<?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>InformationTheory on SiBlog</title><link>https://sinimite.work/tags/informationtheory/</link><description>Recent content in InformationTheory on SiBlog</description><image><title>SiBlog</title><url>https://sinimite.work/images/og-default.svg?v=20260525-210321</url><link>https://sinimite.work/images/og-default.svg?v=20260525-210321</link></image><generator>Hugo -- 0.156.0</generator><language>zh-cn</language><lastBuildDate>Fri, 27 Mar 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://sinimite.work/tags/informationtheory/rss.xml" rel="self" type="application/rss+xml"/><item><title>LLM 中的熵：从训练到推理到产品的统一语言</title><link>https://sinimite.work/posts/entropy-in-llm/</link><pubDate>Fri, 27 Mar 2026 00:00:00 +0000</pubDate><guid>https://sinimite.work/posts/entropy-in-llm/</guid><description>熵（Entropy）贯穿 LLM 的整个生命周期——训练用交叉熵做 loss，推理用 temperature 控制输出熵，评估用 perplexity 衡量质量，产品层面用 entropy 信号检测幻觉。本文从一个转型 AI 工程师的视角，梳理熵在 LLM 中的完整图谱。</description></item></channel></rss>