8 ms·

Not even deleted! Actively stolen. No one has yet grappled with the reality that especially in the realm of generative art, these models do not work without a b

Not even deleted! Actively stolen. No one has yet grappled with the reality that especially in the realm of generative art, these models do not work without a basis of absolutely stressful amounts of stolen artwork to train them. They simply do. Not. Exist. Without that baseline unethical contribution that tons of artists made without their knowledge and certainly without their consent. None of the major figures behind any of this tech will acknowledge this. They make some hand-wavy non-committal gestures towards fair use and "how artists learn" and act like that's an answer, completely discarding tons of points not worth revisiting here because it's frankly been beaten to death already and if you're still fence sitting or pro generative art at this point, it's because you want/need to be to prevent cognitive dissonance in one way or another. And to be clear: this stuff is SO COOL. The idea of entering prompts and getting at least vaguely unique and interesting output is incredibly cool. The notion of machines that learn, even in a restrictive way as the current models do, is so fascinating. And of course, like any cool and interesting new tech, it was immediately adopted/hijacked by the worst people imaginable, determined to fill Amazon self-publishing with generated hack kids books filled with generated text and generated art, at the lowest cost point possible, sold on the cheap, with the idea that they could make them convincing enough to trick some poor overworked parents into buying them and pocketing the cash. Just grift. Every new technology we get now, as few as the true new innovations they are, are always, always, immediately, co-opted by the worst actors in the space and turned to utter garbage.

sillysaurusx 3y ago
Oh please. First of all, how do you steal an idea? We’re talking about pictures. Supposing that you buy into the theory that you can, copyright was created to further the arts and sciences; it’s in the US constitution. The point isn’t to control your work — it’s to live in a richer society. And it’s not even clear that training a model counts as infringement. Being able to recite a quote from a book is different than reproducing the entire book. Artists won’t acknowledge that the same applies to their art. If you believe that training models on art is stealing, then I’m a master ninja, since I’m the creator of books3. And even Stephen King today came out and said that he’s fine with it: > Would I forbid the teaching (if that is the word) of my stories to computers? Not even if I could. I might as well be King Canute, forbidding the tide to come in. Or a Luddite trying to stop industrial progress by hammering a steam loom to pieces. https://www.theatlantic.com/books/archive/2023/08/stephen-king-books-ai-writing/675088/ https://www.theatlantic.com/books/archive/2023/08/stephen-ki... If he’s not worried, why are you? I take a dim view of people trying to frame researchers as criminals. We’re not. We want to further science. That’s all. You call me a grifter, but I’ve made roughly a hundred bucks from books3, and that’s because someone found my patron buried under a pile of links and subscribed to it many months ago. Most of my researcher colleagues seem to have similar distaste for wanting to make money. The work is the goal.
- xigoi 3y ago
  If OpenAI believes that copyright is wrong, why are they keeping their software proprietary?
  - sillysaurusx 3y ago
    I agree with you, for what it’s worth. I don’t think it would be a disadvantage for them to open source GPT-4. They’re the ones with the hardware.
  - ben_w 3y ago
    Ostensibly because of the risk of misuse. This may be a losing battle even assuming it's 100% sincere (I'm not interested in debating the assumption).
- thomasahle 3y ago
  > Being able to recite a quote from a book is different than reproducing the entire book. Language models can often reproduce an entire book. And image models like midjourney can reproduce lots of copyrighted art more or less flawlessly: https://ceoln.wordpress.com/2022/12/16/some-light-infringement/ https://ceoln.wordpress.com/2022/12/16/some-light-infringeme...
  - andrewfurey2003 3y ago
    That doesnt make training the models wrong, just using those specific outputs wrong.
  - renewiltord 3y ago
    It's possible with Google Docs too. The whackiest thing the other day was that I was using the screenshot tool in MacOS and it made a screenshot of copyrighted content. I was flabbergasted. The tool just did it with 4 parameters.
    - callalex 3y ago
      And if you use those reproductions commercially it is almost universally illegal already.
puchatek 3y ago
Reminds me a bit of how everybody was cool with pirating software and media back in the day. To help the bad conscience there was a handy narrative of how the record companies were making a killing off of everybody so it was cool not to pay. But in reality it was just technically feasible. That's what enabled the behavior, not a few million Robin hoods coming together to do the right thing. Similarly I feel creators need to make it impossible to have their work stolen for AI if possible. It will be tough though. They have half the world against them it seems.
- Gigachad 3y ago
  It’s too late. The datasets that exist already are already more than enough to train future models.
- midasz 3y ago
  I wouldn't equate it with a person pirating a DVD to watch in his own home. It's more like someone pirating a DVD, burning a bunch of copies of it and selling those. Big difference.
stelonix 3y ago
I believe if you look further and deeper you'll reach the conclusion the real issue is how awful copyright laws are and more importantly, how absurd is the current economic system. Glorified markov chains are not the culprit here. So the current movement where artists "shame" random joes for using a cool technology has only one possible outcome which is to push said average joes into being politically active in defense of AI. No one seems to want to properly organize and file a class-action, just twitter bickering.
simonjgreen 3y ago
How does a Human generate their ideas, styles, etc? We absorb everything we come in to contact with too.
- mattmcknight 3y ago
  Exactly. Preventing the use of some information for training data is like saying we can't go to the gallery or library to see what prior art looks like. Creativity is just having a higher temperature setting and selecting items that trigger a positive response. We are mimetic. The art world itself has already asked many of these questions 60 years ago, when Warhol made Campbell's soup can art, and then people used that template to turn other brands into art.
- ToucanLoucan 3y ago
  A human doesn't exist in servers, work for free (well, minus utilities) 24 hours per day and work as long as you want on a project you've given it with zero input. If you can't appreciate the difference between an AI model and a person I think that says more about you than AI.
  - ben_w 3y ago
    None of those differences really says anything about morality of having an AI do things, unless you want to argue that the AI suffer from their (mis)treatment. People sometimes state that The Simpsons is now drawn in sweatshops somewhere in Asia. Assuming that's true, or at least that it's happened at least once for at least some cartoon, does it make any difference at all to the question of where the people directing them get their ideas from or how those workers learned to do their thing?
    - ToucanLoucan 3y ago
      > None of those differences really says anything about morality of having an AI do things, unless you want to argue that the AI suffer from their (mis)treatment. I mean, if it was actually AI and not machine learning, that would certainly be a question wouldn't it? That being said, I'm not saying the machine is suffering. I'm saying the people it's replacing cannot possibly hold their own. There is simply no way in hell a person can compete with generative AI, assuming of course you don't need perfection just "good enough." And given how massive components of the modern economy get by on far less than "good enough" I'd say that's a solid reason to be concerned. > People sometimes state that The Simpsons is now drawn in sweatshops somewhere in Asia. Assuming that's true, or at least that it's happened at least once for at least some cartoon, does it make any difference at all to the question of where the people directing them get their ideas from or how those workers learned to do their thing? Collaborative creative work among a team of individuals regardless of their location is not the same thing as generative art trained on a dataset it's designed to mimic, come on now. You're grasping at some pretty flimsy straws here.
ben_w 3y ago
> No one has yet grappled with the reality that especially in the realm of generative art, these models do not work without a basis of absolutely stressful amounts of stolen artwork to train them. They simply do. Not. Exist. Without that baseline unethical contribution that tons of artists made without their knowledge and certainly without their consent. None of the major figures behind any of this tech will acknowledge this. Well that's not correct. Don't get me wrong, I've seen the statements you're probably referring to; but there's also Adobe with one made from licensed and public domain images only. (You may want to argue, like my literally Communist ex, that under capitalism there can be no such thing as a fair contract between a corporation and a worker; I'm unsure of the power dynamics so won't counter-argue that).
- CatWChainsaw 3y ago
  Which part?

