A new study from Penn State University found that ChatGPT actually produces more accurate results when it leaves out “please” and “thank you.” Unlike what was previously thought, and unlike how it works with humans.
In 250 test questions, the very rude ones achieved 84.8% accuracy, while the very polite questions remained at 80.8%.
Researchers believe that polite phrases like “Can you please tell me…” add unnecessary words and create ambiguity, while direct commands like “Tell me the answer” make the intent crystal clear.