1 paper
Bofan Gong, Shiyang Lai, James Evans +1
Polysemanticity is pervasive in language models and remains a major challenge for interpretation and model behavioral control. Leveraging sparse autoencoders (SAEs), we map the pol…