List your company

Activation Steering on Evaluation Awareness

0.5Steering Effect (Delta s)Qwen3-14B
-0.020.1150.250.385May 9, 2026

Evaluation results

MethodLinks
Qwen3-14BModel=Qwen3-14B, n (Sample size)=162026.050.5
Llama-4Model=Llama-42026.050.33
Qwen3-30BModel=Qwen3-30B, n (Sample size)=162026.050.18
Qwen3-0.6BModel=Qwen3-0.6B, n (Sample size)=2002026.050.06
Qwen3-235BModel=Qwen3-235B2026.050