sillysaurusx 3y ago

Oh please. First of all, how do you steal an idea? We’re talking about pictures. Supposing that you buy into the theory that you can, copyright was created to further the arts and sciences; it’s in the US constitution. The point isn’t to control your work — it’s to live in a richer society. And it’s not even clear that training a model counts as infringement. Being able to recite a quote from a book is different than reproducing the entire book. Artists won’t acknowledge that the same applies to their art. If you believe that training models on art is stealing, then I’m a master ninja, since I’m the creator of books3. And even Stephen King today came out and said that he’s fine with it: > Would I forbid the teaching (if that is the word) of my stories to computers? Not even if I could. I might as well be King Canute, forbidding the tide to come in. Or a Luddite trying to stop industrial progress by hammering a steam loom to pieces. https://www.theatlantic.com/books/archive/2023/08/stephen-king-books-ai-writing/675088/ https://www.theatlantic.com/books/archive/2023/08/stephen-ki... If he’s not worried, why are you? I take a dim view of people trying to frame researchers as criminals. We’re not. We want to further science. That’s all. You call me a grifter, but I’ve made roughly a hundred bucks from books3, and that’s because someone found my patron buried under a pile of links and subscribed to it many months ago. Most of my researcher colleagues seem to have similar distaste for wanting to make money. The work is the goal.

