Hallucination Detection on WikiBio GPT-3.5-Turbo-Instruct (test)
92.5AUC-PR (Nonfactual)SelfCheckGPT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SelfCheckGPTVariant=w/NLI2024.02 | 92.5 | 58.47 | 70.55 | |
| CEGtop-k=62024.02 | 92.31 | 70.24 | 77.59 | |
| SelfCheckGPTVariant=w/Prompt2024.02 | 91.16 | 68.37 | 72.64 | |
| FocusBackbone=LLaMA-65B2024.02 | 89.94 | 64.9 | 74.08 | |
| FocusBackbone=LLaMA-30B2024.02 | 89.79 | 65.69 | 73.64 | |
| HalluDetectorCM=28, CFA=962024.02 | 86.45 | 61.96 | 74.82 | |
| HalluDetectorCM=14, CFA=212024.02 | 82.42 | 57.01 | 70.54 | |
| SelfCheckGPTVariant=w/BERTScore2024.02 | 81.96 | 44.23 | 59.31 |