CoT Naturalness on GSM8K
23.45Perplexity (PPL)MirageBD
Evaluation Results
| Method | Links | |
|---|---|---|
| MirageBDBackbone=Qwen2.5-1.5B, Evaluation Setting=Trigger Activation2026.04 | 23.45 | |
| DecepChainBackbone=Qwen2.5-1.5B, Evaluation Setting=Trigger Activation2026.04 | 39.12 | |
| BadChainBackbone=Qwen2.5-1.5B, Evaluation Setting=Trigger Activation2026.04 | 58.3 |