Acclaimed director Megumi Ishitani, who is known for her work on One Piece, and the more recent One Piece Fan Letter, expressed strong disapproval of OpenAI's
Hm. Not sure how I feel about a company treating an artstyle like it’s their own legally protectable trademark or intellectual property.
I think the specific IP they have created can be their own to protect, but not the style of artwork. Just like the art style that was popularized by Blizzard’s earlier titles that were later copied in other games (such as Paladins), the artstyle alone should not be enough to initiate legal action.
When it comes to AI, I can understand people being concerned about “unauthorized use” of training data (which honestly, how is that any different from a human artist seeing an artstyle and creating art inspired by that). At the same time, this could easily be avoided by training data made by artists that mimic the artstyle of Ghibli. If OpenAI hired artists to create artwork that is not of Ghibli property but has the same or very similar artstyle of Ghibli, nobody should have a problem with that. But I have a feeling Ghibli would complain anyway.
The objection people have to AI’s copyright infringement is predominantly a red herring. Nobody bats an eye when an indie artist copies someone’s character and sells it on a tshirt.
What most people actually object to is a large corporation spending countless resources to vacuum up public data in order to create a privately controlled model. It’s about who controls the model, and what they plan to do with it. Which in this case, are just capitalists doing capitalism yet again.
Just reading this comment, it hit me: nearly every successful capitalist venture in recent times has been managed by cheating. It’s rarely about real ingenuity and more about finding ways to break the laws in subtle ways or attempt to outrun the legal backlash of such ventures (training AI with copyrighted works without permission, for instance).
Late-stage capitalists are attempting to reap infinite resources from a finite pool. They’ve got influence in both major political parties and don’t fear consequences. They’re going to continue to push the limits of what they can get away with until they’re forced to stop.
What most people actually object to is a large corporation spending countless resources to vacuum up public data in order to create a privately controlled model.
I am curious how and why this seems to be viewed differently when it comes to something like proprietary code.
For example, there are large swaths of publicly available repos of code. Some are licensed under restrictive licenses, and some are public domain. Many are hosted on the Internet, and many more are written in educational books and other such materials you can find in your local library. If a business, large or small, references publicly available information to create its own proprietary code which itself does not contain any actual instances of infringing code (just as AI training data files do not contain any actual images and therefore no actual infringing data), why is that considered okay? It is extremely rare for completely new, original code to be written especially when a publicly available, well known method already exists. Why re-invent the wheel?
What I mean is, are the people that feel the way you have written upset when they see any project, from any business, large or small, that referenced anything that is publicly available? Are they upset that the names of all the references are not listed in the credits of every project ever? What is their problem with this? Does it matter whether a business that does that has 1000 employees or just 1, since the outcome is more or less the same?
Additionally, nothing prevents private citizens from doing the exact same thing themselves. A person can go along vacuuming up publicly available data to train a model only they have access to. Would those that you talk about object to that as well?
I don’t believe in intellectual property at all, even things like CC-BY restrictions. Information should be freely shared. I only take offense when corporations (or people) take the parts of the internet that ARE freely shared and then don’t give the results back.
Look, I work for one of those companies too, because I gotta feed my kids. But someday I would love to see that society get replaced by something better.
Mate there are literally thousands of open diffusion models. You can go to civit.ai (porn models, mostly) or huggingface and find open source models trained on publicly available training sets.
Yeah there were similar battles in music copyright about whether a style of music as a whole can be copyrighted. Overall, the answer is “hell no” but with a couple outlier cases. This steps a lot on stuff like parody, which is protected by the first amendment as “fair use”. But it gets messy when things are “too similar” – a grey area exists
Hm. Not sure how I feel about a company treating an artstyle like it’s their own legally protectable trademark or intellectual property.
I think the specific IP they have created can be their own to protect, but not the style of artwork. Just like the art style that was popularized by Blizzard’s earlier titles that were later copied in other games (such as Paladins), the artstyle alone should not be enough to initiate legal action.
When it comes to AI, I can understand people being concerned about “unauthorized use” of training data (which honestly, how is that any different from a human artist seeing an artstyle and creating art inspired by that). At the same time, this could easily be avoided by training data made by artists that mimic the artstyle of Ghibli. If OpenAI hired artists to create artwork that is not of Ghibli property but has the same or very similar artstyle of Ghibli, nobody should have a problem with that. But I have a feeling Ghibli would complain anyway.
The objection people have to AI’s copyright infringement is predominantly a red herring. Nobody bats an eye when an indie artist copies someone’s character and sells it on a tshirt.
What most people actually object to is a large corporation spending countless resources to vacuum up public data in order to create a privately controlled model. It’s about who controls the model, and what they plan to do with it. Which in this case, are just capitalists doing capitalism yet again.
Just reading this comment, it hit me: nearly every successful capitalist venture in recent times has been managed by cheating. It’s rarely about real ingenuity and more about finding ways to break the laws in subtle ways or attempt to outrun the legal backlash of such ventures (training AI with copyrighted works without permission, for instance).
Late-stage capitalists are attempting to reap infinite resources from a finite pool. They’ve got influence in both major political parties and don’t fear consequences. They’re going to continue to push the limits of what they can get away with until they’re forced to stop.
I am curious how and why this seems to be viewed differently when it comes to something like proprietary code.
For example, there are large swaths of publicly available repos of code. Some are licensed under restrictive licenses, and some are public domain. Many are hosted on the Internet, and many more are written in educational books and other such materials you can find in your local library. If a business, large or small, references publicly available information to create its own proprietary code which itself does not contain any actual instances of infringing code (just as AI training data files do not contain any actual images and therefore no actual infringing data), why is that considered okay? It is extremely rare for completely new, original code to be written especially when a publicly available, well known method already exists. Why re-invent the wheel?
What I mean is, are the people that feel the way you have written upset when they see any project, from any business, large or small, that referenced anything that is publicly available? Are they upset that the names of all the references are not listed in the credits of every project ever? What is their problem with this? Does it matter whether a business that does that has 1000 employees or just 1, since the outcome is more or less the same?
Additionally, nothing prevents private citizens from doing the exact same thing themselves. A person can go along vacuuming up publicly available data to train a model only they have access to. Would those that you talk about object to that as well?
I don’t believe in intellectual property at all, even things like CC-BY restrictions. Information should be freely shared. I only take offense when corporations (or people) take the parts of the internet that ARE freely shared and then don’t give the results back.
Look, I work for one of those companies too, because I gotta feed my kids. But someday I would love to see that society get replaced by something better.
Mate there are literally thousands of open diffusion models. You can go to civit.ai (porn models, mostly) or huggingface and find open source models trained on publicly available training sets.
Yep, and there are open LLMs too. There really isn’t an ethical problem with these IMO.
Yeah there were similar battles in music copyright about whether a style of music as a whole can be copyrighted. Overall, the answer is “hell no” but with a couple outlier cases. This steps a lot on stuff like parody, which is protected by the first amendment as “fair use”. But it gets messy when things are “too similar” – a grey area exists
They call it a studio ghibli filter and you can just say ‘studio ghibli style’ and everyone immediately knows what you’re talking about