Nazemian v. NVIDIA Corporation

Background

The plaintiffs are Abdi Nazemian, Brian Keene, Stewart O’Nan, and other proposed class members who are authors of books that allegedly were pirated and included in a training dataset used by NVIDIA Corporation for its NEMO Megatron-GPT large language models.

The complaint identified Books3, which the opinion describes as a copy of “The Pile,” as a database allegedly used for the training material. The plaintiffs later sought discovery about NVIDIA’s possible use of six other databases containing pirated material. They argued that proposed search terms concerning those databases were relevant and would not impose an undue burden on NVIDIA. They also argued that NVIDIA had limited their ability to identify other datasets by declining to identify the datasets used to train its models or to run searches designed to find those datasets.

Judge Kim denied the request for discovery concerning the additional databases. Judge Kim ruled that, because the plaintiffs alleged that the models were trained on Books3, discovery could appropriately be limited to that database. Judge Kim also found that the plaintiffs were not entitled to information about any dataset used by NVIDIA and that the request lacked sufficient boundaries.

Motion for Relief

The plaintiffs timely asked the district court for relief from Judge Kim’s non-dispositive pretrial order. A non-dispositive pretrial order resolves an issue during the case without finally deciding the claims. The plaintiffs argued that Judge Kim committed legal error by limiting discovery to information known at the pleading stage and abused her discretion by failing to address the plaintiffs’ relevance arguments despite the asserted lack of burden on NVIDIA.

The plaintiffs also argued that limiting discovery to databases they already knew NVIDIA used conflicted with copyright-infringement precedent and undermined the purpose of discovery. NVIDIA responded that the plaintiffs had not shown that Judge Kim’s order was clearly erroneous or contrary to law.

Legal Standard

Under Federal Rule of Civil Procedure 72(a) and 28 U.S.C. § 636(b)(1)(A), a district court must defer to a magistrate judge’s non-dispositive order unless it is clearly erroneous or contrary to law. A decision is contrary to law if it applies an incorrect legal standard or fails to consider an element of the applicable standard. Factual findings are reviewed under the clearly erroneous standard, while legal conclusions are reviewed independently.

Federal Rule of Civil Procedure 26(b)(1) permits discovery of nonprivileged information relevant to a claim or defense and proportional to the needs of the case. Proportionality considers factors including the importance of the issues, the amount in controversy, the parties’ access to information and resources, the importance of the discovery, and whether its burden or expense outweighs its likely benefit.

Court’s Analysis

The court held that Judge Kim did not commit legal error or abuse her discretion by refusing to compel searches concerning the six additional databases. The court concluded that the additional discovery was based on infringement theories beyond the scope of the allegations in the complaint.

The court noted that courts in the district had routinely denied discovery concerning theories exceeding the scope of the operative complaint. It held that the magistrate judge acted within her discretion by limiting discovery to the specific infringement allegations made in the complaint. The court also rejected the plaintiffs’ argument that Judge Kim’s order improperly limited the proposed class of authors whose works might appear in the other databases, explaining that the limits on the proposed class arose from the plaintiffs’ own operative complaint.

Disposition

The court denied the plaintiffs’ motion for relief from the non-dispositive pretrial order.