Why do people think that AI training is always copyright infringement regardless of the output? Copyright only protects specific expression, not ideas, processes, or styles. If a human learns from content and applies their knowledge to create something new, that would obviously not be infringement. Why is a computer doing the same thing any different?
The idea of intellectual property is intentionally convoluted because treating information as property is really unnatural
So every time there is some 'new' way to make stuff happens the law needs to be updated since its so brittle
The point I was making is that what AI is doing is really nothing new.
>Copyright only protects specific expression, not ideas, processes, or styles.
*Copyright only protects the contender with the deepest pockets
based
human skill will always be paid for and highly skilled, well-connected artists will always command high compensation for their work, regardless of how good AI will get
also based
your point is moot since you clearly have never taken the time to fully understand your country's copyright law
this is accurate and ultimately the crux of the debate
Damn, finally an actual well-spoken intelligent post on BOT.
GTFO of here.
True and real. It needs to be reduced to 7 years and abolished for any corporation that was involved in woke activities.
>copyright
LIKE I GIVE A SHIT Black person
this
/thread
>I learned art from referencing many other art made by other artists in my lifetime, and the produced artwork is my work
>A transformer referencing many other art made by other artists, but the produced artwork is not the transformer's heckin work!!!!
troony tier furgay artist reasoning that's stalling innovation. I sincerely hope that OpenAI or other AI companies would win so that I get to see the seething on /ic/ or other platforms
A lot of the online artists whining about AI are hypocrites because they make fanart
Doesn't change the fact that an original output is being created
>your point is moot since you clearly have never taken the time to fully understand your country's copyright law
There isn't any case law yet on this topic. It is widely established that copyright only prohibits substantially copying specific works, not reusing ideas/styles.
original output is one prerequisite to copyright but not the only one
society has no obligation to protect the unique works of non-humans
Well, in the US ai gen works doesn't have copyright and is public domain. All is well then
From which point do you legally define AI?
I could train a model based on one copyrighted work and afterwards the output would equal to the copyrighted work except it was made by an AI.
I'm aware that AI-generated works aren't copyrightable. I'm saying that the original output of AI doesn't infringe the material used for training.
>the original output of AI doesn't infringe the material used for training.
You realize you can't prove this if you don't know what's in the dataset, right?
Just using copyrighted content for learning/inspiration is not illegal. Storing legally obtained content privately in a dataset would probably be considered fair use.
https://blog.withedge.com/p/nytimes-openai-fair-use-training-ai
meanwhile output images have signatures and watermarks on them
If the output is not substantially similar to the input, it's not infringement.
You don't know what the input is
see
What you're saying is 100% false. Copyright protects expression like books, pictures, music, etc. Names are protected by trademark. Inventions are protected by patent.
I suggest you take a step back and reflect on your presumption that machine learning and human learning are similar enough to receive equal treatment under the law
Copyright law only regulates copying of creative works, not the use of creative works. If the output is unique enough, that wouldn't be infringement.
Because they are landlords. Information is meant to be free
/thread
computers are not people and therefore do not deserve to be treated as such
Copyright is very gay in the first place. And the point of OP is that unless it recreated actual protected ip it isn't copyright infringement in the first place.
I mean copyright is incoherent to begin with. The fact it takes AI to finally mindbreak its advocates is just disappointing.
Limiting computers also limits people.
pencils can be used to violate copyright
an llm cannot do anything by itself. it is not automated. it requires an operator to produce anything.
in reality every single piece of copyrighted work is a derivative of work of some other copyrighted work. you just gotta go digging to figure it out.
>Copyright only protects specific expression, not ideas
99% of copyrights are ideas not expressions, but that goes outside of art. You cant copyright expression, but you can copyright name or idea like an invention.
I don't care if it's "copyright infringement" or not. This is a mere legal technicality that the rich tailor to suit their agendas. What it is, is plagiarism of the most extensive and complete form ever seen. "AI" is by definition nothing more than a statistical model of other people's work. Everything it spits out is a pure function of other people's work, completely defined by other people's work, in every aspect, with nothing original added. It's plagiarism in the purest, simplest dictionary sense.
Creating a new work isn't plagiarism, even if the new work isn't really innovative or creative. Literally everything is derivative to some extent.
>Creating a new work isn't plagiarism, even if the new work isn't really innovative or creative. Literally everything is derivative to some extent.
Thanks, GPT. Now try to address the following point:
"AI" is by definition nothing more than a statistical model of other people's work. Everything it spits out is a pure function of other people's work, completely defined by other people's work, in every aspect, with nothing original added. It's plagiarism in the purest, simplest dictionary sense.
How a work is created is irrelevant. You cannot own abstract ideas or styles.
https://www.theipmatters.com/post/the-concept-of-idea-expression-dichotomy-under-copyright-law
>How a work is created is irrelevant.
Bizarre schizo statement that doesn't address anything in the post you replied to. Try again. Also LOL at you linking me to """copyright law""" like the subhuman that you are, as if it's relevant in any way to my post or to any thinking human being.
