>slaves
Ain't nobody want that. What these developments do is they remove the need for people. One tiny demographic at a time, you become unneeded and unwanted, just marked for death. A slow one, of your own making, while you are made to feel guilt and shame.
Yes, that because this isn't "we". "we" didn't create and don't own any of these technologies. They will be rented to us in limited form if it makes the owners profit.
The Internet is indeed used as a tool for invigilation, yes. The fact that it's everywhere and that I need to use it unless I want to cut all ties with family and friends is not something I have control over.
Why? Because you will no longer be paid to contribute nothing to society?
>Oh no, my art job is redundant, I'm a slave! I have to actually contribute real value to society to get paid and enjoy its benefits!
What tech does is it takes the worthless eaters like artists and it reminds them how useless they always were. Basically you are a cotton picking dumbfuck being replaced during the industrial revolution.
A buckbroken wagecuck or a sperglord denying the fact that his meat calculator job will also be easily replacable. Either way, acquire class consciousness. Until then, meds.
Why? Because you will no longer be paid to contribute nothing to society?
>Oh no, my art job is redundant, I'm a slave! I have to actually contribute real value to society to get paid and enjoy its benefits!
What tech does is it takes the worthless eaters like artists and it reminds them how useless they always were. Basically you are a cotton picking dumbfuck being replaced during the industrial revolution.
>NOOOO YOU CAN'T MAKE THAT ART! THAT'S INTELLECTUAL PROPERTY THEFT! >USE THE ROBOT! IT'S NOT SUBJECT TO IP LAWS
every human would be capable of achieving beauty if it wasn't illegal to. information is the ultimate evil in this world unless you're an """AI""".
This raises an interesting question: does ai generated content have to be treated with the same rules? Like e-boi porn isn't illegal, but cp is, so how would you rule ai generated cp?
Technically ai didn't generate all these images, they just rearrange pixel patterns from the dataset they've learnt, so any "generated" cp is a mixture of various existing cp.
By that logic, yes it should be illegal, since ai could never generate cp if it never seen one.
>Technically ai didn't generate all these images, they just rearrange pixel patterns from the dataset they've learnt, so any "generated" cp is a mixture of various existing cp.
you have no idea what you're talking about, dunce. read about the topic and then come back hopefully knowing more.
Maybe five years ago, but these new AI are sophisticated to the point where they can clearly merge concepts in unique ways. You think OP's picture is drawn only from the AI's understanding of hippos in hot tubs? Of course not. It's drawn from its understanding of hippos and hot tubs separately.
>By that logic, yes it should be illegal, since ai could never generate cp if it never seen one.
Co-Pilot IS NOT ILLEGAL
STOP TRYING TO PRETEND THE SOURCE MATERIAL MATTERS!
This AI shit unironically mind blowing. Where do we go from there?
It cant create porn so it's useless
This raises an interesting question: does ai generated content have to be treated with the same rules? Like e-boi porn isn't illegal, but cp is, so how would you rule ai generated cp?
e-boi is already barely legal in the US. The law on that is pretty vague and it's illegal in most of europe. Realistic looking e-boi porn would definitely be illegal. People in USA have actually got convicted for having e-boi. The law in USA is: if the general population thinks that the work clearly lack any literary, artistic, political, or scientific value. Yes, it's that subjective. Porn is not considered freedom of expression in USA.
Find me a person in the US who was convicted for it who didn't also have real stuff in possession. The only time they give a shit is if it's something they can tack onto a more serious charge.
Nowhere
And when it all comes crashing down, and you come crawling back to artists for work, they'll all be sure to tack on an nice extra 'thinking this AI could do shit better than them' fee
what I don't understand with most of the machine learning stuff is that the AI is learning from resources that got created by humans like books, pictures etc. so in the end it can only create things that are in some way mixed or derived from what already exists and has been created by humans; not really new stuff. so in context of creating profound new, really creative stuff, it simply won't work
No, we've already seen that AI can extrapolate outside training data. For example, GPT-3 can be asked math questions, that was never present in the training data, and it will answer them correctly, because it has learned generalized arithmetic from the examples that does exist in the training data.
You could say that this is still not new since it's still just within the concept of arithmetic which was present in the training data, but if you speak that broadly then it's the same for most humans. We also almost never generate something new, we are just inspired from existing art or existing culture and extrapolate on that. Otherwise it wouldn't even be good art, because if it has no connection to existing concepts, it has no meaning to us.
>For example, GPT-3 can be asked math questions, that was never present in the training data, and it will answer them correctly, because it has learned generalized arithmetic from the examples that does exist in the training data.
hard to believe, since there are many examples of GPT-3 not understanding even some simple math (wrong results) problems when you change the numbers around
Shut right the fuck up. You're spouting bullshit. You don't know what you're talking about. GPT stands for "generative pre-trained transformer" - there is no way for them to escape their prior training by their very nature. You can look at some of the prior frame animations that people have posted of DALLE-2 and figure this out:
DALLE-2 will eventually converge towards data in its set in an orderly manner, and will even make transformations using other parameters in its training set, but coherence is not intelligent inference.
Given a 100 question math test of basic arithmetic questions, its safe to say with near guaranteed certainty that mathematician will answer the test with 100% accuracy. There is a possibility that even GPT-3 could do that, but eventually, given enough generations, GPT-3 will make a what a human will interpret as an "objective mistake" simply because the model is entirely probabilistic. Humans, if you get reductive enough, also work on probabilities, but they're exponentially more accurate. Give the test to a mathematician 100 times, and there is an extremely high probability he will 100% all of those tests as well. GPT-3, and most transformers, fail as generations compound.
The frustrating thing about generative transformers is that they are an easy, boring problem to solve for AI researchers because "error rates" in the arts are much more forgiving. However, teaching an AI to accurately navigate a city while driving a gas tanker, at this current time, is nearly impossible with current hardware, because the margins of error are so low.
I'm not conflating. Read the post I replied to, moron. Also, the original DALLE used a modified version of GPT-3 for text encoding, and DALLE-2 very likely uses that same modified model, but the AI primitives used for CLIP and GPT-3 are fundamentally the same, just using different types of data. Fuck right off.
Your understanding is too simplistic and rigid. If there was "no way for them to escape their prior training", then they would not be able to do anything other than word-for-word copy their training material. But they don't, they can rearrange the words in their training material into entirely new text, hence they have escaped their training material by definition.
Sure, the output is probabilistic based on their training, but as said, that could in fact enough be to generate new things, because there's a good argument to be made that that's how humans work as well. All of our thoughts and ideas could in theory be extrapolated from past experience, memories and world-model. We don't know enough about neurology and AI to be sure, the systems are already too complex for us to make these kinds of black/white statements you are making.
>For example, GPT-3 can be asked math questions, that was never present in the training data, and it will answer them correctly, because it has learned generalized arithmetic from the examples that does exist in the training data.
hard to believe, since there are many examples of GPT-3 not understanding even some simple math (wrong results) problems when you change the numbers around
Then where do you think those artists get their training and resources and ideas from if not from exisitng works of others. 'Profound' is subjective, but nothing is actually 'new'
However "Screenshot of Raka and Reki from Haibane Renmei" goes entirely sideways. Including Yoshitoshi ABe's name gets me a warning for no discernible reason.
can't put the genie in the bottle, this train has no brakes and it's just going to get even more realistic and even better and eventually it will be open to the public without any restrictions.
this makes no sense.
scaling hypothesis naturally is limited by the size of the dataset.
gpt-3 is already trained on the whole internet. you won't easily get more data.
unclip encoder/decoder is already trained on 650M/250M images.
