The real tragedy of the burning of the Library of Alexandria? That all those books couldn’t be used to train LLMs before they went up in flames.
Driving the news: A number of Canadian used booksellers have seen an increased number of mysterious bulk order requests, per a new CBC News report, which they suspect are coming from companies looking to train AI models with the materials before destroying them.
Catch-up: Concerns about AI companies buying, scanning, and then shredding old books first came to the fore in January, when a tranche of documents unsealed as part of a US$1.5 billion copyright lawsuit against Anthropic revealed the company’s secret book-nabbing plot.
“Project Panama is our effort to destructively scan all the books in the world,” read one supervillain-sounding memo. “We don’t want it to be known that we are working on this.”
Concerns grew further last month after a 404 Media report tracked an order of rare books placed by Amazon to a warehouse where thousands were scanned for AI training, then shredded.
Why it matters: Experts believe that the literary capabilities of large-language models degrade when they’re trained on AI-generated texts, which now make up a large percentage of written content on the web. This has created an insatiable appetite from AI companies for older texts — guaranteed to be devoid of AI influence — so they can keep improving models.
Our take: It’s not the destruction of these specific physical books we find alarming — one bookseller told CBC News the books they sold were unwanted and losing them was “not a great loss to civilization” — it’s the growing disregard for human creation as fodder for automation.—QH
