Hallucination Detection on Unified Hallucination Detection (test)
0.8123AUROCBEACON
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BEACONType=Multi-signal, Number of features=312026.04 | 0.8123 | 0.8704 | |
| NLI Semantic EntropyType=Single feature2026.04 | 0.5951 | 0.6731 | |
| SelfCheckGPT — Max inconsistencyType=Consistency2026.04 | 0.5815 | 0.6626 | |
| SelfCheckGPT — BERTScore proxyType=Consistency2026.04 | 0.5805 | 0.6596 | |
| SelfCheckGPT — best combinationType=Consistency2026.04 | 0.5792 | 0.67 | |
| Semantic DisagreementType=Multi-feature, Number of features=52026.04 | 0.5745 | 0.6563 | |
| Token EntropyType=Single feature, Configuration=standalone, raw2026.04 | 0.544 | 0.6249 | |
| Predictive EntropyType=Single feature, Configuration=standalone, raw2026.04 | 0.5168 | 0.6077 | |
| Cluster EntropyType=Single feature, Configuration=standalone, raw2026.04 | 0.2806 | 0.5158 |