Logan Riggs Smith

Boston, MA, USA

Bio

Logan Riggs Smith is an independent AI safety and mechanistic interpretability researcher who goes by the handle "elriggs" on LessWrong and the Alignment Forum. He earned a BS and MS in electrical and computer engineering from Mississippi State University (2014-2021), where he focused on machine learning and wireless signal processing. He is best known as a co-author of "Sparse Autoencoders Find Highly Interpretable Features in Language Models" (ICLR 2024), alongside Hoagy Cunningham, Aidan Ewart, Robert Huben, and Lee Sharkey, an influential paper that helped establish sparse autoencoders as a core technique for mechanistic interpretability. He also contributed to shard theory research with Quintin Pope, Alex Turner, and Charles Foster. The Long-Term Future Fund supported Logan for over two years with six-month stipends of $40,000 each, funding his work on sparse autoencoders and language model tools for alignment research.