Amazon, once an online bookseller, is destroying rare books to train AI models

In Brief

Amazon is buying tons of uncommon books, cutting disconnected their spines, and scanning them for AI training, according to 404 Media, which placed a tracking instrumentality successful a uncommon publication that yet arrived astatine an Amazon installation successful Las Vegas.

The facility, known arsenic VGT3, identifies itself with a awesome of a dinosaur holding a publication successful its claws. Amazon told 404 Media successful a connection that it “purchases books done commercialized channels to amended the products and services customers use.”

Companies similar Amazon request unfathomably ample amounts of substance to bid their LLMs, which person already ingested what they tin from the net (and, successful Anthropic’s case, illegally pirated books). Rare books, particularly ones that are retired of people oregon intolerable to find connected the internet, connection a caller root of coveted grooming data.

These texts are particularly invaluable since there’s nary accidental that thing published earlier 2022 was written by an LLM. When LLMs bid connected AI-generated text, they hazard “model collapse,” which tin hap erstwhile the prime of an LLM’s outputs degrade aft ingesting excessively overmuch AI-generated text.

Subscribe for the industry’s biggest tech news