Amazon is actively sourcing rare books to train its artificial intelligence models. Recent investigations tracked shipments of rare texts directly to a facility in Las Vegas identified as VGT3, where the books are processed for data extraction.
These physical volumes are destroyed during the process. The books have their spines cut so they can be scanned into digital formats. This practice highlights the growing demand for clean data sets as existing online resources are exhausted or already saturated with machine-generated content.
Language models face quality degradation when they train on synthetic output. By using rare, out-of-print books published before 2022, companies secure high-quality, human-written text that ensures the integrity of the training data. This strategy provides a fresh source of information for large language models that have moved beyond general web scraping.
Amazon stated that it acquires these materials through standard commercial channels to improve its products and services. The scale of this operation underscores how far companies will go to maintain the quality of their AI systems as competition in the space intensifies.