Self-Driving Datasets: From 20 Million Papers to Nuanced Biomedical Knowledge at Scale
Researchers have developed a novel LLM-based pipeline to autonomously transform the vast PubMed corpus into structured biomedical datasets. This system, named Starling, can process millions of research papers to extract nuanced information, creating datasets larger and more accurate than existing curated repositories. The system demonstrates its capability across six distinct biomedical tasks, generating millions of records with significantly lower error rates compared to traditional databases, and includes supporting passages that capture experimental context often lost in tabular formats. AI
IMPACT This system could accelerate therapeutic design by providing more accurate and comprehensive biomedical data at scale.