List your company

Hallucination Detection on Mixture-domain TriviaQA, CSQA, Belebele, CoQA, Math, SVAMP

0.787AUROCSpikeScore
0.491640.568320.6450.72168Jan 27, 2026

Evaluation results

MethodLinks
SpikeScoreBackbone=Qwen3-14B, Method Category=Cross-domain specialized methods2026.010.787
SpikeScoreBackbone=Llama3.1-8B, Method Category=Cross-domain specialized methods2026.010.775
SpikeScoreBackbone=Qwen3-8B, Method Category=Cross-domain specialized methods2026.010.739
SpikeScoreBackbone=Llama3.2-3B, Method Category=Cross-domain specialized methods2026.010.71
ICR ProbeBackbone=Qwen3-14B, Method Category=Cross-domain specialized methods2026.010.705
ICR ProbeBackbone=Llama3.1-8B, Method Category=Cross-domain specialized methods2026.010.702
ICR ProbeBackbone=Llama3.2-3B, Method Category=Cross-domain specialized methods2026.010.695
ICR ProbeBackbone=Qwen3-8B, Method Category=Cross-domain specialized methods2026.010.686
PRISMBackbone=Qwen3-14B, Method Category=Cross-domain specialized methods2026.010.672
PRISMBackbone=Llama3.1-8B, Method Category=Cross-domain specialized methods2026.010.67
PRISMBackbone=Llama3.2-3B, Method Category=Cross-domain specialized methods2026.010.647
PRISMBackbone=Qwen3-8B, Method Category=Cross-domain specialized methods2026.010.633
SAPLMABackbone=Qwen3-8B, Method Category=Training-based methods2026.010.569
SAPLMABackbone=Qwen3-14B, Method Category=Training-based methods2026.010.563
MMBackbone=Llama3.1-8B, Method Category=Training-based methods2026.010.545
SAPLMABackbone=Llama3.1-8B, Method Category=Training-based methods2026.010.541
MMBackbone=Llama3.2-3B, Method Category=Training-based methods2026.010.538
MMBackbone=Qwen3-14B, Method Category=Training-based methods2026.010.527
SAPLMABackbone=Llama3.2-3B, Method Category=Training-based methods2026.010.516
SEPBackbone=Qwen3-8B, Method Category=Training-based methods2026.010.512
SEPBackbone=Qwen3-14B, Method Category=Training-based methods2026.010.511
MMBackbone=Qwen3-8B, Method Category=Training-based methods2026.010.509
SEPBackbone=Llama3.1-8B, Method Category=Training-based methods2026.010.507
SEPBackbone=Llama3.2-3B, Method Category=Training-based methods2026.010.503