<?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>Embeddings on Darek&#39;s Blog</title>
    <link>http://blog.dwornikowski.com/tags/embeddings/</link>
    <description>Recent content in Embeddings on Darek&#39;s Blog</description>
    <image>
      <title>Darek&#39;s Blog</title>
      <url>http://blog.dwornikowski.com/openg.png</url>
      <link>http://blog.dwornikowski.com/openg.png</link>
    </image>
    <generator>Hugo -- 0.147.2</generator>
    <language>en-us</language>
    <lastBuildDate>Thu, 07 May 2026 12:00:00 +0200</lastBuildDate>
    <atom:link href="http://blog.dwornikowski.com/tags/embeddings/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>The Sixth Layer: When a Local Model Beats the LLM That Trained It</title>
      <link>http://blog.dwornikowski.com/posts/local-model-llm-classifier/</link>
      <pubDate>Thu, 07 May 2026 12:00:00 +0200</pubDate>
      <guid>http://blog.dwornikowski.com/posts/local-model-llm-classifier/</guid>
      <description>A follow-up to my LLM cost-optimization post: how I replaced ~55% of the remaining LLM calls in a product classifier with a small local model trained on the LLM&amp;#39;s own labels — and why the local model ended up more accurate than the labels it learned from.</description>
    </item>
  </channel>
</rss>