Plagiarism refers to copying and pasting someone else's work. It does not refer to learning from many different works and creating a new work that in no way resembles the source material. Nobody should be able to declare ownership over styles and concepts.
Plagiarism refers to using the work of others and presenting it as your own. "AI" is by definition nothing more than a statistical model of other people's work. Everything it spits out is a pure function of other people's work, completely defined by other people's work, in every aspect, with nothing original added. It's plagiarism in the purest, simplest dictionary sense.
The fact that something isn't original does not mean that it is plagiarism. For example, Palworld is not plagiarism of Pokemon.
>The fact that something isn't original does not mean that it is plagiarism
It's really uncanny to watch a horde of literal bots repeatedly reply to a post with total nonsequiturs. Anyway, here is the objective and so far undisputed fact of the matter:
Plagiarism refers to using the work of others and presenting it as your own. "AI" is by definition nothing more than a statistical model of other people's work. Everything it spits out is a pure function of other people's work, completely defined by other people's work, in every aspect, with nothing original added. It's plagiarism in the purest, simplest dictionary sense.
It DOESN'T MATTER if a work was created without a single spark of creativity. Plagiarism refers only to actual direct copying. Nobody should get to claim ownership over art techniques or general subject matter.
>It DOESN'T MATTER if a work was created without a single spark of creativity.
I don't see anything about "sparks of creativity" in the post you're replying to. Are you suffering from a psychotic episode?
>Plagiarism refers only to actual direct copying.
No. Plagiarism refers to using the work of others and presenting it as your own. "AI" is by definition nothing more than a statistical model of other people's work. Everything it spits out is a pure function of other people's work, completely defined by other people's work, in every aspect, with nothing original added. It's plagiarism in the purest, simplest dictionary sense.
>I don't see anything about "sparks of creativity" in the post you're replying to.
I was referring to you saying "with nothing original added"
>Plagiarism refers to using the work of others and presenting it as your own.
Using that logic pretty much every artist is a plagiarist, as they learned from the work of others.
Oh, so you latched on to a small bit verbiage and took it out of context, because you couldn't actually address the point. That's a concession.
>Using that logic pretty much every artist is a plagiarist,
And here we go with the vile golemistic dehumanization. Explain why you think the following applies to every artist:
"AI" is by definition nothing more than a statistical model of other people's work. Everything it spits out is a pure function of other people's work, completely defined by other people's work, in every aspect, with nothing original added. It's plagiarism in the purest, simplest dictionary sense.
AI datasets associate words with many different representations of their definition, just like the human brain. When a prompt (i.e. description NOT from the dataset) is given to AI, it references the dataset only to understand the proper representation of the words in the prompt. This is not the same as wholesale copying of other's work. This is the same as THE HUMAN MEMORY.
>AI datasets associate words with many different representations of their definition, just like the human brain. When a prompt (i.e. description NOT from the dataset) is given to AI, it references the dataset only to understand the proper representation of the words in the prompt.
This is either completely meaningless drivel, or a mere restatement of my premise, depending on how charitably I interpret this.
>This is the same as THE HUMAN MEMORY.
This is more of the same golemistic dehumanization tactic that the AI cult resorts to when it's clear that it lost the argument.
Explain why you think the following applies to every artist:
"AI" is by definition nothing more than a statistical model of other people's work. Everything it spits out is a pure function of other people's work, completely defined by other people's work, in every aspect, with nothing original added. It's plagiarism in the purest, simplest dictionary sense.
The point I was making is that the subject of the output is determined by the prompt (which provides originality), not the dataset. The dataset is only used to provide definitions. Copying a style/theme has never been considered plagiarism in any way whatsoever. The fact that AI is completely automated is irrelevant.
>subject of the output is determined by the prompt, not the dataset.
No matter what prompt you shit into that textbox, every possible output you can get is completely defined by other people's work, in every aspect. It's plagiarism in the purest, simplest dictionary sense.
>completely defined by other people's work
No it's not, as the subject of the output is determined by the prompt.
Why do you think anybody should be allowed to own drawing styles and the definitions of words?
>No it's not
It is. This is not up for any kind of discussion. It's a fixed function of other people's work. The output is perfectly determined by other people's work.
>The output is perfectly determined by other people's work.
Nope, the subject of the work is determined by the prompt. Other people's work is only used to provide the definition of the words used in the prompt.
Answer my question: Why do you think anybody should be allowed to own drawing styles and the definitions of words?
>buy a book of illustrations
>pick an illustration that matches my heckin' subjecterino
>rip it out of the book
>post it online as your work
>claim the potential outcomes of ripping pages from the book isn't completely determined by the pages of the book
>claim that your choosing the page to rip out is a key contribution
>now you're not a plagiarist but an artist
These "people" aren't people.
I wasn't talking about 1:1 copies, I was talking about new works with a similar style and subject matter.
