AI Raises Copyright Concerns as Models Mimic Bestsellers
New studies reveal top AI models can generate near-verbatim copies of novels, undermining claims they don't store copyrighted works. Experts warn of serious legal implications.
In a troubling development, recent studies have revealed that the world's leading artificial intelligence (AI) models possess the capability to generate near-verbatim copies of bestselling novels from their training data. This discovery raises fresh questions about the industry's long-standing assertion that their systems do not store copyrighted works.
The series of studies, conducted across various AI research groups, including OpenAI, Google, Meta, Anthropic, and xAI, have uncovered a concerning trend: these large language models (LLMs) are able to memorize far more of their training data than previously acknowledged.
This newfound
Source: Ars Technica
