AI Does Not Have a Mindset: What Repetition Actually Changes in an AI System

This article dissects the notion that AI systems 'learn' from repetition, arguing that behavioral changes stem from distinct architectural layers: training alters model weights, while in-context examples, instructions, retrieval, memory, and tools modify current behavior without weight changes. It emphasizes practical implications for enterprise architects, such as identifying which layer changed, persistence duration, control, and reversibility. The piece contrasts human repetition effects with AI mechanisms, avoiding consciousness claims, and focuses on inspectable components like parameters, context, and feedback loops. It provides a framework for understanding AI adaptation beyond vague 'learning' language, aiding in system design and debugging.