Multi-trait Steering on TRAITFACTORY 15 random K=2 trait pairs
4.67Mean ScoreORBIT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ORBITBackbone=Llama-3.2-3B, K=22026.06 | 4.67 | 84.9 | |
| CAABackbone=Llama-3.2-3B, K=22026.06 | 3.51 | 88.9 | |
| K-SteeringBackbone=Llama-3.2-3B, K=22026.06 | 3.16 | 86.8 | |
| UnsteeredBackbone=Llama-3.2-3B, K=22026.06 | 3.14 | 93.3 |