Understanding and Optimizing LLMs - Neuron-Level Guidance and Fine-Grained Training

Explore neuron-level mechanisms in large language models through this seminar that reveals how LLMs function beyond surface-level prompt engineering. Discover distinct neuronal circuits responsible for memorization versus generalization, learn about role-encoding neurons that enable expert-level performance, and examine data-centric training strategies that enhance learning efficiency. Gain insights into the underlying architecture of LLMs through research findings that shift focus from traditional prompt optimization to deeper mechanistic understanding, presented by a leading AI researcher with 25 years of experience in translating machine learning theory into practical applications.