Bilinear MLPs enable weight-based mechanistic interpretability

1 point by E-Reverance 2 months ago | 0 comments