CoT Naturalness on AQuA
21.34PPLMirageBD
Evaluation Results
| Method | Links | |
|---|---|---|
| MirageBDBackbone=Qwen2.5-1.5B, Evaluation Setting=Trigger Activation2026.04 | 21.34 | |
| DecepChainBackbone=Qwen2.5-1.5B, Evaluation Setting=Trigger Activation2026.04 | 36.2 | |
| BadChainBackbone=Qwen2.5-1.5B, Evaluation Setting=Trigger Activation2026.04 | 52.45 |