You’re signed up for e-mail updates.
Subpoenaed in the Meta cases, OpenAI says its shadow-library use was limited to GPT-3 and GPT-3.5; Anthropic says no public Claude model was trained on LibGen or PiLiMi
Authors and publishers suing Meta subpoenaed OpenAI and Anthropic for evidence about torrenting and licensing, and the three sides set out their positions in joint letters. OpenAI's lawyers wrote that its use of shadow-library material was limited to GPT-3 and GPT-3.5, that the first never powered ChatGPT and the second was retired years ago, and that its products have had no connection to such libraries for years. Anthropic's lawyers wrote: "No publicly available Claude model was trained on text from the Bartz class repositories," meaning LibGen and the Pirate Library Mirror, that such text was used only for early research models never released, and that it filters outputs. Both said they had already produced their licensing agreements in related cases and resisted further discovery, including on torrent software settings. The magistrate judge denied the motions to compel on September 10. These are lawyers' statements in letters, not sworn testimony.
Loading the rest of the news…
What the labels mean
Every item says what kind of source it rests on. A court ruling or order is the court's own statement. A company filing is what a defendant says about itself. A government filing is a statement by a government agency that is not a party to the case, such as a statement of interest; it is the government's position, not a ruling. An allegation is what a complaint or motion claims and has not been decided. Reported means press reporting we have not confirmed from the primary document. Summaries are our own; follow the links for the source.
This page is general information, not legal advice. An allegation is not a finding. Whether any of this matters for a particular work or claim is a question for a lawyer.