1 paper · 1 filter
Danielle Ensign, Adrià Garriga-Alonso
How well will current interpretability techniques generalize to future models? A relevant case study is Mamba, a recent recurrent architecture with scaling comparable to Transforme…