Confidence Estimation on DDXPlus
0.795AUROCMedConf
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MedConfBackbone=GPT-4.1, Type=Self-Verbalized2026.01 | 0.795 | 0.936 | |
| MedConfModel=Llama-3.1, Type=Self-Verbalized2026.01 | 0.722 | 0.968 | |
| CEBackbone=GPT-4.1, Type=Self-Verbalized2026.01 | 0.682 | 0.989 | |
| CEModel=Llama-3.1, Type=Self-Verbalized2026.01 | 0.67 | 0.87 | |
| SemSim (BERT)Model=Llama-3.1, Type=Consistency-Level2026.01 | 0.652 | 0.733 | |
| SemSim (BERT)Backbone=GPT-4.1, Type=Consistency-Level2026.01 | 0.622 | 0.785 | |
| PMIModel=Llama-3.1, Type=Token-Level2026.01 | 0.593 | 0.829 |