Researchers have developed a method for using large language models to extract nuanced information from research articles. They tested four workflows, including having the model create its own prompts and discovering new references on its own. While the model performed well in some cases, it struggled with interpreting scientific context and nuance. To overcome this, the researchers propose an auditable division of labor where experts specify the evidence standard, models cross-check repeated extractions, and human judges resolve disputed cases. --- Why it matters: This matters to AI engineers because it shows how large language models can be used to scale up data curation in scientific research without sacrificing expert oversight. It also highlights the importance of designing workflows that balance model capabilities with human judgment. Source: https://arxiv.org/abs/2608.19025