Amazon's Surprising Use of Rare Books to Train AI Sparks Debate

Amazon has been discovered acquiring and scanning rare books to train its AI models, a practice revealed by 404 Media through tracking a shipment with a tracking device. This shipment ended at Amazon's facility in Las Vegas, named VGT3, which features a symbol of a dinosaur holding a book.

Key Points:

• Amazon claims it purchases these books through commercial channels to enhance its customer-related products and services.
• Large Language Models (LLMs) require substantial amounts of text, which they typically source from the internet. However, now they're turning to rare books as they've exhausted much of the internet's available content.
• Rare books provide a valuable source of training data as they are likely devoid of AI-generated content, helping to mitigate the risk of "model collapse," a phenomenon where AI models degrade in output quality when trained on too much AI-generated text.
• This practice underscores the ongoing debate about ethical AI training and the exploitation of rare, valuable texts for commercial benefit.