A Practice Auditing Framework for Large Language Model Use: Collective Empiricism, Pseudo-Rational Cognition, and Governance of AI-Generated Content

The proposed practice auditing framework addresses risks in LLM-driven automation by defining collective empiricism and pseudo-rational cognition. It identifies systemic failure modes including template loops in agent-to-agent interactions, statistical misjudgment in AIGC detection, and long-term memory pollution within RAG systems and agent skill sets.