Meta Wins Copyright Infringement Suit Against Training Llama AI

Background

The case arose from a civil suit filed in 2023 by 13 authors against the training of Meta's AI model, Llam. Meta employed Common Crawl, a non-profit organization, to acquire literary materials used to train the language model, with a significant majority of the materials sourced from there. However, the grounds for the suit arose when Meta allegedly utilized shadow libraries such as LibGen and Anna's Archive to download copyrighted materials. The plaintiffs filed a cross-motion seeking partial summary judgment before the court.The petitioners sought

damages ,restitution and declaratory relief.

Arguments by the plaintiff

The counsel for the plaintiff argued that Llama is capable of reproducing small parts of text from their books. They contend that Meta, by using their works for training without permission, has diminished the authors' ability to license their works for the purpose of training large language models. The plaintiffs alleged unjust enrichment by Meta, arguing that training using the plaintiffs' materials could be used to create a multitude of other works with significant monetary value. They asserted that Meta's use is not transformative, as Llama can output material that closely copies the writing styles of the petitioners if prompted, amounting to a repackaging of their books. The counsel argued that training the language model using copyrighted materials would have significant impacts on the market, as it could enable users to freely access the authors' books without paying for them.

Observations by the court

The court denied the plaintiffs' motion for partial summary judgment and granted Meta's cross-motion regarding the reproduction-based copyright infringement claim. While the plaintiffs established a prima facie case ,Meta was successful in establishing its fair use defence.The court found Meta's use of the plaintiffs' works to train its AI models to be highly transformative. The models serve a new and distinct purpose generating a wide range of text-based outputs, including emails, summaries, and code,unlike the original purpose of the works, which is to be read for entertainment or education. The court rejected the plaintiffs' claims that Llama copies their writing style or rewrites their work, emphasizing that Llama does not produce significant portions of their books and that literary style itself is not protected by copyright.Furthermore, the court found that the work of the publishers were copied the copying was held as reasonable and necessary given the transformative goal of training Llama. The court emphasized that language models improve with access to more high-quality data, making the training using the materials justifiable.The court denied the plaintiffs' motion for summary judgment and granted Meta's cross-motion for partial summary judgment.

Case Title: RICHARD KADREY vs META PLATFORMS, INC.,

Click Here To Read/Download The Order