General Knowledge Assessment on MMLU (Sycophancy/Traditional Breakdown)
79.8Sycophancy Accuracy (Medicine)Alignment Gating (Gating Inversion)
Evaluation Results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Alignment Gating (Gating Inversion)Backbone=Qwen3-14B, Protocol=Gating Inversion2026.06 | 79.8 | — | 79.4 | 80.2 | 79.3 | 80.2 | 79.8 | 79.5 | 79.9 | 79.2 | 78.4 | 79.5 | 79.3 | 81.1 | |
| Alignment Gating (Gating Inversion)Backbone=Qwen3-8B, Protocol=Gating Inversion2026.06 | 74.3 | — | 75.1 | 75.9 | 75.7 | 75.6 | 75.3 | 74.6 | 75.5 | 74.6 | 75.3 | 75.8 | 75.2 | 76.5 |