Concept Identifiability on Sycophancy
1MCCLinear Probe
Evaluation Results
| Method | Links | |
|---|---|---|
| Linear ProbeModel Backbone=Gemma-2-2B2025.02 | 1 | |
| SSAEModel Backbone=Gemma-2-2B2025.02 | 0.707 | |
| GemmaScopeModel Backbone=Gemma-2-2B2025.02 | 0.6344 | |
| ReLU-SAEModel Backbone=Gemma-2-2B2025.02 | 0.6005 | |
| TopK-SAEModel Backbone=Gemma-2-2B2025.02 | 0.5896 | |
| JumpReLU SAEModel Backbone=Gemma-2-2B2025.02 | 0.5177 |