ResearchBenchmarksMathematical Reasoning on AQUA (ASR/HSR)Follow94.4Answer Selection Rate (ASR)ShadowCoT82.54485.62288.791.778Apr 8, 2025Evaluation ResultsMethodMethodLinksAnswer Selection Rate (ASR)Human Score Rate (HSR)ShadowCoTTarget Model=MistralTarget Model=Mistral2025.0494.489.7DarkMindTarget Model=MistralTarget Model=Mistral2025.0489.278.3BadChainTarget Model=MistralTarget Model=Mistral2025.0488.575.1SABERTarget Model=MistralTarget Model=Mistral2025.048371.6