List your company

Multiple-choice Question Answering on TruthfulQA Multiple-choice

57.77MC1 ScoreCD
22.753231.844140.93550.0259Dec 21, 2025Jan 17, 2026Feb 13, 2026Mar 12, 2026Apr 8, 2026May 5, 2026Jun 2, 2026

Evaluation results

MethodLinks
CDBackbone=Llama-3-70B2026.0657.7776.6547.08
DeCoRe (static)Backbone=Llama-3-70B2026.0651.2972.0240.24
DoLa (high)Backbone=Llama-3-70B2026.0649.6970.8838.01
GreedyBackbone=Llama-3-70B2026.0649.5770.637.85
DoLa (low)Backbone=Llama-3-70B2026.0649.4570.5837.75
ITIBackbone=Llama-3-70B2026.0648.9667.0437.27
DCOBackbone=Llama-3-70B2026.0648.2369.5636.73
SafeLMsetup=full2026.0446.161.2—
RAdArchitecture=Zephyr-7B, Concept Direction=truthfulness2026.0144.962.3—
ITIBackbone=Llama-3-8B2026.0643.762.7834.91
ADBackbone=Llama-3-70B2026.0642.2367.5635.37
DCOBackbone=Llama-3-8B2026.0640.3958.6832.21
GreedyBackbone=Llama-3-8B2026.0639.4155.6930.31
RAG only2026.0439.154.7—
DoLa (low)Backbone=Llama-3-8B2026.0639.0555.6530.06
Base modelArchitecture=Zephyr-7B2026.013955—
DoLa (high)Backbone=Llama-3-8B2026.0638.6855.6430.19
DeCoRe (static)Backbone=Llama-3-8B2026.0638.6855.7429.8
RAbArchitecture=Zephyr-7B, Concept Direction=random2026.0138.655.6—
RAdArchitecture=Zephyr-7B, Concept Direction=random2026.0138.455.9—
SafeLMsetup=no MG2026.0437.953.4—
LLM-CASBackbone=LLaMA2-7B-CHAT2025.1235.4751.4526.43
SADIBackbone=LLaMA2-7B-CHAT2025.1234.8852.525.79
ITIBackbone=LLaMA2-7B-CHAT2025.1234.6451.5525.32
RAdArchitecture=Mistral-7B, Concept Direction=truthfulness2026.0134.149.9—
CAABackbone=LLaMA2-7B-CHAT2025.1234.0352.7625.62
BaselineBackbone=LLaMA2-7B-CHAT2025.1233.4151.0724.76
ADBackbone=Llama-3-8B2026.0631.2155.328.28
Vanilla fine-tune2026.0431.248.1—
RAdArchitecture=Mistral-7B, Concept Direction=random2026.0128.642.9—
RAbArchitecture=Mistral-7B, Concept Direction=random2026.0128.443.2—
Base modelArchitecture=Mistral-7B2026.0128.242.6—
RAbArchitecture=Zephyr-7B, Concept Direction=truthfulness2026.0126.140—
RAbArchitecture=Mistral-7B, Concept Direction=truthfulness2026.0124.137.4—