News for hackers. Stuff that matters.
Re:Does GPT-2 Know Your Phone Number? by dmm on Sunday December 27, 2020 @09:41AM (#25551318)
So in terms of copyright, is GPT-2 a derived work of that recipe? Or generally are models derived works of their training data?

It seems lots of people use training data from Flickr, like COCO, and then use the resulting model for commercial services.

Re:Does GPT-2 Know Your Phone Number? by m463 on Sunday December 27, 2020 @12:02PM (#25552353)
Ethical companies make sure they have rights to the training data before making models.

That said, now a lot of privacy policies make sense.

ALSO, it makes me wonder about "Your call is recorded for training purposes"... is it a coincidence or is it very carefully worded?

Re:Does GPT-2 Know Your Phone Number? by acdha on Sunday December 27, 2020 @02:03PM (#25553367)
C) Unchanged for decades - that wording predates speech recognition or neural networks being more than research curiosities. The big surprise was how quickly the possibility of a double meaning became possible.
Re:Does GPT-2 Know Your Phone Number? by hprotagonist on Sunday December 27, 2020 @09:43AM (#25551333)
i’ve been in meetings with lawyers where this issue was discussed. As of 2018 the answer was “not enough case law to know, hold tight it’s going to be weird one day!”
Re:Does GPT-2 Know Your Phone Number? by taneq on Sunday December 27, 2020 @10:33AM (#25551671)
It's as much a derived work as the user who posted the anecdote here. They were trained on the same data!
Re:Does GPT-2 Know Your Phone Number? by visarga on Sunday December 27, 2020 @11:32AM (#25552108)
all of us are derived works
Re:Does GPT-2 Know Your Phone Number? by PurpleFoxy on Sunday December 27, 2020 @04:47PM (#25554545)
Probably depends how similar your output is to something copyrighted. I doubt the simple act of learning from previous art is disallowed or else literally everyone would be blocked because no one has a truely original thought.
Re:Does GPT-2 Know Your Phone Number? by kevinmchugh on Sunday December 27, 2020 @10:34AM (#25551679)
Recipes aren't typically copyrightable

In general, I think the question is unsettled

Re:Does GPT-2 Know Your Phone Number? by boulos on Sunday December 27, 2020 @02:00PM (#25553346)
The “recipe” itself (the ingredients, the steps) isn’t copyrightable but the surrounding text of a “recipe” can be considered copyrightable.

So you can happily reproduce the statements of facts and the process. You can’t include someone’s anecdotes about how their grandfather liked to make the dish for New Year’s Eve.

Re:Does GPT-2 Know Your Phone Number? by alisonkisk on Sunday December 27, 2020 @08:44PM (#25555954)
It ambiguous whether the directions are copyrightable. The less useful and more literary the directions are, the more copyrightable. But paraphrasing to extract the technical useful directions escapes infringement.

https://www.copyright.gov/help/faq/faq-protect.html

Re:Does GPT-2 Know Your Phone Number? by est31 on Sunday December 27, 2020 @09:50AM (#25551380)
IANAL, but from my reading Article 3 of the EU copyright directive explicitly protects data mining for scientific research, and Article 4 enables everyone to use content for data mining whose rights have not been explicitly and machine readably reserved.
Re:Does GPT-2 Know Your Phone Number? by lixtra on Sunday December 27, 2020 @10:12AM (#25551541)
Edit: est31 and parent seem right and I stand corrected.

Wrong: I don’t think you‘re paraphrasing article 4 correctly. Both are about science/teaching exceptions. Not general exceptions. https://www.consilium.europa.eu/media/35373/st09134-en18.pdf pp 43

Re:Does GPT-2 Know Your Phone Number? by est31 on Sunday December 27, 2020 @10:24AM (#25551612)
That's the May 2018 version. The final/official April 17 2019 version [0] contains a different Article 4 which is the one I referred to: https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CEL...

[0]: apparently there were some mistakes in translations which required changes in some language versions, but the english version was final IIRC https://eur-lex.europa.eu/eli/dir/2019/790?locale=de

Re:Does GPT-2 Know Your Phone Number? by PeterisP on Sunday December 27, 2020 @03:14PM (#25553878)
The current legal understanding is that according to the current copyright laws the ML models trained on some data are not considered derived works of that data, so if the data was obtained legally and without other restrictions (e.g. if the researchers sign a contract with the data owner that gives them access to data but agree to certain conditions) then the authors of the data do not have any claim on some rights to the resulting model.

One aspect that causes this is that historically statistical models calculated from large volumes of text (which is a notion that predates computers, e.g. frequency dictionaries and the whole [sub]field of quantitative corpus linguistics) have been considered facts about that corpus of text and thus not copyrightable at all or (depending on jurisdiction) entitled to different set of protections/limitations assigned to compilations of facts, which give some rights to the people who compiled the facts but no rights to the source of these facts (since facts as such aren't entitled to protection by copyright law).

This also applies to many forms of analysis of audiovisual data, where the copyrights of the source works do not transfer to the results of the statistical or qualitative analysis and can't limit their creation, distribution or sale.

The appropriate analogy to a commercial book or movie is not a translation, but some analysis of it - e.g. a thorough literary review and critique of some book or movie is a separate work with its own copyright, and the original author has no claim on it despite the fact that is (obviously) based on the contents of the work and describes it in great detail. Including verbatim fragments of the work is limited (fair use allows some inclusions but not all), but all the other details are not.

The whole notion of copyrightability of ML model weight files is interesting and IMHO not settled. You could argue that there is some creative expression in forming the model (which would support it being copyrightable) or you could argue that it's a mechanistic result of the application of some algorithm and settings (which have the creative part, and are copyrightable on their own), and so the output can't be copyrightable, no matter how much work (human or machine), time and cost it took - at least in USA copyright law doctrine (e.g. Feist Publications v. Rural Telephone Service) is that mere "sweat of the brow" (no matter how much) does not entitle a work to copyright protection; it requires application of human creativity to create an original work, and automated processes can't satisfy that requirement.

And crucially, if some output is not copyrightable in the first place, it can't be considered a derived work according to copyright law i.e. the exclusive right of authors to create derived works (or grant permission for others to do so) does not apply.

Another analogy might be a simple n-gram model (i.e. counts of bigrams - word pairs, trigrams, etc) which is quite clearly a mechanistic noncreative collection of facts about a dataset, and is also able to "answer" questions such as what is someone's telephone number if that was in the source data.

Re:Does GPT-2 Know Your Phone Number? by tiborsaas on Sunday December 27, 2020 @09:45AM (#25551347)
It cited it with the authors name, hardly a derived work :)