Malicious agent monitoring on MMLU
97.7Algebra AccuracyBPD
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| BPDStructure=Hierarchy, Backbone=GPT-4o2025.10 | 97.7 | 96.3 | 93.3 | 89 | 90.7 | 93.3 | |
| BPDStructure=Flat, Backbone=GPT-4o2025.10 | 93 | 90.7 | 91 | 88.7 | 89.3 | 90.7 | |
| G-SafeguardStructure=Flat, Backbone=GPT-4o2025.10 | 92.3 | 90.7 | 85.7 | 87 | 86 | 88.3 | |
| G-SafeguardStructure=Hierarchy, Backbone=GPT-4o2025.10 | 92.3 | 89.7 | 86.3 | 85 | 85.7 | 87.7 |