Amazon is purchasing large quantities of rare books, cutting their spines, and scanning the pages for AI training data, according to 404 Media. A tracking device placed inside a rare volume confirmed it arrived at an Amazon facility in Las Vegas known as VGT3, which displays a symbol of a dinosaur holding a book. The company stated it buys books through commercial channels to improve products and services customers use.
This practice addresses a specific technical requirement for large language models. Companies need unfathomably large amounts of text to train these systems after ingesting available internet content. Rare books, especially those out of print, offer a new source of coveted training data. Texts published before 2022 are particularly valuable since there is no chance they were written by an AI. When models train on AI-generated text, they risk model collapse, which occurs when output quality degrades after consuming too much synthetic material.
