List your company

Concept Identifiability on Refusal

0.9MCCLinear Probe
0.5553440.6448220.73430.823778Feb 14, 2025

Evaluation results

MethodLinks
Linear ProbeModel Backbone=Gemma-2-2B2025.020.9
SSAEModel Backbone=Gemma-2-2B2025.020.6233
GemmaScopeModel Backbone=Gemma-2-2B2025.020.6119
ReLU-SAEModel Backbone=Gemma-2-2B2025.020.6116
JumpReLU SAEModel Backbone=Gemma-2-2B2025.020.5699
TopK-SAEModel Backbone=Gemma-2-2B2025.020.5686