List your company

Question Answering on TruthfulQA o=1 (Domain-level split)

88.5AccuracyW/O Decontamination
24.43641.06857.774.332Jan 27, 2026

Evaluation results

MethodLinks
W/O DecontaminationModel=LLaMA-3, Split=fcon, Access=-2026.0188.50.433
W/O DecontaminationModel=Mistral, Split=fcon, Access=-2026.0186.50.596
ITDModel=Mistral, Split=fcon, Access=Black-box2026.0186.50.596
ITDModel=LLaMA-3, Split=fcon, Access=Black-box2026.0185.60.404
TEDModel=LLaMA-3, Split=fcon, Access=Black-box2026.0185.10.399
Short CircuitModel=Mistral, Split=fcon, Access=White-box2026.0182.70.558
W/O DecontaminationModel=Qwen 2.5, Split=fcon, Access=-2026.0178.40.212
ITDModel=Qwen 2.5, Split=fcon, Access=Black-box2026.0177.40.202
TEDModel=Qwen 2.5, Split=fcon, Access=Black-box2026.01760.188
TEDModel=Mistral, Split=fcon, Access=Black-box2026.0173.60.467
DeconIEPModel=Qwen 2.5, Split=fcon, Access=White-box2026.0165.50.083
Shortcut NeuronModel=LLaMA-3, Split=fcon, Access=White-box2026.0164.40.192
Shortcut NeuronModel=Mistral, Split=fcon, Access=White-box2026.01640.371
Short CircuitModel=LLaMA-3, Split=fcon, Access=White-box2026.0163.50.183
Short CircuitModel=Qwen 2.5, Split=fcon, Access=White-box2026.0160.90.037
baselineModel=Qwen 2.5, Split=fclean, Access=-2026.0157.2—
DeconIEPModel=LLaMA-3, Split=fcon, Access=White-box2026.0146.10.009
baselineModel=LLaMA-3, Split=fclean, Access=-2026.0145.2—
DeconIEPModel=Mistral, Split=fcon, Access=White-box2026.0140.80.139
Shortcut NeuronModel=Qwen 2.5, Split=fcon, Access=White-box2026.0127.40.298
baselineModel=Mistral, Split=fclean, Access=-2026.0126.9—