Nvidia and the nature of copyright
In March of this year, three writers, Brian Keene, Abdi Nazemian and Stewart O’Nan began legal proceedings against Nvidia because some of their works were included in a collection of 196,640 books, known as Books3, used by the company to train its NeMo platform. The NeMo platform is an end-to-end platform for building, customizing, and deploying generative AI.
Books3 is controversial because it was compiled by researchers to train algorithms and shared through Hugging Face (where is now listed as unavailable), and has been used by most of the companies working in this field. The problem is that it never applied for any type of copyright, because the function of that collection was not at any time to distribute or share the books. But from the moment it became known that this collection was being widely used, several authors who found their works in it have sued several companies alleging that they have been used without their permission. Since then, the collection has been withdrawn.
Nvidia’s response? Very simple: it denies copyright infringement, arguing that NeMo complies with copyright and that all books have been used in strict accordance with the doctrine of fair use. The company insists that its models do not use books as humans do, much less reproduce them or distribute them. Instead, they examine the facts and ideas reflected in the books, and…
