Scientific research is crucial but hindered by complexity and slow progress. To address this, a ResearchAgent powered by a language model generates research ideas and refines them based on literature. It connects information through an academic graph and a knowledge store, mining entities shared across papers. ReviewingAgents provide feedback aligned with human preferences, enhancing idea generation. Experimental validation across disciplines shows the ResearchAgent’s ability to generate novel and valid research ideas. This innovative approach streamlines the research process, potentially revolutionizing the field.
https://arxiv.org/abs/2404.07738