一门带你从基础到前沿、系统掌握现代 Agent 架构与实践的课程。不需要机器学习背景,帮助你理解、构建并驾驭下一代 AI Agent。
Abstract
Introduction to Modern Agentic Systems: The Way Towards RSI
This course provides a comprehensive, end-to-end journey from the foundations of modern AI to the design and construction of advanced agentic systems. We start from Transformers, ChatGPT and RLHF, and explore how models evolve from generating answers to reasoning and taking action. Through the perspectives of Model, Data, and Harness, you will learn to build modern agentic systems, from tool use and context engineering to memory, agentic search, multi-agent systems, loop engineering, graph engineering, and proactive agents. We also cover continuous learning, self-evolving agents, and the road towards Recursive Superintelligence (RSI). No prior machine learning background is required, but basic knowledge of math, statistics, and programming is helpful.
课程介绍
Introduction to Modern Agentic Systems: The Way Towards RSI 是一门从零基础出发、一路讲到最前沿 Agent 系统的课程。目标是对现代 Agentic system 的构建过程与进展进行系统化介绍。
与传统的机器学习课程不同,我们首先会压缩神经网络的发展历史,从 Transformers、ChatGPT 与 RLHF 开始,理解模型如何从”生成答案”进化到”推理与行动”。
而后,我们会从模型、数据、harness 三个角度构建多个现代 Agentic system。我们将从简单的 Tool Use、Context Engineering 开始,延伸到 Memory、Agentic Search、Multi-Agent 系统的构建,并引入 Loop Engineering、Graph Engineering 与 Proactive Agent。
最后,我们会介绍 Agent 如何通过持续学习、自我改进来展望 Self-Evolving Agent 的未来,并给出构建当前先进水平 Agent 系统的基础设施与工具集。
课程不要求机器学习基础,但具备本科基础水平的数学/统计/编程基础会有助于学习。此外,课程会要求实操构建 Agent,所以电脑与合适的 AI 编程工具将是必要的。
“不只是学习 AI,而是一起走向更智能的未来。” —— Modern Agentic Systems