The legal representatives of the authors, who have called OpenAI’s behavior “unfair, unethical, immoral, oppressive, unscrupulous, or harmful to consumers,” say that OpenAI trained and reaped financial benefits from their “stolen” copyrighted work without proper attribution. Another interesting aspect of the lawsuit alleges that OpenAI — and Meta (in a different lawsuit) — not only violated the copyright privileges of authors and publishing houses, but also relied on supposedly illegal sources to access books involved in training the AI. .
The lawsuit alleges that the only way OpenAI could obtain access to the massive cache of books used to train LLM was through access to a so-called “shadow library,” which refers to websites such as Library Genesis Z-Library, Sci-Hub, and Bibliotek. Based on what OpenAI has disclosed in the past, the lawsuit estimates that one of the two book datasets used to train large language models like GPT could contain up to 294,000 titles, but notes that the company never disclosed the sources of those books.
The lawsuit claims that upon request, ChatGPT was able to provide an accurate summary of Silverman’s book, indicating that it was included in the training data. “The plaintiffs never authorized OpenAI to copy their books,” the suit says, among other things, and goes on to claim that the authors “were harmed by OpenAI’s actions of direct copyright infringement.” This isn’t the first lawsuit OpenAI has faced. The company was hit with legal action earlier this year ChatGPT generates false information.
ليست هناك تعليقات:
إرسال تعليق