

<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Yuxuan Zhang — Blog</title>
  <subtitle>Notes on agentic AI, LLM training, and coding agents.</subtitle>
  <link href="https://yuxuan.world/feed.xml" rel="self" type="application/atom+xml"/>
  <link href="https://yuxuan.world/blog/" rel="alternate" type="text/html"/>
  <updated>2026-07-10T00:00:00+08:00</updated>
  <id>https://yuxuan.world/blog/</id>
  <author><name>Yuxuan Zhang</name><uri>https://yuxuan.world</uri></author>
  
  <entry>
    <title>AutoResearch：把研究循环变成可证伪的系统</title>
    <link href="https://yuxuan.world/blog/autoresearch/" rel="alternate" type="text/html"/>
    <published>2026-07-10T00:00:00+08:00</published>
    <updated>2026-07-10T00:00:00+08:00</updated>
    <id>https://yuxuan.world/blog/autoresearch/</id>
    <summary>AutoResearch 不是一个 benchmark，也不是 RSI 的同义词。这篇用可复现的研究循环、评测器边界与保留集验证，解释什么才算可信的自主研究。</summary>
  </entry>
  
  <entry>
    <title>Harness Engineering：从提示词到可问责的智能体循环</title>
    <link href="https://yuxuan.world/blog/harness-engineering/" rel="alternate" type="text/html"/>
    <published>2026-07-10T00:00:00+08:00</published>
    <updated>2026-07-10T00:00:00+08:00</updated>
    <id>https://yuxuan.world/blog/harness-engineering/</id>
    <summary>Harness 不是 prompt 的同义词。本文用工具、状态、权限、预算、轨迹和 evaluator 六个平面，说明如何设计并评估长程智能体系统。</summary>
  </entry>
  
  <entry>
    <title>网页 Agent 环境：把 Browser Use 变成可验证的 RL</title>
    <link href="https://yuxuan.world/blog/web-agent-environments/" rel="alternate" type="text/html"/>
    <published>2026-07-10T00:00:00+08:00</published>
    <updated>2026-07-10T00:00:00+08:00</updated>
    <id>https://yuxuan.world/blog/web-agent-environments/</id>
    <summary>一份面向 ClawBench V2 与 WebsiteBench 的设计说明：用自托管状态、动作拦截、独立 verifier 与证据轨，把网页 Agentic RL 变成可验证的实验。</summary>
  </entry>
  
  <entry>
    <title>RSI：条件性的理论，不是已达成的里程碑</title>
    <link href="https://yuxuan.world/blog/rsi/" rel="alternate" type="text/html"/>
    <published>2026-07-02T00:00:00+08:00</published>
    <updated>2026-07-02T00:00:00+08:00</updated>
    <id>https://yuxuan.world/blog/rsi/</id>
    <summary>递归自我改进（RSI）不是一个已经实现的二元标签。本文从 Good、Gödel Machine 与当前受限自我优化系统出发，说明应当如何严谨地提出、验证和限制 RSI 主张。</summary>
  </entry>
  
  <entry>
    <title>Agent Research Environments：到底在测什么</title>
    <link href="https://yuxuan.world/blog/env/" rel="alternate" type="text/html"/>
    <published>2026-07-02T00:00:00+08:00</published>
    <updated>2026-07-02T00:00:00+08:00</updated>
    <id>https://yuxuan.world/blog/env/</id>
    <summary>把 MLS-Bench、AutoLab、EdgeBench、SForge、Terminal-Bench、Harbor、Browser Use 与 Computer Use 放在同一张测量地图上：它们各自负责什么、如何反馈、哪些结论不能互相迁移。</summary>
  </entry>
  
  <entry>
    <title>Coding Interview 速查：Classic Hot 100 + 手写 MHA / Conv</title>
    <link href="https://yuxuan.world/blog/coding/" rel="alternate" type="text/html"/>
    <published>2026-04-29T00:00:00+08:00</published>
    <updated>2026-04-29T00:00:00+08:00</updated>
    <id>https://yuxuan.world/blog/coding/</id>
    <summary>面试速查：2026 年 4 月的 Classic Hot 100 题型地图，以及可手写、可解释的 PyTorch multi-head attention、1D convolution 和 2D convolution 模板。</summary>
  </entry>
  
  <entry>
    <title>LLM RL 核心算法</title>
    <link href="https://yuxuan.world/blog/rl/" rel="alternate" type="text/html"/>
    <published>2026-04-29T00:00:00+08:00</published>
    <updated>2026-04-29T00:00:00+08:00</updated>
    <id>https://yuxuan.world/blog/rl/</id>
    <summary>LLM RL 三大核心算法速查：PPO、DPO、GRPO，以及 reward model、critic、online RL、offline preference optimization 的区别。</summary>
  </entry>
  
  <entry>
    <title>Agent Harness Engineering: A Source-Referenced Comparison of Coding Agents</title>
    <link href="https://yuxuan.world/blog/harness/" rel="alternate" type="text/html"/>
    <published>2026-04-23T00:00:00+08:00</published>
    <updated>2026-04-23T00:00:00+08:00</updated>
    <id>https://yuxuan.world/blog/harness/</id>
    <summary>A bilingual, interactive deep-dive into hermes-agent, claw-code, codex, opencode, openclaw, and pi — with flowcharts tied to source files and selected line references.</summary>
  </entry>
  
</feed>
