from Hacker News

Bilinear MLPs enable weight-based mechanistic interpretability

by E-Reverance on 5/8/25, 4:24 AM with 0 comments