清华新研究，Nature+Science双杀！

Core Insights - The research from Tsinghua University reveals a paradox in the AI for Science field, where AI accelerates individual scientific productivity but narrows collective scientific focus, leading to a phenomenon known as "group climbing" [3][4][31]. Group 1: Research Findings - The study analyzed 250 million scientific papers and found that while AI tools have enabled scientists to publish 3.02 times more papers and receive 4.84 times more citations, the overall breadth of scientific exploration has decreased by 4.63% [3][26][29]. - The research indicates that the collective interaction among scientists from different fields has reduced by 22%, suggesting a trend towards concentration and a lack of innovative vitality in research [29]. Group 2: Methodology - The research team utilized a combination of expert annotation and large-scale language model inference to identify AI-augmented research, achieving a high accuracy score of 0.875 [12][13]. - A comprehensive dataset covering 41.3 million research papers from 1980 to 2025 was created, serving as a benchmark for understanding the systematic impact of AI on scientific research [14]. Group 3: Implications - The findings suggest that the current AI models lack generalizability, leading to a strong "scientific intelligence gravity" effect that draws researchers towards a few popular research areas, thereby solidifying existing scientific exploration paths [31][32]. - The research concludes that while AI enhances local efficiency, it struggles to drive innovation across the entire research chain and multiple disciplines [35].