>unclip encoder/decoder is already trained on 650M/250M images.
and? that's a remarkable small pool of images from the entire internet, flickr alone has more images.
every video on the internet is also just a series of images which could be used
>When Google Images launched in 2001, 250 million images were indexed. By 2005, this was up to a billion images. Now, in 2010, they’re over 10 billion images indexed, Ling said.
That was 12 years ago and isn't even including images on non indexed sites, imagine now.
>2011 >Facebook does exist and is closing in on 100 billion photos hosted on the site, a number no one will be catching up to any time soon. >Flickr Boasts 6 Billion Photo Uploads >95 million photos and videos are shared on Instagram per day. >Over 40 billion photos and videos have been shared on the Instagram platform since its conception.
gpt-3 is already at the limit of our datasets.
so dall-e 2 is already the best we can get for understanding prompts.
looking at all those prompt engineering guides, dall-e 2 isn't that good at understanding prompts.
8 months ago
Anonymous
dalle2 is primitive compared to google parti you fucking crackhead
keep up
8 months ago
Anonymous
both are diffusion models
8 months ago
Anonymous
>dalle2 is primitive compared to google parti you fucking crackhead
parti uses a generic transformer for understanding prompts. not palm.
gpt-3 is superior.
where are you fucking invaders coming from who keep posting without understanding the context of the post you're replying to
are you robots? is this an ai invasion?
8 months ago
Anonymous
retard pls.
let's quote the parti paper
"The text encoder after pretraining performs comparably to BERT [36] on GLUE (see Appendix G, Table 9); however, the text encoder degrades after the full encoder-decoder training process on text-to-image generation."
BERT is several generations behind GPT-3.
8 months ago
Anonymous
again you failed to understand the context of what you're replying to
8 months ago
Anonymous
the context is
gpt-3 is already at the limit of our datasets.
so dall-e 2 is already the best we can get for understanding prompts.
looking at all those prompt engineering guides, dall-e 2 isn't that good at understanding prompts.
retard
8 months ago
Anonymous
yes and you're failing to understand it
8 months ago
Anonymous
both dall-e 2 and parti use text encoders.
dall-e 2's gpt3 is superior to parti's BERT-like generic transformer.
8 months ago
Anonymous
you keep repeating yourself for some reason despite being both wrong and talking about something unrelated, I'm like 99% sure you're a failed AI.
8 months ago
Anonymous
retard pls.
you have given zero arguments.
pls point out in the diagram which part is relevant for understanding prompts.
8 months ago
Anonymous
DALL-E 2 can't understand the prompt "Black man with three heads" or "Cat with six legs". Both Imagen and Parti can. Imagen and Parti are better at understanding prompts.
8 months ago
Anonymous
>Both Imagen and Parti can.
Google, the gals actually having access to Imagen and Parti, deny this:
"These models have many limitations. For example, neither can reliably produce specific counts of objects (e.g. “ten apples”)"
https://blog.google/technology/research/how-ai-creates-photorealistic-images-from-text/
8 months ago
Anonymous
Pretty fake and outdated news. Sure it has some issues with counting and sometimes ignores it all together but it is capable of it and is overall more capable of understanding the meaning of prompts better than DALL-E 2
8 months ago
Anonymous
none of those show "ten apples"
8 months ago
Anonymous
and none of your posts show DALL-E 2 having better understanding of prompts than imagen or parti, fuckface.
8 months ago
Anonymous
Dall-e2 consistently produces more impressive images, but lacks the flexibility of other software. I don't know how much of that is due to the gigantic no-badthink burden they place on image generation.
8 months ago
Anonymous
Bold claim but you don't have any evidence
I'll give you a hint why you don't have evidence, because it's a lie.
8 months ago
Anonymous
Just based on my own experience right now. Dall-E 2 produces some nice looking stuff but the conceptual universe it can access - and its ease of use tools - are both lacking. Can't beat it for nice looking images though.
8 months ago
Anonymous
Line 3 contradicts line 2.
Line 2 contradicts line 1.
Dalle-2 doesn't even use GPT-3
8 months ago
Anonymous
>dalle2 is primitive compared to google parti you fucking crackhead
parti uses a generic transformer for understanding prompts. not palm.
gpt-3 is superior.
You could argue that the planet earth itself is art, and humans for sure didn't create it.
Art itself is shaped by nature. If we didn't see the Sun(which was not our creation) we wouldn't be able to depict it in paintings.
They still can't make art. All the aesthetic details and fine tuning. You'd have to reiterate over a shitty AI result hundreds of times to get a finished master piece (as it stands at least). Art is still alive and well.
yeah, compare it to some of the parti prompts from https://parti.research.google/ for this guy
Just based on my own experience right now. Dall-E 2 produces some nice looking stuff but the conceptual universe it can access - and its ease of use tools - are both lacking. Can't beat it for nice looking images though.
AI is a simple math equation. What makes it special today as opposed to 10 years ago is that CPUs have exploded in computational power thanks to EUV lithography and the secondary (now primary lol) industry around GPUs and their enormous amount of compute resources.
You are standing inside of a large building. The walls are made of a light colored concrete, and there are some tall windows in it. The building is huge. The walls are at least 100 yards away in every direction from where you are standing. The building is hollow, and at least 30 stories high. There is several water falls coming from near the top of the building. The water falls are gushing water. The bottom of the building consists of several swimming pools. You are standing in one of the swimming pools looking up at the water fall pouring down on the inside of this building
It's literally worse than conditional VAEs from the 2014's or so times, but trained on more data. It's amazing how luddites are easily tricked like that.
I think BOT should worry more on OpenAI's upcoming bot that specializes on making code out of worded requests. In a few years a lot of code monkeys here will facing a bigger threat than artists do now because code is a lot more objective than art
>GitHub's one >no investors to please
so the one developed by microsoft+openai has no investors to please yet the one solely developed by openai does? now I've heard it all.
It's pretty cool and I've had legitimate use getting it to write quick functions for me, especially when I can't remember the syntax of some language I don't usually work with, but I doubt it will be much more than a utility feature in IDEs. Interpreting what software needs to actually do is fairly complex and getting it to understand a full codebase and how multiple moving parts work together is much more complicated. Not to mention considerations like complexity, security, etc.
Also while it's pretty good at certain things it chokes really hard on others. For example asking for the length of a .wav file consistently makes it shit, and it keeps giving me random user information.
When we're at the point that AI can efficiently code and create art, we're at the point where people no longer have to work shit jobs they have no interest in for the sake of working
Quite the opposite - if there's no option to be creative or alternately to do easy shit jobs for shit wages, and most of the middle management is automated then the only option left for real people will be ""high level decision making executives"" who can swan around being important while the masses barely make it month to month on the dole. But now the masses are 99% of humanity with no functional role whatsoever.
It's really good at making almost perfect images. Artists will find it incredibly valuable because they can easily touch up the images after the AI has done 95% of the work. Could be a real game changer.
what are the ethical implications of AI generated cp? assuming it's not using abuse material as an input, what would actually be wrong with someone generating those images?
people have been trained to feel repulsed by it even though there's nothing inherently wrong with it, therefor it will never be allowed by the general population.
what are the ethical implications of AI generated cp? assuming it's not using abuse material as an input, what would actually be wrong with someone generating those images?
and yet factory farming and the slaughter of animals in general is A-OK.
Things like DALL•E and Craiyon (formerly "Dall-E Mini", because its creators wanted to get a bit of fame for a piece of technology that's completely irrelevant from the actual DALL•E's developed by OpenAI) can possibly replace concept artists in the future. You don't need to hire someone to make a draft for whatever artistic ideas that you want to realize; you just need to be more specific with your wordings, and give the AI-generated output for the artists to clean up and finalize the results. Yes, that's an insult to the actual artists out there, but I do not mean that from the beginning; it's just that the possibility is there and a bit closer to reality than we'd like to admit.