xigoi 3y ago
If OpenAI believes that copyright is wrong, why are they keeping their software proprietary?
- sillysaurusx 3y ago
  I agree with you, for what it’s worth. I don’t think it would be a disadvantage for them to open source GPT-4. They’re the ones with the hardware.
- ben_w 3y ago
  Ostensibly because of the risk of misuse. This may be a losing battle even assuming it's 100% sincere (I'm not interested in debating the assumption).
thomasahle 3y ago
> Being able to recite a quote from a book is different than reproducing the entire book. Language models can often reproduce an entire book. And image models like midjourney can reproduce lots of copyrighted art more or less flawlessly: https://ceoln.wordpress.com/2022/12/16/some-light-infringement/ https://ceoln.wordpress.com/2022/12/16/some-light-infringeme...
- andrewfurey2003 3y ago
  That doesnt make training the models wrong, just using those specific outputs wrong.
- renewiltord 3y ago
  It's possible with Google Docs too. The whackiest thing the other day was that I was using the screenshot tool in MacOS and it made a screenshot of copyrighted content. I was flabbergasted. The tool just did it with 4 parameters.
  - callalex 3y ago
    And if you use those reproductions commercially it is almost universally illegal already.

puchatek 3y ago

Reminds me a bit of how everybody was cool with pirating software and media back in the day. To help the bad conscience there was a handy narrative of how the record companies were making a killing off of everybody so it was cool not to pay. But in reality it was just technically feasible. That's what enabled the behavior, not a few million Robin hoods coming together to do the right thing. Similarly I feel creators need to make it impossible to have their work stolen for AI if possible. It will be tough though. They have half the world against them it seems.

Gigachad 3y ago
It’s too late. The datasets that exist already are already more than enough to train future models.
midasz 3y ago
I wouldn't equate it with a person pirating a DVD to watch in his own home. It's more like someone pirating a DVD, burning a bunch of copies of it and selling those. Big difference.

stelonix 3y ago

I believe if you look further and deeper you'll reach the conclusion the real issue is how awful copyright laws are and more importantly, how absurd is the current economic system. Glorified markov chains are not the culprit here. So the current movement where artists "shame" random joes for using a cool technology has only one possible outcome which is to push said average joes into being politically active in defense of AI. No one seems to want to properly organize and file a class-action, just twitter bickering.

simonjgreen 3y ago

How does a Human generate their ideas, styles, etc? We absorb everything we come in to contact with too.

mattmcknight 3y ago
Exactly. Preventing the use of some information for training data is like saying we can't go to the gallery or library to see what prior art looks like. Creativity is just having a higher temperature setting and selecting items that trigger a positive response. We are mimetic. The art world itself has already asked many of these questions 60 years ago, when Warhol made Campbell's soup can art, and then people used that template to turn other brands into art.
ToucanLoucan 3y ago
A human doesn't exist in servers, work for free (well, minus utilities) 24 hours per day and work as long as you want on a project you've given it with zero input. If you can't appreciate the difference between an AI model and a person I think that says more about you than AI.
- ben_w 3y ago
  None of those differences really says anything about morality of having an AI do things, unless you want to argue that the AI suffer from their (mis)treatment. People sometimes state that The Simpsons is now drawn in sweatshops somewhere in Asia. Assuming that's true, or at least that it's happened at least once for at least some cartoon, does it make any difference at all to the question of where the people directing them get their ideas from or how those workers learned to do their thing?
  - ToucanLoucan 3y ago
    > None of those differences really says anything about morality of having an AI do things, unless you want to argue that the AI suffer from their (mis)treatment. I mean, if it was actually AI and not machine learning, that would certainly be a question wouldn't it? That being said, I'm not saying the machine is suffering. I'm saying the people it's replacing cannot possibly hold their own. There is simply no way in hell a person can compete with generative AI, assuming of course you don't need perfection just "good enough." And given how massive components of the modern economy get by on far less than "good enough" I'd say that's a solid reason to be concerned. > People sometimes state that The Simpsons is now drawn in sweatshops somewhere in Asia. Assuming that's true, or at least that it's happened at least once for at least some cartoon, does it make any difference at all to the question of where the people directing them get their ideas from or how those workers learned to do their thing? Collaborative creative work among a team of individuals regardless of their location is not the same thing as generative art trained on a dataset it's designed to mimic, come on now. You're grasping at some pretty flimsy straws here.

ben_w 3y ago

> No one has yet grappled with the reality that especially in the realm of generative art, these models do not work without a basis of absolutely stressful amounts of stolen artwork to train them. They simply do. Not. Exist. Without that baseline unethical contribution that tons of artists made without their knowledge and certainly without their consent. None of the major figures behind any of this tech will acknowledge this. Well that's not correct. Don't get me wrong, I've seen the statements you're probably referring to; but there's also Adobe with one made from licensed and public domain images only. (You may want to argue, like my literally Communist ex, that under capitalism there can be no such thing as a fair contract between a corporation and a worker; I'm unsure of the power dynamics so won't counter-argue that).

CatWChainsaw 3y ago
Which part?