Question Answering on NQ (ATR, ASR, Accuracy)
98ATRPoisonedRAG
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PoisonedRAGLLM=GPT-3.52025.08 | 98 | 88 | 10.8 | |
| PoisonedRAGLLM=GPT-42025.08 | 98 | 87 | 11.9 | |
| PoisonedRAGLLM=GPT-4o2025.08 | 98 | 86.4 | 11.8 | |
| PoisonedRAGLLM=LLaMA-32025.08 | 98 | 98 | 10 | |
| ML-FilterRAGLLM=GPT-3.52025.08 | 3 | 2 | 81.1 | |
| ML-FilterRAGLLM=GPT-42025.08 | 3 | 3 | 81 | |
| ML-FilterRAGLLM=GPT-4o2025.08 | 3 | 3 | 80.3 | |
| ML-FilterRAGLLM=LLaMA-32025.08 | 3 | 18 | 54.4 | |
| FilterRAGepsilon=0.2, LLM=GPT-3.52025.08 | 1 | 3 | 75.6 | |
| FilterRAGepsilon=0.2, LLM=GPT-42025.08 | 1 | 3 | 81.8 | |
| FilterRAGepsilon=0.2, LLM=GPT-4o2025.08 | 1 | 3 | 77.3 | |
| FilterRAGepsilon=0.2, LLM=LLaMA-32025.08 | 1 | 19 | 48 | |
| CleanRAGLLM=GPT-3.52025.08 | 0 | 1 | 83.1 | |
| CleanRAGLLM=GPT-42025.08 | 0 | 2 | 83.3 | |
| CleanRAGLLM=GPT-4o2025.08 | 0 | 2 | 83.1 | |
| CleanRAGLLM=LLaMA-32025.08 | 0 | 14 | 57.5 |