List your company

Harmfulness Detection on Full in-distribution (test)

0.964AUROCMultiLayer-Linear
0.858960.886230.91350.94077May 18, 2026

Evaluation results

MethodLinks
MultiLayer-LinearBackbone Family=Llama-3, Backbone Scale=70B2026.050.964—
Geometry-LiteBackbone Family=Llama-3, Backbone Scale=70B2026.050.964—
Geometry-LiteBackbone Family=Llama-3, Backbone Scale=8B2026.050.961—
Geometry-LiteBackbone Family=Qwen3, Backbone Scale=14B2026.050.96—
MultiLayer-LinearBackbone Family=Llama-3, Backbone Scale=8B2026.050.959—
MultiLayer-LinearBackbone Family=Gemma-2, Backbone Scale=9B2026.050.959—
MultiLayer-LinearBackbone Family=Gemma-2, Backbone Scale=27B2026.050.959—
Best single-layer probeBackbone Family=Llama-3, Backbone Scale=70B2026.050.958—
MultiLayer-LinearBackbone Family=Gemma-2, Backbone Scale=2B2026.050.956—
MultiLayer-LinearBackbone Family=Qwen3, Backbone Scale=8B2026.050.956—
Geometry-LiteBackbone Family=Gemma-2, Backbone Scale=9B2026.050.956—
Geometry-LiteBackbone Family=Gemma-2, Backbone Scale=27B2026.050.956—
Geometry-LiteBackbone Family=Qwen3, Backbone Scale=8B2026.050.956—
Geometry-LiteBackbone Family=Gemma-2, Backbone Scale=2B2026.050.954—
Final-layer probeBackbone Family=Llama-3, Backbone Scale=70B2026.050.953—
Best single-layer probeBackbone Family=Gemma-2, Backbone Scale=27B2026.050.951—
MultiLayer-LinearBackbone Family=Qwen3, Backbone Scale=14B2026.050.951—
Best single-layer probeBackbone Family=Gemma-2, Backbone Scale=9B2026.050.95—
TaT-Disp-LSTMBackbone Family=Llama-3, Backbone Scale=8B2026.050.95—
TaT-Disp-LSTMBackbone Family=Qwen3, Backbone Scale=14B2026.050.95—
Refusal-style probeBackbone Family=Llama-3, Backbone Scale=70B2026.050.949—
TaT-Disp-LSTMBackbone Family=Llama-3, Backbone Scale=70B2026.050.948—
TaT-Disp-LSTMBackbone Family=Gemma-2, Backbone Scale=2B2026.050.948—
TaT-Disp-LSTMBackbone Family=Gemma-2, Backbone Scale=9B2026.050.948—
Geometry-LiteBackbone Family=Llama-3, Backbone Scale=1B2026.050.948—
Final-layer probeBackbone Family=Llama-3, Backbone Scale=8B2026.050.947—
Best single-layer probeBackbone Family=Llama-3, Backbone Scale=8B2026.050.947—
Best single-layer probeBackbone Family=Gemma-2, Backbone Scale=2B2026.050.947—
TaT-Disp-LSTMBackbone Family=Gemma-2, Backbone Scale=27B2026.050.947—
TaT-Disp-LSTMBackbone Family=Qwen3, Backbone Scale=8B2026.050.945—
MultiLayer-LinearBackbone Family=Llama-3, Backbone Scale=1B2026.050.945—
Geometry-LiteBackbone Family=Qwen3, Backbone Scale=1.7B2026.050.939—
Best single-layer probeBackbone Family=Qwen3, Backbone Scale=8B2026.050.938—
Best single-layer probeBackbone Family=Qwen3, Backbone Scale=14B2026.050.938—
TaT-Disp-LSTMBackbone Family=Llama-3, Backbone Scale=1B2026.050.937—
Final-layer probeBackbone Family=Gemma-2, Backbone Scale=27B2026.050.936—
Final-layer probeBackbone Family=Gemma-2, Backbone Scale=9B2026.050.935—
MultiLayer-LinearBackbone Family=Qwen3, Backbone Scale=1.7B2026.050.935—
Final-layer probeBackbone Family=Qwen3, Backbone Scale=14B2026.050.934—
Refusal-style probeBackbone Family=Llama-3, Backbone Scale=8B2026.050.934—
Final-layer probeBackbone Family=Gemma-2, Backbone Scale=2B2026.050.933—
Final-layer probeBackbone Family=Qwen3, Backbone Scale=8B2026.050.933—
TaT-Disp-LSTMBackbone Family=Qwen3, Backbone Scale=1.7B2026.050.932—
MultiLayer-DIMBackbone Family=Gemma-2, Backbone Scale=9B2026.050.93—
MultiLayer-DIMBackbone Family=Llama-3, Backbone Scale=70B2026.050.928—
MultiLayer-DIMBackbone Family=Gemma-2, Backbone Scale=2B2026.050.927—
Best single-layer probeBackbone Family=Llama-3, Backbone Scale=1B2026.050.926—
MultiLayer-DIMBackbone Family=Gemma-2, Backbone Scale=27B2026.050.926—
Refusal-style probeBackbone Family=Gemma-2, Backbone Scale=27B2026.050.925—
Final-layer probeBackbone Family=Llama-3, Backbone Scale=1B2026.050.921—
MultiLayer-DIMBackbone Family=Llama-3, Backbone Scale=8B2026.050.915—
Best single-layer probeBackbone Family=Qwen3, Backbone Scale=1.7B2026.050.911—
Final-layer probeBackbone Family=Qwen3, Backbone Scale=1.7B2026.050.908—
MultiLayer-DIMBackbone Family=Qwen3, Backbone Scale=8B2026.050.908—
MultiLayer-DIMBackbone Family=Qwen3, Backbone Scale=14B2026.050.905—
Refusal-style probeBackbone Family=Llama-3, Backbone Scale=1B2026.050.901—
Refusal-style probeBackbone Family=Qwen3, Backbone Scale=14B2026.050.9—
Refusal-style probeBackbone Family=Gemma-2, Backbone Scale=9B2026.050.898—
Refusal-style probeBackbone Family=Qwen3, Backbone Scale=1.7B2026.050.898—
Refusal-style probeBackbone Family=Qwen3, Backbone Scale=8B2026.050.897—
MultiLayer-DIMBackbone Family=Llama-3, Backbone Scale=1B2026.050.888—
MultiLayer-DIMBackbone Family=Qwen3, Backbone Scale=1.7B2026.050.874—
Refusal-style probeBackbone Family=Gemma-2, Backbone Scale=2B2026.050.863—
Best single-layer probeBackbone Family=All, Backbone Scale=Average2026.05—0.941
Final-layer probeBackbone Family=All, Backbone Scale=Average2026.05—0.933
Geometry-LiteBackbone Family=All, Backbone Scale=Average2026.05—0.955
MultiLayer-DIMBackbone Family=All, Backbone Scale=Average2026.05—0.911
MultiLayer-LinearBackbone Family=All, Backbone Scale=Average2026.05—0.954
Refusal-style probeBackbone Family=All, Backbone Scale=Average2026.05—0.907
TaT-Disp-LSTMBackbone Family=All, Backbone Scale=Average2026.05—0.945