List your company
ResearchFormerly WizWand

Hallucination Detection on PsiloQA

92.64AUROCCurvature
48.762460.153771.54582.9363May 25, 2026
Updated 5mo ago

Evaluation results

MethodLinks
CurvatureBackbone=Mistral-7B-Instruct-v0.3, First-sentence truncation=true, Forward horizon (w)=72026.0592.64
Val LossBackbone=LLaMA-3.1-8B-Instruct, First-sentence truncation=true, Forward horizon (w)=72026.0592.38
Val LossBackbone=Mistral-7B-Instruct-v0.3, First-sentence truncation=true, Forward horizon (w)=72026.0592.16
IDBackbone=LLaMA-3.1-8B-Instruct, First-sentence truncation=true, Forward horizon (w)=72026.0592.08
FEPoIDBackbone=LLaMA-3.1-8B-Instruct, First-sentence truncation=true, Forward horizon (w)=72026.0592.08
FEPoIDBackbone=Mistral-7B-Instruct-v0.3, First-sentence truncation=true, Forward horizon (w)=72026.0592.07
CurvatureBackbone=LLaMA-3.1-8B-Instruct, First-sentence truncation=true, Forward horizon (w)=72026.0591.6
Val LossBackbone=LlaMA-3.1-8B, Forward horizon (w)=7, Representation extraction method=FST, Model tuning status=non-instruction-tuned (base model)2026.0590.51
CurvatureBackbone=LlaMA-3.1-8B, Forward horizon (w)=7, Representation extraction method=FST, Model tuning status=non-instruction-tuned (base model)2026.0590.49
FEPoIDLLM Backbone=Mistral-7B-Instruct-v0.3, Method Category=Hidden-State Probing2026.0590.31
IDBackbone=LlaMA-3.1-8B, Forward horizon (w)=7, Representation extraction method=FST, Model tuning status=non-instruction-tuned (base model)2026.0589.85
FEPoIDBackbone=LlaMA-3.1-8B, Forward horizon (w)=7, Representation extraction method=FST, Model tuning status=non-instruction-tuned (base model)2026.0589.85
IDBackbone=Mistral-7B-Instruct-v0.3, First-sentence truncation=true, Forward horizon (w)=72026.0589.42
RGNBackbone=Mistral-7B-Instruct-v0.3, First-sentence truncation=true, Forward horizon (w)=72026.0587.97
IDBackbone=LlaMA-3.1-8B base, Forward horizon (w)=7, FST=false2026.0586.55
FEPoIDBackbone=LlaMA-3.1-8B base, Forward horizon (w)=7, FST=false2026.0586.55
CurvatureBackbone=LlaMA-3.1-8B base, Forward horizon (w)=7, FST=false2026.0586.39
RankMEBackbone=Mistral-7B-Instruct-v0.3, First-sentence truncation=true, Forward horizon (w)=72026.0586.17
RGNBackbone=LlaMA-3.1-8B, Forward horizon (w)=7, Representation extraction method=FST, Model tuning status=non-instruction-tuned (base model)2026.0585.83
RankMEBackbone=LLaMA-3.1-8B-Instruct, First-sentence truncation=true, Forward horizon (w)=72026.0585.49
Val LossBackbone=LlaMA-3.2-3B, Forward horizon (w)=7, FST extraction status=without FST2026.0584.98
FEPoIDBackbone=LlaMA-3.2-3B, Forward horizon (w)=7, FST extraction status=without FST2026.0584.83
Val LossBackbone=LlaMA-3.1-8B base, Forward horizon (w)=7, FST=false2026.0584.34
RankMEBackbone=LlaMA-3.1-8B, Forward horizon (w)=7, Representation extraction method=FST, Model tuning status=non-instruction-tuned (base model)2026.0583.98
CurvatureBackbone=LlaMA-3.2-3B, Forward horizon (w)=7, FST extraction status=without FST2026.0582.8
SNRBackbone=Mistral-7B-Instruct-v0.3, First-sentence truncation=true, Forward horizon (w)=72026.0582.74
