Researchers have developed a new AI model called RetroDFM-R for predicting chemical reactions. Unlike previous methods that rely on pattern matching, RetroDFM-R uses large language models and reinforcement learning to provide transparent and interpretable results. The model achieves high accuracy in predicting chemical pathways and can even reconstruct complex synthetic routes for real-world pharmaceuticals and materials. Its reasoning is explicit and human-interpretable, making it more trustworthy and deployable in automated planning. --- Why it matters: This matters because it addresses a key barrier to trust in AI-driven retrosynthetic planning: the lack of transparency and interpretability in previous methods. By providing step-by-step rationale, RetroDFM-R enables researchers to understand and verify its predictions, making it more reliable for practical applications. Source: https://arxiv.org/abs/2507.17448