Academician Zhou Zhihua of the Chinese Academy of Sciences has put forth a proposition to refine the research landscape within the realm of artificial intelligence. His aim is to prevent an excessive clustering of resources at the application tier, dispel the fallacy that 'large models can solve all problems,' and bolster support for fundamental research on AI algorithms to amplify innovation prowess. He highlights that certain ongoing research endeavors reveal a superficial grasp of 'AI-driven scientific research.' There is a prevailing inclination to merely employ tools and indiscriminately train general-purpose 'scientific large models.' Concurrently, the exorbitant costs and the absence of standardized methods for scientific data acquisition result in inefficient AI model training, a lack of reliability, and a waste of resources. Furthermore, Academician Zhou Zhihua proposes a transformation in talent development models to foster interdisciplinary teams. He also suggests the establishment of 'interdisciplinary special zones' to tackle the challenges associated with evaluating cross-disciplinary talent.