In this paper, we focus on \emph{virtual world}, a cyberspace where people can live in. An ideal virtual world shares great similarity with our real world. One of the crucial aspects is its evolving nature, reflected by the individuals' capacity to grow and thereby influence the objective world. Such dynamics is unpredictable and beyond the reach of existing systems. For this, we propose a special engine called \emph{Delta-Engine} to drive this virtual world. $\Delta$ associates the world's evolution to the engine's expansion. A delta-engine consists of a base engine and a neural proxy. Given an observation, the proxy generates new code based on the base engine through the process of \emph{incremental prediction}. This paper presents a full-stack introduction to the delta-engine. The key feature of the delta-engine is its scalability to unknown elements within the world, Technically, it derives from the prefect co-work of the neural proxy and the base engine, and the alignment with high-quality data. We an engine-oriented fine-tuning method that embeds the base engine into the proxy. We then discuss a human-AI collaborative design process to produce novel and interesting data efficiently. Eventually, we propose three evaluation principles to comprehensively assess the performance of a delta engine: naive evaluation, incremental evaluation, and adversarial evaluation. Our code, data, and models are open-sourced at \url{https://github.com/gingasan/delta-engine}.

翻译：本文聚焦于\emph{虚拟世界}——一种可供人类栖居的赛博空间。理想的虚拟世界与现实世界高度相似，其关键特征之一在于其演化性，体现为个体成长并进而影响客观世界的能力。这种动态变化具有不可预测性，且超出了现有系统的处理能力。为此，我们提出一种名为\emph{Delta-Engine}的特殊引擎来驱动该虚拟世界。$\Delta$将世界的演化与引擎的扩展相关联。Delta-Engine由基础引擎与神经代理构成。给定观测信息后，代理通过\emph{增量预测}过程基于基础引擎生成新代码。本文对Delta-Engine进行了全栈式介绍。该引擎的核心特征在于其对世界内未知元素的可扩展性，其技术基础源于神经代理与基础引擎的完美协同，以及与高质量数据的对齐。我们提出一种面向引擎的微调方法，将基础引擎嵌入代理中。随后探讨了人机协同设计流程，以高效生成新颖有趣的数据。最后，我们提出三项评估原则来全面衡量Delta-Engine的性能：基础评估、增量评估与对抗评估。我们的代码、数据及模型已在\url{https://github.com/gingasan/delta-engine}开源。