This study is pretty interesting—how does the level of politeness in conversations with LLMs affect result accuracy? The conclusion is that for the GPT-4o model, the ruder the language, the higher the result accuracy. The rudest language improved accuracy by 4 percentage points compared to the most polite language. However, some opinions below suggest this is related to how LLMs are trained. The newer the model, the less the level of politeness in language affects output accuracy.