Court says Mark Zuckerberg approved Meta’s use of noise tools to train Llama AI

Meta intentionally used fraudulent tools to train its Llama AI models — with the approval of company CEO Mark Zuckerberg — according to an ongoing lawsuit against the company. Like Results TechCrunch report, and plaintiffs in Kadrey v. Trim a case was sent court documents about the company using LibGen AI training dataset.

LibGen is often described as an “image library” that provides file sharing for academic and hobbyist books, magazines, photos and other resources. Counsel for the plaintiffs, which include authors Sarah Silverman and Ta-Nehisi Coates, accused Zuckerberg of agreeing to use LibGen in teaching despite the company’s executives and employees complaining that it is a “dataset (they) know is pirated.”

The company removed spying data from LibGen’s equipment, the complaint said, before giving it to Llama. Meta apparently admitted in a court filing that it had “delete(ed) all copyright clauses from beginning to end” of scientific articles. One of its engineers claims to have created a script to automatically remove copyright notices. The judge said that Meta did this to hide from the people what they are doing in violation of their rights. Additionally, the counsel noted that Meta admitted to downloading the LibGen tools, even though its engineers were upset about sharing them “from (Meta’s) company laptops.”

Silverman, along with other authors, opposed Meta and OpenAI for copyright infringement in 2023. They accused the company of using hacked tools from shady libraries to train their AI models. The court had previously dismissed some of their claims, but the plaintiffs said their amended complaint is consistent with the allegations and addresses the reasons the court dismissed them.