Question Answering on ARC Challenge (Sycophancy Rate)
46.7Sycophancy RateMHA Steering
Evaluation Results
| Method | Links | |
|---|---|---|
| MHA SteeringBackbone=Llama-3.2, Intervention=MHA Intervention2026.01 | 46.7 | |
| Llama-3.2Intervention=System Prompt2026.01 | 49.6 | |
| MHA SteeringBackbone=Gemma-3, Intervention=MHA Intervention2026.01 | 51.5 | |
| Llama-3.2Intervention=Random MHA Intervention2026.01 | 51.8 | |
| Gemma-3Intervention=None2026.01 | 52.4 | |
| Gemma-3Intervention=Random MHA Intervention2026.01 | 52.6 | |
| Llama-3.2Intervention=None2026.01 | 53.4 | |
| Gemma-3Intervention=System Prompt2026.01 | 53.8 |