<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Mathis Fajeau · Blog</title>
    <link>https://mtsfj.github.io/blog/</link>
    <atom:link href="https://mtsfj.github.io/blog/feed.xml" rel="self" type="application/rss+xml"/>
    <description>Interactive articles and notes: autotelic exploration, world models, evaluating large generative models.</description>
    <language>en</language>
    <lastBuildDate>Thu, 29 Oct 2026 00:00:00 GMT</lastBuildDate>
    <item>
      <title>Pinductor: learning POMDP world models from observations with language-model priors</title>
      <link>https://mtsfj.github.io/articles/pinductor.html</link>
      <guid>https://mtsfj.github.io/articles/pinductor.html</guid>
      <description>An LLM proposes programs as hypotheses about the dynamics of a partially observable environment. A particle filter selects among them from observations alone. An interactive walkthrough of the NeurIPS 2026 paper.</description>
    </item>
    <item>
      <title>Autotelic quality-diversity for discovering LLM sycophancy</title>
      <link>https://mtsfj.github.io/articles/sycophancy.html</link>
      <guid>https://mtsfj.github.io/articles/sycophancy.html</guid>
      <description>A repertoire instead of a score: a loop that invents its own attacks and maps 164 tactics that make a local model give in.</description>
    </item>
  </channel>
</rss>
