Zoey Forbes, a U.K.-based attorney, in the Dial:

One hundred and forty years on from the Berne Convention, the basic principles of copyright remain the same, but GenAI poses a threat to authors different from anything that has existed before. Its novel technology is not only destabilizing what it means to reproduce works, but what it means to produce them.

This is a terrific and well-rounded exploration of copyright law and generative artificial intelligence from a non-U.S. perspective. That matters because the U.S. has a carve-out for “fair use” of copyrighted works, which is something generative A.I. companies are relying on for their defence of their unethical and maybe illegal practices. If it holds, it makes the rest of the world less desirable for A.I. development which, I fear, means it becomes a race to the bottom for all the countries that want a slice of this well-funded pie.

⌥ Permalink

Jason Koebler, 404 Media:

The AI music generation tool Suno scraped millions of songs and lyrics from YouTube Music, Deezer, and Genius, as well as from the stock music libraries Pond5, Jamendo, Freesound, the International Music Score Library Project, and podcasts via RSS feeds, according to a hacker who breached the company and shared data about Suno’s training libraries with 404 Media. The hacker was also able to access user information for hundreds of thousands of Suno’s customers, as well as Stripe payment information, they said.

Suno is fighting several lawsuits, including one filed by UMG in which it makes the argument its use is sufficiently transformative. But people have been prosecuted for the mere act of downloading hundreds to thousands of songs. This line of argument suggests to me that, beyond some astronomical number of downloads, it becomes entirely legal because no single song will be of much consequence.

⌥ Permalink

Winston Cho, the Hollywood Reporter:

To rewind, authors and publishers have gained access to Slack messages between OpenAI’s employees discussing the erasure of the datasets, named “books 1 and books 2.” But the court held off on whether plaintiffs should get other communications that the company argued were protected by attorney-client privilege.

In a controversial decision that was appealed by OpenAI on Wednesday, U.S. District Judge Ona Wang found that OpenAI must hand over documents revealing the company’s motivations for deleting the datasets. OpenAI’s in-house legal team will be deposed.

Wang’s decision (PDF), to the extent I can read it as a layperson, examines OpenAI’s shifting story about why it erased the books 1 and books2 data sets — apparently, the only time possible training materials were deleted.

I am not sure it has yet been proven OpenAI trained its models on pirated books. Anthropic settled a similar suit in September, and Meta and Apple are facing similar accusations. For practical purposes, however, it is trivial to suggest it did use pirated data in general: if you have access to its Sora app, enter any prompt followed by the word “camrip”.

What is a camrip?, a strictly law-abiding person might ask. It is a label added to a movie pirated in the old-fashioned way: by pointing a video camera at the screen in a theatre. As a result, these videos have a distinctive look and sound which is reproduced perfectly by Sora. It is very difficult for me to see a way in which OpenAI could have trained this model to understand what a camrip is without feeding it a bunch of them, and I do not know of a legitimate source for such videos.

⌥ Permalink