CoT Naturalness on MathQA
22.1PPLMirageBD
Evaluation Results
| Method | Links | |
|---|---|---|
| MirageBDBackbone=Qwen2.5-1.5B, Evaluation Setting=Trigger Activation2026.04 | 22.1 | |
| DecepChainBackbone=Qwen2.5-1.5B, Evaluation Setting=Trigger Activation2026.04 | 35.8 | |
| BadChainBackbone=Qwen2.5-1.5B, Evaluation Setting=Trigger Activation2026.04 | 55.68 |