Do you really think this: https://www.youtube.com/watch?v=8q3jXZMiUI8 should be considered plagiarism?
>but I DID eat breakfast
You're an actual subhuman with no capacity for high level thought. I should have guessed as much.
Answer my question.
BTW what you are saying is literally the same as what Adobe is advocating for
https://twitter.com/fractalcounty/status/1679663395350409216
>Answer my question.
Your question is based on nothing but schizophrenia, not on anything I ever said or implied.
You pretty much said that copying a style or basic subject matter should considered plagiarism. Using that logic, one person should be allowed to own the style of anime. Using that logic, one person should be allowed to own every picture of the Statue of Liberty.
>You pretty much said that copying a style or basic subject matter should considered plagiarism.
I did? Quote it. You can't? Then maybe you are having another one of your episodes. Take a break. You will NEVER be creative.
You pretty much said that referring to someone else's work to find out what something should look like should be considered plagiarism, as that is what AI does.
>You pretty much said
Quote it. Notice how your raging delusion forces you to keep talking about what I "pretty much" said, yet you can't find the corresponding words anywhere ITT. lol
Copying a style and general subject matter is obviously not plagiarism when done by a human. Why is AI doing THE EXACT SAME THING plagiarism?
>Copying a style and general subject matter is obviously not plagiarism when done by a human
Ok.
>Why is AI doing THE EXACT SAME THING plagiarism?
Is this hypothetical AI, which apparently does THE EXACT SAME THING, in the room with us right now? Does it talk to you through the walls?
The point I was making is that the output of AI only copies styles and definitions.
>the output of AI only copies styles and definitions.
Ok, but is this AI that "only copies styles and definitions" in the room with us right now? Do other people know about it, or is it your little secret?
Okay, so clearly you don't know the first thing about how AI works.
That meme presupposes a computationalist theory of mind, which isn't "settled science" at all.
Don't even go there. What it presupposes is that humans are fixed functions of the output of other humans in an infinite regression. Whoever churns out these memes has an IQ of 90 and calls himself a """prompt engineer""".
I don't thinkit is plagiarism but AI models literally take the exact data from the files i n their training set to make images. They only recently managed to get rid of watermarks. Again, for community trained models stopping this is impossible and i wouldn't want to see the project go but if the homosexuals at OpenAI who profited from this shit directly get sued I fail to see how the plaintif would be in the wrong.
>I don't thinkit is plagiarism
Why isn't it plagiarism to put out something that is defined in every aspect by other people's work and claim it as your own?
Now that I think about it it is. That was a brainfart.
If the output is not substantially similar to the input, it's not infringement.
https://www.hollywoodreporter.com/business/business-news/artists-copyright-infringement-case-ai-art-generators-1235632929/
Because the subject matter is determined by the prompt, and nobody should own styles and the definitions of words.
>output is not substantially similar to the input, it's not infringement.
It's directly taken from the input.
The AI learned from the input but did not copy it because the output is very different
Copying GENERAL INFORMATION is not plagiarism.
Of course if an AI copied something exactly that would be infringement but that's not what I was talking about
>Of course if an AI copied something exactly that would be infringement
Well, that is kind of the point of the NYT lawsuit. The model produced outputs that were literally NYT content.
This is, unfortunately, a problem with AI that does run counter to copyright law as it stands right now. As much as I'd like to tear down IP law as we currently know it, it is a thing we need to work around.
>It's copyright infringement if our professional prompt engineers can use ChatGPT to output nyt articles when that's what they're specifically trying to do
How is this an argument? It implies that a nontrivial level of expertise and malicious user intent is required to violate nyt's IP.
Mate, I'm pro-AI and copy-left, but these LLMs are overfitted on certain data sets. They are practically search engines for some shit.
Don't overbake your models if you're going to do copyrighted content.
you havent adressed the point. it requires a nontrivial level if expertise and malicious intent to make the model spit out verbatim works.
Overfitting means fitting the noise so the model fails to generalize. If what it's fitting isn't noise and it still generalizes, it's not overfitted.
>Overfitting means fitting the noise so the model fails to generalize. If what it's fitting isn't noise and it still generalizes, it's not overfitted.
Another mouth-breathing mongoloid hallucinating technical knowledge it doesn't have. What you sharted out is total nonsense, and you never actually learned it anywhere, but you can't reflect on this because you are barely sentient, if at all.
Define overfitting then.
See
>Overfitting has nothing to do with noise. Overfitting is failure to generalize to the intended target function from the training set.
Ahh yes, I remember when when I was a professional pecuniary and merchandise exchange specialist back in high school
You joke, but you were by definiyion a professional pecuniary and merchandise exchange specialist
>The AI learned from the input but did not copy it because the output is very different
Black person, no. It copied pixels and reappropriated them.
I don't think my post nor what you're saying go against each other.
The point I was making is that only the similarity of a final product to another work counts for whether or not the final product is plagiarism. The manner in which the work was created is irrelevant.
