<?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>Claude on npow</title>
    <link>http://npow.github.io/tags/claude/</link>
    <description>Recent content in Claude on npow</description>
    <generator>Hugo -- 0.146.0</generator>
    <language>en-us</language>
    <lastBuildDate>Mon, 20 Apr 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="http://npow.github.io/tags/claude/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Does Caveman Mode Actually Work? I Benchmarked It (Properly This Time).</title>
      <link>http://npow.github.io/posts/does-caveman-mode-actually-work/</link>
      <pubDate>Mon, 20 Apr 2026 00:00:00 +0000</pubDate>
      <guid>http://npow.github.io/posts/does-caveman-mode-actually-work/</guid>
      <description>Caveman mode makes Claude talk in terse fragments to save tokens. I benchmarked it via the raw API, against a fair terse baseline. The real story: it clearly helps on 4-6 long-form, hurts short-answer 4-7, and the mode-ranking story is underpowered.</description>
    </item>
  </channel>
</rss>
