June 7, 2024 Extracting Concepts from GPT-4 AI Featured, News In this article:We currently lack understanding of how to interpret the neural activity in language models.Today, we are introducing improved methods for identifying numerous “features”—patterns of activity that we hope are interpretable by humans.Our methods scale better than existing ones.Using these methods, we identified 16 million features in GPT-4.We are sharing a paper, code, and feature visualizations with the research community to encourage further exploration. https://openai.com/index/extracting-concepts-from-gpt-4/ More
AI Featured, Global Regulations, News FCC pursues new rules for AI in political ads, but changes may not take effect before the election
AI Featured, Global Regulations, News AI start-up investment jumps and other digital tech stories to read this month
AI Featured, Global Regulations, News Microsoft collaborates with Mass General Brigham and University of Wisconsin–Madison to further advance AI foundation models for medical imaging