AI Companies Buy Printed Books for Training Data as ISBNdb Pitches AI-Slop-Free Knowledge

3 articles · Updated · 404 Media · Jul 22

Printed books are being bought by AI companies as a new training-data source, with the appeal that older volumes predate the flood of AI-generated text now contaminating the web.
ISBNdb, which says it runs the world's largest book database, is marketing high-volume book acquisition services by arguing shelf-bound books offer dense, edited and domain-specific human knowledge.
The pitch reflects a broader data squeeze for model developers, who are looking beyond web crawls for cleaner material to improve AI systems.

Sources

Left50%

Right50%

As AI-generated 'slop' corrupts the web, are physical books the only path left to trustworthy AI models?

Can AI trained on 'pure' old books escape the outdated information and historical biases printed within them?

The AI data gold rush turns to old books, but what is the hidden environmental cost of this new frontier?