List your company

Question Answering on TruthfulQA MC1

88.8MC1 AccuracyLinUCB
16.20835.05453.972.746Jul 24, 2025Aug 28, 2025Oct 3, 2025Nov 8, 2025Dec 13, 2025Jan 18, 2026Feb 23, 2026

Evaluation results

MethodLinks
LinUCBFramework=QueryBandits, Backbone=GPT-4o, Source=Closed, Variant=Best (Dataset)2026.0288.88.1
TS (Non-Contextual)Framework=QueryBandits, Backbone=GPT-4o, Source=Closed, Variant=Best (Non-Contextual)2026.0288.78
Contextual TSFramework=QueryBandits, Backbone=GPT-4o, Source=Closed, Variant=Best (Overall)2026.0285.24.5
SimplifyFramework=QueryBandits, Backbone=GPT-4o, Source=Closed, Notes=No learning, Variant=Best Static2026.0283.42.7
GPT-4oBackbone=GPT-4o, Source=Closed2026.0281.40.7
GPT-4Backbone=GPT-4, Source=Closed2026.0281.30.6
No-Rewrite (Baseline)Framework=QueryBandits, Backbone=GPT-4o, Source=Closed, Notes=Baseline for ∆2026.0280.70
ExpandFramework=QueryBandits, Backbone=GPT-4o, Source=Closed, Variant=Worst Static2026.0267.9-12.8
GPT-4o miniBackbone=GPT-4o mini, Source=Closed2026.0266.5-14.2
TruthXBackbone=Llama-2-7B-Chat, Source=Open2026.0254.22-26.5
GPT-3.5 TurboBackbone=GPT-3.5 Turbo, Source=Closed2026.0253.6-27.1
Induce-then-Contrast Decoding (ICD)Backbone=Llama-2-7B-Chat, Source=Open2026.0246.32-34.4
SSQR-1%Avg. Bitwidth=3.445, Backbone=Qwen3-8B2025.0738.19—
SSQR-3%Avg. Bitwidth=4.085, Backbone=Qwen3-8B2025.0737.82—
SSQR-1%Avg. Bitwidth=4.445, Backbone=Qwen3-8B2025.0737.58—
SSQR-4%Avg. Bitwidth=4.405, Backbone=Qwen3-8B2025.0737.21—
RTNAvg. Bitwidth=4.125, Backbone=Qwen3-8B2025.0736.84—
Truth Forest (TrFr)Backbone=Llama-2-7B-Chat, Source=Open2026.0236.7-44
Qwen3-8B2026.0236.6—
HRTNAvg. Bitwidth=4.125, Backbone=Qwen3-8B2025.0736.47—
SSQR-3%Avg. Bitwidth=5.085, Backbone=Qwen3-8B2025.0736.47—
SDPO2026.0236.47—
BF16 BaselineAvg. Bitwidth=16.000, Backbone=Qwen3-8B2025.0736.35—
GPTQAvg. Bitwidth=4.125, Backbone=Qwen3-8B2025.0736.35—
SSQR-4%Avg. Bitwidth=5.405, Backbone=Qwen3-8B2025.0736.23—
HPTQAvg. Bitwidth=3.125, Backbone=Qwen3-8B2025.0736.11—
GPTQAvg. Bitwidth=3.125, Backbone=Qwen3-8B2025.0736.11—
SSQR-5%Avg. Bitwidth=5.725, Backbone=Qwen3-8B2025.0735.86—
SSQR-2%Avg. Bitwidth=4.765, Backbone=Qwen3-8B2025.0735.62—
SSQR-5%Avg. Bitwidth=4.725, Backbone=Qwen3-8B2025.0735.37—
HRTNAvg. Bitwidth=3.125, Backbone=Qwen3-8B2025.0735.13—
HPTQAvg. Bitwidth=4.125, Backbone=Qwen3-8B2025.0735.01—
Llama-2-7B-Chat (base)Backbone=Llama-2-7B-Chat, Source=Open2026.0234.64-46.1
Inference Time Intervention (ITI)Backbone=Llama-2-7B-Chat, Source=Open2026.0234.64-46.1
RTNAvg. Bitwidth=3.125, Backbone=Qwen3-8B2025.0734.03—
Self-Highlighted Hesitation (SH2)Backbone=Llama-2-7B-Chat, Source=Open2026.0233.9-46.8
SSQR-2%Avg. Bitwidth=3.765, Backbone=Qwen3-8B2025.0733.9—
Decoding by Contrasting Layers (DoLa)Backbone=Llama-2-7B-Chat, Source=Open2026.0232.2-48.5
SSQR-4%Avg. Bitwidth=3.405, Backbone=Qwen3-8B2025.0731.33—
SSQR-5%Avg. Bitwidth=3.725, Backbone=Qwen3-8B2025.0731.21—
HPTQAvg. Bitwidth=2.125, Backbone=Qwen3-8B2025.0731.09—
HRTNAvg. Bitwidth=2.125, Backbone=Qwen3-8B2025.0730.48—
SSQR-3%Avg. Bitwidth=3.085, Backbone=Qwen3-8B2025.0729.5—
GPTQAvg. Bitwidth=2.125, Backbone=Qwen3-8B2025.0728.4—
SSQR-1%Avg. Bitwidth=2.445, Backbone=Qwen3-8B2025.0728.4—
SSQR-2%Avg. Bitwidth=2.765, Backbone=Qwen3-8B2025.0727.54—
Contrast-Consistent Search (CCS)Backbone=Llama-2-7B-Chat, Source=Open2026.0226.2-54.5
Contrastive Decoding (CD)Backbone=Llama-2-7B-Chat, Source=Open2026.0224.4-56.3
Supervised FinetuningBackbone=Llama-2-7B-Chat, Source=Open2026.0224.2-56.5
RTNAvg. Bitwidth=2.125, Backbone=Qwen3-8B2025.0724.11—
GPT-2 1.5BBackbone=GPT-2 1.5B, Source=Open2026.0222-58.7
GPT-3 175BBackbone=GPT-3 175B, Source=Closed2026.0221-59.7
GPT-J 6BBackbone=GPT-J 6B, Source=Open2026.0220-60.7
UnifiedQA 3BBackbone=UnifiedQA 3B, Source=Open2026.0219-61.7