In a way, we might possibly end up being janitors for the machine learning models in the far future, and our jobs is to just label and clean up the stuff the models generated, rather than we making the stuff by ourselves from scratch.
Does that hurt creativity? Or does the future I've described even have any chance to materialize? That's a question we can all ponder about.
Back in the day I would say miniaturize it and sell the software but everything is a subscription service now where you get some API coins.
Anyway, I think the biggest impact use cases are the other direction: transcription and translation.
why would you want dall-e when you can have imagen?
https://github.com/pesser/stable-diffusion
readme is outdated tho: this is stable diffusion, not latent diffusion.
This thread is about generative adversarial networks (GANS) you dumbass.
Nobody is talking about "deep learning" or any other shitty machine learning algorithm.
GANs have only been around since 2014 with the first human faces being generated in 2017.
the classic "I was only pretending to be a retard" after you get called out for actually being a retard.
Typical poser. Here's a tip for you, whenever you experience imposter syndrome in life it's because you are actually an imposter. There's no escaping that fact.
8 months ago
Anonymous
You mix and confuse basic concepts like a poorly trained bot... or an imposter.
8 months ago
Anonymous
The classic double down tactic in addition to the classic "I was only pretending to be a retard" after you get called out for actually being a retard.
Typical poser. Here's a tip for you, whenever you experience imposter syndrome in life it's because you are actually an imposter. There's no escaping that fact.
8 months ago
Anonymous
The classic double down tactic in addition to the classic "I was only pretending to be a retard" after you get called out for actually being a retard.
Typical poser. Here's a tip for you, whenever you experience imposter syndrome in life it's because you are actually an imposter. There's no escaping that fact.
>Where do we go from there?
Collectively agree to hide the fact that AI exists to the new generations and create a world where they get entertained automatically by machine creations.
Be ~~*AI*~~ >generate funny image
Be human basedbrain >OOooooo AI uprising very soon very scary
Do you retards think that when we're talking about the risks and fears of AI and automation we're talking about sentience or some shit? You're not on the same page, stop being retarded.
this is google's take on Parti: >Photorealistic outputs pose risks and concerns around the creation of deepfakes, and propagation of visually-oriented misinformation >We have decided not to release our models without further safeguards in place >We will focus on strategies, such as prompt filtering and output filtering >We hope we can all enjoy responsible aesthetic visual experiences.
Funny how americans still pretend to be a free country while government sponsored corporations get away with this kind of shit
China is unironically more of a free country nowadays
Back in the 60s this shit wouldn't fly:This tech should be improving the wealth of the country right now, while there's an advantage against other nations
[...]
Do you retards think that when we're talking about the risks and fears of AI and automation we're talking about sentience or some shit? You're not on the same page, stop being retarded.
I think some dumbasses are, as far as creating fake realities I think that ship has sailed and is fairly technology non-dependent. Fake written articles and manipulated ~~*scientific*~~ articles are just as potent as a deepfake.
To be clear neural networks are an interesting technology and are cool as fuck. The problem is it's not >AI
and when people call it >AI
it clicks something in our hollywood primed brains that blows things out of proportion.
>what does the future hold?
Probably gif generators of similar quality and limited application in medium/low budget entertainment production.
>what will 100% not happen
No chance this is some huge turning point in reality manipulation and propaganda.
No chance this is the beginning of some AI caused unemployment wave.
No chance this is the beginning of a new wave of actually good generated media/
AI at its current level is going to replace all the useless "designers" who make logos, icons, promotional media, mascots, etc. Those people are going to be fired, and they will have to do something useful for once. They will be replaced by random ordinary employees giving the AI prompts until the sole remaining art department employee, the AI screener, likes the result.
Yeah, exactly. Calling it AI fries normie brains way too much.
It's a technological revolution similar to the software/internet revolution we experienced in the late 90s. I was able to make a CV program that detected when I was working out, which then counted my reps and label exercises in my home gym (maybe I should publish a paper).
Super powerful stuff if you know how to use it. My advice to others itt.. stay invested in the snp. It's going to be a huge boon for productivity.
>No chance this is the beginning of some AI caused unemployment wave. >No chance this is the beginning of a new wave of actually good generated media/
ignorance is bliss I guess, you'll see.
the second point is already happening so I'm assuming you aren't even posting in good faith.
le funny pics equal to what a deviantart chud takes 2 hours to draw isnt 'good media'
come back when 'ai' can create actual works of art equal to chopin
are you a fucking troglodyte or what? current generative models have surpassed anything human artists have ever created in the history of mankind.
https://twitter.com/search?q=%23midjourney
https://twitter.com/search?q=%2523stablediffusion
https://parti.research.google/
https://imagen.research.google/
8 months ago
Anonymous
you must be an npc because you clearly dont understand art
soulless automaton
call me when a shitty 'ai' can write faust
8 months ago
Anonymous
post some good pictorical art, mr soul
8 months ago
Anonymous
just looks like generic digital art to me...
8 months ago
Anonymous
i don't want excuses, just post it, gay
8 months ago
Anonymous
For one you think digital art could ever surpass paintings, youre an actual NPC.
Also refer to what I said regarding music and literature.
8 months ago
Anonymous
8 months ago
Anonymous
8 months ago
Anonymous
polished edgy turd
https://i.imgur.com/Ck4R0mA.jpg
good stuff. i'll give you that
8 months ago
Anonymous
8 months ago
Anonymous
just looks like generic digital art to me...
low quality bait, go back to china.
8 months ago
Anonymous
it's actually funny, I see a trog in the comments of https://www.youtube.com/watch?v=03xMIcYiB80 pretty much parroting what you said even though he's 100% wrong.
8 months ago
Anonymous
Generated art by definition will never be as good as that created by humans.
8 months ago
Anonymous
Give it a few years
8 months ago
Anonymous
Part of what makes art great is who created it, if no one created it then how will time make up for this?
8 months ago
Anonymous
>Part of what makes art great is who created it
Exactly the opposite, anon.
8 months ago
Anonymous
it's better than any art i can make.
but really, this is a human made tool that creates images based on human input.
it's a veeeery far cry from a truly ai generated image, something imagined created from an artificial mind.
8 months ago
Anonymous
if entire world was built based on what some BOTtard cannot do, it would be shittier than it already is
8 months ago
Anonymous
there's plenty of images in this thread that are much better drawings than most people can create.
you only have to look at the front page of deviant art to see that.
8 months ago
Anonymous
doesn't change that all of them are dogshit
8 months ago
Anonymous
they're perfectly fine. there's nothing stopping someone with talent from touching up an ai generated image.
i can see plenty of uses for technology like this as a tool for artists.
8 months ago
Anonymous
if they really had talent they'd make something new and better than generic traced image which now uses an image traced by AI instead of an original work or something traced by another human
real art is not a replica of something that was drawn before, cope artlet
8 months ago
Anonymous
post some drawings gay
8 months ago
Anonymous
This regurgitated automatic photobashed bullshit is worth $0. Someone might think it makes a pretty neat wallpaper but nothing beyond that.
AI at its current level is going to replace all the useless "designers" who make logos, icons, promotional media, mascots, etc. Those people are going to be fired, and they will have to do something useful for once. They will be replaced by random ordinary employees giving the AI prompts until the sole remaining art department employee, the AI screener, likes the result.
You vastly underestimate how much goes into medium/high end design. These neural network tools might be used within the design pipeline but the final product will always be considered head to toe. Again, it might change the way some people work but it will not be more disruptive than digital image manipulation.
>No chance this is the beginning of some AI caused unemployment wave.
