Researchers have found that large language models (LLMs) struggle to integrate information across long-form texts, despite growing context lengths. To assess this, they compared human-written summaries of novels with those generated by nine state-of-the-art LLMs. The study found both stylistic and conceptual differences between human and model-authored summaries, with the latter tending to focus on the ends of texts rather than the narrative's core. This suggests that current LLM attention mechanisms may be hindering their ability to comprehend complex narratives. --- Why it matters: This research matters because it highlights a key limitation in current large language models: their inability to effectively integrate information across long-form texts. Understanding and addressing this issue is crucial for developing more sophisticated AI systems capable of handling complex tasks like narrative comprehension. Source: https://arxiv.org/abs/2604.06416