Amazon has joined a growing list of companies destroying rare books in an effort to gather fresh data for training artificial intelligence systems, according to reporting on the practice.
The extent of the activity came to light after an AirTag was concealed within a rare book. The tracking device followed the book’s journey to a warehouse in Las Vegas, where it was likely destroyed as part of an effort to supply new material to large language models.
Feeding Data to Large Language Models
The demand for source material to train large language models has grown considerably, with AI systems now embedded across news, music and social media. As the appetite for training data increases, physical books have become a source that some companies are willing to destroy in order to digitise their contents.
The book that carried the hidden AirTag was tracked to a Las Vegas warehouse, illustrating how printed material is being routed to facilities where it can be processed and, in many cases, discarded once its contents have been extracted.
The Risk of AI Training on AI Output
Experts have long cautioned about a scenario in which AI systems begin training on data that was itself generated by AI. That threshold appears to have been reached, with a growing volume of online content now produced by artificial intelligence rather than people.
This creates a practical problem. Training AI on AI-generated material has proven to be of limited value, prompting companies to seek out original, human-created sources such as printed books to maintain the quality of the data feeding their models.
The concealment of an AirTag inside a rare book confirmed its route to the Las Vegas warehouse, where it was likely destroyed.
Source
Image: appleinsider.com
