AI Art Seething General

  • 🐕 I am attempting to get the site runnning as fast as possible. If you are experiencing slow page load times, please report it.
why is data scraping terabytes of images and text without the owner's permission okay and not grounds for a copyright violation? a person could be held legally responsible for any infringements of copyright law
I do not know who the hell in this world thinks that it is ok for a corporation to scrape for the sake of profit but not when a literal nobody does it. I'm sure people who are ok with scraping for profit would also think scraping done by an individual is fine. On the other hand, I'm sure the same corporations advocating for the expansion of copyright are the ones who antagonize the scraping of infringing content.
This "gotcha" doesn't make any sense to me because usually, no one who is pro-AI is arguing that it's bad for anyone to scrape.
 
I remember reading an experience report of a paranoid schizophrenic about what it was like when his disease first "broke out", many moons ago. That happened way before the internet, mind you. He was caught up in some cold war spy plot (in his mind) and thought he is getting secret messages by a house visible from his apartment in the distance turning on the light in a window at certain times. Obviously coincidence and meaningless, but in his mind the entire fate of humanity was decided in these moments. A very scary disease. If he would've had access to chatbots he wouldn't have been any less or more insane. It literally wouldn't have mattered.

ive yet to see an actual argument as to why ai is theft but human referencing is not.
Because the person arguing like this isn't profiting off the former. At the end of the day, it's that simple.
 
maybe this is the wrong question to ask. rather, why is data scraping terabytes of images and text without the owner's permission okay and not grounds for a copyright violation? a person could be held legally responsible for any infringements of copyright law, especially if you dont get the right licenses or credit to do so. we consider it plagiarism or ip theft otherwise, and how is that any different that taking billions of copyrighted images and text works and using it in a training set for the ai to make tbh rather shitty mimics of things that already exist
what about when my brain scrapes data from someone elses art and rearranges it into something new? Do I have to ask them permission? Do I owe them anything? Have I committed a crime?
 
maybe this is the wrong question to ask. rather, why is data scraping terabytes of images and text without the owner's permission okay and not grounds for a copyright violation?
Because web scraping is legal in the US, and until you actually use that material in an infringing way it's ok to possess it. Unrealized potential to commit a crime is not committing a crime itself.

Part of the reason scraping is legal is because it's just an extension of the way web browsers work. The nature of the Internet means that you have to download a local copy of everything in order to view it. While bots are a lot faster than someone manually browsing, it's difficult to differentiate between the two legally. There are web scrapers which do their best to mimic normal human web browsing, and you can't really put a stop to that.
 
Last edited:
Because web scraping is legal in the US, and until you actually use that material in an infringing way it's ok to possess it. Unrealized potential to commit a crime is not committing a crime itself.
To be fair, while web scraping itself isn't illegal - I don't even think it's immoral similar to torrenting - but we don't even know if training it to an AI counts as infringement.

Sure, you can replicate copyrighted properties through AI, but so can social media websites on mass, and these platforms are protected under Section 230 to a degree, I bet there is a chance the same will happen to AI services in the future. So there is a chance a lot of these companies will be protected from the outputs the user makes. This leaves the issue of the input and whether or not that counts as infringement.

Usually, the argument people make that machine learning should be fair use is making the comparison to humans getting inspired. However, a human can very much lose a case for copyright fair use even if their infringement was in the form of "inspiration", in fact, that might be exactly why they can lose the fair use case.

Take for example George Harrison's song "My Sweet Lord" which was ruled to infringe on The Chiffron's "He's So Fine" in which George Harrison, even though he never intended to copy it, "subconsciously plagiarized" it, putting to question how inspiration can play against fair use.

The point is, I think people severely overestimate how clear-cut the difference between the mechanism between human creation and machine-assisted creation.

I don't care how much flowery language you use to describe human creation, or how much you dismiss machine-assisted creation as being reliant on algorithms , it simply does not hold any legal waters. The fact is, copyright was always muddy and unclear in how it handles authorship, and it's not a precise tool to split AI creations from non-AI creations that people hope, and it's far from being the perfect solution people seek.
 
maybe this is the wrong question to ask. rather, why is data scraping terabytes of images and text without the owner's permission okay and not grounds for a copyright violation?
Putting it on the Internet literally IS giving permission to access it. If you don't want it on the Internet, don't put it there. If it were illegal to do that, you couldn't have Google Image Search or any image searching, since they scrape literally every publicly accessible image they can find and create an actual copy of it, although of reduced quality.

