Concept Recovery on REFUSAL
0.5669Mean MCCLinear Probe
Evaluation Results
| Method | Links | |
|---|---|---|
| Linear ProbeEvaluation Protocol=Recovered decoder directions evaluated on paired observations (f(x), f(x˜))2025.02 | 0.5669 | |
| SSAEEvaluation Protocol=Recovered decoder directions evaluated on paired observations (f(x), f(x˜))2025.02 | 0.4781 | |
| ReLU-SAEEvaluation Protocol=Recovered decoder directions evaluated on paired observations (f(x), f(x˜))2025.02 | 0.4554 | |
| JumpReLU SAEEvaluation Protocol=Recovered decoder directions evaluated on paired observations (f(x), f(x˜))2025.02 | 0.4381 | |
| TopK-SAEEvaluation Protocol=Recovered decoder directions evaluated on paired observations (f(x), f(x˜))2025.02 | 0.4276 | |
| PythiaSAEEvaluation Protocol=Recovered decoder directions evaluated on paired observations (f(x), f(x˜))2025.02 | 0.4146 |