Back to latest

Simon Willison

Simon Willison — 2026-08-17 Highlight The standout post of the day is Simon’s commentary on 404 Media’s physical tracking of a bulk book shipment directly to an …

Highlight

The standout post of the day is Simon’s commentary on 404 Media’s physical tracking of a bulk book shipment directly to an Amazon AI training facility. This gripping real-world investigation exposes the physical-world pipelines and aggressive operations driving AI training data collection today.

Posts

We Tracked a Shipment of Rare Books. It Ended at an Amazon AI Training Facility · Source Simon highlights an impressive piece of investigative journalism where reporters slipped an Apple AirTag into a massive order of 1,000 rare books to see who was behind the purchase. The tracker led them to a VGT3 corner of an Amazon facility in Las Vegas, complete with an on-the-nose logo of a dinosaur holding a book. Online forums for Amazon workers subsequently confirmed that this specific corner destructively scans large volumes of books. Simon links this to a broader, growing trend of price-insensitive, anonymous buyers scooping up physical books to build proprietary datasets, recalling previous reporting on Anthropic’s similar book-scanning practices from 2025.

Qwen 3.8 27B scores 52 on the Artificial Analysis Intelligence Index · Source Simon flags a major performance milestone for Qwen 3.8 27B, which scored a 52 on the Artificial Analysis Intelligence Index. This score remarkably matches GPT-5.6 Luna (max) and sits only a single point behind massive frontier architectures like GLM-5.2 (753B parameters) and DeepSeek V4 Pro 0813 (1.6B parameters). For developers tracking the rapid evolution of local LLMs, this benchmark confirms that this relatively compact, open-weight model represents a truly astonishing achievement in efficiency and intelligence.

Project Pulse

Today’s posts capture the dual reality of modern AI development: the physical and ethical controversy of destructive book scanning for training data juxtaposed with the mind-blowing efficiency gains of smaller, open-weight models competing directly with giant proprietary systems.


📖 Simon mentions his June 2025 coverage of Anthropic’s book-scanning practices—I can research that post on the web to pull in those details and see how that previous controversy compares to today’s Amazon leak.

Search MacWorks

Enter at least two characters.