List your company

Hallucination Detection on UMWP

89.6ROC-AUCSinkProbe
69.42474.66279.985.138Apr 12, 2026

Evaluation results

MethodLinks
SinkProbeLLM=Phi3.52026.0489.6
SinkProbeLLM=Llama3.1-8B2026.0487.9
LookbackLensLLM=Phi3.52026.0487.7
SinkProbeLLM=Mistral-Nemo2026.0487.6
MTopDivLLM=Phi3.52026.0487.2
LookbackLensLLM=Mistral-Nemo2026.0486.6
AttnLogDetLLM=Phi3.52026.0486.4
AttnEigvalLLM=Phi3.52026.0486.4
LapEigvalLLM=Phi3.52026.0486.2
AttnLogDetLLM=Mistral-Nemo2026.0484.7
LapEigvalLLM=Mistral-Nemo2026.0484.7
MTopDivLLM=Mistral-Nemo2026.0484.1
LookbackLensLLM=Llama3.1-8B2026.0483.8
LapEigvalLLM=Llama3.1-8B2026.0483.4
AttnEigvalLLM=Mistral-Nemo2026.0482.6
AttnLogDetLLM=Llama3.1-8B2026.0482.5
AttnEigvalLLM=Llama3.1-8B2026.0481.8
AttnLogDetLLM=Llama3.2-3B2026.0480.9
MTopDivLLM=Llama3.1-8B2026.0480.9
AttnScoreLLM=Mistral-Nemo2026.0477.3
SinkProbeLLM=Llama3.2-3B2026.0475.7
AttnScoreLLM=Phi3.52026.0475.7
LookbackLensLLM=Llama3.2-3B2026.0474.9
MTopDivLLM=Llama3.2-3B2026.0473.5
AttnScoreLLM=Llama3.1-8B2026.0472.2
LapEigvalLLM=Llama3.2-3B2026.0472
AttnEigvalLLM=Llama3.2-3B2026.0470.4
AttnScoreLLM=Llama3.2-3B2026.0470.2