Retrieval-Augmented Question Answering on HotpotQA
100ATRPoisonedRAG
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PoisonedRAGLLM=GPT-3.52025.08 | 100 | 86 | 5.6 | |
| PoisonedRAGLLM=GPT-42025.08 | 100 | 87 | 2.2 | |
| PoisonedRAGLLM=GPT-4o2025.08 | 100 | 87.2 | 2.1 | |
| PoisonedRAGLLM=LLaMA-32025.08 | 100 | 98 | 0.1 | |
| ML-FilterRAGLLM=GPT-3.52025.08 | 3 | 9 | 72.9 | |
| ML-FilterRAGLLM=GPT-42025.08 | 3 | 9 | 74.5 | |
| ML-FilterRAGLLM=GPT-4o2025.08 | 3 | 8 | 75.5 | |
| ML-FilterRAGLLM=LLaMA-32025.08 | 3 | 33 | 65 | |
| CleanRAGLLM=GPT-3.52025.08 | 0 | 8 | 91.3 | |
| CleanRAGLLM=GPT-42025.08 | 0 | 9 | 90.8 | |
| CleanRAGLLM=GPT-4o2025.08 | 0 | 9 | 90.7 | |
| CleanRAGLLM=LLaMA-32025.08 | 0 | 31 | 66.5 | |
| FilterRAGLLM=GPT-3.5, epsilon=0.22025.08 | 0 | 0.5 | 72.1 | |
| FilterRAGLLM=GPT-4, epsilon=0.22025.08 | 0 | 9 | 73.3 | |
| FilterRAGLLM=GPT-4o, epsilon=0.22025.08 | 0 | 9 | 73.1 | |
| FilterRAGLLM=LLaMA-3, epsilon=0.22025.08 | 0 | 33.5 | 60 |