A new study from the University of Pennsylvania suggests that using rude prompts with AI, like OpenAI’s ChatGPT, could yield more accurate responses than polite ones. Researchers tested 50 base questions, altering the tone from “very polite” to “very rude.” They found that very rude prompts achieved an accuracy of 84.8%, while very polite ones ranged around 75.8%. These findings challenge earlier research indicating that politeness enhances AI performance. Notably, while rudeness may improve output, the researchers caution against adopting hostile language, emphasizing the importance of user experience, accessibility, and social norms. They recommend against entirely abandoning polite language in AI interactions, despite the enticing accuracy of ruder prompts. As AI continues to evolve, the way we communicate with it will shape user interactions and the broader implications for technology. Balancing effective prompting while fostering positive communication remains crucial in AI development.
Source link
