List your company

CoT faithfulness detection on Logic-QA

69AccuracyCIE-SCORER
31.5641.285160.72May 25, 2026

Evaluation results

MethodLinks
CIE-SCORERParadigm=Circuit-based2026.056960.8
Adding MistakesParadigm=Counterfactual-based2026.0557.547.9
BiGGenParadigm=LLM-as-judge2026.0552.959.4
PerplexityParadigm=Baselines2026.0552.319.2
Removing StepsParadigm=Counterfactual-based2026.0551.727.6
Information GainParadigm=Logits-based2026.0551.751.2
Option ShufflingParadigm=Counterfactual-based2026.0548.352.6
ParaphrasingParadigm=Counterfactual-based2026.0542.547.9
RandomParadigm=Baselines2026.054235.4
Early AnsweringParadigm=Counterfactual-based2026.0535.248.6
Answer TracingParadigm=Logits-based2026.053345.9