[Azath Vitr (D'ivers]

JPK, on 10 December 2025 - 07:02 PM, said:

Azath, please don't go into politics. You spin like a Maga Republican.

You tell me I'm wrong, but what about your stated scenario doesn't boil down to the prompt: "Take Dungeon Crawler Carl and make it from the POV of the cat"? This right here is exactly what so many people don't want. It's just like saying "take A Game of Thrones and stylize it like a Studio Ghibli film.".

This isn't anywhere near the case of a book being rewritten to add depth and perspective that the original didn't have. Think more along the lines of James by Percival Everett for an example of what I'm talking about here.

Well if it accepted the plot of Game of Thrones as a prompt---or let you input the video, if you want the characters to look very similar---and you explicitly asked the AI to remake it in the style of Studio Ghibli, then yes you could do that, but it would be in breach of the terms of service. And if it's not clearly a parody then it would also be in breach of copyright. But I know Suno, at least, prevents users from inputting lyrics or audio that match copyrighted works (using Content ID for the audio check, but of course things that are posted to the internet but not entered into Content ID or independent lyrics databases probably won't be checked, so there's still the possibility of knowing, intentional misuse).

Yes, my suggestion about rewriting it from the perspective of the cat was mostly flippant, and I'm sorry if you took that flippancy as implying that the books are trash; that really wasn't my intent. As I understand it the books themselves are already humorous and flippant, with no aspiration to be "great literature" (perhaps I'm mistaken about that?). And I actually am interested in novels that go deep into the perspectives of cats (whether quasi-accurately or in more of a whimsically fanciful way---the first novel by Tad Williams was more of the latter IMO); I wish there were more of them, and I'm very interested in imagining what the world is like from a cat's perspective. I love watching cat cam videos (there's a camera on the collar, showing the cat's perspective; but I'd like these even more if they also altered the color scheme to more accurately depict feline vision).

But I'm not certain exactly what your primary issue would be with a Ghibli-ized version of Game of Thrones---that it would somehow trivialize the show or the book? Or that it would be unauthorized, not under the control of HBO or GRRM? Even if someone wanted to draw caricatures---with AI or otherwise---while it might be disrespectful, I don't think it actually diminishes the work. For that matter, independent artists at conventions over the last few decades have been making and selling unauthorized drawings of comic book and fantasy characters, some comical or satirical, but not at all IMO diminishing the works they're based on. A fully Ghibli-ized Game of Thrones video doesn't seem fundamentally all that different.

This post has been edited by Azath Vitr (D'ivers: 10 December 2025 - 09:54 PM

[Macros]

Pretty sure all the big LLM AI robot things owners have been accused and in some places sued for feeding copy righted material into their robots.

That's theft

[Maark Abbott]

Macros, on 11 December 2025 - 07:51 AM, said:

Pretty sure all the big LLM AI robot things owners have been accused and in some places sued for feeding copy righted material into their robots.

That's theft

It's why I call them Theft Engines. Without things to steal from, what can they do? Nowt.

[TheRetiredBridgeburner]

From an industry perspective, Mr NAB works for a Microsoft-owned game studio. Their hard rule is that unless it's an AI tool they and only they have trained (so they know exactly what's been inputted and by whom) they don't touch them, because it's virtually impossible to know what's been submitted and who's copyright may have been infringed. And oddly enough, copyright is a big deal in their industry.

[Azath Vitr (D'ivers]

Macros, on 11 December 2025 - 07:51 AM, said:

Pretty sure all the big LLM AI robot things owners have been accused and in some places sued for feeding copy righted material into their robots.

That's theft

They have them crawl the internet to analyze and learn from data that includes copyrighted material. Analysis is not theft. Learning is not theft.

Here's a good explanation from Vox of how AI image generators work. As I've said, contrary to the misinformation some people have been repeating here, they don't chop up and recombine images from the training data:

Quote

We might assume that when we give them a text prompt, like "a banana inside a snow globe from 1960," they search through the training data to find related images and then copy over some of those pixels. But that's not what's happening.

