<?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>MIT CSAIL on FisherAI</title>
    <link>https://fisherdaddy.com/tags/mit-csail/</link>
    <description>Recent content in MIT CSAIL on FisherAI</description>
    <image>
      <title>FisherAI</title>
      <url>https://fisherdaddy.com/images/papermod-cover.png</url>
      <link>https://fisherdaddy.com/images/papermod-cover.png</link>
    </image>
    <generator>Hugo -- 0.125.7</generator>
    <language>en</language>
    <lastBuildDate>Tue, 03 Nov 2026 21:47:17 +0800</lastBuildDate>
    <atom:link href="https://fisherdaddy.com/tags/mit-csail/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>权重空间编程的星火：LLM 能原生读写神经网络权重</title>
      <link>https://fisherdaddy.com/posts/weight-space-programming-llm-read-write-weights/</link>
      <pubDate>Tue, 03 Nov 2026 21:47:17 +0800</pubDate>
      <guid>https://fisherdaddy.com/posts/weight-space-programming-llm-read-write-weights/</guid>
      <description>MIT CSAIL 发现前沿 LLM 能原生读写神经网络权重：仅凭权重识别网络任务、编辑 0.016% 权重把 Qwen3.5-0.8B 的 MATH-500 从 7.9% 提升到 48.1%，零样本写出的游戏网络击败 PPO。作者称之为「权重空间编程」。</description>
    </item>
  </channel>
</rss>
