did).
Overall, i think people who will end using AI stuff for business related stuff are suits that don't have an idea what they're doing and only care about profits (look at what Wacom did).
Anonymous at Wed, 17 Jan 2024 22:07:02 UTC No. 970917
>>970542
>The reason why AI is so appealing is great tool to create art for people who lack the skill and money to create their fantasies.
>Only good enthusiastic and evolving artists will remain.
The cognitive dissonance of ai trannies lmao. If people with no skill and money can do it then it doesn’t matter how enthusiastic or willing your are to embrace new tech.
Anonymous at Tue, 23 Jan 2024 03:28:20 UTC No. 971523
>>970628
to talk about technology and art obviously
Anonymous at Tue, 23 Jan 2024 03:35:30 UTC No. 971528
>>969818
Well, Japs are usually pretty far behind when it comes to technology, and having a huge language barrier between them and the AI is also a major factor
Anonymous at Tue, 23 Jan 2024 03:36:44 UTC No. 971529
>>969642
It's actually a lot harder to create known characters than original characters. And saying you aren't making something when you're literally the only reason it exists is mindless.
Anonymous at Wed, 24 Jan 2024 04:03:29 UTC No. 971701
>>971529
>And saying you aren't making something when you're literally the only reason it exists is mindless.
I took a giant elephant shit in my toilet this morning. I made something and I’m literally the only reason it exists.
Anonymous at Wed, 24 Jan 2024 05:32:42 UTC No. 971703
>>970542
But if artists stop creating art, and ai only is feeding itself with her own art doesn't that means that ai image generation would be really homogenised?. I don't think that art nowadays is great, because it's just faggots following thrends, But I have seen the same face over and over again in ai images that it looks like it was made by the same prompter. So ai would probably kill art, but it will make Imagen generation really accessible.
Anonymous at Wed, 24 Jan 2024 12:33:27 UTC No. 971729
>>971703
Tons of people draw for fun. Not everything is about money.
Anonymous at Wed, 24 Jan 2024 15:38:48 UTC No. 971746
>>971729
What?
Tons of people don't clean their hands after going to the toilet too, my point still stands
Anonymous at Wed, 24 Jan 2024 19:08:54 UTC No. 971760
>>971703
The best of the AI's output is leagues above average human output, as long as you score a set of AI generated images in terms of how good they are
it's possible to use synthetic data to improve these models over time by feeding it's own best output back into the training set.
These models will also soon be trained on moving video and thru sensors looking and traversing the real world. They will understand spatial relationships and
gestalt theory and parallax and all those things that are difficult to learn from static images.
They'll be able to take the look of anything and extrapolate that into euclidean space so you can freely move the camera around.
It'll eventually infer the 3D space in 2D images and allow you to look around or move into images.
Basically we're not just fucked but we are magnafucked (that is when you are fucked by magnitudes).
The machines will soon create art that makes what we came up with during our run look primitive in comparison.
Anonymous at Thu, 25 Jan 2024 01:04:00 UTC No. 971783
Anyone not consumed with robot cock, just remember fine are will rise again. Trad chads will inherit the world. Buy some clay.
Anonymous at Thu, 25 Jan 2024 07:54:19 UTC No. 971800
>>971783
You think and AI can't sculpt clay? You're in for a surprise bud.
Robots will perform any physical task better than us because they're not limited to hardware and scales suitable to be manipulated by our fingered hands.
Any physical task they need to do they can design custom appendages specialized to their needs. Just as their minds software is mallable and they'll be able to restructure the
architecture of their brain any way necessary to grow smarter so will their physical bodies be able to come with arbitrarrily shaped geometry optimized for their need.
Robotic sculptors will arrive at the scene starting out near the very top of human capability, but few generations down the line it'll be like us
trying to out-sculpt a 100 armed octopus that squeezes out a finished sculpt of anything by changing it's grip twice.
Anonymous at Thu, 25 Jan 2024 21:03:57 UTC No. 971860
I'm not too worried because I think people of taste will always prefer human made over AI, and I don't think that prompters can ever really compete with real artists.
I think taste and creativity are honed over years working at creating art, and prompters just don't have that.
