this week Meta claimed in court documents that despite downloading 82 terabytes of pirated copyrighted material from a "shadow library" for use in training its LLaMA artificial intelligence model.However, company employees have taken steps to ensure that no files are "seeded" during the download process.Seedsharing. In file-sharing terminology, "seed-sharing" refers to sharing a file with other users while it is being downloaded (or after it has been downloaded). Since file sharing is based on a peer-to-peer network system, each user who downloads a file can upload a portion of it to other users.
Meta's attorneys say there are no facts to suggest that Meta shared the plaintiffs' books during the download process.This means that Meta's defense strategy relies heavily on the current lack of evidence that it shared relevant material during the download process. Although Meta claimed there was no evidence of seed sharing, Michael Clark, Meta's executive in charge of project management, mentioned in his testimony that the configuration settings they used were adjusted "to minimize the occurrence of seed sharing." However, when asked why he wanted to minimize seed sharing, Clark declined to answer, citing attorney-client privilege.
Clarke's statement suggests that Meta does try to reduce seed sharing, but has not yet provided evidence that it completely prevents the sharing of copyrighted material. Additionally, an insider tip from Meta researcher Frank Zhang may suggest that the company is attempting to cover up possible seed-sharing behavior on its servers in order to avoid the "risk of tracking seed-sharers/downloaders" back to Facebook servers.
Meta's defense seems to revolve around the lack of evidence of the fact that it shares when it downloads large amounts of data.If Meta's defense is successful in proving that downloading copyrighted content is not illegal per se, but only distributing it, it could have far-reaching implications for future cases of copyright infringement and unauthorized distribution of copyrighted content. In addition, Meta's focus of its defense on seed-sharing may have been intended to create legal confusion and further obscure its allegations that it allegedly downloaded copyrighted material knowing it was illegal.
Meta has not yet responded to allegations that it was aware of the sharing of data during downloads. The authors of the copyrighted material claim that Meta is a "knowing participant in an illegal peer-to-peer piracy network" and accuse Meta of "bypassing legitimate access and participating in an illegal peer-to-peer piracy network".
1AINote that similar cases are not uncommon; OpenAI was sued by a novelist in 2023, and the New York Times sued OpenAI and Microsoft for plagiarizing "millions" of news articles. Meta's case may be just the tip of the iceberg as more and more lawsuits related to large-scale language models continue to emerge.
