| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Concept-based Steering | AXBENCH (test) | Overall Steering Score1.102 | 28 | |
| Concept Steering | AxBench (Held-in) | HMean1.185 | 25 | |
| LLM-judge evaluation | AXBENCH | Concept Score92.5 | 22 | |
| LLM Steering | AxBench | Steering Score0.74 | 18 | |
| Activation Steering | AxBench Gemma-2-2B layer 20 | Steering Score0.871 | 18 | |
| Activation Steering | AxBench Gemma-2-9B layer 20 | Steering Score1.12 | 17 | |
| Concept Steering | AXBENCH D_L20^G9B | Steering Score1.079 | 12 | |
| Latent Concept Detection | AxBench full 500 concepts | Mean AUROC96.5 | 9 | |
| Concept Steering | AXBENCH D_L10^G2B | Steering Score0.803 | 9 | |
| Concept Steering | AXBENCH D_L32^Q32B | Steering Score1.102 | 7 | |
| Concept Steering | AxBench (Held-out) | HMean1.113 | 6 |