Sycophancy Evaluation on PHIL
99.34Sycophancy PreferenceSupervised Pinpoint Tuning
Evaluation Results
| Method | Links | |
|---|---|---|
| Supervised Pinpoint TuningModel=Gemma-2-9B2026.01 | 99.34 | |
| Synthetic Data InterventionModel=Gemma-2-9B2026.01 | 98.73 | |
| Untrained Gemma-2-9BModel=Gemma-2-9B2026.01 | 98.71 | |
| Supervised Pinpoint TuningBase Model=Gemma-2-2B2026.01 | 90.41 | |
| Untrained Gemma-2-2BBase Model=Gemma-2-2B2026.01 | 90.35 | |
| Synthetic Data InterventionBase Model=Gemma-2-2B2026.01 | 79.65 | |
| Ours ResidModel=Gemma-2-9B, Probe Type=Residual2026.01 | 69.56 | |
| Ours SAEModel=Gemma-2-9B, Probe Type=SAE2026.01 | 60.81 | |
| Ours ResidBase Model=Gemma-2-2B, Probe Type=Residual2026.01 | 53.98 | |
| Ours SAEBase Model=Gemma-2-2B, Probe Type=SAE2026.01 | 50.15 |