<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Quirky Ideas Research</title>
    <link>https://www.quirky-ideas.com/research/</link>
    <description>Technical investigations of AI models, AI hardware, advanced systems, and quantum computing with claims, uncertainty, and ideological priors kept separate.</description>
    <language>en-us</language>
    <lastBuildDate>Mon, 10 Aug 2026 00:00:00 GMT</lastBuildDate>
    <atom:link href="https://www.quirky-ideas.com/research/feed.xml" rel="self" type="application/rss+xml" />
    
    <item>
      <title>When Inference Compute Helps: Search, Verification, and the Cost of Better Answers</title>
      <link>https://www.quirky-ideas.com/research/articles/reasoning-models-inference-compute/</link>
      <guid isPermaLink="true">https://www.quirky-ideas.com/research/articles/reasoning-models-inference-compute/</guid>
      <pubDate>Mon, 10 Aug 2026 00:00:00 GMT</pubDate>
      <description>More inference work can improve selected answers, but only through a complete protocol whose search, verification, stopping, and cost boundaries are disclosed.</description>
    </item>
    
    <item>
      <title>Post-Training Gains Still Depend on the Base Model</title>
      <link>https://www.quirky-ideas.com/research/articles/scaling-beyond-pretraining/</link>
      <guid isPermaLink="true">https://www.quirky-ideas.com/research/articles/scaling-beyond-pretraining/</guid>
      <pubDate>Mon, 10 Aug 2026 00:00:00 GMT</pubDate>
      <description>Public evidence shows large scoped post-training gains, but base-model support, verifier quality, and deployment economics still bound them.</description>
    </item>
    
  </channel>
</rss>
