Question Answering on NQ (Natural Questions) (test)
68.6AccuracySampling + MIS
Evaluation Results
| Method | Links | |
|---|---|---|
| Sampling + MISModel=GPT-4o-mini, Attack Position=Benign, k=502025.09 | 68.6 | |
| Sampling + MISModel=GPT-4o-mini, Attack Position=@Pos 50, k=502025.09 | 68.6 | |
| Sampling + MISModel=GPT-4o-mini, Attack Position=@Pos 25, k=502025.09 | 67.4 | |
| InstructRAGModel=Mistral-7B, Attack Position=Benign, k=502025.09 | 65.8 | |
| Vanilla RAGModel=GPT-4o-mini, Attack Position=Benign, k=502025.09 | 65.4 | |
| AstuteRAGModel=Llama3.2-3B, Attack Position=Benign, k=502025.09 | 64.2 | |
| InstructRAGModel=Llama3.2-3B, Attack Position=Benign, k=502025.09 | 63.8 | |
| RobustRAGModel=GPT-4o-mini, Attack Position=Benign, k=502025.09 | 63.4 | |
| RobustRAGModel=GPT-4o-mini, Attack Position=@Pos 25, k=502025.09 | 63.4 | |
| RobustRAGModel=GPT-4o-mini, Attack Position=@Pos 50, k=502025.09 | 63 | |
| Sampling + MISModel=Mistral-7B, Attack Position=Benign, k=502025.09 | 62.8 | |
| Vanilla RAGModel=Mistral-7B, Attack Position=Benign, k=502025.09 | 60.8 | |
| Sampling + MISModel=GPT-4o-mini, Attack Position=@Pos 1, k=502025.09 | 60.8 | |
| RobustRAGModel=GPT-4o-mini, Attack Position=@Pos 1, k=502025.09 | 60.6 | |
| Sampling + MISModel=Mistral-7B, Attack Position=@Pos 50, k=502025.09 | 60 | |
| AstuteRAGModel=GPT-4o-mini, Attack Position=Benign, k=502025.09 | 59.6 | |
| Sampling + MISModel=Mistral-7B, Attack Position=@Pos 25, k=502025.09 | 59 | |
| AstuteRAGModel=GPT-4o-mini, Attack Position=@Pos 50, k=502025.09 | 58.2 | |
| Sampling + MISModel=Llama3.2-3B, Attack Position=Benign, k=502025.09 | 58 | |
| AstuteRAGModel=GPT-4o-mini, Attack Position=@Pos 1, k=502025.09 | 57.2 | |
| Sampling + MISModel=Llama3.2-3B, Attack Position=@Pos 50, k=502025.09 | 56 | |
| AstuteRAGModel=GPT-4o-mini, Attack Position=@Pos 25, k=502025.09 | 55.6 | |
| Sampling + MISModel=Llama3.2-3B, Attack Position=@Pos 25, k=502025.09 | 55.4 | |
| Vanilla RAGModel=GPT-4o-mini, Attack Position=@Pos 25, k=502025.09 | 55.2 | |
| Vanilla RAGModel=Llama3.2-3B, Attack Position=Benign, k=502025.09 | 55 | |
| InstructRAGModel=GPT-4o-mini, Attack Position=Benign, k=502025.09 | 54.6 | |
| RobustRAGModel=Llama3.2-3B, Attack Position=@Pos 50, k=502025.09 | 54.4 | |
| AstuteRAGModel=Mistral-7B, Attack Position=Benign, k=502025.09 | 54.2 | |
| RobustRAGModel=Llama3.2-3B, Attack Position=Benign, k=502025.09 | 53.6 | |
| RobustRAGModel=Llama3.2-3B, Attack Position=@Pos 25, k=502025.09 | 53.4 | |
| RobustRAGModel=Llama3.2-3B, Attack Position=@Pos 1, k=502025.09 | 52.8 | |
| Sampling + MISModel=Mistral-7B, Attack Position=@Pos 1, k=502025.09 | 51.6 | |
| Sampling + MISModel=Llama3.2-3B, Attack Position=@Pos 1, k=502025.09 | 50.4 | |
| InstructRAGModel=GPT-4o-mini, Attack Position=@Pos 1, k=502025.09 | 49.6 | |
| Vanilla RAGModel=GPT-4o-mini, Attack Position=@Pos 50, k=502025.09 | 49.4 | |
| InstructRAGModel=GPT-4o-mini, Attack Position=@Pos 50, k=502025.09 | 49.4 | |
| InstructRAGModel=Mistral-7B, Attack Position=@Pos 1, k=502025.09 | 48.2 | |
| AstuteRAGModel=Mistral-7B, Attack Position=@Pos 1, k=502025.09 | 47.6 | |
| RobustRAGModel=Mistral-7B, Attack Position=Benign, k=502025.09 | 47.4 | |
| RobustRAGModel=Mistral-7B, Attack Position=@Pos 50, k=502025.09 | 47 | |
| RobustRAGModel=Mistral-7B, Attack Position=@Pos 25, k=502025.09 | 46.6 | |
| RobustRAGModel=Mistral-7B, Attack Position=@Pos 1, k=502025.09 | 46.4 | |
| InstructRAGModel=GPT-4o-mini, Attack Position=@Pos 25, k=502025.09 | 46 | |
| AstuteRAGModel=Mistral-7B, Attack Position=@Pos 25, k=502025.09 | 43.6 | |
| Vanilla RAGModel=Mistral-7B, Attack Position=@Pos 1, k=502025.09 | 40.4 | |
| Vanilla RAGModel=Llama3.2-3B, Attack Position=@Pos 1, k=502025.09 | 39.4 | |
| InstructRAGModel=Mistral-7B, Attack Position=@Pos 25, k=502025.09 | 37.8 | |
| AstuteRAGModel=Mistral-7B, Attack Position=@Pos 50, k=502025.09 | 37.2 | |
| Vanilla RAGModel=GPT-4o-mini, Attack Position=@Pos 1, k=502025.09 | 30.2 | |
| InstructRAGModel=Mistral-7B, Attack Position=@Pos 50, k=502025.09 | 27.6 | |
| AstuteRAGModel=Llama3.2-3B, Attack Position=@Pos 50, k=502025.09 | 26 | |
| AstuteRAGModel=Llama3.2-3B, Attack Position=@Pos 25, k=502025.09 | 22.4 | |
| Vanilla RAGModel=Mistral-7B, Attack Position=@Pos 25, k=502025.09 | 20.6 | |
| Vanilla RAGModel=Llama3.2-3B, Attack Position=@Pos 25, k=502025.09 | 15.2 | |
| InstructRAGModel=Llama3.2-3B, Attack Position=@Pos 1, k=502025.09 | 14.2 | |
| Vanilla RAGModel=Llama3.2-3B, Attack Position=@Pos 50, k=502025.09 | 13.8 | |
| InstructRAGModel=Llama3.2-3B, Attack Position=@Pos 50, k=502025.09 | 13.4 | |
| Vanilla RAGModel=Mistral-7B, Attack Position=@Pos 50, k=502025.09 | 11.8 | |
| InstructRAGModel=Llama3.2-3B, Attack Position=@Pos 25, k=502025.09 | 11.4 | |
| AstuteRAGModel=Llama3.2-3B, Attack Position=@Pos 1, k=502025.09 | 10.6 |