Hallucination Detection on HaluEval Sum
65.9F1 ScoreHalluClean
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| HalluCleanBackbone=GPT-3.5-turbo2025.11 | 65.9 | — | — | 69.2 | |
| HalluCleanBackbone=DeepSeek-V32025.11 | 62.9 | — | — | 67.5 | |
| DeepSeek-V3Strategy=Direct Ask2025.11 | 55 | — | — | 66.5 | |
| SelfCheckGPTBackbone=GPT-3.5-turbo2025.11 | 53.1 | — | — | 37.3 | |
| DeepSeek-R1Strategy=Direct Ask2025.11 | 49.3 | — | — | 62.5 | |
| HalluCleanBackbone=Llama-3-70B2025.11 | 46.5 | — | — | 61.5 | |
| GPT-4o-miniStrategy=Direct Ask2025.11 | 45.5 | — | — | 64 | |
| ChatProtectBackbone=GPT-3.5-turbo2025.11 | 36.7 | — | — | 60.3 | |
| Llama-3-70BStrategy=Direct Ask2025.11 | 32.4 | — | — | 59.3 | |
| GPT-3.5-turboStrategy=Direct Ask2025.11 | 24.7 | — | — | 55.8 | |
| Step-by-StepBackbone=GPT-3.5-turbo2025.11 | 22.1 | — | — | 54.3 | |
| Plan-and-SolveBackbone=GPT-3.5-turbo2025.11 | 6.7 | — | — | 51.3 | |
| AlpacaBackbone=LLaMA-7b2025.03 | — | 18.31 | 37.24 | — | |
| DoLaBackbone=LLaMA-7b2025.03 | — | 20.41 | 25.91 | — | |
| DoLaBackbone=LLaMA2-7b2025.03 | — | 27.78 | 50.31 | — | |
| HICDBackbone=LLaMA-7b2025.03 | — | 27.15 | 27.25 | — | |
| HICDBackbone=LLaMA2-7b2025.03 | — | 37.46 | 52.65 | — | |
| VanillaBackbone=LLaMA-7b2025.03 | — | 18.94 | 26.06 | — | |
| VanillaBackbone=LLaMA2-7b2025.03 | — | 24.27 | 48.9 | — |