This will sound pretentious, but 2D art is very hard. The amount of time it takes to go from zero skill to actually good is years. Beginner artists suck fucking dick, they are so bad. Really dedicated diligent students and get to a decent quality in a year, but they'd have to be practicing full time. There's almost no profession that takes so much time and skill to go from shit to "okay". And then to become truly great can take your whole life. Along the way, artists will be looking at everything. Buckles, pipes, zippers, fabrics, clouds, teeth, chairs, screws. Looking at everything with interest. Great artists need to know how everything works, how a steam engine works, how back muscles work, how trees move in the wind.
As good as AI can be, prompters can't ever approach this. They've never heard of energy preservation, they don't know about atmospheric effects. Prompters can get lucky when the machine does all this for them, but they won't notice when it gets it wrong, and they know when they should purposely get it wrong.
And this is born out by the fact that all they make is different anime girls. Anime robot girl, anime elf girl, trending on art station, ruan jia. They don't know, and they'll never know what they don't know.
So there will always always always be a value in real artists, but capitalism might not necessarily agree with that, but there will always be a way to survive as an artist I think.
What really makes me sad is the idea of kids who have ideas in their head resorting to AI because they can't face the difficulty curve in becoming an artist. And they would be robbed of an amazing life.
Anonymous at Sat, 27 Jan 2024 05:00:57 UTC No. 972003
>>969626
>artists getting pissy over sour grapes
GOOD.
Anonymous at Sun, 28 Jan 2024 23:11:57 UTC No. 972171
>>969626
That's SHUTTERSTOCK and turbosquid, and if their tool is even a minuscule of a percentage better than the free AIs that make 3D models today... than their ai is better than 99.999% of all of the 4chan users right now
Anonymous at Sun, 28 Jan 2024 23:26:27 UTC No. 972175
>>972171
Man fuck these pieces of shit. Shows what they think about artists, no wonder with how they commercialise art
Anonymous at Sat, 3 Feb 2024 04:21:45 UTC No. 972814
>>972175
>>>>/g/entlemen go out of their way to create tools that let artist create things that would take entire teams and millions of dollars in minutes
>noooo this is literally genocide
why are artfags like this?
Anonymous at Mon, 19 Feb 2024 23:09:58 UTC No. 974732
>>972171
>and if their tool is even a minuscule of a percentage better than the free AIs that make 3D models today.
it won't be
the only lab making consistent breakthroughs is openai. used to also include google, but not so much these days.
until openai shows off point-e 2, efforts from other places will mostly be jokes. unfortunately, when point-e 2/3 arrive they will be leaps comparabale to sora.
Anonymous at Tue, 20 Feb 2024 01:43:49 UTC No. 974752
>>969984
Yes they can with new lasso tools that target and regenerate specific areas.
Anonymous at Tue, 20 Feb 2024 09:07:47 UTC No. 974806
>>969626
I shit and piss on AI. AI may be "better" than me, but I'd rather not have to rely on a proprietary machine for my artistic needs. AI tools can be swept from down under you and leave you on your ass at any time, whereas artistic skill, no matter how humble, cannot be taken away from you.
Also AI will inevitably come for 3D too, already is. Just keep grinding, don't fall for the AI trap, AI is a crutch and if anyone ever decides to kick it from down under you you are fucked. Be independent
Anonymous at Tue, 20 Feb 2024 20:54:33 UTC No. 974902
What's weird is that, despite the massive rapid development of ai, I rarely see it adopted by people in my life. Dall-e 3 went public and SD content specific local models have been around for a long time, become easily available to the masses...and people didn't really adopt them. You could hook up a chat bot trained on a specific person to a voice clone trained on their voice and then port it through a rig of their face, and create a full facsimile digital person, right now, and people just aren't. The world seems unfazed and apathetic to this technology outside of some diehard supporters and haters. People are just opting not to use it despite the fact it's free and it's productive. You have an image generator that can make multitudes of high quality images accurate to your prompts, and most people simply aren't using it. OpenAI is losing hundreds of millions of dollars from a complete inability to monetize or profit off of their creations (which is nevertheless irrelevant because Microsoft can subsidize it indefinitely) It's interesting.