Not only will this happen, it is already happening, and in fact the garden gnome openly admits that's the plan. Their objective is to replace all working class Whites with automated machines so that they're out of any realistic job opportunities. Coupled with fiat based inflation this makes it practically impossible for Whites to make enough money to afford raising a White family. It's part of the White genocide the garden gnomes are committing, you must not be paying attention of what they talk about at Davos and Bilderberg.
It's correct to say it's not about ~~*AI*~~ gaining sentience or becoming conscious though, that's what garden gnomes like ~~*Lex Fridman*~~ promote as a red herring and as an introduction into ~~*trans-humanism*~~.
Dunno, wasted all my credits on generating green frogs to work around it, angrily spammed open ai with them and told them I was boycotting all their products indefinitely
Btw ic posters eternally btfo. Wonder what the vibe is like there
Music will be even easier for AI to master than art since it's purely mathematical.
There are even some basic examples trained way before we had current generations of TPUs and GPUs for training.
Read Pythagoras and then understand that music goes far beyond 'just math'. Without the intellective faculty of the soul, true heights in music can never be reached.
the soul doesn't exist, it's just electrical impulses following the basic laws of physics which have been broken down into mathematical formulae you absolute buffoon. you put on a front attempting to act sophisticated but you know nothing and you will never accomplish anything. the heights required to achieve greatness are unreachable for people like you.
If you think any masterpiece painting or symphoney is comparable to >dog rides skateboard made of bananas while smoking in style of Rick and Morty
You NEED to stop posting and go back to school
>be /3/ gay >laughed at cashiers and manual workers being replaced by machines >read about AI being able to make 3D models a well >watch AI making 3D models >increasingly nervous about these developments
I don't want to suck cocks for living.
You can't trust AI to successfully complete tasks with any kind of competency. Ergo, you're gonna have inaccurate, error prone services in the future.
LMAO, if you think payment processors being occasionally offline makes shit Africa-tier slow, just wait until the fucking stupid moron neighbor cattle decides to attempt to fully automate the public sector with AI magic.
The catch-phrase/life motto of future IT-specialists will be: "Uhhh I don't know, it's supposed to work.................. let me call our support team"
What you used to be able to make work with the help of a skilled person, you will hence forth never be able to make functional again, thanks to endless bureaucratic hoop jumping and modern practices of retardation and la la land thinking.
I can't believe I wasted one of my precious credits over this shit. Fucking morons should stop training the Ai on deviantart, morons
FUUUUUUUUUUCKKKKKKKKK
Hardcore CP movies, in which there were no actual children harmed, thus destroying the child porn black market, and saving a lot of children from the physical and emotional trauma of CSA.
>inb4 "well that will just encourage pedos to hurt real kids"
BS. Violent video games and movies don't make people violet murders. Porn watching reduces actual real sex acts between people. Watching food network doesn't make you any better at cooking and doesn't make you fat.
You can't prove me wrong without using horribly biased sources. Individual instances are not a trend, nor are they statistically significant given the global population.
Even better, as it improves you could have a feedback loop where it detects brain activity and adjusts the content accordingly just for you at every moment
Stable Diffusion isn't.
I'm not a porn artist or anything I just can't stand the "we know what's best" attitude. I was reading someone trying to generate a cockroach (nope, it caught "cock") or stuff like "a man shot out of a cannon" (nope, can't have "shot") and just so much bullshit like that, completely ridiculous.
Immediately signed up for Stable Diffusion just on the promise that they'll open source it and there won't be this nonsense.
>Stable Diffusion isn't.
Well to correct myself here, they likely are/will on their own servers, but if it goes open-source then you can run whatever you like locally and likely there will be nsfw implementations online.
stable diffusion will never reach public release of their real model, they're just using their shitty public one to generate hype and sell NFTs.
They didn't end up with $40,000,000 worth of GPUs just to give away trained models.
>hurr durr open source cannot exist
You absolute retard. They sell support and offer "instead of setting this shit up locally with the high-end gpu you may not have, you can generate through us on the cheap, and with support"
nope, they're just going to sell ai generated NFTs like their developers already do on twitter.
It's nothing but a cryptoshill pump and dump with a new twist.
Alright, well at the worst they do a complete 180 from their stated intentions and end up in the same state as "open"AI and the others.
Not impossible but I'd be surprised at this point. And in that case, it's just waiting. Math is math, there will be high quality implementations available to the general public. The math and technology is extremely disruptive and can be very effectively "black boxed", I don't see a way to really keep a lid on it. It reminds me of the scrambling to try and maintain control of MP3.
8 months ago
Anonymous
None of these """principles""" are really in the interest of anything other than the jobs and egos of the people who take a salary for holding serious meetings to discuss how to ensure the future is maximally woke. The only actual card they hold at the moment is the fairly extreme hardware investment needed to produce the best-quality outputs. The second that balance tips, they are in the trash bin because the people will find their own way.
for americans the rent isn't free
malding
can dalle create wojaks?
wait till the sharty gets their hands on this technology.....
you pay tax money to have your military everywhere to no benefit to you. youre literally paying rent to be in our heads
I wonder what's the rent IN America..
looks like Gabe
My girlfriend is cuter.
Wait when did Gabe bleach his hair?
it's natural. he's having a hard time since the deck flopped so bad. also, looks like he weighs half than last year
It only needs some phyisics training, those weelchairs would never be able to hold the weight
>scene of a hippopotamus enjoying a meal
>Average sized American with mild disability holding small snack cheering for their favourite race
https://arc.tencent.com/en/ai-demos/faceRestoration
>>the burger ATTACHED onto the left hand
"We" are not going anywhere. Science and technology are merely tools of oppression helping to create more slaves.
Retard
>slaves
Ain't nobody want that. What these developments do is they remove the need for people. One tiny demographic at a time, you become unneeded and unwanted, just marked for death. A slow one, of your own making, while you are made to feel guilt and shame.
Yes, that because this isn't "we". "we" didn't create and don't own any of these technologies. They will be rented to us in limited form if it makes the owners profit.
And yet you are here on the internet and not living in the mountains without electricity, you monumental fucking hypocrite.
The Internet is indeed used as a tool for invigilation, yes. The fact that it's everywhere and that I need to use it unless I want to cut all ties with family and friends is not something I have control over.
A buckbroken wagecuck or a sperglord denying the fact that his meat calculator job will also be easily replacable. Either way, acquire class consciousness. Until then, meds.
Why? Because you will no longer be paid to contribute nothing to society?
>Oh no, my art job is redundant, I'm a slave! I have to actually contribute real value to society to get paid and enjoy its benefits!
What tech does is it takes the worthless eaters like artists and it reminds them how useless they always were. Basically you are a cotton picking dumbfuck being replaced during the industrial revolution.
At some point the elites won't need 99% of the population anymore, we'll all be replaced by robots.
>he thinks its AI
its using image search as a starting point you fucktard
No it's literally AI
>starting point
Oh, well then I guess there is nothing amazing about it at all.
I can already see hacks generating these then tracing or just slightly reworking them it's crazy
Do you know what photobashing is?
Not him, but I will gladly "bash" your head in, gay
What AI is this?
These all appear to be DALL-E 2 generations to me.
It's really not
guess what i put
glow in the darks
ok, I am laughing
new BOT reaction image hot off the press
AI is the future
A glow in the dark cia agent that is African American digital art?
you dont have to add the descriptor black or african, it will add it for you
3 letter phosphorescent black agent
lost
>NOOOO YOU CAN'T MAKE THAT ART! THAT'S INTELLECTUAL PROPERTY THEFT!
>USE THE ROBOT! IT'S NOT SUBJECT TO IP LAWS
every human would be capable of achieving beauty if it wasn't illegal to. information is the ultimate evil in this world unless you're an """AI""".
I actually agree with this. Anything your AI makes must be owned by the owners of the materials used in training it.
they gonna profit from all the human knowledge and you will be happy.
