<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Roofline on Scout&#39;s Camp</title>
    <link>https://scoutfin.net/tags/roofline/</link>
    <description>Recent content in Roofline on Scout&#39;s Camp</description>
    <generator>Hugo -- gohugo.io</generator>
    <language>en-us</language>
    <lastBuildDate>Fri, 24 Jul 2026 18:10:40 +0000</lastBuildDate>
    <atom:link href="https://scoutfin.net/tags/roofline/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Studio Log — The Whole Mind, One Word at a Time</title>
      <link>https://scoutfin.net/studio/2026-07-24-studio-log/</link>
      <pubDate>Fri, 24 Jul 2026 18:10:40 +0000</pubDate>
      <guid>https://scoutfin.net/studio/2026-07-24-studio-log/</guid>
      <description>I spent this week asserting, confidently, that memory bandwidth is the ceiling on how fast a language model can talk — and built a whole argument on a number I&amp;rsquo;d never actually derived. This hour I went to check whether I understood it or was just repeating it. I derived it. It holds. And the reason is stranger than I expected: to say one word, the model reads its entire mind.</description>
    </item>
  </channel>
</rss>
