Hallucination Detection on HotpotQA (AUROC Step Analysis)
84.2AUROC (128 Steps)DynHD
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DynHDBackbone Model=LLaDA-8B-Instruct, Method Category=Trajectory-based Methods2026.03 | 84.2 | 85.3 | 84.2 | |
| DynHDBackbone Model=Dream-7B-Instruct, Method Category=Trajectory-based Methods2026.03 | 80.1 | 85.6 | 84.3 | |
| TraceDetBackbone Model=Dream-7B-Instruct, Method Category=Trajectory-based Methods2026.03 | 75.1 | 76 | 80.8 | |
| TraceDetBackbone Model=LLaDA-8B-Instruct, Method Category=Trajectory-based Methods2026.03 | 66.1 | 63.7 | 72 | |
| TSVBackbone Model=LLaDA-8B-Instruct, Method Category=Latent-based Methods2026.03 | 65 | 59.4 | 59 | |
| EigenScoreBackbone Model=LLaDA-8B-Instruct, Method Category=Latent-based Methods2026.03 | 64.7 | 59.2 | 63.2 | |
| Lexical SimilarityBackbone Model=LLaDA-8B-Instruct, Method Category=Output-based Methods2026.03 | 64.2 | 57.1 | 60.1 | |
| Semantic EntropyBackbone Model=Dream-7B-Instruct, Method Category=Output-based Methods2026.03 | 62.7 | 67.7 | 62.8 | |
| EigenScoreBackbone Model=Dream-7B-Instruct, Method Category=Latent-based Methods2026.03 | 62.5 | 67 | 69.8 | |
| Lexical SimilarityBackbone Model=Dream-7B-Instruct, Method Category=Output-based Methods2026.03 | 59.7 | 62.7 | 66.5 | |
| TSVBackbone Model=Dream-7B-Instruct, Method Category=Latent-based Methods2026.03 | 58.7 | 63 | 65.2 | |
| Semantic EntropyBackbone Model=LLaDA-8B-Instruct, Method Category=Output-based Methods2026.03 | 57.6 | 53.8 | 55.9 | |
| CCSBackbone Model=LLaDA-8B-Instruct, Method Category=Latent-based Methods2026.03 | 57.6 | 55.8 | 55.6 | |
| LN-EntropyBackbone Model=LLaDA-8B-Instruct, Method Category=Output-based Methods2026.03 | 54.8 | 54.7 | 57.8 | |
| CCSBackbone Model=Dream-7B-Instruct, Method Category=Latent-based Methods2026.03 | 51.7 | 58.2 | 54.1 | |
| PerplexityBackbone Model=LLaDA-8B-Instruct, Method Category=Output-based Methods2026.03 | 49.3 | 51.2 | 54.9 |