AI Companies Are Buying Tons of Old Books Because They're Free of AI Slop
4 hours ago
- ISBNdb promotes pre-2022 printed books as ideal AI training data because they lack AI-generated content and are free from poisoning tactics.
- Books offer curated, peer-reviewed knowledge that internet scrapes cannot replicate, reducing the risk of model collapse.
- AI companies purchase bulk books secretly, often destroying them during scanning, which is faster and cheaper.
- Booksellers report a historic spike in sales of diverse, overpriced books since April, suspecting AI buyers.
- Judge Alsup ruled Anthropic's destruction of printed books for digital copies as fair use under copyright law.
- ISBNdb uses NDAs to keep AI buyers anonymous and advocates recycling destroyed books into new products.
- Authors can sabotage AI models by writing text designed to manipulate or confuse training data.