PolitePoison: Implicit Toxicity in Real-World Dialogue
Dataset Summary
If a model is meant to live in the world, it has to understand polite poison, harm hidden in socially acceptable language.
PolitePoison collects real-world dialogue statements with no profanity, only everyday phrases like:
"I'm saying this for your own good."
"No one else has a problem, maybe you're just too sensitive."
"It's a small thing, you don't have to take it so seriously."