<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>LLM on Subculture Science Research Group&#39;s Blog</title>
    <link>https://www.subcul-science.com/en/tags/llm/</link>
    <description>Recent content in LLM on Subculture Science Research Group&#39;s Blog</description>
    <generator>Hugo</generator>
    <language>en-US</language>
    <lastBuildDate>Sun, 15 Nov 2026 15:30:00 +0900</lastBuildDate>
    <atom:link href="https://www.subcul-science.com/en/tags/llm/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>The Small LLM That Answers Before It Thinks: Four Roles and a Comparison Design</title>
      <link>https://www.subcul-science.com/en/post/20261115umigame-gan-solution/</link>
      <pubDate>Sun, 15 Nov 2026 15:30:00 +0900</pubDate>
      <guid>https://www.subcul-science.com/en/post/20261115umigame-gan-solution/</guid>
      <description>Small LLMs commit to a hunch and stop tracking candidates. This post breaks down the failure modes, describes a loop that grows a creator and a solver in turn, and reports how far a per-turn reward actually moved the numbers.</description>
    </item>
    <item>
      <title>Co-evolving Four Roles, GAN-Style: The Testbed Behind umigame_GAN</title>
      <link>https://www.subcul-science.com/en/post/20261115umigame-gan-tech/</link>
      <pubDate>Sun, 15 Nov 2026 15:20:00 +0900</pubDate>
      <guid>https://www.subcul-science.com/en/post/20261115umigame-gan-tech/</guid>
      <description>A testbed for strengthening trial-and-error reasoning in small LLMs: four competing roles, a domain-independent loop, R-Zero and EDA style rewards, and an MLX LoRA comparison design. Code excerpts and measured numbers from the repository.</description>
    </item>
    <item>
      <title>Inside chatGPT_streamlit: A Search-Enabled Chat UI with Streamlit and LangChain ReAct</title>
      <link>https://www.subcul-science.com/en/post/20261115chatgpt-streamlit-tech/</link>
      <pubDate>Sun, 15 Nov 2026 12:40:00 +0900</pubDate>
      <guid>https://www.subcul-science.com/en/post/20261115chatgpt-streamlit-tech/</guid>
      <description>A tour of chatGPT_streamlit: a LangChain ReAct agent with Google and Wikipedia tools behind a Streamlit chat UI. How the tools, prefix/suffix prompt, and session_state history are wired, plus the version pinning and the gaps between README and code.</description>
    </item>
    <item>
      <title>Cutting the Cost of Knowing What Is Trending on X</title>
      <link>https://www.subcul-science.com/en/post/20261115topic-digestx-solution/</link>
      <pubDate>Sun, 15 Nov 2026 12:30:00 +0900</pubDate>
      <guid>https://www.subcul-science.com/en/post/20261115topic-digestx-solution/</guid>
      <description>Following an X timeline costs three things: finding the right posts, reading them, and summarizing them. How TopicDigestX moves each to a server, the trade-offs, and what is still not built.</description>
    </item>
    <item>
      <title>Fine-tuning a Japanese LLM Without a GPU: One Mac, One Pipeline</title>
      <link>https://www.subcul-science.com/en/post/20261115ft-llm-solution/</link>
      <pubDate>Sun, 15 Nov 2026 11:50:00 +0900</pubDate>
      <guid>https://www.subcul-science.com/en/post/20261115ft-llm-solution/</guid>
      <description>No CUDA GPU, no cloud quota, no separate evaluation scripts. How FT_LLM turns fine-tuning a Japanese LLM into one Mac workflow, and where the approach still falls short.</description>
    </item>
    <item>
      <title>A LoRA Training Pipeline for Japanese LLMs on MLX: LFM2.5-1.2B-JP</title>
      <link>https://www.subcul-science.com/en/post/20261115ft-llm-tech/</link>
      <pubDate>Sun, 15 Nov 2026 11:40:00 +0900</pubDate>
      <guid>https://www.subcul-science.com/en/post/20261115ft-llm-tech/</guid>
      <description>How FT_LLM fine-tunes LFM2.5-1.2B-JP with LoRA on Apple Silicon: config resolution, two execution modes, a chunked loss for long sequences, early stopping, and character-level evaluation, with code excerpts and measured numbers.</description>
    </item>
    <item>
      <title>A Discord Bot That Lives in the Chat: Ollama, discord.js, and a News Pipeline</title>
      <link>https://www.subcul-science.com/en/post/20261115discord-bot-tech/</link>
      <pubDate>Sun, 15 Nov 2026 10:40:00 +0900</pubDate>
      <guid>https://www.subcul-science.com/en/post/20261115discord-bot-tech/</guid>
      <description>How Lobster, a Kansai-dialect Discord bot, is built: one Node process running discord.js and Ollama, persona composition, an instant 👀 acknowledgement, 17 RSS feeds, and interests that change from chat.</description>
    </item>
  </channel>
</rss>
