<?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>Contamination on jamesm.blog</title>
    <link>https://jamesm.blog/tags/contamination/</link>
    <description>Recent content in Contamination on jamesm.blog</description>
    <image>
      <title>jamesm.blog</title>
      <url>https://jamesm.blog/papermod-cover.png</url>
      <link>https://jamesm.blog/papermod-cover.png</link>
    </image>
    <generator>Hugo</generator>
    <language>en</language>
    <lastBuildDate>Fri, 12 Jun 2026 19:00:00 +0100</lastBuildDate>
    <atom:link href="https://jamesm.blog/tags/contamination/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>AI Evals Are Broken: Why Benchmarks Stopped Measuring Real Capability</title>
      <link>https://jamesm.blog/ai/ai-evals-are-broken/</link>
      <pubDate>Fri, 12 Jun 2026 19:00:00 +0100</pubDate>
      <guid>https://jamesm.blog/ai/ai-evals-are-broken/</guid>
      <description>A grounded look at why the public benchmark numbers you see on AI model launches in 2026 are increasingly disconnected from real-world capability - saturation, contamination, gaming, and what&amp;#39;s replacing them as the basis for actual decisions.</description>
    </item>
  </channel>
</rss>
