A Novel Metric for Measuring the Robustness of Large Language Models in Non-adversarial ScenariosSamuel AckermanElla Rabinovichet al.2024EMNLP 2024