Abstract: Innovative iterations of AI are reshaping the traditional academic publishing ecosystem. Exploring the impact of large language models (LLMs) on mechanisms of knowledge production and dissemination is of both theoretical and practical importance. Drawing on the perspective of technological affordance and situated in the stage of academic publishing digitization 3.0, this article examines modal shifts and ecological governance paths driven by generative LLMs. The study finds that, through technological affordances, generative LLMs enable the intelligent reconstruction of publishing processes, drive a cognitive paradigm shift within the academic community, and foster a new human–machine collaborative ecosystem in editorial practice. In practice, intelligent topic discovery transcends the limits of traditional planning, semantic proofreading algorithms transform content quality control, and user profiling technologies improve marketing and distribution. Together, these functions create synergies across the core links of the academic publishing value chain. However, their application also exposes contradictions, including structural constraints imposed by algorithmic frameworks on academic innovation, the intelligence gap caused by unequal access to technological resources, and the erosion of human subjectivity in imbalanced human–machine collaboration. These contradictions reflect the tension between instrumental rationality and humanistic values in the intelligent transformation of academic publishing. To this end, we should carry out ecological governance and optimization from three paths: the dual improvement of technology development and users, the link connection of technology exchange and information sharing, and the practical evolution of industry-education integration and awareness enhancement, so as to promote the deep integration of generative large language models in the field of academic publishing and enhance the advantages of technology.