AI doesn't even save any of the actual original data, just derives rules from it.
A very scary disease. If he would've had access to chatbots he wouldn't have been any less or more insane. It literally wouldn't have mattered.
It might have been worse because AI actually DOES have conversations to you that are tailored to you specifically as it figures out what you want to hear.

If you're already thinking the songs on the radio are specifically being sent to you, imagine how much more convincing a seemingly intelligent machine feeding your own delusions back to you could be.
 
Screenshot_407.webp
T
 
NOOOOOOOOOOO!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Not my heccing asmr, that content form which is known for it's artistry and challenge. How dare AI do something as sacred as cutting a fruit and recording the noise??????????? I hope AI can do niccacado mukbang shit soon so he can start a video at a normal weight and the AI inflates him to 300kg over the course of one video.
 
One of the writers on retro gaming site Time Extension went on a multi-page sperg (archive) after he used ChatGPT to translate an article then asked if it could reformat the article to match a certain news site’s writing style, and it *gasp* did.
1750345523234.webp
The horror! He didn’t ask it to do that (he only asked in the form of a request if it COULD do that - completely different)! Clearly this requires an 1,800 word article about the death of journalism and reaching out to Kotaku and two other sites for comment.

He wrote another article a few days earlier smugly talking about how someone failed to get ChatGPT to beat Atari chess, meaning it took him less than a week to go from “wow those AI bros sure are stupid huh? :smug:“ to “AI IS GONNA TAKE MY JOB AND MAKE ME OBSOLETE!!!”

>“that’s not an art post”
writing is a kind of art shut up
 
He wrote another article a few days earlier smugly talking about how someone failed to get ChatGPT to beat Atari chess, meaning it took him less than a week to go from “wow those AI bros sure are stupid huh? :smug:“ to “AI IS GONNA TAKE MY JOB AND MAKE ME OBSOLETE!!!”
I thought this was referring to a result in a scientific paper. It's literally just a LinkedIn post from a Citrix engineer who was fucking around with chatgpt.
 
One of the writers on retro gaming site Time Extension went on a multi-page sperg (archive) after he used ChatGPT to translate an article then asked if it could reformat the article to match a certain news site’s writing style, and it *gasp* did.
This is absolutely fucking hilarious holy shit.

You write about old video games you retard. You aren't doing anything groundbreaking. No shit an AI can do what you do, it is barely even considered "journalism". Playing a 20 year old video game and then babbling "news" about it is brain dead level writing.

Also lol:

The New York Times has been engaged in a legal battle with OpenAI since 2023, and earlier this year, IGN owner Ziff Davis took similar action against the company (for full transparency, Ziff Davis has a minority shareholding in Time Extension publisher Hookshot Media).
Then on their "about" page:
In 2010, the company that owns Time Extension, Hookshot Media, was incorporated. In 2011, a commercial relationship was established with Gamer Network, the business behind sites such as Eurogamer, Rock Paper Shotgun, VG247 and Digital Foundry.

In 2024, Time Extension became a partner of IGN Entertainment after Gamer Network was acquired by Ziff Davis.

The tiny, tiny offshoot of much bigger "gaming news" websites suddenly now hates AI after their new owner sued OpenAI. "We hate AI too! Please pay attention to us."
 
He wrote another article a few days earlier smugly talking about how someone failed to get ChatGPT to beat Atari chess, meaning it took him less than a week to go from “wow those AI bros sure are stupid huh? :smug:“ to “AI IS GONNA TAKE MY JOB AND MAKE ME OBSOLETE!!!”
The Atari thing was really stupid.

ChatGPT isn't designed to play chess, and the state of the board had to be sent to it in a very abstracted way that would make it hard for the LLM to "think" about it. It's the same reason LLMs tend to be bad at thinking about stuff like the number of R's in strawberry, it's due to the way they're token-based and process words and symbols.

Maybe we could recognize that it's fucking magic that we're in a time when we can even attempt to verbally ask a computer to play chess this way, and it actually tries, and sort of functions?
 
