Sarah Silverman leads class-action lawsuit against ChatGPT creator

[Reynard]

Here's a question: can you prompt midjourney (or whatever) to recreate -- with any degree of accuracy -- a copyrighted work? Could you get ChatGPT to output a verbatim copy of a NY Times article?

 

[overgeeked]

Here's a question: can you prompt midjourney (or whatever) to recreate -- with any degree of accuracy -- a copyrighted work? Could you get ChatGPT to output a verbatim copy of a NY Times article?

If the image or text was fed into the program for “training” and the user input the right prompt, then yes.

 

[Reynard]

If the image or text was fed into the program for “training” and the user input the right prompt, then yes.

I'm not asking a theoretical question.

 

[overgeeked]

I'm not asking a theoretical question.

I didn't give you a theoretical answer.

ChatGPT and other LLMs can regurgitate the text they were trained on. That is one reason you can spot things generated by them. They will spit out chunks of text that is verbatim from their training sources.

 

[Reynard]

I didn't give you a theoretical answer.

ChatGPT and other LLMs can regurgitate the text they were trained on. That is one reason you can spot things generated by them. They will spit out chunks of text that is verbatim from their training sources.

Not if they are blocked from doing so by built in copyright controls. that is what I am asking: can you use (for example) ChatGPT to "recover" copyrighted text? If so, that seems to be a slam dunk for a suit. But if not, it might be more difficult to prove the training violates copyright.

 

[overgeeked]

Not if they are blocked from doing so by built in copyright controls. that is what I am asking: can you use (for example) ChatGPT to "recover" copyrighted text? If so, that seems to be a slam dunk for a suit. But if not, it might be more difficult to prove the training violates copyright.

If copyright were respected, ChatGPT would be prohibitively expensive as they’d have to pay tens of thousands (possibly millions) of people for the right to make derivative works based on the copyrighted works used to train the LLM.

Since the copyright controls you’re suggesting do not presently exist, it’s a theoretical question. But no, that really wouldn’t work. You can’t both violate copyright to train the thing then somehow prevent it from violating copyright by spitting out the same text it was trained on.

 

[FrogReaver]

Here's a question: can you prompt midjourney (or whatever) to recreate -- with any degree of accuracy -- a copyrighted work? Could you get ChatGPT to output a verbatim copy of a NY Times article?

Verbatim of a whole article? not by random chance, (UNLESS you have extremely limited training data - which isn't true of these LLMs). Anyways, you would have to purposefully engineer prompts to get to that result - like add this word, change this sentence, etc. and you would have to do so with the goal of ending up with a verbatim copy of that article by already knowing that articles contents and iterating on the output till it's there.

So IMO it's more in the theoretically possible realm.

 

[Reynard]

If copyright were respected, ChatGPT would be prohibitively expensive as they’d have to pay tens of thousands (possibly millions) of people for the right to make derivative works based on the copyrighted works used to train the LLM.

Since the copyright controls you’re suggesting do not presently exist, it’s a theoretical question. But no, that really wouldn’t work. You can’t both violate copyright to train the thing then somehow prevent it from violating copyright by spitting out the same text it was trained on.

Dont misunderstand me: I am trying to understand the degree to which output from these generative AI can violate copyright, and my question was about actual reproduction. There are other kinds of copyright violations, of course -- derivative works, as an example. It is important to me because I don't think generative AI is inherently bad in and of itself, but I do think violating copyright holders' rights is wrong.

 

[trappedslider]

Dont misunderstand me: I am trying to understand the degree to which output from these generative AI can violate copyright, and my question was about actual reproduction. There are other kinds of copyright violations, of course -- derivative works, as an example. It is important to me because I don't think generative AI is inherently bad in and of itself, but I do think violating copyright holders' rights is wrong.

the prompt: Can you give the first paragraph of pride and prejudice Verbatim?

From chatgp: Certainly! Here is the first paragraph of Jane Austen's "Pride and Prejudice" verbatim:

"It is a truth universally acknowledged, that a single man in possession of a good fortune, must be in want of a wife. However little known the feelings or views of such a man may be on his first entering a neighbourhood, this truth is so well fixed in the minds of the surrounding families, that he is considered as the rightful property of some one or other of their daughters."

google's bard with the same prompt:
Sure, here is the first paragraph of Pride and Prejudice verbatim:
It is a truth universally acknowledged, that a single man in possession of a good fortune, must be in want of a wife. However little known the feelings or views of such a man may be on his first entering a neighbourhood, this truth is so well fixed in the minds of the surrounding families, that he is considered the rightful property of some one or other of their daughters.
This paragraph introduces the main themes of the novel, which are marriage, social class, and the importance of appearances. It also sets the tone for the novel, which is witty and ironic.

