ChatGPT has proven itself to be a terrible comedian. So as not-so-successful comedians sometimes do, it stole material. And it had better get itself one of those AI lawyers, because it’s going to court.
Comedian Sarah Silverman and authors Richard Kadrey and Christopher Golden have filed two copyright infringement lawsuits, one against Facebook parent Meta Platforms and one against ChatGPT-maker OpenAI for using their work to train chatbots.
The Meta complaint alleges that Silverman’s memoir The Bedwetter, Golden’s novel Ararat, and Kadrey’s novel Sandman Slim were part of data sets Meta used to train its AI model, LLaMA. Meta said in a research paper that it partly used data from the Books3 section of ThePile from EleutherAI, which is a publicly available data set for training large language models.
The complaint says that Books3 contains torrented content from “notorious shadow library websites.” It offers proof in Exhibit B, where the copyright management information for the three books in question appears within the publicly available data for Books3.
As for OpenAI, it trained on a large chunk of the internet that has been archived starting in 2011 called Common Crawl; WebText 2, which contains outbound links from Reddit that received three or more karma points; Books 1, which are free novels available on the internet; Books 2, an online book repository of unknown content; and the English version of Wikipedia. What is included in Books 2 has been the largest unknown, but the exhibits in the complaint lay out some of what could be in there.
Exhibit B of that complaint details a back-and-forth between prompts given to ChatGPT and answers supplied by the chatbot asking for summaries of Silverman’s memoir The Bedwetter, Golden’s novel Ararat, and Kadrey’s novel Sandman Slim, “something only possible if ChatGPT was trained on Plaintiffs’ copyrighted works,” states the complaint.
Silverman, Kadrey, and Golden are represented by the Joseph Saveri Law Firm and Matthew Butterick, who have filed a prior AI-related lawsuit centered on copyright infringement. In a statement on their LLM Litigation site, Saveri and Butterick say that since the release of ChatGPT, they’ve heard complaints from writers, authors, and publishers. They state: “Much of the material in the training datasets used by OpenAI and Meta comes from copyrighted works—including books written by Plaintiffs—that were copied by OpenAI and Meta without consent, without credit, and without compensation.”
Meta declined to comment. OpenAI and Sarah Silverman did not immediately respond to a request for comment.