<?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>Physical Reasoning on PaperMoon&#39;s blog</title>
    <link>https://milknocandy.github.io/tags/physical-reasoning/</link>
    <description>Recent content in Physical Reasoning on PaperMoon&#39;s blog</description>
    <image>
      <title>PaperMoon&#39;s blog</title>
      <url>https://milknocandy.github.io/%3Clink%20or%20path%20of%20image%20for%20opengraph,%20twitter-cards%3E</url>
      <link>https://milknocandy.github.io/%3Clink%20or%20path%20of%20image%20for%20opengraph,%20twitter-cards%3E</link>
    </image>
    <generator>Hugo -- 0.154.3</generator>
    <language>en</language>
    <lastBuildDate>Sun, 05 Apr 2026 17:58:05 +0800</lastBuildDate>
    <atom:link href="https://milknocandy.github.io/tags/physical-reasoning/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>When VLMs Become Cognitive Mimics, Not Physical Reasoners: A QuantiPhy Study</title>
      <link>https://milknocandy.github.io/posts/2026-03-23-quantiphy/</link>
      <pubDate>Mon, 23 Mar 2026 16:42:46 +0800</pubDate>
      <guid>https://milknocandy.github.io/posts/2026-03-23-quantiphy/</guid>
      <description>&lt;div class=&#34;paperbox&#34;&gt;
    &lt;div class=&#34;pb-item&#34;&gt;
        &lt;span class=&#34;pb-key&#34;&gt;TOPIC&lt;/span&gt;
        &lt;span class=&#34;pb-sep&#34;&gt;&lt;/span&gt;
        &lt;span class=&#34;pb-val&#34;&gt;Quantitative Physical Understanding&lt;/span&gt;
    &lt;/div&gt;
    &lt;div class=&#34;pb-item&#34;&gt;
        &lt;span class=&#34;pb-key&#34;&gt;WHY READ&lt;/span&gt;
        &lt;span class=&#34;pb-sep&#34;&gt;&lt;/span&gt;
        &lt;span class=&#34;pb-val&#34;&gt;Exposes that top VLMs guess physical quantities from memory (pre-trained world knowledge) rather than measure from video, with rigorous tests to diagnose this failure.&lt;/span&gt;
    &lt;/div&gt;
    &lt;div class=&#34;pb-item&#34;&gt;
        &lt;span class=&#34;pb-key&#34;&gt;TAKEAWAY&lt;/span&gt;
        &lt;span class=&#34;pb-sep&#34;&gt;&lt;/span&gt;
        &lt;span class=&#34;pb-val&#34;&gt;Current VLMs are cognitive mimics not physical reasoners, so build systems that arbitrate between perception and memory rather than forcing pure end to end inference. (Context Learning, Agentic AI)&lt;/span&gt;
    &lt;/div&gt;
    &lt;div class=&#34;pb-links&#34;&gt;
        &lt;span class=&#34;pb-org&#34;&gt;Stanford University, UST&lt;/span&gt;
        &lt;div class=&#34;pb-link-group&#34;&gt;&lt;a href=&#34;https://arxiv.org/abs/2512.19526&#34; target=&#34;_blank&#34; class=&#34;pb-link&#34;&gt;📄 Paper&lt;/a&gt;&lt;a href=&#34;https://github.com/Paulineli/QuantiPhy&#34; target=&#34;_blank&#34; class=&#34;pb-link&#34;&gt;💻 Code&lt;/a&gt;&lt;a href=&#34;https://github.com/Paulineli/QuantiPhy&#34; target=&#34;_blank&#34; class=&#34;pb-link&#34;&gt;🌐 Project&lt;/a&gt;&lt;a href=&#34;https://github.com/Paulineli&#34; target=&#34;_blank&#34; class=&#34;pb-link&#34;&gt;👤 Author&lt;/a&gt;
        &lt;/div&gt;
    &lt;/div&gt;
&lt;/div&gt;

&lt;hr&gt;
&lt;h2 id=&#34;-1-motivation--problem&#34;&gt;🚀 1 Motivation &amp;amp; Problem&lt;/h2&gt;
&lt;p&gt;Humans understand the physical world through structured mathematical abstractions. From Isaac Newton’s formulation of universal gravitation inspired by a falling apple, to modern physics, quantitative laws enable precise reasoning about the dynamics of the real world. In contrast, although state-of-the-art AI systems demonstrate remarkable capabilities in mathematical reasoning, programming, and scientific writing, enabling artificial intelligence to &lt;u&gt;&lt;i&gt;ground its understanding in the physical world&lt;/i&gt;&lt;/u&gt; remains a fundamental and unresolved challenge. This limitation poses a critical barrier to deploying AI systems in real-world, embodied environments.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
