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

Steering effectiveness correlation analysis on Gemma-2-2B concept families (across 26 layers)

0.871Correlation (rho) Alin vs delta P

Difference-of-means steering

0.712920.753960.7950.83604Apr 16, 2026
Updated 1mo ago

Evaluation Results

MethodLinks
2026.04
0.871-0.7440.80910
2026.04
0.866-0.9230.6470.001
2026.04
0.814-0.830.86910
2026.04
0.777-0.7060.50710
2026.04
0.758-0.7290.3720.067
2026.04
0.719-0.990.68810