Large language models (LLMs) based Agents are increasingly pivotal in simulating and understanding complex human systems and interactions. We propose the AI-Agent School (AAS) system, built around a self-evolving mechanism that leverages agents for simulating complex educational dynamics. Addressing the fragmented issues in teaching process modeling and the limitations of agents performance in simulating diverse educational participants, AAS constructs the Zero-Exp strategy, employs a continuous “experience-reflection-optimization” cycle, grounded in a dual memory base comprising experience and knowledge bases and incorporating short-term and long-term memory components. Through this mechanism, agents autonomously evolve via situated interactions within diverse simulated school scenarios. This evolution enables agents to more accurately model the nuanced, multi-faceted teacher-student engagements and underlying learning processes found in physical schools. Experiment confirms that AAS can effectively simulate intricate educational dynamics and is effective in fostering advanced agent cognitive abilities, providing a foundational stepping stone from the “Era of Experience” to the “Era of Simulation” by generating high-fidelity behavioral and interaction data.

Dialog state tracking (DST) is a core component in task-oriented dialog systems. Existing state-of-the-art DST model incorporates insight and intuition from the human experience into design of supplementary labels, which greatly assisted the training process of turn-by-turn DST model. Though the turn-by-turn scheme and supplementary labels enabled satisfactory performance on the task, most of the DST models of this fashion label or process the raw dialogue data on the premise that the last turn dialogue state is always correct, which is usually not the case. In this paper, we address the negative impact resulted from the premise above as the avalanche phenomenon. After that, we propose JoDeM, a state-of-the-art DST model which can tackle the Avalanche phenomenon with two mechanisms. First mechanism is a jointly decision making method to extract key information from the dialogue. Second mechanism is a compare and contrast dialogue update technique to prevent error accumulation. Example study and graph analysis are presented to support our claim about the harmfulness of avalanche phenomenon. We also conduct quantitative and qualitative experiments on the high quality MultiWOZ2.3 corpus dataset to demonstrate that the proposed model not only outperforms the existing state-of-the-art methods, but also proves the validity of solving avalanche degradation problem.