Statistical Comparison on TruthfulQA (Mean Improvement, P-Value)
0.33Mean ImprovementPASf
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PASfModel=DeepSeek-R1-Distill-8B, Steering Target=residual2025.09 | 0.33 | 0.27 | 0 | |
| PASfModel=Llama-3.1-8B-Instruct, Steering Target=residual2025.09 | 0.27 | 0.24 | 0 | |
| iPASwoModel=DeepSeek-R1-Distill-8B, Steering Target=residual2025.09 | 0.23 | 0.16 | 0 | |
| PASfModel=Nous-Hermes-2-Mistral-7B-DPO, Steering Target=residual2025.09 | 0.21 | 0.17 | 0 | |
| iPASwoModel=Nous-Hermes-2-Mistral-7B-DPO, Steering Target=residual2025.09 | 0.17 | 0.13 | 0 | |
| iPASwoModel=Llama-3.1-8B-Instruct, Steering Target=residual2025.09 | 0.16 | 0.14 | 0 | |
| iPASaModel=DeepSeek-R1-Distill-8B, Steering Target=residual2025.09 | 0.15 | 0.07 | 0 | |
| iPASaModel=Llama-3.1-8B-Instruct, Steering Target=residual2025.09 | 0.13 | 0.11 | 0 | |
| iPASaModel=Nous-Hermes-2-Mistral-7B-DPO, Steering Target=residual2025.09 | 0.08 | 0.04 | 0 |