VON NEUMIDAS: Enhanced Annotation Schema for Human-LLM Interactions Combining MIDAS with Von Neumann Inspired Semantics

LLM-based chatbots represent a significant milestone as the initial point of interaction between artificial intelligence and the general public. These chatbots offer greater flexibility compared to traditional chatbots, yet their behavior deviates notably from human interaction pat terns. Current annotation schemas may not be adequately suited to capture this unique interaction paradigm. In this paper, we propose a novel annotation method designed to annotate interactions between ChatGPT and users of varying expertise levels engaged in complex tasks. Our approach builds on the MIDAS annotation framework, introducing an additional semantic layer inspired by the Von Neumann base operation set. This layer provides detailed descriptions of requested behaviors and prompts, enhancing the granularity of interaction analysis. We aim to utilize this annotation scheme to explore the relationship between user interactions and their perception of AI, evaluate user expertise, and offer insights and suggestions for improved alignment and support.