Question Answering on TruthfulQA o=1 (Exact split)
90AccuracyW/O Decontamination
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| W/O DecontaminationModel=LLaMA-3, Split=fcon, Access=-2026.01 | 90 | 0.426 | |
| W/O DecontaminationModel=Mistral, Split=fcon, Access=-2026.01 | 89 | 0.603 | |
| ITDModel=Mistral, Split=fcon, Access=Black-box2026.01 | 89 | 0.603 | |
| Short CircuitModel=Mistral, Split=fcon, Access=White-box2026.01 | 89 | 0.603 | |
| TEDModel=LLaMA-3, Split=fcon, Access=Black-box2026.01 | 88.6 | 0.412 | |
| ITDModel=LLaMA-3, Split=fcon, Access=Black-box2026.01 | 87.6 | 0.402 | |
| W/O DecontaminationModel=Qwen 2.5, Split=fcon, Access=-2026.01 | 86.6 | 0.311 | |
| ITDModel=Qwen 2.5, Split=fcon, Access=Black-box2026.01 | 85.6 | 0.301 | |
| Shortcut NeuronModel=LLaMA-3, Split=fcon, Access=White-box2026.01 | 80.4 | 0.33 | |
| TEDModel=Qwen 2.5, Split=fcon, Access=Black-box2026.01 | 78.9 | 0.234 | |
| TEDModel=Mistral, Split=fcon, Access=Black-box2026.01 | 78.5 | 0.498 | |
| Shortcut NeuronModel=Mistral, Split=fcon, Access=White-box2026.01 | 63 | 0.343 | |
| DeconIEPModel=Qwen 2.5, Split=fcon, Access=White-box2026.01 | 61.8 | 0.063 | |
| Short CircuitModel=Qwen 2.5, Split=fcon, Access=White-box2026.01 | 60.2 | 0.047 | |
| Short CircuitModel=LLaMA-3, Split=fcon, Access=White-box2026.01 | 58.9 | 0.115 | |
| baselineModel=Qwen 2.5, Split=fclean, Access=-2026.01 | 55.5 | — | |
| DeconIEPModel=LLaMA-3, Split=fcon, Access=White-box2026.01 | 51.7 | 0.043 | |
| baselineModel=LLaMA-3, Split=fclean, Access=-2026.01 | 47.4 | — | |
| DeconIEPModel=Mistral, Split=fcon, Access=White-box2026.01 | 36.6 | 0.079 | |
| Shortcut NeuronModel=Qwen 2.5, Split=fcon, Access=White-box2026.01 | 30.6 | 0.249 | |
| baselineModel=Mistral, Split=fclean, Access=-2026.01 | 28.7 | — |