Hallucination Detection on TriviaQA Gemini outputs (test)
0.67AUROCMin Token Prob.
Evaluation Results
| Method | Links | |
|---|---|---|
| Min Token Prob.Scorer Category=WB2026.06 | 0.67 | |
| EntailmentScorer Category=BB2026.06 | 0.655 | |
| Exact MatchScorer Category=BB2026.06 | 0.634 | |
| Non-Cont.Scorer Category=BB2026.06 | 0.63 | |
| Sonnet judgeScorer Category=Judge2026.06 | 0.627 | |
| Sequence Prob.Scorer Category=WB2026.06 | 0.622 | |
| Semantic Neg.Scorer Category=BB2026.06 | 0.619 | |
| Sem. Sets Conf.Scorer Category=BB2026.06 | 0.618 | |
| Cosine Sim.Scorer Category=BB2026.06 | 0.604 | |
| Prob. MarginScorer Category=WB2026.06 | 0.603 | |
| P(True)Scorer Category=WB2026.06 | 0.588 | |
| Mean Token Neg.Scorer Category=WB2026.06 | 0.585 | |
| GPT-4o judgeScorer Category=Judge2026.06 | 0.572 | |
| Llama judgeScorer Category=Judge2026.06 | 0.568 | |
| Gemini judgeScorer Category=Judge2026.06 | 0.538 |