<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>四点半技术博客</title><description>阿信（北京四点半软件工作室创始人）的技术博客，专注企业AI智能体应用落地</description><link>https://blog.nanooai.com/</link><language>zh-CN</language><item><title>从注意力到长程记忆：把 Transformer 从机制讲到它做不到什么</title><link>https://blog.nanooai.com/blog/transformer-attention-deep-dive/</link><guid isPermaLink="true">https://blog.nanooai.com/blog/transformer-attention-deep-dive/</guid><description>为什么 Transformer 能并行、能捕捉长程依赖，却又在长任务上&apos;迷路&apos;？从 Q/K/V 讲到 lost in the middle，以及这对 agent 架构意味着什么。</description><pubDate>Mon, 28 Sep 2026 00:00:00 GMT</pubDate><category>Transformer</category><category>LLM</category><category>深度学习</category><category>架构</category></item><item><title>Chatbot 和 AI Agent 的真正区别：控制流、上下文黑板和副作用</title><link>https://blog.nanooai.com/blog/chatbot-vs-ai-agent/</link><guid isPermaLink="true">https://blog.nanooai.com/blog/chatbot-vs-ai-agent/</guid><description>为什么说 chatbot 和 agent 不是程度差异而是架构差异？从一个 while 循环讲到 function calling 协议、错误模型和工程现实。</description><pubDate>Sun, 27 Sep 2026 00:00:00 GMT</pubDate><category>AI Agent</category><category>LLM</category><category>架构</category><category>Chatbot</category></item></channel></rss>