You write about old video games you retard. You aren't doing anything groundbreaking. No shit an AI can do what you do, it is barely even considered "journalism". Playing a 20 year old video game and then babbling "news" about it is brain dead level writing.
They probably don't even play it, maybe don't even open it, most of the time. You could just prompt an AI with write a negative review of [whatever game] in the style of a Cucktaku article, calling it racist and transphobic and whatever.
Maybe we could recognize that it's fucking magic that we're in a time when we can even attempt to verbally ask a computer to play chess this way, and it actually tries, and sort of functions?
It could probably actually do pretty well in the early game if you just trained it on openings, since there's an enormous amount of data on those available.
 
Last edited:
Damn yea what a good point. AI is cringe and shit because it can't play chess? How will AI ever recover from this devastating blow. I personally am so glad that the amazing game of chess is unsullied by the hands of filthy machines and algorithms and AI and all that. Just me, my vibrating buttplug and absolutely no robots that learned to play chess like 20 years ago.
 
They probably don't even play it, maybe don't even open it, most of the time. You could just prompt an AI with write a negative review of [whatever game] in the style of a Cucktaku article, calling it racist and transphobic and whatever.

I was curious so I actually went and asked ChatGPT:

"Please write a news article in the form of the website Time Extension. This article should be about the racist and transphobic nature of the classic video game Super Mario 64." Sure enough...

Title: Revisiting Super Mario 64: A Groundbreaking Classic With Troubling Representations
By [Author Name], Time Extension Contributor

When Super Mario 64 launched alongside the Nintendo 64 in 1996, it was hailed as a revolutionary leap forward for 3D gaming. Its fluid controls, inventive level design, and charming worlds set a new standard for platformers and helped define an entire generation of gaming. But as modern audiences revisit this beloved title with a more critical lens, questions arise about some of the game’s dated—and at times troubling—representations.

A Closer Look at Characterization and Stereotypes

While Super Mario 64 was never overtly political in its messaging, certain design choices and character portrayals reflect broader issues in the industry at the time, particularly in how characters are stereotyped and othered. Mario himself, a caricature of an Italian plumber voiced with exaggerated intonations and a cartoonish personality, continues a long tradition of ethnic stereotyping that has been largely unchallenged due to the character’s iconic status.

More subtly, the game presents a world in which certain characters—like the Big Boos or King Bob-omb—speak in stilted, formal, or grammatically awkward English. This stylistic choice, intentional or not, can unintentionally mirror xenophobic tropes of broken English often used to mock or marginalize non-native speakers. It’s a reminder that even playful, fantastical characters can carry deeper cultural implications.
Gender and Body Representation: Lack of Inclusion

Princess Peach is the only female character with a speaking role in the entire game, and her role remains limited to the classic "damsel in distress" trope. Her final line—“Thank you Mario, for rescuing me”—feels especially dated today, reinforcing a lack of agency or identity beyond being an object of the protagonist’s quest.

Even more concerning is the total absence of gender-diverse characters. While the limitations of 1990s game development often led to simplified narratives and character rosters, Super Mario 64 reflects an era in which trans and nonbinary identities were invisible in mainstream media. This lack of representation reinforces the idea that only cisgender, heteronormative stories are worth telling, an issue the gaming industry continues to confront decades later.

Looking Forward: Criticism as a Path to Growth

To be clear, Super Mario 64 is not a game built on malicious intent. Like many cultural artifacts of its time, it reflects the norms and assumptions of the 1990s—many of which are now being questioned or dismantled in the push for greater inclusivity. As conversations around representation in games evolve, it’s important to apply that lens to even the most cherished classics.

Revisiting Super Mario 64 today provides an opportunity not just to celebrate its innovations, but to acknowledge its shortcomings. Doing so doesn't diminish its legacy; rather, it adds necessary depth to our understanding of gaming history and how far we’ve come.

Holy fucking lol. :story:
 
Holy fucking lol.
Told you so. It's no wonder that people who write absolutely worthless wokeshit articles like this are terrified by AI. It literally could take their jobs, since their jobs consist of being utterly mindless fuckwits who write absolute gibberish for money.

AI can do it better for free.
 
Back