Saturday, 5 April, 2025

Study Suggests OpenAI's AI Models Memorize Copyrighted Content

OpenAI's AI models, including GPT-4, may have memorized portions of copyrighted materials. Researchers from the University of Washington, University of Copenhagen, and Stanford developed a method to detect "memorized" content by analyzing the models' ability to predict uncommon words in text snippets. Findings suggest that GPT-4 retained segments from fiction books and New York Times articles, raising concerns about the use of copyrighted data in AI training.