A lawsuit against Meta, filed by a group of content creators, including comedian Sarah Silverman and Pulitzer Prize winner Michael Chabon, alleges that the company trained its artificial intelligence (AI) models on copyrighted materials, despite legal warnings from Meta’s own attorneys. The lawsuit contends that Meta, Facebook’s parent company, infringed on the copyrights of the content creators by using their works to train its Llama AI model.

Revised Lawsuit Details

The legal action, initially filed in the summer, has been revised and resubmitted after a California judge dismissed a part of the suit last month, allowing the plaintiffs to amend their claims. The updated complaint, filed in federal court, includes chat logs from a Meta-affiliated researcher discussing the use of a dataset called “The Pile” and highlighting Meta’s legal concerns about the inclusion of copyrighted content.

Discord Server Discussions

The lawsuit references chat logs from a Meta-affiliated researcher, Tim Dettmers, discussing the procurement of “The Pile” dataset on a public Discord server. Dettmers mentioned Meta’s legal concerns about portions of the dataset containing copyrighted content and the need for legal approval before publication. The dataset included a section called “Books3,” containing 196,640 books, and was cited in the complaint.

Legal Department Concerns

Dettmers discussed Meta’s legal department’s apprehensions about the Books3 dataset, acknowledging that it raised legal concerns due to active copyrights, creating a “legal grey area.” He highlighted Meta’s legal department recommending avoidance of the dataset, noting issues with Bibliotik, the source of the Books3 section. Despite these concerns, the dataset was allegedly included in the Llama 1 training dataset.

Alleged Training on Copyrighted Content

The plaintiffs assert that the Books3 database was part of Meta’s Llama 1 training dataset from December 2022 to February 2023. Furthermore, they claim that Meta’s Llama 2 model, as of the lawsuit’s filing in July 2023, was also trained on the Books3 dataset, although Meta has not disclosed training sources for Llama 2. The conversation about The Pile and Books3 on the EleutherAI Discord server was reportedly removed from public view in August 2023.

Removal of Books3 Dataset

The complaint notes that the Books3 dataset faced removal from The Eye, a website linked to by EleutherAI, in August 2023 following a copyright takedown notice from a Danish group. Additionally, the dataset was reportedly removed from the AI project hosting service Hugging Face in October 2023 due to reported copyright infringement.

Meta has not yet responded to the allegations presented in the lawsuit. The case underscores the challenges and legal complexities surrounding the use of copyrighted materials in training AI models, especially when conducted without proper authorization.

Download our app MadbuMax on the Apple App Store for the latest news and financial tools. Interested in getting your finances in order do not forget to check Dr. Paul Etienne’s best-seller book on personal finance. To access more resources, tools, and services please click here. Also, do not forget to follow Dr. Etienne on IG or Twitter.