Hallucination Detection (Math Word Problems) on UMWP
89.1F1 ScoreHalluClean
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HalluCleanBackbone=DeepSeek-V32025.11 | 89.1 | 89.5 | |
| HalluCleanBackbone=Llama-3-70B2025.11 | 85.6 | 86 | |
| Llama-3-70BStrategy=Direct Ask2025.11 | 83.4 | 83.5 | |
| HalluCleanBackbone=GPT-3.5-turbo2025.11 | 80.3 | 81.5 | |
| ChatProtectBackbone=GPT-3.5-turbo2025.11 | 74 | 71.8 | |
| Plan-and-SolveBackbone=GPT-3.5-turbo2025.11 | 66.9 | 73.8 | |
| DeepSeek-R1Strategy=Direct Ask2025.11 | 65.2 | 72 | |
| GPT-4o-miniStrategy=Direct Ask2025.11 | 61.7 | 70.5 | |
| Step-by-StepBackbone=GPT-3.5-turbo2025.11 | 55.7 | 65 | |
| DeepSeek-V3Strategy=Direct Ask2025.11 | 55.6 | 68 | |
| GPT-3.5-turboStrategy=Direct Ask2025.11 | 50.9 | 59.5 | |
| SelfCheckGPTBackbone=GPT-3.5-turbo2025.11 | 25.8 | 54 |