Brief

Researchers have developed PolyBench, a comprehensive benchmark dataset and training methodology for large language models (LLMs) focused on polymer design tasks. This dataset, comprising over 125,000 tasks and leveraging a knowledge base of millions of data points, aims to equip LLMs with the specific knowledge and reasoning capabilities needed for polymer science. Experiments demonstrate that smaller language models trained with PolyBench's knowledge-augmented reasoning distillation method can outperform similar-sized models and compete with larger, closed-source LLMs on polymer-related challenges, showing promise for advancing AI in scientific discovery. AI

Researchers have introduced SciPaths, a new benchmark designed to forecast the pathways to scientific discovery by identifying enabling contributions and their dependencies on prior work. Unlike existing benchmarks that focus on simpler tasks like citation prediction, SciPaths requires models to reason backward from a target contribution to the necessary building blocks. Evaluations of current frontier and open-weight language models show that even the best models struggle with this complex reasoning, achieving only a 0.189 F1 score, indicating that accurately recovering methodological dependencies remains a significant challenge. AI