<?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>知识 on Blog</title><link>https://icelemon.top/knowledge/</link><description>Recent content in 知识 on Blog</description><generator>Hugo</generator><language>zh-CN</language><lastBuildDate>Mon, 19 Oct 2026 00:00:00 +0800</lastBuildDate><atom:link href="https://icelemon.top/knowledge/index.xml" rel="self" type="application/rss+xml"/><item><title>37% 法则，究竟在什么条件下最优</title><link>https://icelemon.top/knowledge/secretary-problem/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/secretary-problem/</guid><description>先观察，再选择：秘书问题的解法，以及它能够回答的问题。</description></item><item><title>奥卡姆剃刀：少加假设，也要接受检验</title><link>https://icelemon.top/knowledge/occams-razor-and-overfitting/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/occams-razor-and-overfitting/</guid><description>从一串猜测到一条拟合曲线，简约的价值在于减少没有证据支撑的自由度。</description></item><item><title>从一片噪声，到一张图像</title><link>https://icelemon.top/knowledge/diffusion-models/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/diffusion-models/</guid><description>扩散模型怎样把生成图像，拆成一连串可以学习的去噪任务。</description></item><item><title>德国坦克问题：从几个编号估计看不见的总数</title><link>https://icelemon.top/knowledge/german-tank-problem/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/german-tank-problem/</guid><description>最大编号提供一个下限，抽样模型帮助估计下限之外还藏着多少。</description></item><item><title>队友报点，为什么你当时没听见</title><link>https://icelemon.top/knowledge/inattentional-deafness/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/inattentional-deafness/</guid><description>视觉任务占用注意资源时，已经出现的声音可能没有进入觉察。</description></item><item><title>多修一条路，为什么可能让所有人更慢</title><link>https://icelemon.top/knowledge/braess-paradox/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/braess-paradox/</guid><description>布雷斯悖论揭示了个人最优路线与整个路网效率之间的差别。</description></item><item><title>汉长安与唐长安，为什么不是同一座城</title><link>https://icelemon.top/knowledge/han-tang-changan/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/han-tang-changan/</guid><description>同一个城名，连接着不同的位置、规模与城墙。</description></item><item><title>荆江的河道，为什么让防洪更困难</title><link>https://icelemon.top/knowledge/jingjiang-river/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/jingjiang-river/</guid><description>高水位、曲流与水沙变化，共同塑造了平原河段的防洪难题。</description></item><item><title>咖啡因，为什么让不同的人反应不同</title><link>https://icelemon.top/knowledge/caffeine-and-individual-differences/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/caffeine-and-individual-differences/</guid><description>受体反应、肝脏代谢与日常耐受，共同影响清醒和睡眠。</description></item><item><title>模型能力突然出现，可能发生了什么</title><link>https://icelemon.top/knowledge/emergence-and-measurement/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/emergence-and-measurement/</guid><description>规模规律、评分方式与延迟泛化，分别解释了不同的变化。</description></item><item><title>你脑中的旋律，对方未必听得见</title><link>https://icelemon.top/knowledge/curse-of-knowledge/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/curse-of-knowledge/</guid><description>熟悉一件事之后，我们容易省略别人理解它所需要的前提。</description></item><item><title>闪电最亮的那一下，为什么向上传播</title><link>https://icelemon.top/knowledge/lightning-return-stroke/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/lightning-return-stroke/</guid><description>从向下的先导到向上的回击，把一次雷电放慢来看。</description></item><item><title>熵增，为什么常见的状态更容易出现</title><link>https://icelemon.top/knowledge/entropy-and-irreversibility/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/entropy-and-irreversibility/</guid><description>从硬币的排列到气体混合，理解微观状态、宏观规律与开放系统。</description></item><item><title>稳定匹配：为什么先提议的一方更有利</title><link>https://icelemon.top/knowledge/stable-matching/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/stable-matching/</guid><description>同一份偏好名单，换一组人发出提议，怎样得到不同的稳定结果。</description></item><item><title>信息有多少，取决于原来有多不确定</title><link>https://icelemon.top/knowledge/information-and-surprise/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/information-and-surprise/</guid><description>从猜数到深空通信，理解比特、信息熵、压缩与纠错。</description></item><item><title>幸存者偏差：长寿个例为什么不能代表总体</title><link>https://icelemon.top/knowledge/survivorship-bias/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/survivorship-bias/</guid><description>看见了谁，也要追问哪些人已经离开了可见的样本。</description></item><item><title>一美元拍卖，怎样把竞争推向两败俱伤</title><link>https://icelemon.top/knowledge/dollar-auction/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/dollar-auction/</guid><description>每一步都像是在减少损失，连续下去却可能让损失越来越大。</description></item><item><title>以牙还牙：简单规则怎样维持合作</title><link>https://icelemon.top/knowledge/tit-for-tat/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/tit-for-tat/</guid><description>先合作，再回应上一轮的选择。它的优势，要放在重复互动里理解。</description></item><item><title>运气的形状，怎样被算出来</title><link>https://icelemon.top/knowledge/shape-of-luck/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/shape-of-luck/</guid><description>一次结果难以预知，许多可能结果的结构却可以计算。</description></item><item><title>真正的随机，为什么看起来不够随机</title><link>https://icelemon.top/knowledge/what-randomness-looks-like/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0800</pubDate><guid>https://icelemon.top/knowledge/what-randomness-looks-like/</guid><description>偏爱的数字、连续的硬币与重复的歌单，揭示了直觉和随机模型的距离。</description></item></channel></rss>