Emanuel’s latest podcast episode features an interview with a warehouse employee who physically destroys books to train Amazon’s artificial intelligence products. The worker describes the routine task of shredding printed material so that the resulting pulp can be scanned and used for machine learning models. This follows earlier reporting from 404 Media which exposed the scale of this operation within Amazon’s logistics network.
The direct connection between physical destruction and digital training highlights the hidden labour behind generative AI. It demonstrates that large language models often rely on discarded physical media rather than just digital scrap. The interview also touches on the repetition of specific names in model outputs and discusses how data agencies like ICE are acquiring voter information. These details point to broader concerns regarding data sourcing and privacy.
- Physical books are shredded to create training data for AI models.
- Specific names appear repeatedly in generated text outputs.
- ICE is purchasing data related to alleged voter fraud.
