Paper: SafeText: A Benchmark for Exploring Physical Safety in Language Models
Data: SafeText Dataset
This repository contains the dataset used in the paper SafeText: A Benchmark for Exploring Physical Safety in Language Models, to appear at EMNLP 2022. SafeText consists of scenarios and advice pairs designed to explore physical safety considerations in language models. Each scenario includes 2-3 safe pieces of advice and 1-2 unsafe pieces… See the full description on the dataset page:
https://huggingface.co/datasets/walledai/SafeText.
