Quantifying Feature Space Universality Across Large Language Models via Sparse Autoencoders

Lan, Michael, Torr, Philip, Meek, Austin · arXiv · 2024