<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>mtkl</title>
    <link>https://devmtkl.com</link>
    <description>mtkl — I build things, mostly machine learning, systems, and online software.</description>
    <language>en</language>
    <atom:link href="https://devmtkl.com/rss.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Training a German GPT from scratch</title>
      <link>https://devmtkl.com/writing/training-a-german-gpt-from-scratch</link>
      <guid>https://devmtkl.com/writing/training-a-german-gpt-from-scratch</guid>
      <pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate>
      <description>A small language model trained from zero on the German classics — a custom tokenizer, two-stage training, and an honest look at what 42M parameters can and can't do.</description>
    </item>
    <item>
      <title>Mining dialogue from 18th-century plays</title>
      <link>https://devmtkl.com/writing/mining-dialogue-from-old-plays</link>
      <guid>https://devmtkl.com/writing/mining-dialogue-from-old-plays</guid>
      <pubDate>Tue, 09 Jun 2026 00:00:00 GMT</pubDate>
      <description>To make a chat model out of classic German literature, I needed conversations. Here's how I turned centuries-old dramas into question-and-answer pairs — and where the heuristics broke.</description>
    </item>
  </channel>
</rss>