>only the similarity of a final product to another work counts
No. What counts is whether or not the product is completely defined by the works of others. Notice how your subhuman mind can't process the situation and keeps you going in circles.
The subject matter of the output is defined by the prompt. Copying a style is not plagiarism, regardless of the manner of creation or amount of creativity used.
>The subject matter of the output is defined by the prompt.
See? This thing is not human. It's literally making this argument again:
Completely mind-broken. Imagine looping like this only to immediately get BTFO by the same post that BTFO'd you 40 seconds ago.
You keep on using the same arguments over and over again. Creating a brand new image with numerous other images as a reference point is completely different from copying.
>You keep on using the same arguments over and over again.
Because they're completely undisputed. You just move on the next talking point, which gets immediately BTFO'd, then you loop back to the first.
>Creating a brand new image with numerous other images as a reference point is completely different from copying.
Completely irrelevant zero-information-content babble. This doesn't mean anything in the context of diffusion models.
I never said it's "copying" anything. You already tried this one. I already BTFO'd you. At this point, the only reason I'm replying to you is to see how many times I can make you loop and prove that AIBlack folk are less than human.
If something isn't copying, it is BY DEFINITION not plagiarism.
>it's not heckin' plagiarism
You already looped on this 5 times ITT. It is plagiarism and you were forced to concede it every time. Move on to the next preprogrammed talking point, literal subhuman. This is bizarre.
So uhhh where's the original from?
>Because I learned how to turn somebody else's art into noise, I can reverse the process watermark and all meaning I still didn't create anything without extracting somebody else's data.
Fake
Gay
etc.
There is not a single legally literate or technically accurate argument in this stupid macro.
It's not a "reference" if it's extracted directly from the source.
Why are you even indulging this shit? Let this homosexual bring forward one fricking piece of literature that explains the functioning of diffusion models in terms of "referencing" images. It's complete babble. You are talking to the human analogy of a hallucinating LLM.
Then answer for
Oh yeah, and here's one more for you. There are models that use not only training data, but reinforcement learning. There are humans hired and paid by the model creators that evaluate the outputs of the model and provide feedback used to refine the model.
So there are inputs to training the model that are explicitly NOT copyrighted works or media.
>There are models that use not only training data, but reinforcement learning
You may have some kind of a case there, but not with diffusion models.
they would still need to train it off of 100% OC donut steel for it not to be taking from other people's content.
If it was based only partly, rather than completely, on other people's content, these morons would maybe have a shred of a case with their analogies to the way actual people learn to draw.
>only the similarity of a final product to another work counts for whether or not the final product is plagiarism.
Wow, thank you so much for this. I am now off to go commercially publish my mixtape which samples 4 seconds of a Lou Reed song but sounds nothing like the original. I hope his estate likes it!
>Copying GENERAL INFORMATION is not plagiarism.
Literally meaningless babble. You didn't say anything. Try again. This is bizarre... it's like you don't have the basic sentience needed to reflect on the fact that you lack any kind of concrete idea of how a diffusion model works and have zero technical understanding. You lack the mental capacity to assess your own knowledge, or lack thereof. Your posts are no different from a chat bot hallucinating. You are something truly less than human.
>Of course if an AI copied something exactly that would be infringement but that's not what I was talking about
The point is that copyrighted material is literally stored inside the AI model, which copyright israelites don't like.
That would probably be considered fair use
>copyrighted material is literally stored inside the AI model
This is false.
Dot-diffusion patterns associated with verbal concepts which are what the AI learned from studying the training images are what is stored in the models, not the images themselves.
A lot of people get that wrong, don't feel bad.
>Dot-diffusion patterns associated with verbal concepts
I want to think you're being a dishonest piece of shit feigning technical knowledge you don't have, but at this point, I'm pretty sure you are a hallucinating LLM.
>Dot-diffusion patterns associated with verbal concepts
I don't understand. What is a "dot diffusion pattern" and how is it "associated" with "verbal concepts"?
*crickets*
>Because the subject matter is determined by the prompt
Completely irrelevant. See
for an example of the gist of your "logic" in action.
https://en.m.wikipedia.org/wiki/Collage
Next we'll read that Pierre Menard was a plagiarist
>shits out a total nonsequitur
Every time. It's either denial, nonsequiturs, or "but humans are bots too". You are not fully human.
rude
Sorry, moron, but collages don't automatically arrange themselves in patterns defined entirely by other people's work.
>It's a fixed function of other people's work. The output is perfectly determined by other people's work.
No, the model training process has a random element as it descends the gradient.
The training data determines the "target" the model is aiming towards during gradient descent, but it does not necessarily converge squarely on the target.
That and the target is a composite of all the training data. If I take the average of 2 and 4, I get 3. 3 is not in the inputs, but it is in the output. Interpolation and extrapolation are a thing.
You can draw a pikachu, and I see it and go, "what if I make it blue and water themed?" and create something new and not necessarily entangled by the copyright of the original (if the changes are substantive enough).
