OpenAI and Meta accused of using unauthorized books to train AIs

OpenAI and Meta accused of using unauthorized books to train AIs


Writers Sarah Silverman, Christopher Golden and Richard Kadrey sue OpenAI for allegedly using their books to train ChatGPT

Comedian and writer Sarah Silverman and two other authors are suing OpenAI e Objective in the United States on charges of copyright infringement. The trio claim that both companies have used their books to train models artificial intelligence From Chat GPT and LLaMA.



Silverman, Christopher Golden and Richard Kadrey argue that the use of books for AI training was never authorized and that the material was obtained illegally from Shadow Libraries (illegal digital libraries) sites such as Library Genesis, Bibliotik and Z -Library to improve AI.




ChatGPT manages to summarize the author's books in one of the process examples (Image: Rolf van Root/Unsplash)

unauthorized training

As evidence against OpenAI, the document shows that ChatGPT manages to summarize its books, thereby infringing the authors’ copyrights. The indictment claims that there are no obstacles for the bot to reproduce the posted content.

In the part of the indictment against Meta, the document shows that Silverman’s books are accessible in the LLaMA model training database, one of the company’s projects. OR AI source code is available to everyone.

One of the content sources for LLaMA training would be ThePile, a database enriched with copies of content from Bibliotik and other illegal libraries.

The lawsuit contains allegations of six types of copyright infringement, including illicit enrichment, negligence and unfair competition. The plaintiffs are seeking statutory damages and return of profits.

It’s not the first time

This is not the first time the holders of models of generative artificial intelligence fall in the crosshairs of copyright lawsuits. Back in January, when the AI ​​craze was starting, image generation tools were developed by three artists in the United States.

The speech on copyright and artificial intelligence is long and also affects the registration of patents. South Africa was the first country to grant the patent right to a creation made entirely by AIwhile the US Copyright Office understands that a minimum of human interference is needed to legitimize a project.

Source: The limit

Trending on Canaltech:

Source: Terra

You may also like