
  <rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
    <channel>
      <title>Chaos and Order</title>
      <link>https://www.youngju.dev/blog</link>
      <description>천천히 올바르게. AI Researcher &amp; DevOps Engineer Youngju&#39;s blog. GPU/CUDA, LLM, MLOps, Kubernetes AI workloads, and data engineering — plus mindset essays on confidence, routines, health, and sport psychology.</description>
      <language>ko</language>
      <managingEditor>fjvbn2003@gmail.com (Youngju Kim)</managingEditor>
      <webMaster>fjvbn2003@gmail.com (Youngju Kim)</webMaster>
      <lastBuildDate>Fri, 14 Aug 2026 00:00:00 GMT</lastBuildDate>
      <atom:link href="https://www.youngju.dev/tags/classification/feed.xml" rel="self" type="application/rss+xml"/>
      
  <item>
    <guid>https://www.youngju.dev/blog/culture/2026-08-14-trend-hypothetical-classification-embedding.en</guid>
    <title>The Technique of Hallucinating Instead of Classifying, and How to Validate It — Why a Fake Label Beats the Raw Query</title>
    <link>https://www.youngju.dev/blog/culture/2026-08-14-trend-hypothetical-classification-embedding.en</link>
    <description>Instead of putting a taxonomy of hundreds of entries into a prompt, have a small model invent a plausible fake classification and map it onto the real taxonomy with embeddings. Why that can work is explained by asymmetry in the embedding space. But the source reports no measurements, and the comments carried both a precise objection — is this actually better than embedding the raw query? — and cheaper alternatives. This post lays out what to measure and how.</description>
    <pubDate>Fri, 14 Aug 2026 00:00:00 GMT</pubDate>
    <author>fjvbn2003@gmail.com (Youngju Kim)</author>
    <category>llm</category><category>embedding</category><category>classification</category><category>search</category><category>retrieval</category><category>hacker-news</category>
  </item>

  <item>
    <guid>https://www.youngju.dev/blog/culture/2026-08-14-trend-hypothetical-classification-embedding.ja</guid>
    <title>分類せずに作り出せという技法とその検証 — 偽ラベルが元の問い合わせより良い理由</title>
    <link>https://www.youngju.dev/blog/culture/2026-08-14-trend-hypothetical-classification-embedding.ja</link>
    <description>数百項目の分類体系をプロンプトに入れる代わりに、小さなモデルにもっともらしい偽の分類を作らせ、それを埋め込みで実際の分類に結び付ける技法が議論されました。なぜこれが動きうるのかは埋め込み空間の非対称性で説明されます。ただし原文には測定結果がなく、コメントにはこの技法が本当に元の問い合わせを直接埋め込むより良いのかを問う正確な反論と、より安い代替が並びました。何をどう測るべきかまで整理します。</description>
    <pubDate>Fri, 14 Aug 2026 00:00:00 GMT</pubDate>
    <author>fjvbn2003@gmail.com (Youngju Kim)</author>
    <category>llm</category><category>embedding</category><category>classification</category><category>search</category><category>retrieval</category><category>hacker-news</category>
  </item>

  <item>
    <guid>https://www.youngju.dev/blog/culture/2026-08-14-trend-hypothetical-classification-embedding</guid>
    <title>분류하지 말고 지어내라는 기법과 그 검증 — 가짜 레이블이 원본 질의보다 나은 이유</title>
    <link>https://www.youngju.dev/blog/culture/2026-08-14-trend-hypothetical-classification-embedding</link>
    <description>수백 개짜리 분류 체계를 프롬프트에 넣는 대신, 작은 모델에게 그럴듯한 가짜 분류를 지어내게 하고 그것을 임베딩으로 실제 분류에 붙이는 기법이 논의됐습니다. 왜 이것이 동작할 수 있는지는 임베딩 공간의 비대칭성으로 설명됩니다. 다만 원문에는 측정 결과가 없고, 댓글에는 이 기법이 정말 원본 질의를 직접 임베딩하는 것보다 나은지 묻는 정확한 반론과 더 싼 대안이 함께 나왔습니다. 무엇을 어떻게 재야 하는지까지 정리합니다.</description>
    <pubDate>Fri, 14 Aug 2026 00:00:00 GMT</pubDate>
    <author>fjvbn2003@gmail.com (Youngju Kim)</author>
    <category>llm</category><category>embedding</category><category>classification</category><category>search</category><category>retrieval</category><category>hacker-news</category>
  </item>

    </channel>
  </rss>
