arxiv Can Large Language Models Understand Context?