Fairness in Large Language Models and Generative AI

Generative models move fairness from comparing error rates on labeled decisions to constructing evidence about open-ended outputs; that evidence is valid only when anchored in a concrete use, tested against the model's own variability and reported with its assumptions.