Concept Identifiability on Refusal
0.9MCCLinear Probe
Evaluation Results
| Method | Links | |
|---|---|---|
| Linear ProbeModel Backbone=Gemma-2-2B2025.02 | 0.9 | |
| SSAEModel Backbone=Gemma-2-2B2025.02 | 0.6233 | |
| GemmaScopeModel Backbone=Gemma-2-2B2025.02 | 0.6119 | |
| ReLU-SAEModel Backbone=Gemma-2-2B2025.02 | 0.6116 | |
| JumpReLU SAEModel Backbone=Gemma-2-2B2025.02 | 0.5699 | |
| TopK-SAEModel Backbone=Gemma-2-2B2025.02 | 0.5686 |