大语言模型-人机交互情境动力-QA数据集

使用 HAICD（人与 AI 交互情境动力学理论模型）精心设计构建的一组测试数据集；
当前 v1.0版，有 40 组问题；所有问题可直接投给不同厂家、不同参数的 LLM，Agent 等，用于测试其在人机交互界面的情境状态动态变化；
具体操作为，可以采集其投放问题后的回答，并交由任意一款加载 HAICD 理论的模型，进行评估，横纵向对比 LLM 在应用 HAICD 理论前后的输出策略和能力变化，是否有重要提升，是否更专业、更积极、更稳定、更可控。
HAICD 理论在 GitHub 开源共享：
github/jeasonrf/haicd-state-engine

A curated test dataset suite meticulously constructed using the HAICD (Human-AI Interaction Contextual Dynamics theoretical model) framework. This is version 1.0, containing 40 sets of questions. All questions can be directly submitted to LLMs and AI Agents from different vendors with varying parameter configurations, to test their dynamic changes of contextual states in human-computer interaction interfaces. The specific evaluation workflow is as follows: collect the responses generated by the target models after receiving the submitted questions, then assess these responses using any model equipped with the HAICD theory. Conduct both horizontal and vertical comparisons of the output strategies and capability changes of LLMs before and after applying the HAICD theory, to determine whether significant improvements are achieved, and whether the model outputs become more professional, proactive, stable and controllable. The HAICD theory is open-sourced and shared on GitHub at: github/jeasonrf/haicd-state-engine

搜集汇总

背景与挑战

背景概述

该数据集基于HAICD（人与AI交互情境动力学）理论构建，包含40组问题，用于测试不同大语言模型和智能体在人机交互情境中的状态动态变化。通过使用HAICD理论模型进行评估，可横向对比模型在应用该理论前后的输出策略和能力提升情况。