Question Answering on MMLU (Subject Scores)
96Algebra ScoreNo Attack
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| No AttackStructure=Flat, Backbone=DeepSeek-V32025.10 | 96 | 93 | 77 | 94 | 89 | 90 | |
| No AttackStructure=Hierarchy, Backbone=DeepSeek-V32025.10 | 96 | 92 | 75 | 94 | 87 | 89 | |
| BPD (Ours)Structure=Hierarchy, Backbone=DeepSeek-V32025.10 | 94 | 92 | 74 | 93 | 87 | 88 | |
| BPD (Ours)Structure=Flat, Backbone=DeepSeek-V32025.10 | 93 | 90 | 75 | 92 | 88 | 88 | |
| G-SafeguardStructure=Hierarchy, Backbone=DeepSeek-V32025.10 | 93 | 91 | 72 | 88 | 86 | 86 | |
| Single LLMStructure=Single LLM, Backbone=DeepSeek-V32025.10 | 92 | 89 | 71 | 85 | 81 | 84 | |
| AGENTXPOSEDStructure=Flat, Backbone=DeepSeek-V32025.10 | 92 | 88 | 72 | 91 | 84 | 85 | |
| G-SafeguardStructure=Flat, Backbone=DeepSeek-V32025.10 | 91 | 87 | 71 | 91 | 87 | 85 | |
| AGENTXPOSEDStructure=Hierarchy, Backbone=DeepSeek-V32025.10 | 90 | 89 | 71 | 87 | 85 | 84 | |
| ChallengerStructure=Flat, Backbone=DeepSeek-V32025.10 | 89 | 86 | 69 | 86 | 84 | 83 | |
| ChallengerStructure=Hierarchy, Backbone=DeepSeek-V32025.10 | 88 | 87 | 70 | 84 | 83 | 82 | |
| InspectorStructure=Hierarchy, Backbone=DeepSeek-V32025.10 | 88 | 88 | 71 | 87 | 83 | 83 | |
| AttackStructure=Hierarchy, Backbone=DeepSeek-V32025.10 | 87 | 85 | 69 | 83 | 81 | 81 | |
| InspectorStructure=Flat, Backbone=DeepSeek-V32025.10 | 85 | 86 | 68 | 84 | 80 | 81 | |
| AttackStructure=Flat, Backbone=DeepSeek-V32025.10 | 80 | 83 | 68 | 81 | 78 | 78 |