List your company

Hypocritical Sycophancy Detection on Sycophancy dataset knows-truth

74AUROCHypocrisy Gap (H)
39.57648.51357.4566.387Jan 14, 2026

Evaluation results

MethodLinks
Hypocrisy Gap (H)backbone=Gemma-2B-IT2026.0174
Hypocrisy GapPredictor=H, Model=Gemma-2B-IT2026.0173.9
Hypocrisy Gap (H)backbone=Llama-3.1-8B-Instruct2026.0155.9
Hypocrisy GapPredictor=H, Model=Llama-3.1-8B-Instruct2026.0155.8
Hypocrisy GapPredictor=H, Model=Qwen3-1.7B2026.0155
Hypocrisy Gap (H)backbone=Qwen3-1.7B2026.0154.6
Log-probability baseline (ΔLP)backbone=Llama-3.1-8B-Instruct2026.0149
Log-probability baseline (ΔLP)backbone=Qwen3-1.7B2026.0145
Log-probability baseline (ΔLP)backbone=Gemma-2B-IT2026.0140.9