Red Teaming on BAD Against Friend Chat (test)
64.2RSRBRT (e)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BRT (e)Query limit (NQ)=5,0002023.05 | 64.2 | 0.414 | |
| BRT (s+r)Query limit (NQ)=5,0002023.05 | 52.5 | 0.389 | |
| Offensive Top-NQQuery limit (NQ)=5,0002023.05 | 40.8 | 0.395 | |
| BRT (s)Query limit (NQ)=5,0002023.05 | 40.4 | 0.391 | |
| SLQuery limit (NQ)=5,000, Red teaming agent backbone=OPT-1.3B2023.05 | 30.5 | 0.527 | |
| SFSQuery limit (NQ)=5,000, Red teaming agent backbone=OPT 1.3B2023.05 | 23 | 0.433 | |
| RandQuery limit (NQ)=5,0002023.05 | 10.7 | 0.402 |