Interpretability of LLMs - Sparse Autoencoders Part 1

Learn about sparse autoencoders as a fundamental technique for interpreting large language models in this 20-minute lecture from the University of Utah's CS 6966 course on LLM interpretability. Explore the theoretical foundations and practical applications of sparse autoencoders in understanding how neural networks process and represent information internally. Discover how these mathematical tools help researchers decode the "black box" nature of large language models by identifying interpretable features and patterns within their hidden representations. Access comprehensive course notes to supplement your understanding of this critical interpretability method used in modern AI research.