AI companies are reportedly acquiring large quantities of old books, specifically those published before 2022. The rationale behind this strategy is to obtain training data that is free from the "contamination" of modern AI-generated text, often referred to as "slop." This approach aims to leverage the inherent cleanliness of pre-LLM era physical books as a valuable data source for training AI models. AI
IMPACT AI companies are seeking out pre-2022 physical books as a strategy to acquire cleaner training data, avoiding the issue of AI-generated text contamination.
RANK_REASON The item discusses a trend and strategy related to AI training data, but does not announce a new product, research, or policy.
Read on Mastodon — fosstodon.org →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
