ResearchBenchmarksConditional-independence (belief sufficiency) testing on Diab (Diabetes Patient Records)Follow0.2695CMILlama0.0092920.0768460.14440.211954Feb 6, 2026Evaluation ResultsMethodMethodLinksCMICMI 95% CI RangeLogLoss Improvement (CB: A~p vs A~(p,θ))LogLoss Improvement 95% CI (CB: A~p vs A~(p,θ))LogLoss Improvement (CB: A~(p,x) vs A~(p,x,θ))LogLoss Improvement 95% CI (CB: A~(p,x) vs A~(p,x,θ))LlamaModel=LlamaModel=Llama2026.020.2695—0.1233—-0.0026—GPT-HighModel=GPT-HighModel=GPT-High2026.020.0461—0.0022—0.0043—DeepSeekModel=DeepSeekModel=DeepSeek2026.020.0351—-0.0243—0.0001—GPT-MinModel=GPT-MinModel=GPT-Min2026.020.0193—0.0284—0—