
  <rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
    <channel>
      <title>Koko Blog</title>
      <link>https://adefioye.github.io/blog</link>
      <description>Writing about AI, deep learning, and software engineering.</description>
      <language>en-us</language>
      <managingEditor>address@yoursite.com (Abdulhakeem Adefioye)</managingEditor>
      <webMaster>address@yoursite.com (Abdulhakeem Adefioye)</webMaster>
      <lastBuildDate>Thu, 16 Apr 2026 00:00:00 GMT</lastBuildDate>
      <atom:link href="https://adefioye.github.io/feed.xml" rel="self" type="application/rss+xml"/>
      
  <item>
    <guid>https://adefioye.github.io/blog/spd-1</guid>
    <title>SPD 1: Analysis of component activations of deep toy model of superposition</title>
    <link>https://adefioye.github.io/blog/spd-1</link>
    <description>Analysing patterns of component activations on 2-6L tied and untied SPD-based decomposed TMS models.</description>
    <pubDate>Thu, 16 Apr 2026 00:00:00 GMT</pubDate>
    <author>address@yoursite.com (Abdulhakeem Adefioye)</author>
    <category>llm</category><category>residual stream</category><category>mlp</category><category>concept detection</category><category>gpt2-small</category>
  </item>

  <item>
    <guid>https://adefioye.github.io/blog/snmf-1</guid>
    <title>SNMF 1: How does the type of activations and number of top activating tokens affect concept detection in LLMs?</title>
    <link>https://adefioye.github.io/blog/snmf-1</link>
    <description>A small reproduction and extension of the concept detection results of (Shafran et al., 2025) paper.</description>
    <pubDate>Sun, 22 Feb 2026 00:00:00 GMT</pubDate>
    <author>address@yoursite.com (Abdulhakeem Adefioye)</author>
    <category>llm</category><category>residual stream</category><category>mlp</category><category>concept detection</category><category>gpt2-small</category>
  </item>

  <item>
    <guid>https://adefioye.github.io/blog/keen-paper-repro</guid>
    <title>Reproduction of the KEEN paper on estimation of knowledge on QA dataset</title>
    <link>https://adefioye.github.io/blog/keen-paper-repro</link>
    <description>A small reproduction of the main thesis of the KEEN (Gottesman &amp; Geva, 2024) paper.</description>
    <pubDate>Fri, 16 Jan 2026 00:00:00 GMT</pubDate>
    <author>address@yoursite.com (Abdulhakeem Adefioye)</author>
    <category>linear probe</category><category>llm</category><category>residual stream</category><category>mlp</category><category>attention</category>
  </item>

  <item>
    <guid>https://adefioye.github.io/blog/listwise-loss</guid>
    <title>Listwise loss does not consistently makeup for retrieval performance drop induced by InfoNCE loss</title>
    <link>https://adefioye.github.io/blog/listwise-loss</link>
    <description>Replicating Tamber et al. with Nomic-Embed and ModernBERT-Embed shows that conventional contrastive learning can fail, but not uniformly across embedding models.</description>
    <pubDate>Wed, 12 Nov 2025 00:00:00 GMT</pubDate>
    <author>address@yoursite.com (Abdulhakeem Adefioye)</author>
    <category>dense retrieval</category><category>contrastive learning</category><category>reranking</category><category>distillation</category><category>synthetic data</category>
  </item>

    </channel>
  </rss>
