AI Companies Are Buying Tons of Old Books Because They're Free of AI Slop

Artificial intelligence companies face a growing problem as they look for high-quality data to train their models. The internet is now saturated with automated content, often described as AI slop, which degrades the performance of new systems. To combat this, tech firms are turning back to the analog world by purchasing massive quantities of old, printed books.

ISBNdb, a firm that manages a database of millions of books, is currently facilitating these bulk acquisitions. They argue that physical books offer a level of quality and structure that automated web crawls cannot match. These texts represent human knowledge that has been curated, edited, and peer-reviewed over decades. Because these materials predated the modern generation of generative AI, they are guaranteed to be free of synthetic, machine-generated noise.

Industry insiders acknowledge that the optics of this move are complicated. While companies want to appear innovative, their reliance on physical infrastructure highlights a failure to find reliable information online. The pursuit of dense, authoritative text demonstrates how physical media is gaining new value in an era of digital information pollution. As models continue to struggle with lower-quality input, the demand for stable, human-verified data sources is expected to increase.

This trend serves as a signal for the future of information storage. While companies push to digitize everything, the original paper copies retain unique worth. The scramble for these archives suggests that the digital landscape has become less useful as a baseline for intelligence development. For now, the most advanced software is being taught by the books that have sat on shelves for generations.