Another thing is that the primary things generative ai does currently is allow content to be created far faster, and reduce the cost of creating that content drastically. But we're already mired in content. 99.99% of all content has zero views or playtime. People have a finite capacity regarding their ability to consume media. I think that what you're going to see as ai media ramps up is that people are going to get desensitized to media in general and consume less of it. We already see most people totally bored with ai generative tools.
Anonymous at Tue, 20 Feb 2024 20:57:03 UTC No. 974903
>>969626
I'm not scared until AI can make a better likeness from reference than I can, and im not great
it's probably just a matter of time but for now it's safe
Anonymous at Tue, 20 Feb 2024 21:08:22 UTC No. 974906
>>974902
It will be interesting to see if we change how we consume content in response to this. The assumption is that people will just come home, go "computer, make me a 2.5 hour legend of zelda movie in the style of Studio Ghibli meets Killer is Dead starring Michael Cera" and enjoy immediate, personalized content tailored to their specific prompts and tastes. Just immediate gratification for you and only you at any time for any reason featuring anything.
But I don't think that's going to happen. There's a lot of content which is only relevant to people because it's real. Sports games, concerts, weddings or funerals or holiday videos, photos of friends etc. You aren't going to go "computer: create a video of my wedding night" or "computer: make a picture of me and my grandma at the Antiques Roadshow". We share media with our friends, they share it with us, we watch it together simultaneously, we discuss it, we go back and rewatch things, we share our thoughts on them. I don't see a culture where people stop doing all of that and only enjoy instant gratification. You said it yourself with the Dall-E stuff. People can generate any image they want RIGHT NOW, but they don't. It's not their inclination.
Anonymous at Tue, 20 Feb 2024 21:10:28 UTC No. 974908
>>969919
Normies hate ai lel. Also most ai channels are propped up by ai viewers
Anonymous at Tue, 20 Feb 2024 21:37:41 UTC No. 974918
>>974902
3d isn't actually time consuming or difficult in the technical sense, it's difficult for the vast amounts of knowledge that needs to be acquired and called on for projects. Once that threshold is met, 1 motived person can make a full scene to their design within a few hours-days. 3d Artists aren't using AI outside of rendering/compositing because they already have everything they need.
Anonymous at Tue, 20 Feb 2024 21:39:41 UTC No. 974919
>>974918
Yes anon but people who aren't 3d artists at all will be able to simply prompt 3d deliverables
Anonymous at Tue, 20 Feb 2024 21:44:24 UTC No. 974922
>>974919
How would they know if an AI object is usable in a scene if they know nothing about 3d?
Anonymous at Tue, 20 Feb 2024 21:45:46 UTC No. 974923
>>974922
The generative ai could just output the entire scene itself anon. You won't need to generate 3d assets you can just generate the entire either animation or game or program.
Anonymous at Tue, 20 Feb 2024 21:50:14 UTC No. 974925
>>969626
The limit of A.I depends on what it was build on.
Very specific characters of some obscure video game will not be generated unless you feed it.
So a character designer has to make the character first before feeding it to a.i to mass produce. Like any factory made goods.
If anything, you're going to see a flood of artist moving to character designers/concept art in the future.
Anonymous at Tue, 20 Feb 2024 21:53:56 UTC No. 974926
>>974925
This only applies to narrow models like stable diffusion anon. GPT-4 and later 5 are multi modal world models which can successfully be trained on their own outputs and understand multiple different disciplines and output types simultaneously. You'll be able to design a character purely with text, then use that output as a portion of your next prompt i.e. "make this character again but from behind and with blue hair'.
There is nothing it can't do given enough developmental time.
Anonymous at Tue, 20 Feb 2024 22:05:42 UTC No. 974929
>>974926
>There is nothing it can't do given enough developmental time.
tons of stuff, like accurate voice cloning, it cant do
Anonymous at Tue, 20 Feb 2024 22:06:23 UTC No. 974930
>>974929
And just days ago we said it can't do video. We have no reason to think it won't be able to indefinitely.
Anonymous at Tue, 20 Feb 2024 22:22:56 UTC No. 974935
>>974930
you're living in a sci fi fantasy. This is exactly what I warned you tech bros about a couple of months ago. That your entire goal was to reach some singularity where you become a ball of energy and just beam off into the far reach of the galaxy. You need strong ass meds
Anonymous at Tue, 20 Feb 2024 22:33:04 UTC No. 974939
>>974923
How would they know if the output is good if they know nothing about any of those things?
