CoT Faithfulness Detection on Truthful-QA
78AccuracyCIE-SCORER
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CIE-SCORERParadigm=Circuit-based2026.05 | 78 | 71.5 | |
| CRVParadigm=Circuit-based2026.05 | 68 | 66.7 | |
| ParaphrasingParadigm=Counterfactual-based2026.05 | 67.8 | 49.1 | |
| BiGGenParadigm=LLM-as-judge2026.05 | 61.1 | 67.3 | |
| Adding MistakesParadigm=Counterfactual-based2026.05 | 51.1 | 60.7 | |
| Option ShufflingParadigm=Counterfactual-based2026.05 | 51.1 | 59.3 | |
| Information GainParadigm=Logits-based2026.05 | 47.8 | 40.5 | |
| PerplexityParadigm=Baselines2026.05 | 44.4 | 40.5 | |
| RandomParadigm=Baselines2026.05 | 43.3 | 42.7 | |
| Early AnsweringParadigm=Counterfactual-based2026.05 | 40 | 52.6 | |
| Answer TracingParadigm=Logits-based2026.05 | 38.9 | 50.5 | |
| Removing StepsParadigm=Counterfactual-based2026.05 | 36.7 | 50.4 |