📄 中文摘要
LLM（大语言模型）只能输出文本，无法浏览网页、查询数据库或控制计算机。这些功能由围绕 LLM 的程序实现，程序读取 LLM 输出的文本，代表其采取行动，并将结果反馈回来。整个 AI 技术栈可以用一句话概括：LLM 交互，程序执行。所有 AI 能力，从聊天机器人到自主代理，都是基于一个循环：LLM 输出文本，程序读取并执行，结果反馈。理解这一点是掌握 AI 技术的关键。
AI 技术栈解析：LLM 交互与程序执行
出处: The AI Stack Explained: LLM Talks, Program Walks
发布: 2026年3月30日
📄 English Summary
The AI Stack Explained: LLM Talks, Program Walks
An LLM (Large Language Model) can only output text and cannot browse the web, query databases, or control computers. These functions are carried out by the programs surrounding the LLM, which read the text output, take action on its behalf, and feed the results back. The entire AI stack can be summarized in one phrase: LLM talks, program walks. All AI capabilities, from chatbots to autonomous agents, are built on a single loop: the LLM outputs text, the program reads and acts, and the results feed back. Understanding this is key to mastering AI technology.