Anonymous at Tue, 20 Feb 2024 22:35:54 UTC No. 974940
>>974939
They'll just rip all dvds that ever existed with makemkv and then train off it, unlawfully of course because china is doing it too
Anonymous at Wed, 21 Feb 2024 03:39:02 UTC No. 974984
sexooo
Anonymous at Wed, 21 Feb 2024 05:29:39 UTC No. 974994
>>974992
Cris, can you guide me through your reasoning as to why it's ok make a thread where you talk to yourself about AI on a board dedicated to 3D. Especially given that AI can't into 3D yet (if ever).
Help me understand how your wonderful brain works.
Anonymous at Wed, 21 Feb 2024 07:37:47 UTC No. 974999
>>974994
Is everyone cris to you?
Anonymous at Wed, 21 Feb 2024 08:12:44 UTC No. 975002
>>974999
Cris is an idea more than a single person. Cris is an impressionable guy who's being exploited for entertainment by jannies who probably need distraction to keep their mind off the damage they did to their own bodies.
Anonymous at Wed, 21 Feb 2024 19:13:58 UTC No. 975028
>>974806
Artistic skill can be easily taken away by illness or any damage to you
Also you don’t need to use any proprietary AIs, just use the local ones that no one will ever be able to take away
Anonymous at Thu, 22 Feb 2024 12:09:57 UTC No. 975147
>>972814
kek you answered yourself. You are saying that will axe entire teams of artists.
Anonymous at Thu, 22 Feb 2024 22:03:16 UTC No. 975201
>>974806
Absolute fucking retard with a take as stupid as he is.
>hurr I must keep grinding using outdated programs and techniques, that'll surely make me competitive in a field where speed is king
Do you think employers will look at your garbage models which take 100 times longer to make than if you'd used AI and be impressed?
>hurr what if they take it down and we don't got AI no more?!!!
Yeah? They're gonna take away freely available code scattered across the web and the shit I've got installed locally, too? Maybe they'll go after Photoshop and Blender next! Get real, nigger.
Adapt or die, you've got about 5000 years of written history to show you what happens to luddites like you.
Anonymous at Fri, 23 Feb 2024 17:38:59 UTC No. 975299
>>969772
>SD is capable of generating generic, uninspired garbage en masse and little more.
this is such cope
Anonymous at Fri, 23 Feb 2024 19:38:55 UTC No. 975317
>>975299
>neural net also doesn't contain any algorithms, it's just a large heap of weights, millions of if-else instructions
holy fuck I never thought I'd see such level of confident retardation
That's what happens when you had the """creatives""" trying to understand anything math related
Anonymous at Sun, 25 Feb 2024 09:31:03 UTC No. 975597
>>969626
Ai art isnt fucking over anyone retard. If some brazillian in a shack, prompting salimichan is enough to drain you of any confidence, you werent gonna make it anyway.
it seems so apocalyptic cause everyone who didnt like drawing but did the grind cause theyre social outcasts and think its better than an actual job, all got their willpower shit on with the realization that no... this isn't something you do for others:
create for yourself instead of being a fucking sheep
Anonymous at Wed, 28 Feb 2024 16:05:37 UTC No. 975915
>>969626
>With the prevalence of AI art fucking over 2D artists, anyone else relieved to be a 3D one?
Time is ticking, friendo. Only a matter of time. You are replacable.
Anonymous at Wed, 28 Feb 2024 16:17:50 UTC No. 975916
The key is for artists to work with the ai, have it generate your starter image, and then touch it up yourself. We know that it's not perfect as of yet, and we know that whatever it creates will not be a perfect representation of what the artist had in mind. But it can be like that first sentence in "finding forester" where from the beginning provided by another source, the artist can work with that starter to create something of their own.
Anonymous at Wed, 28 Feb 2024 16:20:42 UTC No. 975917
>>975916
It's just another tool for creators to work with, and it's still improving, soon it will be a better master of 2d drawings, and from there we will see ai 3d art before we begin seeing animations done with ai.