Key points are not available for this paper at this time.
Shortcuts often hinder the robustness of large language models.
Du et al. (Thu,) studied this question.