Amazon Cuts Rare Book Spines to Train AI
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Amazon buys rare books, cuts off their spines, and scans the pages for AI training, according to 404 Media, which placed a tracking device in a book that landed at an Amazon facility.
- 404 Media tracked the GPS-tagged book to Amazon's Las Vegas facility VGT3, identified by a dinosaur holding a book in its claws; Amazon told the outlet it "purchases books through commercial channels to improve the products and services customers use."
- LLMs require unfathomably large text datasets and have already ingested what they can from the internet, making rare and out-of-print books — unscannable from the web — a coveted new training source.
- Pre-2022 texts are especially valuable because LLMs training on their own AI-generated output risk "model collapse," the documented degradation of outputs after ingesting too much synthetic text.
- The article notes Anthropic separately has "illegally pirated books" used for AI training, reflecting how the industry's voracious data appetite has pushed it toward both physical destruction and unlicensed copying.
Why it matters: Amazon is physically destroying rare books — slicing spines to scan pages — to feed models that have already exhausted internet-sourced text. The practice reveals how desperate labs are for pre-2022 data and that 'commercial channels' purchases now include one-of-a-kind cultural artifacts, with no public accounting for which titles are lost.
Ask SkimNews
