<?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>Adaptive Compute on Cognaptus</title>
    <link>https://cognaptus.com/tags/adaptive-compute/</link>
    <description>Recent content in Adaptive Compute on Cognaptus</description>
    <generator>Hugo -- 0.145.0</generator>
    <language>en-us</language>
    <lastBuildDate>Tue, 10 Feb 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://cognaptus.com/tags/adaptive-compute/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Confidence Is Not Truth, But It Can Steer: When LLMs Learn When to Stop</title>
      <link>https://cognaptus.com/blog/2026-02-10-confidence-is-not-truth-but-it-can-steer-when-llms-learn-when-to-stop/</link>
      <pubDate>Tue, 10 Feb 2026 00:00:00 +0000</pubDate>
      <guid>https://cognaptus.com/blog/2026-02-10-confidence-is-not-truth-but-it-can-steer-when-llms-learn-when-to-stop/</guid>
      <description>A mechanism-first reading of CoRefine, a confidence-guided controller that uses token-level confidence traces to allocate test-time compute more intelligently.</description>
    </item>
  </channel>
</rss>