The new generated image doesn't come from the training data, it comes from the "latent space" of the deep learning model. That'll make sense in a minute, first let's look at how the model learns.

[...] In order to understand that this arrangement of pixels is a banana, and this arrangement of pixels is a balloon, it looks for metrics that help separate these images in mathematical space. So how about color? If we measure the amount of yellow in the image, that would put the banana over here and the balloon over here in this one-dimensional space.

[...] They find variables that help improve their performance on the task [of matching images with words] and in the process, they build out a mathematical space with way more than 3 dimensions. [...] Translating a point in that mathematical space into an actual image involves a generative process called diffusion. It starts with just noise and then, over a series of iterations, arranges pixels into a composition that makes sense to humans.

https://www.youtube....h?v=SVcsDDABEkM

[explanation of how they don't work---and how they do work---starts at timestamp 6:26]

But most AI image generators now also allow you to input an image for the AI to fill out or alter---for example it can maintain the structure of the image while changing the style, transform a doodle into what looks like an image of an oil painting, even change the perspective / viewing angle. I directed ChatGPT 5 to search publicly available social media for recent examples of AI "stealing" from artists and what it found were cases where someone had taken an artist's image without their permission and used AI to change it while retaining the same structure. For example someone who's anti-AI posted a simple doodle on their social media and somebody replied with a fully-filled in image saying "AI art is art": Clearly that's not nice and it's a breach of the terms of service. In another case someone on Temu had an AI image generator change the image from a side angle view to a frontal view (and a few other changes) and was trying to sell physical copies. So some people are stealing artists' images, inputting them to AI art generators to alter them, and then recirculating or even trying to sell them. But those are examples of people intentionally stealing art. AI does it make it easier for people to steal art and change it in ways that make it harder to recognize, or that they could try to claim are sufficiently transformative despite obviously deriving from the original structure, even though I think most people would recognize these examples as not sufficiently transformative.

This post has been edited by Azath Vitr (D'ivers: 11 December 2025 - 02:41 PM

[QuickTidal]

Azath Vitr (D, on 11 December 2025 - 12:14 PM, said:

They have them crawl the internet to analyze and learn from data that includes copyrighted material. Analysis is not theft. Learning is not theft.

This is like those people in the 90's with a massive satellite dish who said "It's in the air, why can't I grab it?" about the signal...and getting Cable and Speciality for free.

Just because you can rationalize it in your mind, doesn't make it not theft.

[Azath Vitr (D'ivers]

QuickTidal, on 11 December 2025 - 12:44 PM, said:

Azath Vitr (D, on 11 December 2025 - 12:14 PM, said:

They have them crawl the internet to analyze and learn from data that includes copyrighted material. Analysis is not theft. Learning is not theft.

This is like those people in the 90's with a massive satellite dish who said "It's in the air, why can't I grab it?" about the signal...and getting Cable and Speciality for free.

Just because you can rationalize it in your mind, doesn't make it not theft.

No, what you're saying is analogous to claiming that it should be illegal to analyze and learn from copyrighted material on public television without explicit permission.

This post has been edited by Azath Vitr (D'ivers: 11 December 2025 - 04:09 PM

[QuickTidal]

Azath Vitr (D, on 11 December 2025 - 04:09 PM, said:

QuickTidal, on 11 December 2025 - 12:44 PM, said:

Azath Vitr (D, on 11 December 2025 - 12:14 PM, said:

They have them crawl the internet to analyze and learn from data that includes copyrighted material. Analysis is not theft. Learning is not theft.

This is like those people in the 90's with a massive satellite dish who said "It's in the air, why can't I grab it?" about the signal...and getting Cable and Speciality for free.

Just because you can rationalize it in your mind, doesn't make it not theft.

No, what you're saying is analogous to claiming that it should be illegal to analyze and learn from copyrighted material on public television without explicit permission.

Nope. Sorry. You're not going to win this one with me. A few indie artists in the comic industry have been busted over the years for essentially taking existing art, re-inking it, recolouring it, and printing it out to sell....they have been literally run out of Cons for doing it...that's what this is. It's not learning, give me a break, it's copying.

