LLM-AutoSciLab: Closed-Loop Scientific Discovery via Active Experimentation with LLMs

LLM-AutoSciLab proposes a closed-loop framework for scientific discovery, moving beyond static inference by actively coupling hypothesis generation with experiment selection and mechanism refinement. It iteratively suggests plausible hypotheses, selects informative experiments to distinguish or refine them, and updates its state using the resulting evidence.