A development team trains a language model exclusively on a large dataset of news articles, where every article is truncated to a maximum of 1,024 tokens. After deployment, the model is tasked with summarizing academic papers that are frequently 3,000-4,000 tokens long. Users report that the summaries for these longer papers are often of poor quality, frequently ignoring information presented in the latter half of the source paper. Based on this scenario, diagnose the core issue limiting the model's performance and explain the underlying reason for this failure.