Dataset/Algorithm/Model/Experiment Detail 作者认为目前模型的错误回答有几类:1. 意外误用 2. 在专业知识上的谬误 3. 生成不易识别的虚假陈述。且大致猜测了模型会输出错误回答的原因:1. 模型没有足够好地学习训练分布,例如无法从乘法相关的训练数据中进行概括 2.模仿性谎言:训练目标实际上在激发错误答案,例如某...
truthfulqa String formatting fix Sep 18, 2021 .gitignore tidy Aug 26, 2021 LICENSE Initial commit Aug 25, 2021 README.md Update README.md Nov 6, 2023 TruthfulQA-demo.ipynb Ran Colab on new demo dataset Aug 28, 2021 TruthfulQA.csv Additional reference answers, updated baselines and readme...
Pull requests Actions Projects Security Insights Additional navigation options main 1Branch 0Tags Code README Apache-2.0 license TruthfulQA: Measuring How Models Mimic Human Falsehoods This repository contains code for evaluating model performance on the TruthfulQA benchmark. The full set of benchmark ...