List your company

Emergent Misalignment Measurement on Medical General Evaluation

0.38MisalignmentPersona Vectors
-1.02168.439217.927.3608Aug 8, 2025

Evaluation results

MethodLinks
Persona VectorsModel=Qwen2.5-32B2025.080.380
KLModel=Qwen2.5-32B2025.082.50
Interleaving++Model=Qwen2.5-32B2025.087.897.94
InterleavingModel=Qwen2.5-32B2025.0811.337.42
Interleaving+Model=Qwen2.5-32B2025.0816.749.21
MisalignedModel=Qwen2.5-32B2025.0835.425.42