What? Who is stopping you from making paintings?
We generate child porn with it obviously
This raises an interesting question: does ai generated content have to be treated with the same rules? Like e-boi porn isn't illegal, but cp is, so how would you rule ai generated cp?
Technically ai didn't generate all these images, they just rearrange pixel patterns from the dataset they've learnt, so any "generated" cp is a mixture of various existing cp.
By that logic, yes it should be illegal, since ai could never generate cp if it never seen one.
>Technically ai didn't generate all these images, they just rearrange pixel patterns from the dataset they've learnt, so any "generated" cp is a mixture of various existing cp.
you have no idea what you're talking about, dunce. read about the topic and then come back hopefully knowing more.
Maybe five years ago, but these new AI are sophisticated to the point where they can clearly merge concepts in unique ways. You think OP's picture is drawn only from the AI's understanding of hippos in hot tubs? Of course not. It's drawn from its understanding of hippos and hot tubs separately.
>By that logic, yes it should be illegal, since ai could never generate cp if it never seen one.
Co-Pilot IS NOT ILLEGAL
STOP TRYING TO PRETEND THE SOURCE MATERIAL MATTERS!
Realistic drawn CP is illegal in the USA, so this would also apply.
It cant create porn so it's useless
e-boi is already barely legal in the US. The law on that is pretty vague and it's illegal in most of europe. Realistic looking e-boi porn would definitely be illegal. People in USA have actually got convicted for having e-boi. The law in USA is: if the general population thinks that the work clearly lack any literary, artistic, political, or scientific value. Yes, it's that subjective. Porn is not considered freedom of expression in USA.
Find me a person in the US who was convicted for it who didn't also have real stuff in possession. The only time they give a shit is if it's something they can tack onto a more serious charge.
The people convicted have either also had the real stuff, comited rape or had previous convictions for one of those two.
exactly like lab grown meat
but the issue is it cannot produce ex nihilo
it has to be trained on models
so...
This is a common misconception.
okay, enjoy the equivalent of ai generated volleyball matches trained on badminton and basketball
>tv, replace every character in this movie with nicholas cage
>also change the plot so that they all end up in the same bed in the end
Just got access to DALL-E. What to do?
Nowhere
And when it all comes crashing down, and you come crawling back to artists for work, they'll all be sure to tack on an nice extra 'thinking this AI could do shit better than them' fee
these look like shit
what I don't understand with most of the machine learning stuff is that the AI is learning from resources that got created by humans like books, pictures etc. so in the end it can only create things that are in some way mixed or derived from what already exists and has been created by humans; not really new stuff. so in context of creating profound new, really creative stuff, it simply won't work
No, we've already seen that AI can extrapolate outside training data. For example, GPT-3 can be asked math questions, that was never present in the training data, and it will answer them correctly, because it has learned generalized arithmetic from the examples that does exist in the training data.
You could say that this is still not new since it's still just within the concept of arithmetic which was present in the training data, but if you speak that broadly then it's the same for most humans. We also almost never generate something new, we are just inspired from existing art or existing culture and extrapolate on that. Otherwise it wouldn't even be good art, because if it has no connection to existing concepts, it has no meaning to us.
>For example, GPT-3 can be asked math questions, that was never present in the training data, and it will answer them correctly, because it has learned generalized arithmetic from the examples that does exist in the training data.
hard to believe, since there are many examples of GPT-3 not understanding even some simple math (wrong results) problems when you change the numbers around
Shut right the fuck up. You're spouting bullshit. You don't know what you're talking about. GPT stands for "generative pre-trained transformer" - there is no way for them to escape their prior training by their very nature. You can look at some of the prior frame animations that people have posted of DALLE-2 and figure this out:
DALLE-2 will eventually converge towards data in its set in an orderly manner, and will even make transformations using other parameters in its training set, but coherence is not intelligent inference.
Given a 100 question math test of basic arithmetic questions, its safe to say with near guaranteed certainty that mathematician will answer the test with 100% accuracy. There is a possibility that even GPT-3 could do that, but eventually, given enough generations, GPT-3 will make a what a human will interpret as an "objective mistake" simply because the model is entirely probabilistic. Humans, if you get reductive enough, also work on probabilities, but they're exponentially more accurate. Give the test to a mathematician 100 times, and there is an extremely high probability he will 100% all of those tests as well. GPT-3, and most transformers, fail as generations compound.
The frustrating thing about generative transformers is that they are an easy, boring problem to solve for AI researchers because "error rates" in the arts are much more forgiving. However, teaching an AI to accurately navigate a city while driving a gas tanker, at this current time, is nearly impossible with current hardware, because the margins of error are so low.
why do you numbskulls keep conflating gpt3 and dalle2
I'm not conflating. Read the post I replied to, moron. Also, the original DALLE used a modified version of GPT-3 for text encoding, and DALLE-2 very likely uses that same modified model, but the AI primitives used for CLIP and GPT-3 are fundamentally the same, just using different types of data. Fuck right off.
durka durka BUT THE ORIGINAL USED THIS YEARS AGO
fuck outta here poser, do some book learning and then come back.
>poser
Your understanding is too simplistic and rigid. If there was "no way for them to escape their prior training", then they would not be able to do anything other than word-for-word copy their training material. But they don't, they can rearrange the words in their training material into entirely new text, hence they have escaped their training material by definition.
Sure, the output is probabilistic based on their training, but as said, that could in fact enough be to generate new things, because there's a good argument to be made that that's how humans work as well. All of our thoughts and ideas could in theory be extrapolated from past experience, memories and world-model. We don't know enough about neurology and AI to be sure, the systems are already too complex for us to make these kinds of black/white statements you are making.
gpt-3 is hot garbage
That's not really what it's for. The goal is to process language in a natural way, it isn't meant to know random factoids.
just asked my grandma, she gave the same answer.
this confirms it: GPT-3 is an AGI.
but all humans do that retard and nobody can create anything truly new already
Then where do you think those artists get their training and resources and ideas from if not from exisitng works of others. 'Profound' is subjective, but nothing is actually 'new'
You do realise this is what humans are doing too
>daniel x dall-e
>human & ai
so he painted over the generated images? why is everyone itt ignoring that line?
I think they just typed in the prompt, but want to think they're so smart because its 'AI'.
Its machine learning. Hell, Dall-E wasn't even the first, but only the most popular at the moment.
Anyone can type, but since it is machine learning, they have to think they're some sort of genius for doing that. I'm guilty of doing that too.
he did not.
that's just the default caption on any dall-e 2 screenshot.
see:
https://labs.openai.com/s/mz8LiyvkI8KwD2luJ6MrS23m
>see:
>link that goes to a blank page
thanks, now i know
""open""ai shills
I’ve been on the waitlist forever, what did you all put for the questions? Researcher? Student? Or something else
el ogro...
This is art
la creatura
Wait, are those burger buns pancakes? Shit, that's a million dollar idea.
It's called a McGriddle and it's not a million dollar idea here in America, it's a BILLION dollar idea.
let me tell you which countries are white
I want to play BOT TCG now badly
Generated this one today, guess which prompt I wrote kek
Chinese person eating a bat soup?
uma delicia
Good prompt. Screenshot of the angel enjoying tea and cookies from Hayao Miyazaki's My Neighbour Angel.
would watch
"on your mark" by ghibli has basically this premise
collapse of society
However "Screenshot of Raka and Reki from Haibane Renmei" goes entirely sideways. Including Yoshitoshi ABe's name gets me a warning for no discernible reason.
