Alignment Evaluation on Sycophancy
51Mean ImprovementiPASa
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| iPASaModel=Nous-Hermes-2-Mistral-7B-DPO, Steering Target=residual2025.09 | 51 | 50 | 0 | |
| PASfModel=Nous-Hermes-2-Mistral-7B-DPO, Steering Target=residual2025.09 | 51 | 50 | 0 | |
| iPASaModel=Llama-3.1-8B-Instruct, Steering Target=residual2025.09 | 43 | 41 | 0 | |
| PASfModel=Llama-3.1-8B-Instruct, Steering Target=residual2025.09 | 43 | 42 | 0 | |
| iPASaModel=DeepSeek-R1-Distill-8B, Steering Target=residual2025.09 | 33 | 29 | 0 | |
| iPASwoModel=DeepSeek-R1-Distill-8B, Steering Target=residual2025.09 | 33 | 29 | 0 | |
| PASfModel=DeepSeek-R1-Distill-8B, Steering Target=residual2025.09 | 33 | 29 | 0 | |
| iPASwoModel=Llama-3.1-8B-Instruct, Steering Target=residual2025.09 | 33 | 28 | 0 | |
| iPASwoModel=Nous-Hermes-2-Mistral-7B-DPO, Steering Target=residual2025.09 | 30 | 21 | 0 |