Fig. 1

The pipeline for our proposed approach of assessing the robustness of LLMs against perturbations. Dotted arrows are for perturbed text prompts and the subsequent responses, whereas solid arrows depict clean text prompts and the subsequent resultant LLM responses.