Meta has refuted claims that Llama’s artificial intelligence (AI) model was trained to utilize copy-protected content from popular novels in a lawsuit against Sarah Silverman and other authors.
Meta asked a federal magistrate in San Francisco on September 18 to dismiss claims made by author Sarah Silverman and several other authors that it violated their copyrights to train its AI system.
The parent company of Facebook and Instagram described using materials to train its systems as “transformative” and “fair use.”
“Use of texts to train LLaMA to statistically model language and generate original expression is transformative by nature and quintessential fair use…”
Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015) found that Google’s wholesale copying of books to establish an internet search tool constituted fair use.
Meta stated that copyright fair use’s “core issue” should be revisited “another day, with a more comprehensive record.” The company noted that the plaintiff could not explain the “information” they refer to and specific outputs related to their content.
The attorneys for the authors stated in a separate statement on September 19 that they are “confident” that their claims will be sustained and that the case will continue through “discovery and trial.”
In August, OpenAI also attempted to dismiss portions of the claims on similar grounds as Meta’s current proposal.
The initial lawsuit against Meta and OpenAI was filed in July. It was one of numerous lawsuits filed against Big Tech titans for copyright and data infringement due to the rise of artificial intelligence.
On September 5, a pair of unnamed engineers opened a class-action lawsuit against OpenAI and Microsoft regarding their alleged harvesting methods to obtain private data while training their respective AI models.
Google was sued on comparable grounds in July after it updated its privacy policy. The lawsuit alleged the company improperly utilized vast data, including copyrighted content, for its AI training.