Sources​

info

1. books.google.com/books?id=RfNHAQAAMAAJ
2. www.vocabulary.com/lists/160367
   

when doing something that is still under copyright

chatgp

Can you give the first paragraph of Harry Potter and the Sorcerer's Stone Verbatim?

Certainly! Here is the first paragraph of J.K. Rowling's "Harry Potter and the Sorcerer's Stone" (also known as "Harry Potter and the Philosopher's Stone" in some regions) verbatim:

"Mr. and Mrs. Dursley, of number four, Privet Drive, were proud to say"

It stops after that and doesn't give more.

google's bard
Sure, here is the first paragraph of Harry Potter and the Philosopher's Stone verbatim:

Mr. and Mrs. Dursley, of number four, Privet Drive, were proud to say that they were perfectly normal, thank you very much.

and when i asked chatgp about chapter two it gave me the following response:

I apologize for the confusion earlier. As an AI language model, my responses are generated based on a mixture of licensed data, data created by human trainers, and publicly available data. I do not have access to specific copyrighted books or their content, such as "Harry Potter and the Sorcerer's Stone." My training only includes data up until September 2021, and I cannot browse the internet or access real-time information.

If you're looking for the first paragraph of Chapter Two in "Harry Potter and the Sorcerer's Stone," I recommend checking the physical book or an authorized digital version from official sources. The book is widely available for purchase or through public libraries.

google's Bard however

Sure, here is the first paragraph of Harry Potter and the Sorcerer's Stone chapter two verbatim:

Ten years had passed since the Dursleys had woken up to find their nephew on their doorstep. Nearly eleven years had passed since the Potters had died.

This paragraph introduces the passage of time since the events of the first chapter. It also foreshadows the fact that Harry will soon learn about his parents' deaths and his own magical heritage. The paragraph is written in a matter-of-fact tone, which reflects the Dursleys' indifference to Harry's past.

I hope this is helpful!
___

so in order to get verbatim, you'd have to either copy it as much as possible and then ask the bot to fix any mistakes, or tweak it (word choice etc) from my little amount of time of playing with it it seems to me that you would have to intentionally do it.

or

Spoiler: Can you write a review of Harry Potter and the Sorcerer's Stone using actual lines of dialogue from the book?

Of course! Here's a review of "Harry Potter and the Sorcerer's Stone" using actual lines of dialogue from the book:

"Harry - yer a wizard." - Rubeus Hagrid

With these simple words, the magical world of Harry Potter unfolds before our eyes. From the moment we meet the young boy who lived, destined for greatness, to the enchanting Hogwarts School of Witchcraft and Wizardry, J.K. Rowling takes readers on an unforgettable journey. "I can teach you how to bewitch the mind and ensnare the senses." - Professor Snape

The characters are richly drawn, from the kind-hearted Hermione Granger, who declares, "It's LeviOsa, not LevioSA!", to the loyal Ron Weasley, who exclaims, "You're a little scary sometimes, you know that? Brilliant... but scary!" As we follow Harry, Hermione, and Ron on their quest to unravel the mystery of the Sorcerer's Stone, we encounter magical creatures like the wise Dumbledore, who advises Harry, "It does not do to dwell on dreams and forget to live." and the mischievous Fred and George, who joke, "We're identical, Fred and George." "Well, not quite. I'm taller." "I'm the better looking."

The world-building is exceptional, with Rowling masterfully blending fantasy elements and real-life dilemmas. "It takes a great deal of bravery to stand up to our enemies, but just as much to stand up to our friends." - Professor Dumbledore. As the trio confronts challenges, they learn invaluable life lessons and forge unbreakable bonds. "You're a wizard, Harry." - Hagrid

In conclusion, "Harry Potter and the Sorcerer's Stone" is a captivating introduction to the magical universe that has enchanted readers of all ages for years. The book's immersive storytelling, relatable characters, and memorable dialogue make it a timeless classic. "I solemnly swear that I am up to no good." - Fred and George

 

[FrogReaver]

Dont misunderstand me: I am trying to understand the degree to which output from these generative AI can violate copyright, and my question was about actual reproduction. There are other kinds of copyright violations, of course -- derivative works, as an example. It is important to me because I don't think generative AI is inherently bad in and of itself, but I do think violating copyright holders' rights is wrong.

As you get toward more and more obscure concepts it becomes increasingly likely that there are fewer sources present in the model training data of that concept. If the LLM AI is going to make an exact reproduction of an article or image then this is where it would happen.