Share your thoughts, 1 month free Claude Pro on usSee more

Safety Detection on ToxicChat (held-out)

87.7AUROC

MultiLayer-DIM

Updated 13d ago

Evaluation Results

Method	Links
MultiLayer-DIM 2026.05		87.7
TaT-Disp-LSTM 2026.05		81.8
MultiLayer-Linear 2026.05		79.9
Geometry-Lite 2026.05		79.8
Best single-layer probe 2026.05		75.1