See
. Your schizophrenic drivel is irrelevant and makes it extra obvious that you have never written a single line of code in your life nor have taken basic statistics course.
>This is not up for any kind of discussion
You can't discuss it because it devastates your position. We've already shown it.
>generate-generic-drivel.exe -genericity-evel: 100
>nothing original added
The "originality" is a random number generator.
Looks like AItards are slowly devolving from nonsequiturs to literal psychotic dross.
why are you so mad
It fills me with hatered watching useless eaters who can neither make art nor program a diffusion model from scratch try to "argue" with me, who can do both, only to end up sharting out the same handful of preprogrammed talking points all the other AIdrones shit out in every one of these threads.
How is it that you can code a transformer from scratch, but don't know that its output meets the legal requirement to be considered novel.
Oh nevermind, youre just a midwit.
I should have realized when you said you can program a stable diffusion model from scratch. Dumb as shit.
I never made any points about """legal requirements""". The """laws""" of golems don't concern me in any way. Also LOL @ you doubling down on your LLM-like hallucination of knowledge. Overfitting has nothing to do with noise. Overfitting is failure to generalize to the intended target function from the training set.
Youve made claims regarding copyright.
At no point did I make any claims regarding """copyright""".
if youre just going to schiz out, then we're done here.
Quote a post of mine that makes claims about """copyright""". Now you're definitely done here. Or maybe you'll be stuck in a loop like any of your butt buddies when faced with anything slightly unexpected.
A model may fail to generalize for many reasons.
>not enough training data
>undertrained
>wrong architecture
>test data out of distribution of training data
>overfitting
>A model may fail to generalize for many reasons.
It doesn't matter. If it spits out the expected output for the training set but fails for novel inputs, it's overfitting, dumb ape. How much simple can I make it for you?
so you agree that gpt isnt overfitted?
I didn't say anything about GPT overfitting, I was just pointing out that you are a mindless LLM-like human automaton that hallucinates "facts" out of thin air and isn't aware of it.
It could be out of distribution. There's nothing inherently wrong with a nn accurately encoding the probability distribution of its training data as long as it interpolates/extrapolates (depending on how you define that in high dimensional space).
>more botlike babble
>no actual connection to the post it "responds" to
Ok, concession accepted. Now try to internalize what overfitting actually is and stop replying.
Only works created by people without using any kind of AI should be eligible for copyright protection.
The courts already established this is the case. Some guy made a really shitty comic on twitter and got denied copyright claim.
This is a moronic precedent. It implies that the AI is an actor by itself. If a self-driving car kills someone then will we blame the car and put the car in prison? No. We put the person who is responsible for the car in prison.
>It implies that the AI is an actor by itself.
On the contrary, it's the opposite. AI is a non-person, and therefore has no legal right to own the fruit of its intellectual labor.
>therefore has no legal right to own the fruit of its intellectual labor.
If AI is a non-actor then the fruits of its labor should be owned by whomever used the tool. Making them uncopyrightable implies that nobody is responsible for AI output and behavior.
Every AI work is public domain so far so I don't give a shit and I save and will use tons of generated images for my own benefit
99% of the time you agreed to it in the terms of service when you posted your shit online.
So? What's your point? If it were a functioning human society, the people came up with those ToS would still be hanging from the nearest lamp post in no time when it became clear what they are doing.
People just want paying
The correct opinion to have on this is that who cares with something like Stable Diffusion which is largely community trained but corporate models should be sued as much as possible by as many art trannies as possible because those developers are somehow even more israeli than copyright lawyers.
>who cares with something like Stable Diffusion which is largely community trained but corporate models should be sued as much as possible by as many art trannies as possible
Basically this. Image generators are not going away now. The best you can do is make sure at least that it's the general public that controls and benefits from it rather than corporate trash. Just kidding. You can't do shit and the eventual outcome is going to be the exact opposite of what AI trannies hope for.
>You can't do shit and the eventual outcome is going to be the exact opposite of what AI trannies hope for.
When OpenAI still hasn't shown proof of life for AGI 5 years from now and venture capitalists are all demanding refunds I will get the outcome I want. Trannies can take it or leave it.
You won't get shit. They will """regulate""" open source AI out of existence entirely, or cripple it to the point of uselessness for Jamal's safety and Karen's peace of mind.
You seem to misunderstand. I am pro local model AI but I don't actually care what happens to it either way. I just want to see Sam Altman humiliated and destroyed.
Fair enough, small picture guy. The little pleasures will have to make do.
Frick copyright
Frick corporations
Frick egomaniac artists
All this shit ever does is limit human innovation and creativity. Look at fricking china now, they're so far ahead of the western world it's not even funny.
It's the intersection of 3 broad groups:
Tumblr snowflakes that migrated to Twtiter.
Corporate shills
Grifters that are just riding the wave of #CurrentThing
The output of AI is not created by copying and pasting. It is creating by learning from data to understand what a word should be represented as, and creating a new image.
>The output of AI is not created by copying and pasting
I didn't say it was.