RGNBackbone=LLaMA-3.1-8B-Instruct, First-sentence truncation=true, Forward horizon (w)=72026.0581.12
Lexical SimilarityBackbone=LLaMA-3.1-8B-Instruct, First-sentence truncation=true, Forward horizon (w)=72026.0580.61
RankMEBackbone=LlaMA-3.1-8B base, Forward horizon (w)=7, FST=false2026.0580.32
SNRBackbone=LLaMA-3.1-8B-Instruct, First-sentence truncation=true, Forward horizon (w)=72026.0580.01
SNRBackbone=LlaMA-3.1-8B, Forward horizon (w)=7, Representation extraction method=FST, Model tuning status=non-instruction-tuned (base model)2026.0579.99
RGNBackbone=LlaMA-3.2-3B, Forward horizon (w)=7, FST extraction status=without FST2026.0579.9
IDBackbone=LlaMA-3.2-3B, Forward horizon (w)=7, FST extraction status=without FST2026.0579.9
EigenScoreBackbone=LLaMA-3.1-8B-Instruct, First-sentence truncation=true, Forward horizon (w)=72026.0579.41
Pred. EntropyBackbone=LLaMA-3.1-8B-Instruct, First-sentence truncation=true, Forward horizon (w)=72026.0578.8
FEPoIDLLM Backbone=LlaMA-3.1-8B-Instruct, Method Category=Hidden-State Probing2026.0578.62
RGNBackbone=LlaMA-3.1-8B base, Forward horizon (w)=7, FST=false2026.0578.16
SNRBackbone=LlaMA-3.1-8B base, Forward horizon (w)=7, FST=false2026.0576.77
Val LossBackbone=LlaMA-3.2-1B, Forward horizon (w)=3, FST extraction status=without FST2026.0575.97
Pred. EntropyBackbone=Mistral-7B-Instruct-v0.3, First-sentence truncation=true, Forward horizon (w)=72026.0575.74
EigenScoreBackbone=Mistral-7B-Instruct-v0.3, First-sentence truncation=true, Forward horizon (w)=72026.0575.17
RankMEBackbone=LlaMA-3.2-3B, Forward horizon (w)=7, FST extraction status=without FST2026.0574.69
FEPoIDBackbone=LlaMA-3.2-1B, Forward horizon (w)=3, FST extraction status=without FST2026.0574.13
Semantic EntropyBackbone=LLaMA-3.1-8B-Instruct, First-sentence truncation=true, Forward horizon (w)=72026.0573.92
CurvatureBackbone=LlaMA-3.2-1B, Forward horizon (w)=3, FST extraction status=without FST2026.0573.55
Lexical SimilarityBackbone=Mistral-7B-Instruct-v0.3, First-sentence truncation=true, Forward horizon (w)=72026.0569.07
SNRBackbone=LlaMA-3.2-3B, Forward horizon (w)=7, FST extraction status=without FST2026.0568.78
Semantic EntropyBackbone=Mistral-7B-Instruct-v0.3, First-sentence truncation=true, Forward horizon (w)=72026.0566.85
LIDBackbone=LLaMA-3.1-8B-Instruct, First-sentence truncation=true, Forward horizon (w)=72026.0566.48
LIDBackbone=Mistral-7B-Instruct-v0.3, First-sentence truncation=true, Forward horizon (w)=72026.0566
RGNBackbone=LlaMA-3.2-1B, Forward horizon (w)=3, FST extraction status=without FST2026.0565.97
IDBackbone=LlaMA-3.2-1B, Forward horizon (w)=3, FST extraction status=without FST2026.0565.97
RankMEBackbone=LlaMA-3.2-1B, Forward horizon (w)=3, FST extraction status=without FST2026.0564.41
SNRBackbone=LlaMA-3.2-1B, Forward horizon (w)=3, FST extraction status=without FST2026.0561.26
LN-Pred. EntropyBackbone=LLaMA-3.1-8B-Instruct, First-sentence truncation=true, Forward horizon (w)=72026.0557.42
LN-Pred. EntropyBackbone=Mistral-7B-Instruct-v0.3, First-sentence truncation=true, Forward horizon (w)=72026.0550.45