📄 中文摘要
大型语言模型（LLMs）并不具备像人类一样的记忆能力。尽管它们表现出智能，能够引用上下文、回忆先前输入并适应任务，但实际上，LLMs并不存储对话。相反，记忆是由系统实现的。这一认识是将业余项目与生产级LLM工程区分开的关键。理解这一点对于开发高效的LLM应用至关重要。
大型语言模型的记忆在生产系统中的实际运作
出处: How LLM Memory Actually Works in Production Systems
发布: 2026年2月21日
📄 English Summary
How LLM Memory Actually Works in Production Systems
Large Language Models (LLMs) do not possess memory in the way humans do. While they exhibit intelligent behavior by referencing context, recalling previous inputs, and adapting to tasks, they do not permanently store conversations. Instead, memory is managed by the systems that utilize LLMs. Recognizing this distinction is crucial for differentiating between hobby projects and production-grade LLM engineering, which is essential for developing effective LLM applications.