骂得越狠，ChatGPT回答越准，PSU研究实锤，狂飙84%准确率

Core Insights - A recent study from Penn State University reveals that using ruder prompts leads to higher accuracy in responses from ChatGPT, with a surprising accuracy rate of 84.8% for very rude prompts compared to 80.8% for very polite ones [1][15]. Group 1: Research Findings - The study created a dataset of 50 foundational questions across various fields, reformulated into five levels of politeness: very polite, polite, neutral, rude, and very rude [1][11]. - ChatGPT-4o was tested with a total of 250 prompts, and the results showed that ruder prompts consistently outperformed polite ones in terms of accuracy [1][15]. - The accuracy rates for different politeness levels were as follows: very polite (80.8%), polite (81.4%), neutral (82.2%), rude (82.8%), and very rude (84.8%) [15][16]. Group 2: Methodology - The researchers employed a paired sample t-test to assess the statistical significance of the accuracy differences across various politeness levels [1][14]. - Each question was presented to ChatGPT-4o with specific instructions to ensure that it answered independently of previous context, focusing solely on the multiple-choice format [1][13]. Group 3: Implications and Future Research - The findings suggest that the tone of prompts significantly influences the performance of large language models (LLMs), indicating that politeness may not enhance response quality as previously thought [1][19]. - Future research may explore the emotional weight of polite phrases and their impact on LLM performance, as well as the concept of perplexity in relation to prompt effectiveness [1][21].