ResearchBenchmarksPreference Learning on Anthropic HH-RLHF+VI Preference (test)Preference Learning on Anthropic HH-RLHF+VI Preference (test)Follow64Overall AccuracyMC-STL58.860.1561.562.85Jan 10, 2026Evaluation resultsMethodMethodLinksOverall AccuracyGroup AccuracyCalibration SlopeCalibration BiasItem (1-EMD)MC-STL2026.0164641.02-0.010.61phi2026.0160590.940.030.58Maj(Y)2026.0159560.940.020.58