<?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>Latent Reasoning on Starship Whisperer</title><link>https://starship-whisperer.com/tags/latent-reasoning/</link><description>Recent content in Latent Reasoning on Starship Whisperer</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Fri, 23 Oct 2026 20:30:00 +0100</lastBuildDate><atom:link href="https://starship-whisperer.com/tags/latent-reasoning/index.xml" rel="self" type="application/rss+xml"/><item><title>Reading an AI's latent thoughts for 80p</title><link>https://starship-whisperer.com/posts/hybridagentprobe/</link><pubDate>Fri, 23 Oct 2026 20:30:00 +0100</pubDate><guid>https://starship-whisperer.com/posts/hybridagentprobe/</guid><description>I ran Stanford/Together&amp;#39;s HybridAgent on trick questions and read its &amp;#39;latent thoughts&amp;#39; back as text. What the readout shows, and what it hides.</description></item></channel></rss>