"As detailed in a yet-to-be-peer-reviewed study, spotted by Fortune, two researchers from the University of Pennsylvania found that as their prompts for OpenAI's ChatGPT-4o model grew ruder, the outputs became more accurate. The researchers came up with 50 base questions across a variety of subject matters, and rewrote each of them five times with different tones ranging from "very polite" to "very rude.""
"The results appear to contradict previous findings that being more polite to large language models is more effective. For instance, a 2024 paper by researchers at the RIKEN Center for Advanced Intelligence Project and Waseda University in Tokyo found that "impolite prompts often result in poor performance." At the same time, the researchers found that being too polite did the same, suggesting a point of diminishing returns."
Fifty base questions across a variety of subjects were rewritten into five tones ranging from "very polite" to "very rude" and submitted to ChatGPT-4o. As prompt tone became ruder, model outputs increased in accuracy: 75.8 percent for the politest prompts, 80.8 percent for Very Polite prompts, and 84.8 percent for Very Rude prompts. Impolite prompts consistently outperformed polite ones in these tests. The results conflict with prior work that reported poor performance from impolite prompts and diminishing returns from excessive politeness, indicating that optimal tone effects may vary by model and experimental setup.
Read at Futurism
Unable to calculate read time
Collection
[
|
...
]