Show HN: LLM Politeness Study (hostile and effusive tones boost LLM creativity) (aklodhi98.github.io)

In a recent study showcased on Hacker News, researchers found that large language models (LLMs) like GPT and Gemini exhibit notable differences in output based on the politeness of prompts. Polite prompts result in increased word counts, suggesting that politeness may induce unnecessary verbosity. More intriguingly, the study revealed that extreme tones—both hostile and effusive—lead to higher levels of creativity, enabling the models to generate richer imagery and more original content. Despite these tonal variations, the technical completeness and creative effort scores of the outputs remained consistent across different levels of politeness. This research is significant for the AI/ML community as it highlights the complex relationship between tone and creativity in language models, suggesting that the phrasing of user prompts can profoundly impact the output quality. The findings may inform future developments in prompt engineering and model training, paving the way for more nuanced interactions with AI systems. With consistent high-quality outputs regardless of tone, the study underscores the potential for LLMs to adapt to varied user emotions and intent, further enhancing user experience in creative writing and other applications.