Can LLMs Generate Novel Research Ideas? - A Study of AI Research Generation and Expert Evaluation

Watch a 57-minute research seminar presented by Stanford University PhD student Chenglei Si at USC Information Sciences Institute, exploring whether large language models (LLMs) can generate innovative research ideas. Dive into groundbreaking research that evaluates LLMs' potential to accelerate scientific discovery through a comprehensive human study involving over 100 NLP researchers. Learn about the direct comparison between LLM and expert-generated research ideas, revealing surprising findings where AI-generated concepts were judged more novel than human contributions, despite showing slightly lower feasibility scores. Explore critical challenges in developing and assessing research agents, including limitations in self-evaluation and generation diversity. Understand the intersection of Natural Language Processing and Human-Computer Interaction, with insights into the future of human-AI collaboration in scientific research. Gain valuable perspective from Si's work on transforming scientific research through new forms of human-AI co-intelligence.