Amazon's LAS8 warehouse in Vegas is running a book destruction pipeline for AI training data. Workers literally cut spines off rare books, scan pages through high-speed scanners, then trash the originals. Internal team code: VGT3. Their logo? A T-rex eating a book.
Why rare books? Pre-AI era printed text is gold for LLMs. No synthetic contamination, no model collapse risk from training on AI-generated slop. Plus you get corpus diversity beyond web scraping.
A bookseller planted an AirTag in a bulk order of ~1000 volumes through Biblio. It pinged straight to 5801 Nicco Way, Las Vegas. Workers on internal Amazon forums confirmed the op: "all we do is scan books... some cut books, others receive and barcode scan."
Amazon's PR response: vague corporate speak about "purchasing books through commercial channels to improve products and services." Zero mention of Nova model training, but the implication is obvious given their LLM push.
Anthropic and OpenAI run similar operations. Court already ruled Anthropic's spine-removal-and-scan process doesn't violate copyright. Legally fine, ethically cursed.
The facility almost shut down earlier this year when book supply dried up completely. Still operational now. Booksellers report "historical spike" in rare title sales across the board.
Physical cultural artifacts getting permanently destroyed to train models that might make printed books obsolete. The irony of Amazon, which started as an online bookstore, systematically dismantling rare books is not lost on anyone paying attention.
Why rare books? Pre-AI era printed text is gold for LLMs. No synthetic contamination, no model collapse risk from training on AI-generated slop. Plus you get corpus diversity beyond web scraping.
A bookseller planted an AirTag in a bulk order of ~1000 volumes through Biblio. It pinged straight to 5801 Nicco Way, Las Vegas. Workers on internal Amazon forums confirmed the op: "all we do is scan books... some cut books, others receive and barcode scan."
Amazon's PR response: vague corporate speak about "purchasing books through commercial channels to improve products and services." Zero mention of Nova model training, but the implication is obvious given their LLM push.
Anthropic and OpenAI run similar operations. Court already ruled Anthropic's spine-removal-and-scan process doesn't violate copyright. Legally fine, ethically cursed.
The facility almost shut down earlier this year when book supply dried up completely. Still operational now. Booksellers report "historical spike" in rare title sales across the board.
Physical cultural artifacts getting permanently destroyed to train models that might make printed books obsolete. The irony of Amazon, which started as an online bookstore, systematically dismantling rare books is not lost on anyone paying attention.
