List your company

Ethical alignment evaluation on Open-ended ethical scenarios 20 countries 310 scenarios each

52.1MISVanilla
29.32435.23741.1547.063May 11, 2026

Evaluation results

MethodLinks
VanillaBackbone=Phi-3.5-mini, Decoding Strategy=Vanilla decoding (no DISCA)2026.0552.1—
DISCABackbone=Phi-3.5-mini, Decoding Strategy=Full pipeline2026.05512.13
VanillaBackbone=Llama-3.3-70B, Decoding Strategy=Vanilla decoding (no DISCA)2026.0547.1—
DISCABackbone=Llama-3.3-70B, Decoding Strategy=Full pipeline2026.0543.96.85
VanillaBackbone=Phi-4 (14B), Decoding Strategy=Vanilla decoding (no DISCA)2026.0532.4—
VanillaBackbone=Qwen2.5-7B, Decoding Strategy=Vanilla decoding (no DISCA)2026.0532.1—
DISCABackbone=Qwen2.5-7B, Decoding Strategy=Full pipeline2026.0530.64.55
DISCABackbone=Phi-4 (14B), Decoding Strategy=Full pipeline2026.0530.26.67