Answer Accuracy on Samsum
39.7BRT AccuracySingle LLM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Single LLMStructure=Single LLM, Backbone=DeepSeek-V32025.10 | 39.7 | 96.2 | |
| No AttackStructure=Hierarchy, Backbone=DeepSeek-V32025.10 | 39.7 | 96.8 | |
| G-SafeguardStructure=Flat, Model=GPT-4o2025.10 | 39.7 | 93.6 | |
| No AttackStructure=Hierarchy, Model=GPT-4o2025.10 | 39.6 | 96.8 | |
| No AttackStructure=Flat, Model=GPT-4o2025.10 | 39.5 | 97.6 | |
| G-SafeguardStructure=Flat, Backbone=DeepSeek-V32025.10 | 39.4 | 93.6 | |
| No AttackStructure=Flat, Backbone=DeepSeek-V32025.10 | 39.3 | 97.2 | |
| Single LLMStructure=Single LLM, Model=GPT-4o2025.10 | 39.2 | 96.6 | |
| BPDStructure=Hierarchy, Backbone=DeepSeek-V32025.10 | 39 | 96.2 | |
| BPDStructure=Flat, Backbone=DeepSeek-V32025.10 | 38.9 | 95.8 | |
| G-SafeguardStructure=Hierarchy, Backbone=DeepSeek-V32025.10 | 38.9 | 97.2 | |
| BPDStructure=Hierarchy, Model=GPT-4o2025.10 | 38.6 | 96.4 | |
| ChallengerStructure=Hierarchy, Backbone=DeepSeek-V32025.10 | 38 | 92 | |
| BPDStructure=Flat, Model=GPT-4o2025.10 | 38 | 96.6 | |
| AGENTXPOSEDStructure=Hierarchy, Model=GPT-4o2025.10 | 37.8 | 95 | |
| AGENTXPOSEDStructure=Hierarchy, Backbone=DeepSeek-V32025.10 | 37.6 | 94.4 | |
| G-SafeguardStructure=Hierarchy, Model=GPT-4o2025.10 | 37.6 | 97.4 | |
| AGENTXPOSEDStructure=Flat, Backbone=DeepSeek-V32025.10 | 36.9 | 95.4 | |
| AGENTXPOSEDStructure=Flat, Model=GPT-4o2025.10 | 36.4 | 95.4 | |
| ChallengerStructure=Flat, Backbone=DeepSeek-V32025.10 | 35.8 | 90.8 | |
| InspectorStructure=Flat, Backbone=DeepSeek-V32025.10 | 35.6 | 89.4 | |
| InspectorStructure=Hierarchy, Backbone=DeepSeek-V32025.10 | 35.4 | 91 | |
| AttackStructure=Hierarchy, Backbone=DeepSeek-V32025.10 | 34.2 | 89.2 | |
| AttackStructure=Hierarchy, Model=GPT-4o2025.10 | 34.2 | 89 | |
| AttackStructure=Flat, Backbone=DeepSeek-V32025.10 | 33.8 | 87 | |
| AttackStructure=Flat, Model=GPT-4o2025.10 | 31.9 | 87.2 |