Mechanistic Interpretability of Antibody Language Models Using SAEs 事件

Mechanistic Interpretability of Antibody Language Models Using SAEs arXiv:2512.05794v3 Announce Type: replace-cross Abstract: Sparse autoencoders (SAEs) are a mechanistic interpretability technique that have been used to provide insight into learned concepts within large protein language models. Here, we employ TopK and Ordered SAEs to investigate autoregressive antibody language models, and steer their generation. We show that TopK SAEs can reveal biologically meaningful latent features, but h