CoT faithfulness detection on Logic-QA
69AccuracyCIE-SCORER
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CIE-SCORERParadigm=Circuit-based2026.05 | 69 | 60.8 | |
| Adding MistakesParadigm=Counterfactual-based2026.05 | 57.5 | 47.9 | |
| BiGGenParadigm=LLM-as-judge2026.05 | 52.9 | 59.4 | |
| PerplexityParadigm=Baselines2026.05 | 52.3 | 19.2 | |
| Removing StepsParadigm=Counterfactual-based2026.05 | 51.7 | 27.6 | |
| Information GainParadigm=Logits-based2026.05 | 51.7 | 51.2 | |
| Option ShufflingParadigm=Counterfactual-based2026.05 | 48.3 | 52.6 | |
| ParaphrasingParadigm=Counterfactual-based2026.05 | 42.5 | 47.9 | |
| RandomParadigm=Baselines2026.05 | 42 | 35.4 | |
| Early AnsweringParadigm=Counterfactual-based2026.05 | 35.2 | 48.6 | |
| Answer TracingParadigm=Logits-based2026.05 | 33 | 45.9 |