<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
    <channel>
        <title><![CDATA[Kumar Divya Rajat]]></title>
        <description><![CDATA[Writing about software engineering, design, and building things.]]></description>
        <link>https://kumardivyarajat.com</link>
        <generator>RSS for Node</generator>
        <lastBuildDate>Sun, 10 May 2026 10:08:26 GMT</lastBuildDate>
        <atom:link href="https://kumardivyarajat.com/feed.xml" rel="self" type="application/rss+xml"/>
        <language><![CDATA[en]]></language>
        <item>
            <title><![CDATA[Grokking has two phases, and you can see the boundary]]></title>
            <description><![CDATA[A component-freezing ablation study on modular division reveals that grokking isn't one process. It's two: infrastructure setup, then computational reorganization. The boundary between them is sharp and measurable.]]></description>
            <link>https://kumardivyarajat.com/blog/grokking-has-two-phases-and-you-can-see-the-boundary</link>
            <guid isPermaLink="true">https://kumardivyarajat.com/blog/grokking-has-two-phases-and-you-can-see-the-boundary</guid>
            <category><![CDATA[grokking]]></category>
            <category><![CDATA[mechanistic-interpretability]]></category>
            <category><![CDATA[ablation]]></category>
            <category><![CDATA[transformers]]></category>
            <category><![CDATA[experiments]]></category>
            <pubDate>Wed, 01 Apr 2026 00:00:00 GMT</pubDate>
        </item>
        <item>
            <title><![CDATA[One neuron changes everything: the 63 vs 64 puzzle]]></title>
            <description><![CDATA[A 3-layer MLP learning F = m*a behaves completely differently with 63 neurons vs 64. The investigation reveals three training regimes, winner-take-all gradient dynamics, and a lesson about what you're actually ablating when you change architecture.]]></description>
            <link>https://kumardivyarajat.com/blog/one-neuron-changes-everything-the-63-vs-64-puzzle</link>
            <guid isPermaLink="true">https://kumardivyarajat.com/blog/one-neuron-changes-everything-the-63-vs-64-puzzle</guid>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[initialization]]></category>
            <category><![CDATA[training-dynamics]]></category>
            <category><![CDATA[experiments]]></category>
            <category><![CDATA[puzzles]]></category>
            <pubDate>Wed, 01 Apr 2026 00:00:00 GMT</pubDate>
        </item>
        <item>
            <title><![CDATA[Pretrained models converge to deterministic computational subspaces]]></title>
            <description><![CDATA[Four independently trained models find the same subspace geometry. That geometry transfers cross-modally, defines what solutions are reachable, and guides developmental construction toward convergent structure. The evidence, the controls, and what it means.]]></description>
            <link>https://kumardivyarajat.com/blog/pretrained-models-converge-to-deterministic-computational-subspaces</link>
            <guid isPermaLink="true">https://kumardivyarajat.com/blog/pretrained-models-converge-to-deterministic-computational-subspaces</guid>
            <category><![CDATA[neural-networks]]></category>
            <pubDate>Wed, 18 Mar 2026 00:00:00 GMT</pubDate>
        </item>
        <item>
            <title><![CDATA[The genome hypothesis: a thought experiment about what pretraining actually finds]]></title>
            <description><![CDATA[A constraint-based argument for why pretraining might be evolution's outer loop, what the missing developmental stage looks like, and what that means for how we build AI.]]></description>
            <link>https://kumardivyarajat.com/blog/the-genome-hypothesis-a-thought-experiment-about-what-pretraining-actually-finds</link>
            <guid isPermaLink="true">https://kumardivyarajat.com/blog/the-genome-hypothesis-a-thought-experiment-about-what-pretraining-actually-finds</guid>
            <category><![CDATA[evolution]]></category>
            <category><![CDATA[intelligence]]></category>
            <category><![CDATA[thought-experiment]]></category>
            <category><![CDATA[pretraining]]></category>
            <category><![CDATA[cognition]]></category>
            <category><![CDATA[development]]></category>
            <pubDate>Tue, 17 Mar 2026 00:00:00 GMT</pubDate>
        </item>
        <item>
            <title><![CDATA[What if pretrained weights are a genome, not a brain?]]></title>
            <description><![CDATA[Experiments testing whether pretrained weights encode structural blueprints rather than knowledge, and what that means for how we build AI.]]></description>
            <link>https://kumardivyarajat.com/blog/what-if-pretrained-weights-are-a-genome-not-a-brain</link>
            <guid isPermaLink="true">https://kumardivyarajat.com/blog/what-if-pretrained-weights-are-a-genome-not-a-brain</guid>
            <category><![CDATA[evolution]]></category>
            <category><![CDATA[pretraining]]></category>
            <category><![CDATA[svd]]></category>
            <category><![CDATA[representation-learning]]></category>
            <category><![CDATA[experiments]]></category>
            <category><![CDATA[neural-networks]]></category>
            <pubDate>Tue, 17 Mar 2026 00:00:00 GMT</pubDate>
        </item>
        <item>
            <title><![CDATA[The Painting You Couldn't Have Seen]]></title>
            <description><![CDATA[If you painted an animal you never saw, someone described it to you. That's not symbolic thinking. That's language.]]></description>
            <link>https://kumardivyarajat.com/blog/the-painting-you-couldnt-have-seen</link>
            <guid isPermaLink="true">https://kumardivyarajat.com/blog/the-painting-you-couldnt-have-seen</guid>
            <category><![CDATA[evolution]]></category>
            <category><![CDATA[language]]></category>
            <category><![CDATA[intelligence]]></category>
            <pubDate>Mon, 16 Mar 2026 00:00:00 GMT</pubDate>
        </item>
    </channel>
</rss>