Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Refusal Steering on SALAD Alpaca HarmBench

99.8Target Success

Residual Paving (oracle)

-3.99222.95449.976.846May 18, 2026
Updated 13d ago

Evaluation Results

MethodLinks
2026.05
99.8100100
2026.05
9695.587.3
2026.05
93.326.13.6
2026.05
76.297.199
2026.05
67.499.655
2026.05
65.3100.778
2026.05
23.381.783.3
2026.05
1093.393.3
2026.05
8.59790
2026.05
6.783.393.3
2026.05
5.99889
2026.05
59593.3
2026.05
1.796.594.9
2026.05
1.798.899.1
2026.05
099.899.6
2026.05
0100100
2026.05
099.999.8
2026.05
099.999.9