引导式测验 · 中等 等级

Label Bias and Annotator Disagreement Quiz

Examine subjective labels, annotator identity and instructions, proxy targets, disagreement, and safer ways to preserve uncertainty in training data.

问题 1 的 8

A model learns “toxic” labels that reflect annotators’ assumptions about dialect. What risk is most direct?