AI RESEARCH
Mind Your Tone: Does Tone Alter LLM Performance?
arXiv CS.AI
•
ArXi:2605.29027v1 Announce Type: new The use of Large Language Models (LLMs) is proliferating, yet their performance is observed to vary based on prompting styles and tones. In this study, we investigate both whether and how tonal variations in prompts lead to disparate LLM accuracy for objective multiple-choice questions. We use two datasets: a 50-base question dataset with five tone variants and a 570-base question MMLU subset spanning 57 subjects with seven tone variants.