Few-Shot Fairness: Unveiling LLM's Potential for Fairness-Aware Classification