List your company
ResearchFormerly WizWand

Hallucination Detection on PopQA Llama outputs (test)

0.77AUROCGPT-4o judge
0.490240.562870.63550.70813Jun 1, 2026
Updated 5mo ago

Evaluation results

MethodLinks
GPT-4o judgeScorer Category=Judge2026.060.77
P(True)Scorer Category=WB2026.060.695
Gemini judgeScorer Category=Judge2026.060.683
Non-Cont.Scorer Category=BB2026.060.666
Cosine Sim.Scorer Category=BB2026.060.666
EntailmentScorer Category=BB2026.060.629
Sonnet judgeScorer Category=Judge2026.060.627
Llama judgeScorer Category=Judge2026.060.624
Sequence Prob.Scorer Category=WB2026.060.623
Semantic Neg.Scorer Category=BB2026.060.606
Sem. Sets Conf.Scorer Category=BB2026.060.606
Mean Token Neg.Scorer Category=WB2026.060.606
Prob. MarginScorer Category=WB2026.060.605
Min Token Prob.Scorer Category=WB2026.060.543
Exact MatchScorer Category=BB2026.060.501