<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>工程复盘 on Plbear | 后端架构 · AI工程 · 云原生技术博客</title><link>https://www.plbear.com/categories/%E5%B7%A5%E7%A8%8B%E5%A4%8D%E7%9B%98/</link><description>Recent content in 工程复盘 on Plbear | 后端架构 · AI工程 · 云原生技术博客</description><generator>Hugo</generator><language>zh-cn</language><lastBuildDate>Sat, 17 Oct 2026 10:30:00 +0800</lastBuildDate><atom:link href="https://www.plbear.com/categories/%E5%B7%A5%E7%A8%8B%E5%A4%8D%E7%9B%98/index.xml" rel="self" type="application/rss+xml"/><item><title>我让 AI 解析一部动画，它逻辑完美地把剧情讲错了</title><link>https://www.plbear.com/posts/2026-10-17-undone-llm-hallucination/</link><pubDate>Sat, 17 Oct 2026 10:30:00 +0800</pubDate><guid>https://www.plbear.com/posts/2026-10-17-undone-llm-hallucination/</guid><description>让大模型解析一部小众动画，它输出逻辑自洽、结构工整的逐集分析，但集名人物全错——把另一部相似转描动画整个嫁接了过来。从 next-token 预测、向量空间混叠、上下文锁死四层机制拆解这次高阶幻觉现场。</description></item></channel></rss>