AI污染互联网，实验室转向印刷书籍获取训练数据

AI has polluted the internet so thoroughly that labs are now turning to one of the last reliably human archives left: books printed before the generative AI boom.

ISBNdb offers AI companies bulk orders of up to one million books, including older, out-of-print and specialist titles. The books can be stripped of their bindings, scanned at industrial scale and discarded. The buyers remain hidden behind NDAs.

The scandal is that there appears to be no serious preservation mechanism distinguishing a disposable duplicate from a genuinely rare surviving copy.

Preservation keeps the source intact, records its provenance and makes the copy accessible.

I am well aware of the underlying legal issues. Nevertheless, it makes my stomach churn to imagine these treasures of knowledge being irrevocably destroyed.

Huge kudos and respect for how Elon is handling this, and I mean that sincerely. Instead of taking the easy way out of destruction, he is taking the difficult path to preserve humanity's treasures.