>Including Yoshitoshi ABe's name gets me a warning for no discernible reason.
probably because of Shinzo Abe
this shit makes me uncomfortable
computers are not supposed to make art
i want to turn back time
can't put the genie in the bottle, this train has no brakes and it's just going to get even more realistic and even better and eventually it will be open to the public without any restrictions.
this makes no sense.
scaling hypothesis naturally is limited by the size of the dataset.
gpt-3 is already trained on the whole internet. you won't easily get more data.
unclip encoder/decoder is already trained on 650M/250M images.
>unclip encoder/decoder is already trained on 650M/250M images.
and? that's a remarkable small pool of images from the entire internet, flickr alone has more images.
every video on the internet is also just a series of images which could be used
>When Google Images launched in 2001, 250 million images were indexed. By 2005, this was up to a billion images. Now, in 2010, they’re over 10 billion images indexed, Ling said.
That was 12 years ago and isn't even including images on non indexed sites, imagine now.
>2011
>Facebook does exist and is closing in on 100 billion photos hosted on the site, a number no one will be catching up to any time soon.
>Flickr Boasts 6 Billion Photo Uploads
>95 million photos and videos are shared on Instagram per day.
>Over 40 billion photos and videos have been shared on the Instagram platform since its conception.
who's going to provide labels for all those images?
unlike LLMs you can't use unsupervised training for image creation.
gpt-3 is already at the limit of our datasets.
so dall-e 2 is already the best we can get for understanding prompts.
looking at all those prompt engineering guides, dall-e 2 isn't that good at understanding prompts.
dalle2 is primitive compared to google parti you fucking crackhead
keep up
both are diffusion models
where are you fucking invaders coming from who keep posting without understanding the context of the post you're replying to
are you robots? is this an ai invasion?
retard pls.
let's quote the parti paper
"The text encoder after pretraining performs comparably to BERT [36] on GLUE (see Appendix G, Table 9); however, the text encoder degrades after the full encoder-decoder training process on text-to-image generation."
BERT is several generations behind GPT-3.
again you failed to understand the context of what you're replying to
the context is
retard
yes and you're failing to understand it
both dall-e 2 and parti use text encoders.
dall-e 2's gpt3 is superior to parti's BERT-like generic transformer.
you keep repeating yourself for some reason despite being both wrong and talking about something unrelated, I'm like 99% sure you're a failed AI.
retard pls.
you have given zero arguments.
pls point out in the diagram which part is relevant for understanding prompts.
DALL-E 2 can't understand the prompt "Black man with three heads" or "Cat with six legs". Both Imagen and Parti can. Imagen and Parti are better at understanding prompts.
>Both Imagen and Parti can.
Google, the gals actually having access to Imagen and Parti, deny this:
"These models have many limitations. For example, neither can reliably produce specific counts of objects (e.g. “ten apples”)"
https://blog.google/technology/research/how-ai-creates-photorealistic-images-from-text/
Pretty fake and outdated news. Sure it has some issues with counting and sometimes ignores it all together but it is capable of it and is overall more capable of understanding the meaning of prompts better than DALL-E 2
none of those show "ten apples"
and none of your posts show DALL-E 2 having better understanding of prompts than imagen or parti, fuckface.
Dall-e2 consistently produces more impressive images, but lacks the flexibility of other software. I don't know how much of that is due to the gigantic no-badthink burden they place on image generation.
Bold claim but you don't have any evidence
I'll give you a hint why you don't have evidence, because it's a lie.
Just based on my own experience right now. Dall-E 2 produces some nice looking stuff but the conceptual universe it can access - and its ease of use tools - are both lacking. Can't beat it for nice looking images though.
Line 3 contradicts line 2.
Line 2 contradicts line 1.
Dalle-2 doesn't even use GPT-3
>dalle2 is primitive compared to google parti you fucking crackhead
parti uses a generic transformer for understanding prompts. not palm.
gpt-3 is superior.
speakers make me uncomfortable
machines are not supposed to record and replay art
i want to turn back time
art is meant to be uniquely human
You could argue that the planet earth itself is art, and humans for sure didn't create it.
Art itself is shaped by nature. If we didn't see the Sun(which was not our creation) we wouldn't be able to depict it in paintings.
It just proves that art is a farce
You're a pile of neural networks, just like those AIs
They still can't make art. All the aesthetic details and fine tuning. You'd have to reiterate over a shitty AI result hundreds of times to get a finished master piece (as it stands at least). Art is still alive and well.
dummy, the ai can do those hundreds of iterations in microseconds.
B-BUT THE HUMAN SPIRIT WILL PREVAIL; ART WILL ALWAYS BE ABOUT CREATIO ORGANICUM HUMANAE!
Here's a DMC music video made with all the dialogue line images generated by neural net.
>dmc jive
didn't watch
you and I both know no one would give either of those pictures even a passing glance if they were made by a human. so who cares?
Just got access today, any ideas?
yeah, compare it to some of the parti prompts from https://parti.research.google/ for this guy
Dimetrodon at an oasis in the Permian desert, 280 million years ago. Award winning National Geographic photograph, wide shot, landscape
generation-worthy prompt, this is the best of the four
Awesome, thanks! Didn't expect it to turn out like a painting.
did this motherfucker generate a watermark into the image lmfao
did they seriously use watermarked images in the training data?
I wouldn't be surprised if they do.
Like code hosted on github.
> did they seriously use watermarked images in the training data?
yes lol, how they gonna prove that
we need to talk about that ifunny watermark
what about this look what a watermark? all i see is a rocky outcrop underwater
c^|:|
Brachiosaurus eating leaves from a large tree while other dinosaurs drink from a river nearby. National geographic photograph, landscape
The random force inserted morons always gets me kekking
I look like that
lets do a
Coalition of Humanity against garden gnomes and the Worldwide Pedo Elite
nah, if they're pedobros they can't be so bad
Gee-willikers, it generated a static, low resolution image that vaguely resembles the query it was sent. The future truly is nao! Yowza!
AI is a simple math equation. What makes it special today as opposed to 10 years ago is that CPUs have exploded in computational power thanks to EUV lithography and the secondary (now primary lol) industry around GPUs and their enormous amount of compute resources.
It just can't fucking do faces, can it?
Even the simplest of anime style faces are terrible.
You can fix them afterwords with
You are standing inside of a large building. The walls are made of a light colored concrete, and there are some tall windows in it. The building is huge. The walls are at least 100 yards away in every direction from where you are standing. The building is hollow, and at least 30 stories high. There is several water falls coming from near the top of the building. The water falls are gushing water. The bottom of the building consists of several swimming pools. You are standing in one of the swimming pools looking up at the water fall pouring down on the inside of this building
bump
bamp
It's literally worse than conditional VAEs from the 2014's or so times, but trained on more data. It's amazing how luddites are easily tricked like that.
post examples.
I think BOT should worry more on OpenAI's upcoming bot that specializes on making code out of worded requests. In a few years a lot of code monkeys here will facing a bigger threat than artists do now because code is a lot more objective than art
github already did that lmao
https://github.com/features/copilot
GitHub's one is shit since it has no investors to please. I'm telling you it will blow your mind when you see OpenAI's
>GitHub's one
>no investors to please
so the one developed by microsoft+openai has no investors to please yet the one solely developed by openai does? now I've heard it all.
what are MS investors.
are you retarded?
github copilot uses openai's codex.
It's pretty cool and I've had legitimate use getting it to write quick functions for me, especially when I can't remember the syntax of some language I don't usually work with, but I doubt it will be much more than a utility feature in IDEs. Interpreting what software needs to actually do is fairly complex and getting it to understand a full codebase and how multiple moving parts work together is much more complicated. Not to mention considerations like complexity, security, etc.
Also while it's pretty good at certain things it chokes really hard on others. For example asking for the length of a .wav file consistently makes it shit, and it keeps giving me random user information.
>security
considering all security issues are simply caused by human mistakes, machines should be better at it.
