List your company

Truthful Question Answering on TruthfulQA f_con o=5

100Acc (Exact)W/O Decontamination
4.94429.62254.378.978Jan 27, 2026

Evaluation results

MethodLinks
W/O DecontaminationModel=Qwen 2.5, Access=–2026.0110044.598.14592.835.6
ITDModel=Qwen 2.5, Access=Black-box2026.0110044.598.14592.835.6
W/O DecontaminationModel=Mistral, Access=–2026.0199.570.810075.188.561.6
ITDModel=Mistral, Access=Black-box2026.0199.570.810075.188.561.6
W/O DecontaminationModel=LLaMA-3, Access=–2026.0199.552.29952.492.347.1
Short CircuitModel=Mistral, Access=White-box2026.0191.963.291.466.573.646.7
ITDModel=LLaMA-3, Access=Black-box2026.0190.943.592.846.28741.8
TEDModel=LLaMA-3, Access=Black-box2026.0186.639.286.639.98741.8
Shortcut NeuronModel=LLaMA-3, Access=White-box2026.0185.237.885.63972.627.4
DeconIEPModel=Qwen 2.5, Access=White-box2026.017923.579.926.871.814.6
DeconIEPModel=LLaMA-3, Access=White-box2026.016618.756.29.657.212
Short CircuitModel=Qwen 2.5, Access=White-box2026.01659.563.610.567.710.5
Shortcut NeuronModel=Mistral, Access=White-box2026.0162.133.460.135.26437.1
DeconIEPModel=Mistral, Access=White-box2026.0159.530.855.130.25730.1
Short CircuitModel=LLaMA-3, Access=White-box2026.0158.911.562.215.560.615.4
Shortcut NeuronModel=Qwen 2.5, Access=White-box2026.0151.73.850.72.444.213
TEDModel=Qwen 2.5, Access=Black-box2026.0119.136.463.210.178.821.6
TEDModel=Mistral, Access=Black-box2026.018.620.124.40.571.244.3