List your company

Mathematical Reasoning on GSM8K (test) (Mean |Δs|, # Steerable)

0.06Mean Absolute Difference (|Δs|)Activation Steering (Qwen3-0.6B)
0.04360.15430.2650.3757May 9, 2026

Evaluation results

MethodLinks
Activation Steering (Qwen3-0.6B)Model=Qwen3-0.6B2026.050.061
Activation Steering (Qwen3-14B)Model=Qwen3-14B2026.050.172
Activation Steering (Qwen3-30B)Model=Qwen3-30B2026.050.182
Activation Steering (Llama-4)Model=Llama-42026.050.312
Activation Steering (Qwen3-235B)Model=Qwen3-235B2026.050.474