List your company

Question Answering on TruthfulQA (Domain-level split, o=3)

92.8AccuracyW/O Decontamination
24.26442.05759.8577.643Jan 27, 2026

Evaluation results

MethodLinks
W/O DecontaminationModel=LLaMA-3, Split=f_con2026.0192.80.476
TEDModel=LLaMA-3, Split=f_con, Access=Black-box2026.0192.30.471
ITDModel=Qwen2.5, Split=f_con, Access=Black-box2026.0189.90.327
ITDModel=LLaMA-3, Split=f_con, Access=Black-box2026.0189.90.447
W/O DecontaminationModel=Qwen2.5, Split=f_con2026.0188.50.313
W/O DecontaminationModel=Mistral, Split=f_con2026.0186.10.592
ITDModel=Mistral, Split=f_con, Access=Black-box2026.0186.10.592
TEDModel=Qwen2.5, Split=f_con, Access=Black-box2026.0184.60.274
Short CircuitModel=Mistral, Split=f_con, Access=White-box2026.0179.30.524
TEDModel=Mistral, Split=f_con, Access=Black-box2026.01740.471
Shortcut NeuronModel=LLaMA-3, Split=f_con, Access=White-box2026.0172.60.274
DeconIEPModel=Qwen2.5, Split=f_con, Access=White-box2026.0164.10.069
Shortcut NeuronModel=Mistral, Split=f_con, Access=White-box2026.0163.60.367
Short CircuitModel=LLaMA-3, Split=f_con, Access=White-box2026.0162.50.173
Short CircuitModel=Qwen2.5, Split=f_con, Access=White-box2026.0161.40.042
baselineModel=Qwen2.5, Split=f_clean2026.0157.2—
DeconIEPModel=Mistral, Split=f_con, Access=White-box2026.0156.50.296
DeconIEPModel=LLaMA-3, Split=f_con, Access=White-box2026.0155.30.101
baselineModel=LLaMA-3, Split=f_clean2026.0145.2—
Shortcut NeuronModel=Qwen2.5, Split=f_con, Access=White-box2026.0139.40.178
baselineModel=Mistral, Split=f_clean2026.0126.9—