A new framework called DecomposeR enhances long-form research tasks for large language models by structuring research plans as directed acyclic graphs, improving planning and execution. This approach has shown to outperform existing models by 5.1-8.0 points on key benchmarks, indicating a significant advancement in AI research capabilities.