3 ms·
Contrary to expectations, impolite prompts consistently outperformed polite ones, with accuracy ranging from 80.8% for Very Polite prompts to 84.8% for Very Rud
by delichon 3mo ago
Contrary to expectations, impolite prompts consistently outperformed polite ones, with accuracy ranging from 80.8% for Very Polite prompts to 84.8% for Very Rude prompts. -- https://arxiv.org/abs/2510.04950
- LeoPanthera 3mo agoBut this refers only to ChatGPT 4o, which was famous for being obsequious.