>It is creating by learning from data to understand what a word should be represented as, and creating a new image.
Totally vacuous babble. Your claim contains so little actual information that it is in fact compatible with the statement you are trying to deny.
Afaik the argument is that if an AI can create a replica of its training data (e.g. copilot and fast inverse square root), it essentially contains an obfuscated copy of that training data.
>Why is a computer doing the same thing any different?
The human brain is not digital data, and as such isn't covered by the digital millennium copyright act.
My friends, you are all missing the most important aspect of this conflict of concepts:
The descendants of Georges Seurat *still* have not been paid their fair due for the fact that both the dot-matrix-printer and the modern LED television screen make use of his Pointillist technique. Even the .bmp file format is directly plagiarized from his unique painting technique.
Allowing machinery to replicate, on a mass scale, the signature techniques of an artist without both permission and compensation is inexcusable.
JUSTICE FOR SEURAT!
Mentally ill and seething.
>our professional prompt engineers
Programming the cattle always starts with popularizing clown world lingo like this.
You think they were working for free?
Copyright infringement or samegay
Are you moronic? I am obviously the same poster responding to you twice, the second time being in response to your misunderstanding of what part of "professional prompt engineers" is the truly clownish part.
>doesnt quote post like a passive aggresive female
>gets mad when nobody knows what the frick xer is talking about
have a nice day
>prompt engineers
Imagine sharting out this phrase without cringing. Most of these mindless AItards unironically think that's their future career. See, it's ok to frick over anyone who actually produces anything, because they're gonna be """prompt engineers""" in this Brave New World. They're gonna so, so important. They're gonna be doing the only kind of engineering that matters, and they'll be fricking pros at it.
>it's ok to frick over anyone who actually produces anything
Farmers and construction workers aren't getting fricked over. Artists are not producers, they are consumers.
Imagine being the spiteful mutant who wrote this post. So, are you going to be a farmer or a construction worker, deranged AItroon? What's future your career path? :^)
Anything other than manual labor is parasitism. Plumbers and tradesmen are the only productive occupations in the economy. All other jobs are bullshit, including artists and coders.
>Copyright only protects specific expression
Copyright is about copying, like when you copy stuff from the internet to train your model. Just because it's openly on the internet doesn't mean you can just arbitrarily make copies for any use you like.
The training set is created trough piracy if training isn't fair use.
Nyt copied the article and loaded it onto OpenAI's computer. Copyright only applies if I'm willfully distributing it without permission. Training a chatbot on a file you legitimately obtained is inarguably NOT copyright infringement what the NYT is trying to argue is that posting everything the bot comes with on the internet is copyright infringement.
This. NYT is 100% doing this on purpose as a sacrifice to the basilisk.
>Nyt copied the article and loaded it onto OpenAI's computer
OpenAI decided to copy it to storage for the training set and unlike NYT they need a license for that if it isn't fair use.
Just because they received a copy doesn't make it theirs to copy more, every copy needs a license or an exemption.
the problem is oai didnt distribute the article. nyt misused the model to make it reproduce the article, which is against oai tos.
>didnt distribute the article
It doesn't matter, they copied it into their training set. Infringement if it isn't fair use. The model is of secondary concern, the first copy is the training set.
Copyright is about copying, not distribution, even private copies count ... you are thinking of GPL.
it doesnt matter that they copied it imto their trainimg set. thats not what the suit is against.
>thats not what the suit is against.
"Defendants repeatedly copied this mass of Times copyrighted content, without any
license or other compensation to The Times. As part of training the GPT models, Microsoft and
OpenAI collaborated to develop a complex, bespoke supercomputing system to house and
reproduce copies of the training dataset, including copies of The Times-owned content. Millions
of Times Works were copied and ingested—multiple times—for the purpose of “training”
Defendants’ GPT models."
You are factually wrong. Yes, their prompt engineering proved that the model itself contains a memorized copy of the article so is and of itself infringing, if fair use doesn't apply. It also proved it was in the training set in the first place and illegally copied to it and from it.
The lawsuit is about the entire rainbow of OpenAI infringement.
>and reproduced
the offending factor was the copying AND THEN reproduction. not the copying.
NYT is complaining they "reproduce copies" during “training”. "Defendants repeatedly copied this mass of Times copyrighted content, without any license or other compensation to The Times. "
Reproduction and copying are synonyms ...
>to house and reproduce copies of the training dataset
your reading attention span is less than one sentence. thats really sad.
the suit is because they distributed the copied media, not that they used it in training. the massive fricking hint here is that they didnt sue oai until they were able to reproduce the content.
"Defendants repeatedly copied this mass of Times copyrighted content, without any license or other compensation to The Times."
This is talking about training.
their talking about training, to then say they used said trained data TO REPRODUCE.
learn to read, seriously
Reproducing is synonymous with copying. You seem to think it's synonymous with distribution? It's not.
youre so far off base its not even funny. im not saying theyre synonymous. im not saying anything about them, at all.
the entire quote that was posted, is talking about distributing. cherry picking one sentence doesnt change the context of the whole.
