Quiz collegato alla guida · Medio Livello

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.

Domanda 1 di 8

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