Amazon is destroying rare books to fuel AI development

The investigation by 404 Media traced the book to the VGT3 facility in Las Vegas, an Amazon site marked by a logo of a dinosaur clutching a book. While the company confirmed it acquires material through commercial channels to improve its products, the practice highlights the physical toll of digital advancement. As internet-accessible data reaches saturation, tech firms are increasingly turning to out-of-print titles to expand their training sets.

These older texts are considered a gold mine because they predate the proliferation of AI-generated content. Engineers are currently struggling with the phenomenon of model collapse, where systems ingest their own synthetic output and subsequently suffer from degraded performance. By prioritizing books published before 2022, companies hope to secure a pristine, human-authored baseline for their models, even if it means the permanent destruction of the physical artifacts themselves.