JUST Analyzing and learning would be one thing, but it's literally spitting out a product (for lack of a better word) based on that scraping out the other end...if it didn't, there would be no point to it...that's plagiarism and theft, plain and simple.

[Macros]

Azath Vitr (D, on 11 December 2025 - 12:14 PM, said:

Macros, on 11 December 2025 - 07:51 AM, said:

Pretty sure all the big LLM AI robot things owners have been accused and in some places sued for feeding copy righted material into their robots.

That's theft

They have them crawl the internet to analyze and learn from data that includes copyrighted material. Analysis is not theft. Learning is not theft.

No they were sued for deliberately feeding the machines copy righted material.

They had a big puddle of stuff that they all used to feed the first big LLMs, that they fully disclosed.(it has a name even that bundle of stuff but it currently exudes me)
Then came gpt2, less information of source was divulged
Then 3and guess what, LESS information about where it was fed from
And where we are now they disclose fuck all because....why ever would they not say what they've fed it???

But yeah, I'll dig out a recent article for you

[Macros]

Ha, a quick Google throws up a ton of articles, even their ai summary says theyre being sued for theft.

The fact the big boys are now trying to sign deals with large publications shows they know theyre on slippery slopes and want to claw some legitimacy.

Will be irrelevant in a year or two when people realise neural networks is a bullshit lie, the top players are all still glorified LLMs and the circle Jerk of money is one bad day away from an economy crashing fuck you to silicon valley equivalent to the dotcom bang times 10

[Azath Vitr (D'ivers]

QuickTidal, on 11 December 2025 - 04:53 PM, said:

Azath Vitr (D, on 11 December 2025 - 04:09 PM, said:

No, what you're saying is analogous to claiming that it should be illegal to analyze and learn from copyrighted material on public television without explicit permission.

Nope. Sorry. You're not going to win this one with me. A few indie artists in the comic industry have been busted over the years for essentially taking existing art, re-inking it, recolouring it, and printing it out to sell....they have been literally run out of Cons for doing it...that's what this is. It's not learning, give me a break, it's copying.

JUST Analyzing and learning would be one thing, but it's literally spitting out a product (for lack of a better word) based on that scraping out the other end...if it didn't, there would be no point to it...that's plagiarism and theft, plain and simple.

We agree that if it were taking existing art and recoloring it then that would be copyright infringement. But we both know that's not what it's doing (except if users input a copyrighted image in breach of the TOS and ask it to do that, in which case it's the user knowingly stealing---but again I think we'd agree that image generator providers should put a copyright check in place to do their best to prevent users from doing that).

Where we disagree: you're saying that analyzing and learning from public broadcast television to create your own TV show that's different enough from any of the preexisting TV shows to not legally infringe on copyright should still be illegal---because you learned how to do it by analyzing previous shows.

But of course you're fine with the illegal use of IP in fan art and fan fic. Even when artists sell unauthorized images of protected IP at cons. Ironically, with Disney's new deal with OpenAI, the only legal fan art will be through OpenAI:

Quote

Disney is investing $1 billion in OpenAI and will let the startup use characters ​from Star Wars, Pixar and Marvel franchises in ‌its Sora AI video and image generator, a crucial deal that could ‌reshape how Hollywood makes content.

The partnership [...] is a pivotal step in Hollywood's embrace of generative artificial intelligence [...] As part of the agreement, Sora and ChatGPT Images are expected to start generating ‍fan-requested videos using licensed Disney characters in early 2026. [...]
a selection of the videos made by users will be available ‍for streaming ⁠on the Disney+ platform. [...]

The tie-up will also ⁠cover image generation on ChatGPT, drawing from the same Disney intellectual property.

Disney to invest $1 billion in OpenAI, license characters for Sora AI tool

[amphibian]

 Azath Vitr (D, on 11 December 2025 - 05:33 PM, said:

We agree that if it were taking existing art and recoloring it then that would be copyright infringement. But we both know that's not what it's doing (except if users input a copyrighted image in breach of the TOS and ask it to do that, in which case it's the user knowingly stealing---but again I think we'd agree that image generator providers should put a copyright check in place to do their best to prevent users from doing that).

