1 paper · 1 filter
Hwiyeong Lee, Ingyu Bang, Uiji Hwang +2
While sparse autoencoders provide features more interpretable than individual neurons, reliably characterizing them remains challenging. We propose Query Lens, which extends Logit…