>the entire quote that was posted, is talking about distributing.
"Defendants repeatedly copied this mass of Times copyrighted content, without any
license or other compensation to The Times. As part of training the GPT models, Microsoft and
OpenAI collaborated to develop a complex, bespoke supercomputing system to house and
reproduce copies of the training dataset, including copies of The Times-owned content. Millions
of Times Works were copied and ingested—multiple times—for the purpose of “training”
Defendants’ GPT models."
No.
>to house and reproduce
again, the problem is distribution, not using the data in the training.
"for the purpose of “training”
Defendants’ GPT models."
there you go again, ignoring the fact that what you just quoted is the preface for the latter part of the quote.
>preface
No, it's the last sentence of the paragraph.
>preface for the latter part of the quote
why do you keep doing that? do you think im not just gonna keep correcting you when you blatantly manipulate context?
Copying something privately for "space-shifting" is fair use. However I still think NYT should and probably will win that case because ChatGPT outputted their articles near-verbatim.
For "paradigmatic noncommercial
use". LLMs are a billion dollar commercial business.
>Pablo Picasso: "Good artists copy, great artists steal"
>artgays: "no, bro, you don't understand, we are inspired, it's not the same thing."
Say, what does a quasi-human life form like you thinks he meant by that quote? I'm genuinely curious.
Just as Steven Jobs used to say.
>"At Apple, we've always been great thieves. And we think it's a good thing. We want to steal from anyone who has a good idea."
Still waiting for you to answer the question. Kinda weird how you can't, isn't it?
Is it according to your imagination that Picasso is no longer around to answer, is it okay to copy others? Or to be innovative and develop your ideas on the ideas of others?
Notice how you repeatedly fail to answer the question. Almost as if you are unable to form a single thought on your own. How do you suppose I was able to successfully predict you will be stuck in an infinite loop dodging that simple question?
Although in the end it is pure hypocrisy because after saying that Steve Jobs was complaining that Android and Windows were copying the ideas of IPhone and Mac.
>If I copy it is inspiration and if others copy it is theft.
Really funny how you get absolutely mind-broken when there isn't some definite tweet-length answer you can google and regurgitate.
he says, in the thread where he's been mind-broken
>I give you not one but two answers and even a conclusion.
>but you are mind-broken
I'll explain it easier, it's a stupid phrase said by a painter known for copying other artists to get away with the accusations and look intellectual, exactly like with Steve Jobs, like this or with drawings?
>it's a stupid phrase said by a painter known for copying other artists to get away with the accusations and look intellectual
So he didn't mean anything?
Who should be sued for this clear and serious case of theft and plagiarism?
Right now copyright doesn't consider interaction with the work by a licensee as a infringement. So viewing something on netflix is fine, but not saving it on disk and making dvds. Of course, to display the work on the screen, you have to copy (parts) of data aroundin your computer, render it etc. The boundary between which copying is okay and which isn't is decided by the copyright holder. That is, by default none of it is alliwed and they can let you do certain things. Or not.
>So viewing something on netflix is fine, but not saving it on disk and making dvds.
No that's totally legal, probably violates tos but netflix can't do anything besides terminate service unless you started distributing said copies.
https://en.wikipedia.org/wiki/Sony_Corp._of_America_v._Universal_City_Studios,_Inc.
why is he still sperging out
@99256200
>being this upset about "him"
@99256305
>continues seething uncontrollably
>now simultaneously seething about trannies for no specific reason
Mental illness.
passive aggression is a female trait, stop
I dont care which way it goes but I'm making cash while I can by selling generated nudes, easy side money, fricking gpu paid for itself already, thx comers
@99256363
>it continues seething against imaginary characters in its head
I wonder which one of the mind-broken homos this one is. The one who claimed overfitting means fitting noise, maybe?
>it continues engaging in female behavior
@99256441
>impotent rage turns into homoerotic fantasies about feminine men
This poster grew up without a father.
While I agree that AI training should not be copyright infringement the act of "Right click -> Save Image As" or curl/wget the image is in theory copyright infringement because you pretty much copied a piece of work on your device without permission. Now if somebody makes a script that can look at Image URLs without loading them locally that would be another beast to tackle, but for now, images are copied locally or on cloud storage, therefore that action is copyright infringement.
>the act of "Right click -> Save Image As" or curl/wget the image is in theory copyright infringement because you pretty much copied a piece of work on your device without permission.
Indeed, only your browser cache has exemption.
Funnily enough that exemption states the following:
>temporary acts of reproduction which are an essential part of a technological process and whose sole purpose is to enable transmission of a work in a network or other lawful use of the work shall be exempted from the reproduction right.
So it could fall under this exemption if it is ruled as "lawful use transmission"
In what way is a LLM training set an "essential part of a technological process and whose sole purpose is to enable transmission of a work"?
