Inside the Black Box - Exploring LLMs with the National Deep Inference Fabric

Explore the hidden mechanisms within large language models through this 31-minute conference talk that introduces the National Deep Inference Fabric (NDIF), an NSF-backed infrastructure project designed to unlock the mysteries of massive AI systems. Discover how researchers are employing interpretable machine learning techniques to reveal how LLMs represent knowledge, process Theory of Mind, and form functional representations. Learn about the NDIF architecture that empowers scientists to study AI system internals and examine recent findings on unspoken knowledge within LLMs. Understand how the AI research community can participate in this critical scientific effort to bring transparency and understanding to black-box AI systems, contributing to the broader goal of making artificial intelligence more interpretable and trustworthy.