Imagine asking an AI to write a story, not just any story, but one that follows specific instructions. Would it understand the nuances of your request, or would it miss the mark entirely? This question lies at the heart of recent research exploring the ability of Large Language Models (LLMs) to follow instructions when generating story endings. Researchers tackled this challenge by focusing on how well LLMs can write story endings given both a story context and specific instructions. They used an innovative evaluation method. Instead of relying solely on human judgment, they trained a separate “machine reading comprehension” (MRC) model. This MRC model acted as a judge, determining whether the generated ending truly matched the given instructions and story context. This automated approach offers a faster, more objective way to measure how well LLMs follow creative instructions, moving beyond the simpler benchmarks used in traditional NLP tasks. The results revealed some fascinating insights. While LLMs show promise in understanding and adhering to given instructions, there's still room for improvement, especially when compared to human-written endings. The research highlights the importance of developing better metrics for evaluating LLM performance, specifically for creative tasks where simple keyword matching isn't sufficient. By automating the evaluation process, this work opens exciting new avenues for evaluating and refining LLMs' abilities to understand complex narratives and generate creative text that truly aligns with human intent. The implications of this research are far-reaching. As LLMs become increasingly integrated into creative writing tools, understanding their strengths and limitations in following instructions is crucial for developing truly collaborative human-AI writing experiences. This research lays the groundwork for future advancements in this field.