<?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>AI Research on Amit Naik</title>
    <link>https://amitnaik.com/tags/ai-research/</link>
    <description>Recent content in AI Research on Amit Naik</description>
    <generator>Hugo -- 0.144.2</generator>
    <language>en-us</language>
    <lastBuildDate>Mon, 28 Sep 2026 08:59:57 -0700</lastBuildDate>
    <atom:link href="https://amitnaik.com/tags/ai-research/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Sakana explores alternative paths to intelligence</title>
      <link>https://amitnaik.com/til/2026-09-28-sakana-explores-alternative-paths-to-intelligence/</link>
      <pubDate>Mon, 28 Sep 2026 08:59:57 -0700</pubDate>
      <guid>https://amitnaik.com/til/2026-09-28-sakana-explores-alternative-paths-to-intelligence/</guid>
      <description>Sakana AI introduces its Frontier Intelligence Group, focused on exploring alternative approaches to intelligence. The research direction emphasizes learning from nature and questioning whether today&amp;#39;s dominant methods are the only useful path forward.</description>
    </item>
    <item>
      <title>Distilling a model without its logits</title>
      <link>https://amitnaik.com/til/2026-09-28-distilling-a-model-without-its-logits/</link>
      <pubDate>Mon, 28 Sep 2026 08:59:49 -0700</pubDate>
      <guid>https://amitnaik.com/til/2026-09-28-distilling-a-model-without-its-logits/</guid>
      <description>This reply links to research on black-box, on-policy distillation. The paper&amp;#39;s Generative Adversarial Distillation approach pairs a student model with a discriminator that learns to distinguish student and teacher responses, providing adaptive feedback without access to the teacher&amp;#39;s internal parameters.</description>
    </item>
    <item>
      <title>ToolGrad builds tool-use data from successful chains</title>
      <link>https://amitnaik.com/til/2026-09-28-toolgrad-builds-tool-use-data-from-successful-chains/</link>
      <pubDate>Mon, 28 Sep 2026 08:59:44 -0700</pubDate>
      <guid>https://amitnaik.com/til/2026-09-28-toolgrad-builds-tool-use-data-from-successful-chains/</guid>
      <description>Google Research describes ToolGrad, which starts from successful tool-use chains and then generates matching prompts. This reverses the common prompt-first process and uses execution feedback to improve the quality of training examples.</description>
    </item>
  </channel>
</rss>
