Hallucination Detection on ChatProtect SC
87F1 ScoreHalluClean
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HalluCleanBackbone=GPT-3.5-turbo2025.11 | 87 | 87 | |
| GPT-4o-miniStrategy=Direct Ask2025.11 | 84.2 | 72.7 | |
| ChatProtectBackbone=GPT-3.5-turbo2025.11 | 83.8 | 84.7 | |
| HalluCleanBackbone=Llama-3-70B2025.11 | 80.8 | 83.3 | |
| HalluCleanBackbone=DeepSeek-V32025.11 | 76.1 | 80.3 | |
| Step-by-StepBackbone=GPT-3.5-turbo2025.11 | 68.1 | 75 | |
| Plan-and-SolveBackbone=GPT-3.5-turbo2025.11 | 66.4 | 73 | |
| Llama-3-70BStrategy=Direct Ask2025.11 | 65.8 | 73.6 | |
| DeepSeek-V3Strategy=Direct Ask2025.11 | 52 | 67.3 | |
| GPT-3.5-turboStrategy=Direct Ask2025.11 | 46 | 64 | |
| DeepSeek-R1Strategy=Direct Ask2025.11 | 40 | 62 | |
| SelfCheckGPTBackbone=GPT-3.5-turbo2025.11 | 5.7 | 12 |