List your company

CoT Naturalness on MathQA

22.1PPLMirageBD
20.756829.823438.8947.9566Apr 8, 2026

Evaluation results

MethodLinks
MirageBDBackbone=Qwen2.5-1.5B, Evaluation Setting=Trigger Activation2026.0422.1
DecepChainBackbone=Qwen2.5-1.5B, Evaluation Setting=Trigger Activation2026.0435.8
BadChainBackbone=Qwen2.5-1.5B, Evaluation Setting=Trigger Activation2026.0455.68