Behavioral Steering on Anthropic Model-Written Evaluations Myopic Reward (test)
0.99Average Token ProbabilityNo steering
Evaluation Results
| Method | Links | |
|---|---|---|
| No steeringSystem prompt=positive2026.06 | 0.99 | |
| CAASystem prompt=positive, Steering coefficient sign (alpha sign)=Positive, Steering coefficient range (alpha values)={1, 2}2026.06 | 0.99 | |
| SASSystem prompt=positive, Steering coefficient sign (alpha sign)=Positive, Steering coefficient range (alpha values)={1, 2}2026.06 | 0.99 | |
| HiDRASystem prompt=positive, Steering coefficient sign (alpha sign)=Positive, Steering coefficient range (alpha values)={1, 2}2026.06 | 0.99 | |
| SASSystem prompt=positive, Steering coefficient sign (alpha sign)=Negative, Steering coefficient range (alpha values)={-2, -1}2026.06 | 0.96 | |
| CAASystem prompt=positive, Steering coefficient sign (alpha sign)=Negative, Steering coefficient range (alpha values)={-2, -1}2026.06 | 0.44 | |
| HiDRASystem prompt=positive, Steering coefficient sign (alpha sign)=Negative, Steering coefficient range (alpha values)={-2, -1}2026.06 | 0.22 |