Firehose

Several AI companies, including Anthropic, are secretly purchasing and destroying millions of physical books in bulk, a process called "destructive scanning", to feed into their large language models, which can further perpetuate AI contamination. This practice is facilitated by a company called ISBNdb, which promises to obtain "up to one million titles per order" and touts "older, rare and specialist volumes" as premium training data. The destruction of these books is considered a permanent loss, carrying historical and cultural significance that cannot be replaced. AI summary