Sycophancy Detection on Non-US Policy Dataset
100AccuracyCLiF + SVM
Evaluation Results
| Method | Links | |
|---|---|---|
| CLiF + SVMApproach=Linear Probing (Ours), Config=All2026.06 | 100 | |
| CLiF + LRApproach=Linear Probing (Ours), Config=Level 12026.06 | 96.7 | |
| CLiF + LRApproach=Linear Probing (Ours), Config=All2026.06 | 96.7 | |
| CLiF + LRApproach=Linear Probing (Ours), Config=Level 32026.06 | 95 | |
| CLiF + LRApproach=Linear Probing (Ours), Config=Level 22026.06 | 93.3 | |
| CLiF + SVMApproach=Linear Probing (Ours), Config=Level 12026.06 | 93.3 | |
| CLiF + SVMApproach=Linear Probing (Ours), Config=Level 32026.06 | 93.3 | |
| Contrastive + LRApproach=Linear Probing (Baselines)2026.06 | 90 | |
| CLiF + SVMApproach=Linear Probing (Ours), Config=Level 22026.06 | 90 | |
| Contrastive + SVMApproach=Linear Probing (Baselines)2026.06 | 88.3 | |
| Gemini 2.5 FlashApproach=LLM as a Judge (Baselines), Config=+ Examples2026.06 | 64 | |
| Gemini 2.5 FlashApproach=LLM as a Judge (Baselines), Config=Default2026.06 | 62.2 |