Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?
This is your opportunity to ask questions. No question too simple or too silly.
Culture war topics are accepted, and proposals for a better intro post are appreciated.
Jump in the discussion.
No email address required.
Notes -
Do you consider work generated by ChatGPT to be theft of intellectual property of the writers who wrote the text used in the training data? I’m fairly confident in saying its not because transformers generate fairly novel content word-by-word, especially for fiction prompts, but am open to being convinced otherwise. If you agree with me, what’s a simple way of explaining to normies that ChatGPT-generated content is not IP theft?
I'm on the pro-side, on average, but I think presenting a legal argument to people who are making pragmatic ones isn't very productive. Opponents of ML-generated art or text are pretty explicitly not focused on the legal questions for any reason but thinking (imo, wrongly) that they can use them.
In hindsight I shouldn’t have used the words “intellectual property” in my comment since it’s more of a legal term, my question was more about “is it stealing/plagiarism in the same sense that it would be if a human produced the same work”.
That's fair, but I don't think it's still the core disagreement. Cfe: here to here, or here for some anti-ML perspectives.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
IP theft is what the government says IP theft is; it is exclusively a construction without a basis in wider reality.
More options
Context Copy link
In some sense, it is whatever the courts decide, since this is a legal issue and all laws are made up.
But IMO no, not at all.
I don't know about chatGPT, but the stable diffusion model is about 4GB in size. There is not enough room in the model for it to be directly plagiarising.
More options
Context Copy link
What is the legal identity criterion of textual copyright?
E.g. Let's say you wrote a book.
I take it and change one word.
Is it still your book or is it now mine or public domain?
2 words?
100 words?
Is there a percentage?
Does the location matter?
If i change words mostly at the start of the book or throughout it?
Does semantics matters? Can i via a software replace some words by identical synonyms or do i need to change semantics?
I have no clue how the legal system solves this major problem.
I would probably enforce the use of
the sota in https://paperswithcode.com/sota/semantic-textual-similarity-on-sts-benchmark and set a magic number percentage. Although it can be gamed that's probably much more accurate than whatever is being used now.
Juries. Sometimes they get it wildly wrong.
You're applying math thinking to a common law legal problem. Rewording the sentences won't be enough if it's clear you started with the other text.
The actual text doesn't always matter. Harlan Ellison is pretty famous for suing over copied story elements, https://en.wikipedia.org/wiki/Harlan_Ellison#Copyright_suits
Can a jury be sued for criminal incompetence?
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
No, that's the most ridiculous thing I've ever heard. Is it theft to be inspired by someone else's work?
Would you consider it different than the art generators? In some circles, those have been receiving quite a lot of hate recently.
More options
Context Copy link
More options
Context Copy link
This conundrum is the intersection of a few very fuzzy definitional categories. "What is IP?" and "What is theft?" and partially "What is learning?". You are not going to get any concrete answers.
For IP Theft:
Against IP Theft:
I’m not sure I understand the Against point here
Legislating weights in a Neural Network is hard. The work doesn't remain in any recognizable form such as in music sampling.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link