Sarah Silverman leads class-action lawsuit against ChatGPT creator

[Ryujin]

Exactly. If one of our expenses stops being an expense, entertainment in this case, we get richer.

That's a rather odd take-away.

 

[Bohandas]

For clothing? Maybe.

This is taken from fiction but reflects a reality of the last couple of hundred years.


“The reason that the rich were so rich, Vimes reasoned, was because they managed to spend less money.

Take boots, for example. He earned thirty-eight dollars a month plus allowances. A really good pair of leather boots cost fifty dollars. But an affordable pair of boots, which were sort of OK for a season or two and then leaked like hell when the cardboard gave out, cost about ten dollars. Those were the kind of boots Vimes always bought, and wore until the soles were so thin that he could tell where he was in Ankh-Morpork on a foggy night by the feel of the cobbles.

But the thing was that good boots lasted for years and years. A man who could afford fifty dollars had a pair of boots that'd still be keeping his feet dry in ten years' time, while the poor man who could only afford cheap boots would have spent a hundred dollars on boots in the same time and would still have wet feet.

This was the Captain Samuel Vimes 'Boots' theory of socioeconomic unfairness.”
― Terry Pratchett, Men at Arms: The Play

This is why I support the AIs, or at least the ones that are open and downloadable. Instead of buying more and more books and movies and pictures you could buy one computer capable of running high-end AI and be done with it.

 

[Umbran]

At worst the only actionable reproduction is the training dataset itself, anything in the actual program is transformed at least as much as a William S. Burroughs novel.

It is a mistake to think of the program as separate from its data. The training dataset is used to create both its algorithm (usually a complex neural network algorithm), and a resulting reference data set. Once trained, if you feed in a prompt, it is reduced to some number of tokens. The algorithm takes a token, processes it, refers to its reference data, and spits out the next token. Lather, rinse, repeat.

While you speak of transformation, there is an issue that ChatGPT can, with fairly simple prompts, spit out cogent summaries of the works used to train it. The contention is that this indicate that the original work still largely exists within the system.

And in any case it's likely a simple workaround to program the computer to just read from the original documents directly instead of compiling them into a single document and feeding that into the computer

I guess I wasn't clear. The path or method for the data to get into the system is not the issue. The contention is that the dataset itself constitutes an unauthorized derivative work, however it is produced.

"Reading from the original," still entails copying the original work. If you buy the paper book and have the computer optically scan it, that scan is an act of copying. If you feed it a pdf and extract the text stream from that, the text is still reproduced - copied - into the system.

 

[Bohandas]

It is a mistake to think of the program as separate from its data. The training dataset is used to create both its algorithm (usually a complex neural network algorithm), and a resulting reference data set. Once trained, if you feed in a prompt, it is reduced to some number of tokens. The algorithm takes a token, processes it, refers to its reference data, and spits out the next token. Lather, rinse, repeat.

While you speak of transformation, there is an issue that ChatGPT can, with fairly simple prompts, spit out cogent summaries of the works used to train it. The contention is that this indicate that the original work still largely exists within the system.



I guess I wasn't clear. The path or method for the data to get into the system is not the issue. The contention is that the dataset itself constitutes an unauthorized derivative work, however it is produced.

"Reading from the original," still entails copying the original work. If you buy the paper book and have the computer optically scan it, that scan is an act of copying. If you feed it a pdf and extract the text stream from that, the text is still reproduced - copied - into the system.

Again, if training an AI on a work entails copying it than so does creatig a cut-up technique novel based on the same work

 

[FrogReaver]

Again, if training an AI on a work entails copying it than so does creatig a cut-up technique novel based on the same work

the technical process for training an AI literally involves a step where the original work is literally copied into the computer memory.

 

[Umbran]

Again, if training an AI on a work entails copying it than so does creatig a cut-up technique novel based on the same work

Well, no. Training an AI first entails copying the work into memory/storage so that the data can be processed. You cannot, under copyright, do that for business purposes. It requires different licensing.

As for the cut-up technique novel - the cut-up technique is originally done by literally and physically cutting up the original to produce the art, so you technically aren't reproducing it.

If you try to mass-produce a novel like that, and open with, "In a hole in the ground there lived a hobbit." the Tolkien Estate would likely com knocking with a C&D.

 

[KEV777]

Even the people in the entertainment industry are formulaic and interchangable. They have been forever. When I was growing up back in the 1990's and the turn of the century there used to be a joke that the boy bands were all mass produced robots. And now having seen what GPT and Runway can do, I think that joke may have been a little bit unfair to robots.

oh you mean during the days when songs were still actually songs albeit still formulaic? Many of the top 10 songs now are just groves and don't even technically qualify as songs (no chorus, no bridge, etc). With the state of music today AI may actually start to improve things.

Watch this commentary of AI music. I largely agree with Beato here on the ramifications of AI. It's Napster 2.0 right now.


As discussed in the video, Grimes is allowing people to make AI music using her voice - she just wants 50%. Now that's the way forward IMO.

 

[KEV777]

the technical process for training an AI literally involves a step where the original work is literally copied into the computer memory.

Is it legal for a musician to memorize music and play it? What if he then crafts something influenced by it? Are you going to sue him for committing all those notes to memory (copying it to his brain?) Lets not discriminate against silicon based memory.

 

[DaedalusX51]

Is it legal for a musician to memorize music and play it? What if he then crafts something influenced by it? Are you going to sue him for committing all those notes to memory (copying it to his brain?) Lets not discriminate against silicon based memory.

The AI does not have the rights or needs of a human person. What it is doing is due to it's creator's programing and training. It doesn't have thoughts nor feelings. It doesn't own the work that it is creating. We can't compare these things like they are the same when they are not.

 

[KEV777]

The AI does not have the rights or needs of a human person. What it is doing is due to it's creator's programing and training. It doesn't have thoughts nor feelings. It doesn't own the work that it is creating. We can't compare these things like they are the same when they are not.

Print out a bunch of songs and then randomly cut and paste all the bars together in a Frankenstein work of art. AI just automates that process. Perfectly legal. AI is just a tool. The learning / sampling method is irrelevant.