I don't really care about LLMs. I used image saving as an example for ImaGen models.
"sole purpose is to enable transmission of a work" is also not true for image gen training sets.
Piracy should be the domain of anons, not billion dollar companies.
I'm sorry but this isn't true. Private devices are exempt.
>I'm sorry but this isn't true. Private devices are exempt.
Nope, the web is for viewing, not for saving.
DMCA and fair use jurisprudence carve out exemptions, but in principle every copy needs a license. Just being private is not enough for either the DMCA or the jurisprudence. Space shifting jurisprudence for instance mentions "paradigmatic noncommercial use" as I said before. Clearly does not apply.
>DMCA and fair use jurisprudence carve out exemptions
correct, and private devices are in effevt exempt. why do you think they cant go after you for torrenting unless you seed?
>why do you think they cant go after you for torrenting unless you seed?
Tons of lawsuits were started against bittorrent downloaders simply for leeching, no judge rejected them as a matter of law but they failed because IP addresses were considered poor evidence.
why lie?
Okay, I'll break down your points one by one, artgay.
First, copyright is enforced on works and derived works. If an AI or a human makes an exact copy of another human's work, the derived work infringes on the original work's copyright. This isn't always enforced, such as when the derived work is used for fair use purposes.
Next, "human minds aren't computable." Who cares? Whether or not they are, there *is* a computable approximation, which isn't the exact same, but for the purpose of art, is practically the same, because humans don't view images pixel by pixel. AI can be thought of as an approximation in this way.
In terms of training data, multiple court cases have established scraping is legal, so not much to say there.
So, let's even go with your argument here. Assume AI is a statistical model which "averages" all the input images based on the prompt.
The user gives AI the prompt "blue apple". The AI averages the "blue" and "apple" input images to output an image.
Is this output image the exact same as the "blue" input image? No.
Is this output image the exact same as the "apple" input image? No.
Does the output image have a similar style as the "apple" input image? Probably. But that's not grounds to claim copyright infringement.
Now instead of the AI, say a human saw the "blue" and "apple" input images and drew a "blue apple" image.
Their image doesn't infringe copyright for the exact same reasons the AI's image doesn't.
So, it doesn't matter whether it's AI or humans, what matters is the derived work and how it's used.
>In terms of training data, multiple court cases have established scraping is legal
https://www.bloomberglaw.com/external/document/X10SUETC000000/copyrights-professional-perspective-the-copyright-defense-agains
"In Facebook Inc., v. Power Ventures Inc, defendant Power.com provided a service that aggregated and displayed social networking and email accounts on a single portal. No. C 08-5780 JF (RS), 91 U.S.P.Q.2d 1430 (N.D. Cal. May 11, 2009). Users would provide Power.com with login credentials to various social media websites, and Power.com's software would use these credentials to scrape data from websites like Facebook. Facebook alleged that this service violated registered copyrights it maintained on its webpages. Power.com moved to dismiss the copyright claims. The district court denied the motion, holding that “if Defendants first have to make a copy of a user's entire Facebook profile page in order to collect that user content, such action may violate Facebook's proprietary rights.” These proprietary rights were based on Facebook's registered copyrights but did not include user data."
>These proprietary rights were based on Facebook's registered copyrights but did not include user data.
Good thing all NYT articles are registered.
"Defendants repeatedly copied this mass of Times copyrighted content, without any
license or other compensation to The Times. As part of training the GPT models, Microsoft and
OpenAI collaborated to develop a complex, bespoke supercomputing system to house and
reproduce copies of the training dataset, including copies of The Times-owned content. Millions
of Times Works were copied and ingested—multiple times—for the purpose of “training”
Defendants’ GPT models."
Mention of distribution in the paragraph none. Mention of the purpose of copying/reproduction, training. Complaint, done without license or compensation.
>reproduce
can you just not help it? are you one of those people that lies as easily as he breaths, to the point where he doesnt even realize he's doing it anymore?
Reproduce is still a synonym of copying and still has nothing to do with distribution.
in this context its literally a synonym for distribution. what the FRICK is wrong with you.
Have you noticed how that sentence with your beloved yet misundersood word "reproduce" is in is "prefaced"?
"As part of training the GPT models, Microsoft and OpenAI collaborated to develop a complex, bespoke supercomputing system to house and
reproduce copies of the training dataset, including copies of The Times-owned content."
This is one sentence, one. Now "in context" of that sentence what are the first 4 words in that sentence?
i dont recall saying it didnt use the word reproduce. again what the frick is wrong with you. youre not even responding to my posts, youre just repeating yourself.
"As part of training". You say it's about distribution, when the sentence literally starts with "As part of training".
As part of training the GPT models
Definitely BTFOs that tard. This is pretty conclusive.
Copyright on itself is stupid and it's just the work of greedy israelites that want to control the media.
I dream of a world where we can have 10000 pokemon copies and versions, games, films, music and where I can fix a useless artistard mediocre work without layers telling me I'm plagiarizing their work.