<?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>InstructGPT on Yuxia&#39;s Blog</title>
    <link>https://luoyuxia.github.io/tags/instructgpt/</link>
    <description>Recent content in InstructGPT on Yuxia&#39;s Blog</description>
    <generator>Hugo</generator>
    <language>zh</language>
    <lastBuildDate>Sat, 23 May 2026 12:00:00 +0800</lastBuildDate>
    <atom:link href="https://luoyuxia.github.io/tags/instructgpt/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>从 GPT-3 到 ChatGPT：AI 为什么突然像助手了</title>
      <link>https://luoyuxia.github.io/posts/%E4%BB%8E-gpt-3-%E5%88%B0-chatgptai-%E4%B8%BA%E4%BB%80%E4%B9%88%E7%AA%81%E7%84%B6%E5%83%8F%E5%8A%A9%E6%89%8B%E4%BA%86/</link>
      <pubDate>Sat, 23 May 2026 12:00:00 +0800</pubDate>
      <guid>https://luoyuxia.github.io/posts/%E4%BB%8E-gpt-3-%E5%88%B0-chatgptai-%E4%B8%BA%E4%BB%80%E4%B9%88%E7%AA%81%E7%84%B6%E5%83%8F%E5%8A%A9%E6%89%8B%E4%BA%86/</guid>
      <description>从 GPT-3 到 ChatGPT，真正发生的变化不是&amp;rsquo;模型更大了&amp;rsquo;，而是训练目标变了。本文梳理了这段技术演进中的四篇关键论文——GPT-3、Learning to Summarize from Human Feedback、InstructGPT、WebGPT，拆解通用能力如何通过指令微调、人类反馈强化学习和工具增强，被重新塑造成一个可用的助手系统。</description>
    </item>
  </channel>
</rss>
