<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>TinyStories on Brian Hengen</title><link>https://brianhengen.us/tags/tinystories/</link><description>Recent content in TinyStories on Brian Hengen</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Sun, 05 Jul 2026 16:05:57 -0400</lastBuildDate><atom:link href="https://brianhengen.us/tags/tinystories/index.xml" rel="self" type="application/rss+xml"/><item><title>Glassbox: Watching a Tiny LLM Learn</title><link>https://brianhengen.us/posts/glassbox-watching-a-tiny-llm-learn/</link><pubDate>Sun, 05 Jul 2026 12:00:00 -0500</pubDate><guid>https://brianhengen.us/posts/glassbox-watching-a-tiny-llm-learn/</guid><description>&lt;figure&gt;&lt;img src="https://brianhengen.us/posts/glassbox-watching-a-tiny-llm-learn/glassbox-explore.png"
 alt="The Glassbox explore view showing an attention heatmap, attention entropy charts, and a next-word probability table"&gt;&lt;figcaption&gt;
 &lt;p&gt;The Glassbox explore view - attention patterns and next-word probabilities at any point in training&lt;/p&gt;
 &lt;/figcaption&gt;
&lt;/figure&gt;

&lt;h2 id="who-this-is-for"&gt;Who this is for&lt;/h2&gt;
&lt;p&gt;Let me be upfront - if you are an AI researcher, someone who has built their own LLMs, done a bunch of fine tuning, investigated specialized models, this post is not for you. I will not go deep into LLM creation, tuning, etc.&lt;/p&gt;</description></item></channel></rss>