<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
  <title>MATRIX · ScienceClaw-Eval</title>
  <link>https://derekwang2002.github.io/ai-intelligence/en/projects/scienceclaw-eval/</link>
  <description>Material changes in AI models, agents, engineering and infrastructure, with primary sources and recommendations.</description>
  <language>en</language>
  <lastBuildDate>Mon, 02 Nov 2026 16:56:00 GMT</lastBuildDate>
  <atom:link href="https://derekwang2002.github.io/ai-intelligence/en/feeds/projects/scienceclaw-eval.xml" rel="self" type="application/rss+xml"/>
  <item>
    <title>ScienceClaw tests whether scientific agents retain verified improvements</title>
    <link>https://derekwang2002.github.io/ai-intelligence/en/events/ev-20261006-05/</link>
    <guid isPermaLink="false">ev-20261006-05-published</guid>
    <pubDate>Wed, 07 Oct 2026 22:04:39 GMT</pubDate>
    <category>New</category>
    <category>TRIAL</category>
    <description>&lt;p&gt;ScienceClaw-Eval measures whether a fixed model can turn verified task repairs into persistent Skill and Operator changes across sequential scientific work. It spans 23 disciplines and evaluates correctness, improvement, retention, transfer and cost with reset tests. Public code keeps an update only after replay reproduces the repair and independent tasks improve.&lt;/p&gt;&lt;p&gt;Recommendation: TRIAL&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://arxiv.org/abs/2610.08691&quot;&gt;arXiv 2610.08691&lt;/a&gt; · &lt;a href=&quot;https://github.com/beita6969/ScienceClaw&quot;&gt;beita6969/ScienceClaw · GitHub&lt;/a&gt;&lt;/p&gt;</description>
  </item>
</channel>
</rss>