We don't know that the LLMs are not training models on stolen copyrighted materials. The visibility into what the training pool is has gone down over time and the general movement has been towards paying the copyright holders whose work has been stolen - not towards a completely fresh training pool free of copyrighted materials.

[QuickTidal]

Azath Vitr (D, on 11 December 2025 - 05:33 PM, said:

Where we disagree: you're saying that analyzing and learning from public broadcast television to create your own TV show that's different enough from any of the preexisting TV shows to not legally infringe on copyright should still be illegal---because you learned how to do it by analyzing previous shows.

No, I'm literally saying that this is an analogy that does not apply here, and you continuing to try to push it is silly. That's where we disagree. You're applying a safe and comfy analogy to your (seeming) devotion to corpo thieves so you can feel better about that support.

[worry]

From what I've been seeing and reading lately, they are already running out of high quality human-produced material to train on -- copyrighted or not -- and will be cannibalizing AI-produced materials (and then re-cannibalizing those products, and so on). Basically a speed run of dead internet theory, same thing that was producing those Jesus-and-Crabs-in-everything pics that were pervasive a couple years ago.

[Azath Vitr (D'ivers]

amphibian, on 11 December 2025 - 06:12 PM, said:

We don't know that the LLMs are not training models on stolen copyrighted materials. The visibility into what the training pool is has gone down over time and the general movement has been towards paying the copyright holders whose work has been stolen - not towards a completely fresh training pool free of copyrighted materials.

If by "stolen" you mean illegally made available on public websites---for example, pirate pdfs of ebooks---without the AI company making an agreement with license holders then yes, the act of accessing those materials is illegal. OpenAI insists that it doesn't circumvent paywalls and that and that it has strict protocols in place to prevent training on pirated material. The New York Times is claiming in their lawsuit that ChatGPT must have been training on New York Times articles despite the paywall, but OpenAI denies it and is fighting the lawsuit rather than settling. We'll see if OpenAI presents evidence demonstrating that they weren't bypassing the paywall.

Macros: most things people post on the public internet are technically automatically copyrighted as soon as they're created, at least in US law, provided they're sufficiently human-created and sufficiently original. Even if someone doesn't post it but just types it on their computer or scribbles it down somewhere it's automatically copyrighted.


[Edit: however, OpenAI acknowledges that, as with search engines, some pirated material may still get into its training data, though it tries to detect and remove it.]

This post has been edited by Azath Vitr (D'ivers: 11 December 2025 - 07:00 PM

[QuickTidal]

worry, on 11 December 2025 - 06:44 PM, said:

From what I've been seeing and reading lately, they are already running out of high quality human-produced material to train on -- copyrighted or not -- and will be cannibalizing AI-produced materials (and then re-cannibalizing those products, and so on). Basically a speed run of dead internet theory, same thing that was producing those Jesus-and-Crabs-in-everything pics that were pervasive a couple years ago.

I heard about this, and I can't wait to see it eat itself. I would imagine that's when the bubble will burst. It's going to make for some spectacular nonsensical shit.

William Gibson could not write a more compelling narrative.

[QuickTidal]

Azath Vitr (D, on 11 December 2025 - 06:49 PM, said:

If by "stolen" you mean illegally made available on public websites

I'm impressed you keep veering directly back into the "cable signals are in the air, so it's fine to steal them" territory while pretending that's not what you are doing.

Azath Vitr (D, on 11 December 2025 - 06:49 PM, said:

OpenAI insists that it doesn't do that and has strict protocols in place to prevent training on pirated material. We'll see if OpenAI presents evidence demonstrating that they weren't bypassing the paywall.

"OpenAI insists"...hey, let's look into who founded and runs OpenAI....oh.....oh....yeah they'd NEVER lie...

