<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>超超的小站</title><description>用开发者视角理解 AI：解释重要变化，完成真实实验，记录有效方法与失败边界。</description><link>https://chaochao.example.com/</link><language>zh-CN</language><item><title>结构化输出真正改变了什么：从解析 JSON 到约束接口</title><link>https://chaochao.example.com/radar/structured-outputs-in-practice/</link><guid isPermaLink="true">https://chaochao.example.com/radar/structured-outputs-in-practice/</guid><description>结构化输出的价值不只是少写一次 JSON.parse，而是把模型结果变成可以验证、观测和演进的工程接口。</description><pubDate>Tue, 28 Jul 2026 00:00:00 GMT</pubDate><category>结构化输出</category><category>LLM</category><category>可靠性</category></item><item><title>MCP 进入工程现场后，最先要设计的是权限边界</title><link>https://chaochao.example.com/radar/mcp-boundaries/</link><guid isPermaLink="true">https://chaochao.example.com/radar/mcp-boundaries/</guid><description>工具接入协议降低了连接成本，也让权限、确认、审计和失败恢复成为必须先回答的系统问题。</description><pubDate>Tue, 21 Jul 2026 00:00:00 GMT</pubDate><category>MCP</category><category>Agent</category><category>安全</category></item><item><title>把评测理解成反馈系统，而不是发布前的一张成绩单</title><link>https://chaochao.example.com/concepts/evaluation-is-a-feedback-system/</link><guid isPermaLink="true">https://chaochao.example.com/concepts/evaluation-is-a-feedback-system/</guid><description>有效的 LLM 评测连接样本、评分、错误分类和产品决策，并且会随着真实失败持续更新。</description><pubDate>Tue, 14 Jul 2026 00:00:00 GMT</pubDate><category>评测</category><category>LLM</category><category>质量工程</category></item><item><title>搭一条私有笔记到公开站点的最小发布链</title><link>https://chaochao.example.com/builds/building-a-private-publish-pipeline/</link><guid isPermaLink="true">https://chaochao.example.com/builds/building-a-private-publish-pipeline/</guid><description>一次发布链实验：用显式发布门、临时目录和原子替换隔开私有知识库与公开站点。</description><pubDate>Tue, 07 Jul 2026 00:00:00 GMT</pubDate><category>静态站点</category><category>Obsidian</category><category>内容工程</category></item><item><title>如何记录一次可复现的 AI 实验</title><link>https://chaochao.example.com/guides/reproducible-ai-experiment/</link><guid isPermaLink="true">https://chaochao.example.com/guides/reproducible-ai-experiment/</guid><description>一份面向开发者的实验记录清单：固定输入、环境和判定标准，同时诚实保留失败案例与适用边界。</description><pubDate>Tue, 30 Jun 2026 00:00:00 GMT</pubDate><category>实验方法</category><category>可复现</category><category>工程实践</category></item></channel></rss>