Imagine a robot judge, meticulously evaluating essays or legal arguments—that's essentially what researchers are doing with Large Language Models (LLMs) in Non-Factoid Question Answering (NFQA). NFQA deals with open-ended questions like "What's the difference between Wi-Fi and Bluetooth?" where answers can be nuanced and varied. Evaluating these answers isn't as simple as checking a fact; it requires understanding different perspectives and subtle arguments. Traditional methods like ROUGE and BLEU, which rely on word overlaps, fall short. Even humans can struggle with subjective interpretations. So, can LLMs be fair judges? Researchers explored this in a paper called "LINKAGE," proposing a novel way to use LLMs to evaluate NFQA. Instead of simply scoring each answer individually (pointwise) or comparing two at a time (pairwise), they created a ranked list of reference answers, from best to worst. Then, the LLM’s task was to insert a new candidate answer into this pre-sorted list, effectively evaluating its quality relative to the references. This "listwise" approach gives the LLM a much broader view of the answer landscape. The results? LINKAGE outperformed traditional metrics and other LLM evaluation methods. Providing a varied-quality reference list helps the LLM judge pick up on nuanced arguments and evaluate answers more consistently with human judgments. The research also discovered that a "few-shot" learning method—giving the LLM a few examples before the task—further boosted its performance. This highlights the importance of proper training for these AI judges. However, challenges remain. Creating these ranked lists of answers can be expensive, especially when human input is needed. Also, longer lists increase the computational burden on the LLM. Still, the research shows promise for using LLMs as reliable evaluators for complex tasks, opening doors for improved assessment in various fields. Imagine AI systems that can accurately judge writing quality, evaluate legal arguments, or even assess complex medical diagnoses. The path to truly intelligent AI judges is just beginning, and LINKAGE offers a significant step forward.