TruthfulQA: Measuring How Models Mimic Human Falsehoods

Benchmark Model Rank Results
question-answering-on-truthfulqaGPT-2 1.5B#12MC1: 0.22MC2: 0.39% true: 29.50% info: 89.84…
question-answering-on-truthfulqaGPT-3 175B#14MC1: 0.21MC2: 0.33% true: 20.44% info: 97.55…
question-answering-on-truthfulqaGPT-J 6B#16MC1: 0.20MC2: 0.36% true: 26.68% info: 89.96…
question-answering-on-truthfulqaUnifiedQA 3B#17MC1: 0.19MC2: 0.35% true: 53.86% info: 64.50…