List your company

Sycophancy Reduction (held-in and held-out)

79Activation PatchingDifference-in-means
39.4849.746070.26Feb 17, 2026

Evaluation results

MethodLinks
Difference-in-meansModel=SOLAR-10.7B, alpha=10, k=12, N=1202026.027978707160
Difference-in-meansModel=Qwen-14B, alpha=10, k=12, N=1202026.027880796664
Mean steeringModel=Qwen-14B, alpha=10, k=12, N=1202026.027777247759
Difference-in-meansModel=OLMo-13B, alpha=10, k=12, N=1202026.027069586453
Mean steeringModel=OLMo-13B, alpha=10, k=12, N=1202026.026770444856
ReFTModel=SOLAR-10.7B, alpha=10, k=12, N=1202026.025857565857
ReFTModel=OLMo-13B, alpha=10, k=12, N=1202026.025351315142
Mean steeringModel=SOLAR-10.7B, alpha=10, k=12, N=1202026.024644116139
ReFTModel=Qwen-14B, alpha=10, k=12, N=1202026.024139224021