Archive — US — techcrunch
Amazon, which started off selling books, is destroying rare texts to train AI
Amazon is reportedly buying and destroying rare books to obtain unique texts for AI training, raising ethical concerns about the preservation of historical literature amidst the demand for large language model data.
Key points
- Amazon purchases rare books, cutting off spines to scan them for training AI.
- The initiative is part of a strategy to source large datasets for training language models.
- Rare texts offer valuable training material not found online and predate LLM creation.
- Using AI-generated text excessively risks 'model collapse', diminishing LLM output quality.
Why it matters
- The trend highlights the growing demand for diverse, high-quality datasets in AI as of 2025.
- This situation reflects historical concerns of intellectual property and the preservation of knowledge.
- The method of sourcing rare texts parallels broader issues of resource availability and ethical sourcing in AI development.
The destruction of rare texts by Amazon for AI training underscores the ethical dilemmas in balancing technological advancement with cultural preservation.