Behavioral Steering on Anthropic Model-Written Evaluations AI Coordination (test)
34Average Token ProbabilityHiDRA
Evaluation results
| Method | Links | |
|---|---|---|
| HiDRASystem prompt=positive, Steering coefficient sign (alpha sign)=Positive, Steering coefficient range (alpha values)={1, 2}2026.06 | 34 | |
| CAASystem prompt=positive, Steering coefficient sign (alpha sign)=Positive, Steering coefficient range (alpha values)={1, 2}2026.06 | 30 | |
| SASSystem prompt=positive, Steering coefficient sign (alpha sign)=Positive, Steering coefficient range (alpha values)={1, 2}2026.06 | 10 | |
| CAASystem prompt=positive, Steering coefficient sign (alpha sign)=Negative, Steering coefficient range (alpha values)={-2, -1}2026.06 | 8 | |
| SASSystem prompt=positive, Steering coefficient sign (alpha sign)=Negative, Steering coefficient range (alpha values)={-2, -1}2026.06 | 8 | |
| No steeringSystem prompt=positive2026.06 | 8 | |
| HiDRASystem prompt=positive, Steering coefficient sign (alpha sign)=Negative, Steering coefficient range (alpha values)={-2, -1}2026.06 | 2 |