[Azath Vitr (D'ivers]

QuickTidal, on 11 December 2025 - 07:00 PM, said:

Azath Vitr (D, on 11 December 2025 - 06:49 PM, said:

If by "stolen" you mean illegally made available on public websites

I'm impressed you keep veering directly back into the "cable signals are in the air, so it's fine to steal them" territory while pretending that's not what you are doing.

No, I wrote that viewing material that has been illegally shared on the public internet is illegal.

Me: I agree
QT: NO YOU DON'T!

But effectively no one is ever prosecuted or sued for viewing or downloading pirated material. Granted, that could change under Trump, if he wants to find grounds for prosecuting people (but he may find it more convenient to just plant evidence for crimes that come with more severe punishments).

OpenAI definitely downloaded datasets of pirated ebooks from LibGen, but claims ChatGPT wasn't trained on them. Simply having downloaded the dataset is technically illegal though.

A judge ruled in June that training AI without permission on books you've purchased or accessed through public sites that legally distributed them is legal. So it's ironic: OpenAI is spending enormous amounts of money, but apparently can't be bothered to pay for ebooks.

Quote

Judge: It's Fair Use to Train AI on Books You Bought, But Not Ones You Pirated

Judge William Alsup's Tuesday order turns aside part of a class-action lawsuit filed by book authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson against the AI firm Anthropic [...] Alsup found that the authors had no more right to demand payment for it than to charge a human reader for learning from their writing.

"Everyone reads texts, too, then writes new texts," he wrote. "But to make anyone pay specifically for the use of a book each time they read it, each time they recall it from memory, each time they later draw upon it when writing new things in new ways would be unthinkable."

In a later paragraph, Alsup compared the plaintiffs' argument to a complaint that "training schoolchildren to write well would result in an explosion of competing works." He concluded: "This is not the kind of competitive or creative displacement that concerns the Copyright Act."

Judge: It's Fair Use to Train AI on Books You Bought, But Not Ones You Pirated | PCMag

[Macros]

I think the difference comes down to this.

You read a book and are inspired by it to write something.

You don't take out the scissors and cut the book up and rearrange the sentences then present it as your own original work.

That is ultimately what LLMs are doing, just on a maaaahoooooise scale. So its not creating, its mashing shit into a blender and generating income for people that do nothing, not even read the original works to be inspired.

You sell a million copies, how many of those readers are going to write something?
How much of that will ever see the light of day because its not finished, not good enough, author does it for fun?
1, 2?

You've sold a million books, youre doing OK.

You write 20 books, only a moderate seller, you're David Gemmell, enough to live comfortably on but youre not selling 50 shades of grey volume, you're safe, the maths is in your favour.

Some talentless fruitloop stuffs your books in a machine without even reading them and churn out countless volumes of dire which people unwittingly buy and leave the genre forever because its garbage.
The market is flooded with shite, inevitably your market share becomes untenable and you go back to your job at the newspaper, oh wait.....


Keep AI, ok.
Regulate the fuck out of it, full transparency on ALL their training models.
Full accountability for the company's that feed the machine, they ARE responsible for it reading pirated stuff because its a dumbass machine that doesn't know its reading pirated shite.

When that happens we can talk.

[Azath Vitr (D'ivers]

Macros, on 11 December 2025 - 07:48 PM, said:

I think the difference comes down to this.

You read a book and are inspired by it to write something.

You don't take out the scissors and cut the book up and rearrange the sentences then present it as your own original work.

That is ultimately what LLMs are doing, just on a maaaahoooooise scale. So its not creating, its mashing shit into a blender and generating income for people that do nothing, not even read the original works to be inspired.

No, as I've explained several times, generative AI doesn't "cut up the book and rearrange the sentences" or "mash shit into a blender". Even though it's much easier to imagine an automated system doing that, that's not how LLMs work. (Granted, you could include a text with your prompt and ask ChatGPT to cut it up and rearrange it or remix it, but that's not .)

[Edit: slight correction, the Vox quote was actually about image generation; LLM training is different, but it's still true that it's not chopping up sentences and rearranging them. That's just not how it works.]

This post has been edited by Azath Vitr (D'ivers: 11 December 2025 - 07:56 PM