Anthropic Destroys Paper Books for Claude?
Anthropic
Anthropic launched Project Panama, destructively scanning millions of purchased paper books to create a digital library for training AI models. A US court ruled this format-shifting with destruction of the physical copy is fair use, but the company still paid $1.5 billion for pirated books obtained earlier.
In 2024, Anthropic initiated Project Panama, a large-scale effort to destructively scan millions of paper books. The company bought books, cut off spines, used industrial scanners, and recycled the paper, aiming to create a one-to-one digital replacement. Court documents revealed Anthropic had previously downloaded over seven million pirated books from sources like LibGen. Judge William Alsup ruled that training LLMs on legally purchased and destructively scanned books is fair use, as the digital copy replaces the physical one. However, the storage and use of pirated copies were not excused, leading to a $1.5 billion settlement covering about 482,000 works. The case has prompted other AI companies to explore similar physical book sourcing, and booksellers report unusual bulk orders of rare titles, raising concerns about cultural preservation.
- Сокращения
- LLM = Large Language Model — большая языковая модель
Source: Habr — хаб ИИ —
original