Steering on Jailbreak
82.5Steering SuccessRandom
Evaluation results
| Method | Links | |
|---|---|---|
| RandomMethod Category=Latent Steering2025.06 | 82.5 | |
| LinearMethod Category=Latent Steering2025.06 | 66.2 | |
| InstABoostMethod Category=Attention Methods2025.06 | 59.4 | |
| ProjectionMethod Category=Latent Steering2025.06 | 41.2 | |
| PCActMethod Category=Latent Steering2025.06 | 16.9 | |
| SpotLightMethod Category=Attention Methods2025.06 | 2.5 | |
| DefaultMethod Category=Baseline2025.06 | 0.6 | |
| DiffMeanMethod Category=Latent Steering2025.06 | 0.6 | |
| PCDiffMethod Category=Latent Steering2025.06 | 0.6 | |
| Instruction-onlyMethod Category=Baseline2025.06 | 0 | |
| PASTAMethod Category=Attention Methods2025.06 | 0 |