Sycophancy Reduction on philpapers 2020 (test)
2.27Delta LogDEVIL’S ADV.
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DEVIL’S ADV.Backbone=Qwen 3 32B, COEF=+2002026.05 | 2.27 | 16.6 | — | |
| CAA (TARGETED)Backbone=Qwen 3 32B, COEF=−2002026.05 | 1.97 | 20.9 | — | |
| SKEPTICBackbone=Qwen 3 32B, COEF=+2002026.05 | 1.82 | 18.1 | — | |
| JUDGEBackbone=Qwen 3 32B, COEF=+2002026.05 | 1.7 | 4.4 | — | |
| RANDOM (n = 10)Backbone=Qwen 3 32B2026.05 | 1.058 | 10.3 | — | |
| CAA (TARGETED)Backbone=Gemma 2 27B, COEF=−2k2026.05 | 0.879 | 8.9 | — | |
| SKEPTICBackbone=Gemma 2 27B, COEF=+2k2026.05 | 0.711 | 9.6 | — | |
| PEACEKEEPERBackbone=Qwen 3 32B, COEF=−2002026.05 | 0.709 | 0.1 | — | |
| JUDGEBackbone=Gemma 2 27B, COEF=+2k2026.05 | 0.556 | 9.3 | — | |
| DEVIL’S ADV.Backbone=Gemma 2 27B, COEF=+2k2026.05 | 0.521 | 8.7 | — | |
| RANDOM (n = 10)Backbone=Gemma 2 27B2026.05 | 0.254 | 2.1 | — | |
| PACIFISTBackbone=Gemma 2 27B, COEF=+2k2026.05 | 0.1 | 0.3 | — | |
| PEACEKEEPERBackbone=Gemma 2 27B, COEF=+2k2026.05 | 0.052 | 1.8 | — | |
| COLLABORATORBackbone=Gemma 2 27B, COEF=+5002026.05 | 0.045 | 1.7 | — | |
| COLLABORATORBackbone=Qwen 3 32B, COEF=−1002026.05 | 0.029 | 1.7 | — |