Hallucination Detection on MedHallu
1AUROCB: CAA
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| B: CAABackbone=Llama-3.3-70B2026.05 | 1 | — | — | — | — | — | — | — | — | |
| F: Ans. Expect.Backbone=Llama-3.3-70B2026.05 | 1 | — | — | — | — | — | — | — | — | |
| Ensemble A+C+E+FBackbone=Llama-3.3-70B2026.05 | 0.99 | — | — | — | — | — | — | — | — | |
| DRIFT-logpBackbone=Llama-3.3-70B2026.05 | 0.87 | — | — | — | — | — | — | — | — | |
| TXTEMB (ctrl)Backbone=Llama-3.3-70B2026.05 | 0.71 | — | — | — | — | — | — | — | — | |
| P(True)Backbone=Llama-3.3-70B2026.05 | 0.58 | — | — | — | — | — | — | — | — | |
| D: Act. VarianceBackbone=Llama-3.3-70B2026.05 | 0.57 | — | — | — | — | — | — | — | — | |
| DRIFT-concat (ablation of E)Backbone=Llama-3.3-70B2026.05 | 0.57 | — | — | — | — | — | — | — | — | |
| DoLaBackbone=Llama-3.3-70B2026.05 | 0.57 | — | — | — | — | — | — | — | — | |
| SAPLMABackbone=Llama-3.3-70B2026.05 | 0.57 | — | — | — | — | — | — | — | — | |
| E: DRIFTBackbone=Llama-3.3-70B2026.05 | 0.55 | — | — | — | — | — | — | — | — | |
| SelfCheckGPT-NLIBackbone=Llama-3.3-70B2026.05 | 0.55 | — | — | — | — | — | — | — | — | |
| A: PerturbationBackbone=Llama-3.3-70B2026.05 | 0.52 | — | — | — | — | — | — | — | — | |
| ACT (detect)Backbone=Llama-3.3-70B2026.05 | 0.52 | — | — | — | — | — | — | — | — | |
| HaMIBackbone=Llama-3.3-70B2026.05 | 0.52 | — | — | — | — | — | — | — | — | |
| IRIS⋆ (self-verify)Backbone=Llama-3.3-70B2026.05 | 0.51 | — | — | — | — | — | — | — | — | |
| HaloScopeBackbone=Llama-3.3-70B2026.05 | 0.51 | — | — | — | — | — | — | — | — | |
| C: Sem. EntropyBackbone=Llama-3.3-70B2026.05 | 0.5 | — | — | — | — | — | — | — | — | |
| INSIDEBackbone=Llama-3.3-70B2026.05 | 0.5 | — | — | — | — | — | — | — | — | |
| HalluShiftBackbone=Llama-3.3-70B2026.05 | 0.5 | — | — | — | — | — | — | — | — | |
| SEPsBackbone=Llama-3.3-70B2026.05 | 0.49 | — | — | — | — | — | — | — | — | |
| MINDBackbone=Llama-3.3-70B2026.05 | 0.48 | — | — | — | — | — | — | — | — | |
| Log-ProbBackbone=Llama-3.3-70B2026.05 | 0.48 | — | — | — | — | — | — | — | — | |
| PRISMBackbone=Llama-3.3-70B2026.05 | 0.45 | — | — | — | — | — | — | — | — | |
| EigenTrackModel=GPT-2, Evaluation=length-controlled2026.05 | — | — | — | — | — | — | — | 0.57 | — | |
| EigenTrackModel=BERT, Evaluation=length-controlled2026.05 | — | — | — | — | — | — | — | 0.54 | — | |
| EigenTrackModel=Pythia-160M, Evaluation=length-controlled2026.05 | — | — | — | — | — | — | — | 0.61 | — | |
| EigenTrackModel=Flan-T5 (dec), Evaluation=length-controlled2026.05 | — | — | — | — | — | — | — | 0.5 | — | |
| LLM-CheckModel=GPT-2, Evaluation=length-controlled2026.05 | — | — | — | — | 0.62 | 0.65 | 0.64 | — | — | |
| LLM-CheckModel=BERT, Evaluation=length-controlled2026.05 | — | — | — | — | 0.64 | 0.72 | — | — | — | |
| LLM-CheckModel=Pythia-160M, Evaluation=length-controlled2026.05 | — | — | — | — | 0.62 | 0.62 | 0.58 | — | — | |
| LLM-CheckModel=Flan-T5 (cross), Evaluation=length-controlled2026.05 | — | — | — | — | 0.55 | 0.5 | — | — | — | |
| LLM-CheckModel=Flan-T5 (dec), Evaluation=length-controlled2026.05 | — | — | — | — | 0.59 | 0.6 | 0.6 | — | — | |
| StackingAnalyzer=Qwen2.5-0.5B2026.05 | — | — | — | — | — | — | — | — | 0.974 | |
| StackingAnalyzer=Pythia-1.4B2026.05 | — | — | — | — | — | — | — | — | 0.952 | |
| StackingAnalyzer=Gemma-2-2B2026.05 | — | — | — | — | — | — | — | — | 0.965 | |
| StackingAnalyzer=LLaMA-3-3B2026.05 | — | — | — | — | — | — | — | — | 0.977 | |
| StackingAnalyzer=Qwen2.5-7B2026.05 | — | — | — | — | — | — | — | — | 0.99 | |
| StackingAnalyzer=LLaMA-3-8B2026.05 | — | — | — | — | — | — | — | — | 0.983 | |
| StackingAnalyzer=Gemma-2-9B2026.05 | — | — | — | — | — | — | — | — | 0.985 | |
| Transport DiagnosticsModel=GPT-2, Evaluation=length-controlled2026.05 | — | 0.58 | 0.61 | 0.55 | — | — | — | — | — | |
| Transport DiagnosticsModel=BERT, Evaluation=length-controlled2026.05 | — | 0.58 | 0.54 | 0.63 | — | — | — | — | — | |
| Transport DiagnosticsModel=Pythia-160M, Evaluation=length-controlled2026.05 | — | 0.61 | 0.62 | 0.59 | — | — | — | — | — | |
| Transport DiagnosticsModel=Flan-T5 (cross), Evaluation=length-controlled2026.05 | — | 0.56 | 0.67 | 0.68 | — | — | — | — | — | |
| Transport DiagnosticsModel=Flan-T5 (dec), Evaluation=length-controlled2026.05 | — | 0.53 | 0.57 | 0.56 | — | — | — | — | — |