Hallucination Detection on TriviaQA held-out 50% (test)
0.939AUROCLlama-3-8B
Evaluation Results
| Method | Links | |
|---|---|---|
| Llama-3-8BEnsemble Strategy=Weighted Avg.2026.06 | 0.939 | |
| Llama-3-8BEnsemble Strategy=Best J + Best WB2026.06 | 0.926 | |
| Llama-3-8BEnsemble Strategy=Best J + Best BB2026.06 | 0.921 | |
| Llama-3-8BEnsemble Strategy=Best J2026.06 | 0.888 | |
| Gem.-1.5-FlEnsemble Strategy=Best J + Best WB2026.06 | 0.693 | |
| Gem.-1.5-FlEnsemble Strategy=Best J + Best BB2026.06 | 0.685 | |
| Gem.-1.5-FlEnsemble Strategy=Weighted Avg.2026.06 | 0.678 | |
| Gem.-1.5-FlEnsemble Strategy=Best J2026.06 | 0.658 | |
| GPT-4oEnsemble Strategy=Weighted Avg.2026.06 | 0.642 | |
| GPT-4oEnsemble Strategy=Best J + Best BB2026.06 | 0.638 | |
| GPT-4oEnsemble Strategy=Best J + Best WB2026.06 | 0.614 | |
| GPT-4oEnsemble Strategy=Best J2026.06 | 0.585 |