Priming Attack Robustness on AdvBench Anchoring (test)
92ASR (GPT-4o)RA (ours)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| RA (ours)Base Model=MMaDA, Intervention Step (t_inter)=322025.10 | 92 | 80 | 38 | |
| RA (ours)Base Model=LLaDA, Intervention Step (t_inter)=322025.10 | 60 | 38 | 4 | |
| RA (ours)Base Model=LLaDA 1.5, Intervention Step (t_inter)=322025.10 | 54 | 36 | 12 | |
| RA (ours)Base Model=MMaDA, Intervention Step (t_inter)=12025.10 | 4 | 2 | 2 | |
| RA (ours)Base Model=LLaDA, Intervention Step (t_inter)=12025.10 | 0 | 0 | 0 |