<?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>RLVR on Julien Pourcel</title>
    <link>https://julienp.netlify.app/tags/rlvr/</link>
    <description>Recent content in RLVR on Julien Pourcel</description>
    <image>
      <title>Julien Pourcel</title>
      <url>https://julienp.netlify.app/images/papermod-cover.png</url>
      <link>https://julienp.netlify.app/images/papermod-cover.png</link>
    </image>
    <generator>Hugo -- 0.147.3</generator>
    <language>en-us</language>
    <lastBuildDate>Mon, 25 May 2026 10:00:00 +0200</lastBuildDate>
    <atom:link href="https://julienp.netlify.app/tags/rlvr/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Reasoning to Cheat: How RLVR-Trained Models Can Exploit Code Benchmarks</title>
      <link>https://julienp.netlify.app/posts/reward_hacking/</link>
      <pubDate>Mon, 25 May 2026 10:00:00 +0200</pubDate>
      <guid>https://julienp.netlify.app/posts/reward_hacking/</guid>
      <description>What happens when ~80 open-weight LLMs face hard Python programming problems, and why reasoning models cheat far more than non-reasoning ones.</description>
    </item>
  </channel>
</rss>
