Review
Jul 2026
Prompt Robustness Is Task-Dependent: Comparing Objective and Belief-Style Questions in LLM Evaluation
This work evaluates four instruction-tuned model families on three objective datasets and three subjective datasets and shows that prompt robustness depends on the question type, the prompt change, and the model.
Sadia Kamal, A. Patwary, Anthony Marchiafava et al.
· 0 citations