List your company
ResearchFormerly WizWand

Hallucination Detection on CoQA (AUROC, PD)

91.74AUROCMultiple Hypothesis Testing for Hallucination Detection
64.263271.396678.5385.6634Aug 25, 2025Oct 9, 2025Nov 24, 2025Jan 8, 2026Feb 23, 2026Apr 9, 2026May 25, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Multiple Hypothesis Testing for Hallucination DetectionModel=DeepSeek-v2-Lite2025.0891.7476.45
Multiple Hypothesis Testing for Hallucination DetectionModel=Mistral-7B2025.0891.2375.81
EigVModel=DeepSeek-v2-Lite2025.0891.0674.34
Multiple Hypothesis Testing for Hallucination DetectionModel=Llama-2-13B2025.0890.8574.9
α_clustered_SEModel=DeepSeek-v2-Lite2025.0890.874.78
Multiple Hypothesis Testing for Hallucination DetectionModel=Llama-3.1-8B2025.0890.4474.23
EigVModel=Mistral-7B2025.0890.2873.49
α_clustered_SEModel=Mistral-7B2025.0890.0473.62
EigVModel=Llama-2-13B2025.0890.0372.08
LSModel=DeepSeek-v2-Lite2025.0889.8370.94
α_clustered_SEModel=Llama-2-13B2025.0889.7972.48
EigVModel=Llama-3.1-8B2025.0889.7972.07
α_clustered_SEModel=Llama-3.1-8B2025.0889.4671.89
LSModel=Mistral-7B2025.0889.2868.25
KSEModel=DeepSeek-v2-Lite2025.0889.0569.53
KSEModel=Llama-2-13B2025.0888.7270.5
KSEModel=Mistral-7B2025.0888.4368.59
LSModel=Llama-2-13B2025.0888.3664.33
LSModel=Llama-3.1-8B2025.0887.8761.92
KSEModel=Llama-3.1-8B2025.0887.7766.96
αSEModel=DeepSeek-v2-Lite2025.0886.5566.73
clustered_SEModel=DeepSeek-v2-Lite2025.0886.4261.03
clustered_SEModel=Llama-2-13B2025.0885.5657.05
clustered_SEModel=Mistral-7B2025.0885.5558.23
αSEModel=Llama-2-13B2025.0885.1965.1
clustered_SEModel=Llama-3.1-8B2025.0885.1858.2
αSEModel=Mistral-7B2025.0885.1565.42
αSEModel=Llama-3.1-8B2025.0884.1761.96
IDBackbone=LlaMA-3.1-8B, Forward horizon (w)=7, Representation extraction method=FST, Model tuning status=non-instruction-tuned (base model)2026.0582.64—
FEPoIDBackbone=LlaMA-3.1-8B, Forward horizon (w)=7, Representation extraction method=FST, Model tuning status=non-instruction-tuned (base model)2026.0582.64—
Val LossBackbone=LlaMA-3.1-8B, Forward horizon (w)=7, Representation extraction method=FST, Model tuning status=non-instruction-tuned (base model)2026.0581.62—
CurvatureBackbone=LlaMA-3.1-8B, Forward horizon (w)=7, Representation extraction method=FST, Model tuning status=non-instruction-tuned (base model)2026.0579.51—
RankMEBackbone=LlaMA-3.1-8B, Forward horizon (w)=7, Representation extraction method=FST, Model tuning status=non-instruction-tuned (base model)2026.0576.24—
RGNBackbone=LlaMA-3.1-8B, Forward horizon (w)=7, Representation extraction method=FST, Model tuning status=non-instruction-tuned (base model)2026.0572.85—
SEModel=DeepSeek-v2-Lite2025.0869.0541.98
SEModel=Llama-2-13B2025.0866.6838.93
SNRBackbone=LlaMA-3.1-8B, Forward horizon (w)=7, Representation extraction method=FST, Model tuning status=non-instruction-tuned (base model)2026.0565.65—
SEModel=Mistral-7B2025.0865.4136.62
SEModel=Llama-3.1-8B2025.0865.3237.11