Research Paradigms in the Age of AI: From Prompts to Variable Construction / From Method Encapsulation to Knowledge Consolidation

Morning session (10:00 – 11:30)

Research Paradigms in the Age of AI: From Prompts to Variable Construction

Large language models can now carry out much of the executional work in empirical research—data preparation, coding, and literature search—and the quality of their output depends both on the choice of tools and on how researchers organize instructions and materials. The talk begins with an overview of current mainstream models and AI tools, comparing GUI and CLI IDE such as Cursor, Claude Code, and comparable Chinese tools in research tasks, along with basic techniques for prompting and context. The applied section uses the construction of text-based indices from annual reports to illustrate the full pipeline from unstructured text to regression variables. The talk concludes with a brief introduction to Skills and Workflows as reusable configurations for recurring methods and multi-step procedures.

Afternoon session (14:00 – 15:30)

Research Paradigms in the Age of AI: From Method Encapsulation to Knowledge Consolidation

The effectiveness of LLM-assisted research depends largely on whether a researcher’s own methods, procedures, and accumulated knowledge can be effectively invoked by the model. To solve this issue, the talk examines three levels of organization. Skills: packaging domain-specific research methods, together with their supporting corpora, into callable modules, with a worked example illustrating their structure and construction. Workflows: design elements including task decomposition, role separation, quality gates, and human checkpoints, drawing on the operating records of fully automated paper-generation systems such as Paper Factory and Project APE to discuss the limits of research automation. Knowledge bases: the content layer underlying the first two, where measured statistics from 56 public repositories inform writing conventions and maintenance practices that serve both the researcher and the AI as readers. The talk closes with a discussion of turning research needs into working software tools.