Imagine an AI that can translate entire chapters of a novel while seamlessly capturing subtle nuances and complex relationships between characters and plot points. That's the exciting potential of Chapter-to-Chapter (CH2CH) literary translation explored by researchers at the University of Southern California. Traditional machine translation often stumbles when faced with lengthy texts, struggling to maintain coherence and consistency across sentences and paragraphs. Current document-level translation models rely on unrealistic sentence alignments, missing critical context. This research introduces a new paradigm: translating literature at the chapter level. Researchers curated a dataset, 'JAM', comprising 160 classic novels aligned by chapters in both English and Chinese, providing rich context for training AI models. Experiments revealed that Large Language Models (LLMs) excel in CH2CH translation after a two-step fine-tuning process. First, they're trained on sentence-level translations, then fine-tuned on the chapter-level JAM dataset. This allows them to grasp the broader narrative flow and character development within each chapter. However, a significant challenge emerged: LLMs tend to repeat phrases or sentences during long-context generation. This 'repetition problem' underscores the need for advanced decoding strategies that help AI avoid these textual loops while preserving the narrative's integrity. The study also found that decoder-only models, known for efficiency, can effectively handle complex literary translations, outperforming traditional encoder-decoder models on several metrics. This research opens a new chapter in AI-powered literary translation, offering a glimpse into a future where technology can bridge linguistic and cultural gaps with greater depth and understanding. While challenges remain, the ability of LLMs to leverage chapter-level context promises more accurate and engaging translations, potentially unlocking literary treasures for readers worldwide.