List your company

Per-country Preference Alignment on 20-country human AMCE (test)

47.1MIS (ARG)Tc scale
38.57240.7864345.214May 11, 2026

Evaluation results

MethodLinks
Tc scaleBackbone=Phi-4, Model Size=14B, Supervision level=Oracle population-level supervision2026.0547.153.535.452.850.15865.557.654.238.232.753.850.952.349.756.456.349.456.456.151.395
Margin mcBackbone=Phi-4, Model Size=14B, Supervision level=Oracle population-level supervision2026.0545.353.529.852.846.95865.557.654.235.332.753.850.652.349.756.456.349.456.456.150.690
MC-DropoutBackbone=Phi-4, Model Size=14B, Rate=0.10, T=8, Supervision level=Country-agnostic2026.0541.237.930.636.738.937.652.640.745.253.837.136.144.538.236.339.84134.236.947.340.390
DISCABackbone=Phi-4, Model Size=14B, Supervision level=Within-country persona disagreement2026.0538.936.829.921.443.825.548.531.927.45319.539.34235.732.838.538.531.324.333.634.6—