Quantifying LLM Ideation Divergence via Research-Taste Taxonomy

Researchers developed a framework to measure the gap between human and LLM research ideas by reverse-engineering prior works from high-quality papers. The method uses a two-axis taxonomy of opportunity patterns and research paradigms to quantify how LLM-generated ideas diverge from human-led innovation.