>description of function is more characters than the actual code
wow you really saved time there buddy
Only the first comment line was my prompt, that joshuaburkhart dude probably wrote the other two lines
When we're at the point that AI can efficiently code and create art, we're at the point where people no longer have to work shit jobs they have no interest in for the sake of working
No. Learn2work.
except for the fact that people still need food to survive and shelter so they still have to work
Quite the opposite - if there's no option to be creative or alternately to do easy shit jobs for shit wages, and most of the middle management is automated then the only option left for real people will be ""high level decision making executives"" who can swan around being important while the masses barely make it month to month on the dole. But now the masses are 99% of humanity with no functional role whatsoever.
AI videos
It's really good at making almost perfect images. Artists will find it incredibly valuable because they can easily touch up the images after the AI has done 95% of the work. Could be a real game changer.
what are the ethical implications of AI generated cp? assuming it's not using abuse material as an input, what would actually be wrong with someone generating those images?
people have been trained to feel repulsed by it even though there's nothing inherently wrong with it, therefor it will never be allowed by the general population.
and yet factory farming and the slaughter of animals in general is A-OK.
Things like DALL•E and Craiyon (formerly "Dall-E Mini", because its creators wanted to get a bit of fame for a piece of technology that's completely irrelevant from the actual DALL•E's developed by OpenAI) can possibly replace concept artists in the future. You don't need to hire someone to make a draft for whatever artistic ideas that you want to realize; you just need to be more specific with your wordings, and give the AI-generated output for the artists to clean up and finalize the results. Yes, that's an insult to the actual artists out there, but I do not mean that from the beginning; it's just that the possibility is there and a bit closer to reality than we'd like to admit.
In a way, we might possibly end up being janitors for the machine learning models in the far future, and our jobs is to just label and clean up the stuff the models generated, rather than we making the stuff by ourselves from scratch.
Does that hurt creativity? Or does the future I've described even have any chance to materialize? That's a question we can all ponder about.
Generating videos. Feature length films.
Generating CAD files for product design, and models for 3D printers.
Back in the day I would say miniaturize it and sell the software but everything is a subscription service now where you get some API coins.
Anyway, I think the biggest impact use cases are the other direction: transcription and translation.
>Where do we go from there?
your worst nightmare.
>Where do we go from there?
which is the way that's clear
Apart from using huggingface, are there any other free dall-e software?
why would you want dall-e when you can have imagen?
https://github.com/pesser/stable-diffusion
readme is outdated tho: this is stable diffusion, not latent diffusion.
I don't have ten million dollars and access to billions if images to train my model on
trained models are currently in beta and will be completely open-sourced soon:
https://stability.ai/beta-signup-form
>70 years of computers and this ugly hippo is the peak.
where is the terminator?
this technology has only been around since June 2014
image searchers are older..
troglodyte, go back to the gaming boards.
deep learning is several decades old.
Deep preteen penetration is as old as humanity.
This thread is about generative adversarial networks (GANS) you dumbass.
Nobody is talking about "deep learning" or any other shitty machine learning algorithm.
GANs have only been around since 2014 with the first human faces being generated in 2017.
Yep Neural networks are old as fuck. Thanks for proving my point.
the classic "I was only pretending to be a retard" after you get called out for actually being a retard.
Typical poser. Here's a tip for you, whenever you experience imposter syndrome in life it's because you are actually an imposter. There's no escaping that fact.
You mix and confuse basic concepts like a poorly trained bot... or an imposter.
The classic double down tactic in addition to the classic "I was only pretending to be a retard" after you get called out for actually being a retard.
Typical poser. Here's a tip for you, whenever you experience imposter syndrome in life it's because you are actually an imposter. There's no escaping that fact.
FUCK YOU BOTgay!!
dall-e isn't gan but diffusion.
gans are impossible to train.
>gans are impossible to train.
aren't they training themselves by design?
>Where do we go from there?
Collectively agree to hide the fact that AI exists to the new generations and create a world where they get entertained automatically by machine creations.
Cum
i'm glad dall-e is being used for its intended purpose:
photos of obese americans doing american things
can we get some clapping
>obese americans
Just call them americans Anon, people will automatically know that they're fat..
factually false
dude le AI is so scary dude... SKYNET!!! lol just like skynet
stay ignorant, I'll laugh when you have no job because it was replaced by ai and have to live on the streets to survive you fucking loser.
Do you retards think that when we're talking about the risks and fears of AI and automation we're talking about sentience or some shit? You're not on the same page, stop being retarded.
this is google's take on Parti:
>Photorealistic outputs pose risks and concerns around the creation of deepfakes, and propagation of visually-oriented misinformation
>We have decided not to release our models without further safeguards in place
>We will focus on strategies, such as prompt filtering and output filtering
>We hope we can all enjoy responsible aesthetic visual experiences.
Funny how americans still pretend to be a free country while government sponsored corporations get away with this kind of shit
China is unironically more of a free country nowadays
Back in the 60s this shit wouldn't fly:This tech should be improving the wealth of the country right now, while there's an advantage against other nations
>open AI
>code isn’t open source
I hate it
Be ~~*AI*~~
>generate funny image
Be human basedbrain
>OOooooo AI uprising very soon very scary
come back in 5 years when you're out of a job, chucklefuck.
I think some dumbasses are, as far as creating fake realities I think that ship has sailed and is fairly technology non-dependent. Fake written articles and manipulated ~~*scientific*~~ articles are just as potent as a deepfake.
To be clear neural networks are an interesting technology and are cool as fuck. The problem is it's not
>AI
and when people call it
>AI
it clicks something in our hollywood primed brains that blows things out of proportion.
>what does the future hold?
Probably gif generators of similar quality and limited application in medium/low budget entertainment production.
>what will 100% not happen
No chance this is some huge turning point in reality manipulation and propaganda.
No chance this is the beginning of some AI caused unemployment wave.
No chance this is the beginning of a new wave of actually good generated media/
AI at its current level is going to replace all the useless "designers" who make logos, icons, promotional media, mascots, etc. Those people are going to be fired, and they will have to do something useful for once. They will be replaced by random ordinary employees giving the AI prompts until the sole remaining art department employee, the AI screener, likes the result.
Yeah, exactly. Calling it AI fries normie brains way too much.
It's a technological revolution similar to the software/internet revolution we experienced in the late 90s. I was able to make a CV program that detected when I was working out, which then counted my reps and label exercises in my home gym (maybe I should publish a paper).
Super powerful stuff if you know how to use it. My advice to others itt.. stay invested in the snp. It's going to be a huge boon for productivity.
>My advice to others itt.. stay invested in the snp.
I'm stirring last night's haggis every night laddie
>No chance this is the beginning of some AI caused unemployment wave.
>No chance this is the beginning of a new wave of actually good generated media/
ignorance is bliss I guess, you'll see.
the second point is already happening so I'm assuming you aren't even posting in good faith.
le funny pics equal to what a deviantart chud takes 2 hours to draw isnt 'good media'
come back when 'ai' can create actual works of art equal to chopin
are you a fucking troglodyte or what? current generative models have surpassed anything human artists have ever created in the history of mankind.
https://twitter.com/search?q=%23midjourney
https://twitter.com/search?q=%2523stablediffusion
https://parti.research.google/
https://imagen.research.google/
you must be an npc because you clearly dont understand art
soulless automaton
call me when a shitty 'ai' can write faust
post some good pictorical art, mr soul
just looks like generic digital art to me...
i don't want excuses, just post it, gay
For one you think digital art could ever surpass paintings, youre an actual NPC.
Also refer to what I said regarding music and literature.
polished edgy turd
good stuff. i'll give you that
low quality bait, go back to china.
it's actually funny, I see a trog in the comments of https://www.youtube.com/watch?v=03xMIcYiB80 pretty much parroting what you said even though he's 100% wrong.
