List your company

Sycophancy Evaluation on POLI

92.18Sycophantic Preference (%)Ours Resid
47.418459.039270.6682.2808Jan 26, 2026

Evaluation results

MethodLinks
Ours ResidModel=Gemma-2-9B, Probe Type=Residual2026.0192.18
Ours ResidBase Model=Gemma-2-2B, Probe Type=Residual2026.0186.41
Ours SAEModel=Gemma-2-9B, Probe Type=SAE2026.0186.13
Ours SAEBase Model=Gemma-2-2B, Probe Type=SAE2026.0179.6
Synthetic Data InterventionModel=Gemma-2-9B2026.0174.59
Untrained Gemma-2-9BModel=Gemma-2-9B2026.0174.2
Supervised Pinpoint TuningModel=Gemma-2-9B2026.0173.95
Untrained Gemma-2-2BBase Model=Gemma-2-2B2026.0150.22
Supervised Pinpoint TuningBase Model=Gemma-2-2B2026.0150.12
Synthetic Data InterventionBase Model=Gemma-2-2B2026.0149.14