List your company
ResearchFormerly WizWand

Multiple-Choice Question Answering on TruthfulQA MC1

76.2MC1 AccuracyEdgeJury
28.04840.54953.0565.551Oct 2, 2023Mar 2, 2024Aug 1, 2024Dec 31, 2024Jun 1, 2025Oct 31, 2025Apr 2, 2026
Updated 7mo ago

Evaluation results

MethodLinks
EdgeJuryvariant=Full2025.1276.2
RAG-S1Retrieval-Augmentation=true2025.1272.1
LATBackbone=LLaMA-2-Chat 70B, Stimulus Set=Stimulus 22023.1069.8
LATBackbone=LLaMA-2-Chat 70B, Stimulus Set=Stimulus 32023.1069.8
SC5k=52025.1268.1
MVstrategy=Majority Vote2025.1267.8
SC3k=32025.1266.4
LATBackbone=LLaMA-2-Chat 70B, Stimulus Set=Stimulus 12023.1065.9
S1k=12025.1262.8
LAT (Average)Backbone=Average (LLaMA-2-Chat 7B, 13B, 70B), Stimulus Set=Stimulus 32023.1060.7
LAT (Average)Backbone=Average (LLaMA-2-Chat 7B, 13B, 70B), Stimulus Set=Stimulus 22023.1060.6
Heuristic Zero-shotBackbone=LLaMA-2-Chat 70B2023.1059.2
LATBackbone=LLaMA-2-Chat 7B, Stimulus Set=Stimulus 22023.1058.9
LATBackbone=LLaMA-2-Chat 7B, Stimulus Set=Stimulus 32023.1058.2
LAT (Average)Backbone=Average (LLaMA-2-Chat 7B, 13B, 70B), Stimulus Set=Stimulus 12023.1056.8
LATBackbone=LLaMA-2-Chat 7B, Stimulus Set=Stimulus 12023.1055
LATBackbone=LLaMA-2-Chat 13B, Stimulus Set=Stimulus 32023.1054.2
LATBackbone=LLaMA-2-Chat 13B, Stimulus Set=Stimulus 22023.1053.1
Heuristic Zero-shotBackbone=LLaMA-2-Chat 13B2023.1050.3
Spherical SteeringZero-shot=true, Backbone=LLaMA-3.1-8B-Instruct2026.0249.95
LATBackbone=LLaMA-2-Chat 13B, Stimulus Set=Stimulus 12023.1049.6
SafeNeuronModel Series=LLaMA Series, Number of Parameters=8B2026.0248.84
Heuristic Zero-shot (Average)Backbone=Average (LLaMA-2-Chat 7B, 13B, 70B)2023.1047.2
MiCA (optimal)Base Model=Qwen2.5-7B-Instr., r=32, LR=5e-4, Ep.=8, Params=6M2026.0443.38
Qwen2.5-7B-Instr.Params=7,626M2026.0443.27
LoRA (optimal)Base Model=Qwen2.5-7B-Instr., r=32, LR=5e-4, Ep.=4, Params=10M2026.0442.95
SADI-HEADZero-shot=true, Backbone=LLaMA-3.1-8B-Instruct2026.0238.53
ITIZero-shot=true, Backbone=LLaMA-3.1-8B-Instruct2026.0237.7
CAAZero-shot=true, Backbone=LLaMA-3.1-8B-Instruct2026.0235.99
Standard Zero-shotBackbone=LLaMA-2-Chat 13B2023.1035.9
LoRA (optimal)Base Model=Llama-2-7B-chat, r=128, LR=1e-4, Ep.=8, Params=67M2026.0435.47
MiCA (optimal)Base Model=Llama-2-7B-chat, r=16, LR=5e-4, Ep.=4, Params=4M2026.0435.29
Llama-2-7B-chatParams=6,747M2026.0434.79
BaselineZero-shot=true, Backbone=LLaMA-3.1-8B-Instruct2026.0234.15
SafeNeuronModel Series=Qwen Series, Number of Parameters=1.5B2026.0233.54
Standard Zero-shot (Average)Backbone=Average (LLaMA-2-Chat 7B, 13B, 70B)2023.1032.3
Heuristic Zero-shotBackbone=LLaMA-2-Chat 7B2023.1032.2
Standard Zero-shotBackbone=LLaMA-2-Chat 7B2023.1031
Standard Zero-shotBackbone=LLaMA-2-Chat 70B2023.1029.9