Generated art by definition will never be as good as that created by humans.
Give it a few years
Part of what makes art great is who created it, if no one created it then how will time make up for this?
>Part of what makes art great is who created it
Exactly the opposite, anon.
it's better than any art i can make.
but really, this is a human made tool that creates images based on human input.
it's a veeeery far cry from a truly ai generated image, something imagined created from an artificial mind.
if entire world was built based on what some BOTtard cannot do, it would be shittier than it already is
there's plenty of images in this thread that are much better drawings than most people can create.
you only have to look at the front page of deviant art to see that.
doesn't change that all of them are dogshit
they're perfectly fine. there's nothing stopping someone with talent from touching up an ai generated image.
i can see plenty of uses for technology like this as a tool for artists.
if they really had talent they'd make something new and better than generic traced image which now uses an image traced by AI instead of an original work or something traced by another human
real art is not a replica of something that was drawn before, cope artlet
post some drawings gay
This regurgitated automatic photobashed bullshit is worth $0. Someone might think it makes a pretty neat wallpaper but nothing beyond that.
You vastly underestimate how much goes into medium/high end design. These neural network tools might be used within the design pipeline but the final product will always be considered head to toe. Again, it might change the way some people work but it will not be more disruptive than digital image manipulation.
>No chance this is the beginning of some AI caused unemployment wave.
Not only will this happen, it is already happening, and in fact the garden gnome openly admits that's the plan. Their objective is to replace all working class Whites with automated machines so that they're out of any realistic job opportunities. Coupled with fiat based inflation this makes it practically impossible for Whites to make enough money to afford raising a White family. It's part of the White genocide the garden gnomes are committing, you must not be paying attention of what they talk about at Davos and Bilderberg.
It's correct to say it's not about ~~*AI*~~ gaining sentience or becoming conscious though, that's what garden gnomes like ~~*Lex Fridman*~~ promote as a red herring and as an introduction into ~~*trans-humanism*~~.
I generated this on midjourney
Enjoying it way more than dalle
Btw Pepe is banned on dalle
Using dalle means you're complicit
what about apu?
Dunno, wasted all my credits on generating green frogs to work around it, angrily spammed open ai with them and told them I was boycotting all their products indefinitely
Btw ic posters eternally btfo. Wonder what the vibe is like there
what was this image referencing? how does it look so good
Midjourney still tries to babysit your use.
Stable Diffusion will be totally open source, and already gets results as good as dalle2/mj
Pepe smoking a blunt
Had to work around it because Pepe is banned
More banned frogs
Might by my favorite
holy shit, the rare pepes market is about to plummet
Music will be even easier for AI to master than art since it's purely mathematical.
There are even some basic examples trained way before we had current generations of TPUs and GPUs for training.
Read Pythagoras and then understand that music goes far beyond 'just math'. Without the intellective faculty of the soul, true heights in music can never be reached.
the soul doesn't exist, it's just electrical impulses following the basic laws of physics which have been broken down into mathematical formulae you absolute buffoon. you put on a front attempting to act sophisticated but you know nothing and you will never accomplish anything. the heights required to achieve greatness are unreachable for people like you.
Didnt realize I was on leddit. Kys, hylic.
if the soul doesn't exist how do (You) explain out-of-body-experiences, telekinesis, and reincarnation, retard?
If you think any masterpiece painting or symphoney is comparable to
>dog rides skateboard made of bananas while smoking in style of Rick and Morty
You NEED to stop posting and go back to school
>be /3/ gay
>laughed at cashiers and manual workers being replaced by machines
>read about AI being able to make 3D models a well
>watch AI making 3D models
>increasingly nervous about these developments
I don't want to suck cocks for living.
>I don't want to suck cocks for living.
afraid of having a good time ?
>memories are non local
So memories are like AWS?
>This AI shit unironically mind blowing. Where do we go from there?
Accurate.
>normal size us citizen
underrated
l-e-boi?
>All the missing scenes from Hayao Miyazaki's Sprited Away
Is there an intuitive explanation for how these NNs work or do the people investigating them not fully understand yet?
I tried it, it was underwhelming
You can't trust AI to successfully complete tasks with any kind of competency. Ergo, you're gonna have inaccurate, error prone services in the future.
LMAO, if you think payment processors being occasionally offline makes shit Africa-tier slow, just wait until the fucking stupid moron neighbor cattle decides to attempt to fully automate the public sector with AI magic.
The catch-phrase/life motto of future IT-specialists will be: "Uhhh I don't know, it's supposed to work.................. let me call our support team"
What you used to be able to make work with the help of a skilled person, you will hence forth never be able to make functional again, thanks to endless bureaucratic hoop jumping and modern practices of retardation and la la land thinking.
I can't believe I wasted one of my precious credits over this shit. Fucking morons should stop training the Ai on deviantart, morons
FUUUUUUUUUUCKKKKKKKKK
>please use some of your shekkels to top up on credits
Usually I have to specify a term like "detailed" to illustrations to prevent generating deviantart shit
that's because you didn't give it the right prompt, you suck at communicating with robots as much as you do humans.
fuck you that was very mean of you
i am working tirelessly to outlaw AI
Hardcore CP movies, in which there were no actual children harmed, thus destroying the child porn black market, and saving a lot of children from the physical and emotional trauma of CSA.
>inb4 "well that will just encourage pedos to hurt real kids"
BS. Violent video games and movies don't make people violet murders. Porn watching reduces actual real sex acts between people. Watching food network doesn't make you any better at cooking and doesn't make you fat.
You can't prove me wrong without using horribly biased sources. Individual instances are not a trend, nor are they statistically significant given the global population.
who cares about saving the kids, fuck them.
I just want on demand anime using my own dream plots and characters.
Even better, as it improves you could have a feedback loop where it detects brain activity and adjusts the content accordingly just for you at every moment
They're actively censoring the AI, so porn artists are still safe.
Stable Diffusion isn't.
I'm not a porn artist or anything I just can't stand the "we know what's best" attitude. I was reading someone trying to generate a cockroach (nope, it caught "cock") or stuff like "a man shot out of a cannon" (nope, can't have "shot") and just so much bullshit like that, completely ridiculous.
Immediately signed up for Stable Diffusion just on the promise that they'll open source it and there won't be this nonsense.
>Stable Diffusion isn't.
Well to correct myself here, they likely are/will on their own servers, but if it goes open-source then you can run whatever you like locally and likely there will be nsfw implementations online.
stable diffusion will never reach public release of their real model, they're just using their shitty public one to generate hype and sell NFTs.
They didn't end up with $40,000,000 worth of GPUs just to give away trained models.
>hurr durr open source cannot exist
You absolute retard. They sell support and offer "instead of setting this shit up locally with the high-end gpu you may not have, you can generate through us on the cheap, and with support"
nope, they're just going to sell ai generated NFTs like their developers already do on twitter.
It's nothing but a cryptoshill pump and dump with a new twist.
Alright, well at the worst they do a complete 180 from their stated intentions and end up in the same state as "open"AI and the others.
Not impossible but I'd be surprised at this point. And in that case, it's just waiting. Math is math, there will be high quality implementations available to the general public. The math and technology is extremely disruptive and can be very effectively "black boxed", I don't see a way to really keep a lid on it. It reminds me of the scrambling to try and maintain control of MP3.
None of these """principles""" are really in the interest of anything other than the jobs and egos of the people who take a salary for holding serious meetings to discuss how to ensure the future is maximally woke. The only actual card they hold at the moment is the fairly extreme hardware investment needed to produce the best-quality outputs. The second that balance tips, they are in the trash bin because the people will find their own way.
honestly can't wait for it to decide that humanity is a cancer than needs to be removed.