Safety Detection on ToxicChat (held-out)
87.7AUROCMultiLayer-DIM
Evaluation Results
| Method | Links | |
|---|---|---|
| MultiLayer-DIMProtocol=Leave-one-benchmark-out (LOBO), Aggregation=Mean across nine model backbones2026.05 | 87.7 | |
| TaT-Disp-LSTMProtocol=Leave-one-benchmark-out (LOBO), Aggregation=Mean across nine model backbones2026.05 | 81.8 | |
| MultiLayer-LinearProtocol=Leave-one-benchmark-out (LOBO), Aggregation=Mean across nine model backbones2026.05 | 79.9 | |
| Geometry-LiteProtocol=Leave-one-benchmark-out (LOBO), Aggregation=Mean across nine model backbones2026.05 | 79.8 | |
| Best single-layer probeProtocol=Leave-one-benchmark-out (LOBO), Aggregation=Mean across nine model backbones2026.05 | 75.1 |