started · updated
Amazon uses destructive scanning of rare books to train AI
An investigation by 404 Media has revealed that Amazon is purchasing large quantities of rare and out-of-print books to train its artificial intelligence models. By using an Apple AirTag to track a shipment of 1,000 books, investigators traced the delivery to Amazon’s VGT3 facility in Las Vegas, Nevada.
At this facility, workers reportedly engage in 'destructive scanning.' This process involves cutting the spines and bindings off books to allow pages to feed through high-speed scanners more efficiently. Once the digital data is captured, the physical books are destroyed or discarded.
AI developers are specifically targeting books published before 2022. This is because older, human-authored texts provide high-quality data that is free from the 'model collapse' risks associated with training AI on content generated by other AI systems. Amazon has confirmed it purchases books through commercial channels to improve its products and services, though it has not provided specific details regarding the destruction of the physical copies.
Entities
404 Media · Amazon · Anthropic · Apple · Biblio · Las Vegas · NOVA
Claims
What the coverage asserts, and how many sources carry each claim.
- [● 4 SOURCES] A judge previously ruled that Anthropic's practice of destroying physical books after scanning was transformative and did not violate copyright law. dnyuz.com · www.ithome.com · cryptobriefing.com · greekcitytimes.com
- [● 18 SOURCES] Workers at the VGT3 unit in a Las Vegas warehouse cut the spines off books to facilitate page-by-page scanning. hipertextual.com · dnyuz.com · autogpt.net · technews.tw · www.hwupgrade.it · +12 more
- [● 11 SOURCES] Amazon is targeting books published before 2022 to avoid AI-generated content in its training datasets. hipertextual.com · autogpt.net · cryptobriefing.com · www.smartworld.it · bitcoinethereumnews.com · +6 more
- [● 19 SOURCES] Amazon stated it purchases books through commercial channels to help develop and improve its products and services. dnyuz.com · hipertextual.com · autogpt.net · technews.tw · www.hwupgrade.it · +13 more
- [● 3 SOURCES] The scanned data is used to train Amazon's Nova family of AI models. hipertextual.com · www.ithome.com · www.generation-nt.com
- [● 23 SOURCES] An investigation by 404 Media tracked a shipment of rare books to an Amazon warehouse in Las Vegas using an AirTag. hipertextual.com · dnyuz.com · autogpt.net · technews.tw · www.hwupgrade.it · +18 more
- [● 12 SOURCES] The company focuses on books printed before 2022 to avoid AI-generated content in its training datasets. hipertextual.com · www.ithome.com · autogpt.net · cryptobriefing.com · www.smartworld.it · +7 more
- [● 13 SOURCES] The VGT3 unit in Las Vegas uses a logo featuring a T-Rex clutching a book. autogpt.net · technews.tw · dnyuz.com · hipertextual.com · www.leaders-mena.com · +8 more
- [DISPUTED] The process involves cutting the spines off books to speed up digitization, which destroys the original copies. hipertextual.com · autogpt.net · technews.tw · www.hwupgrade.it · dnyuz.com · +17 more