<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Canace 的博客</title>
  
  <subtitle>Tomorrow is another day</subtitle>
  <link href="https://canace.site/atom.xml" rel="self"/>
  
  <link href="https://canace.site/"/>
  <updated>2026-07-12T04:03:04.272Z</updated>
  <id>https://canace.site/</id>
  
  <author>
    <name>Canace</name>
    
  </author>
  
  <generator uri="https://hexo.io/">Hexo</generator>
  
  <entry>
    <title>搞懂 GPT-5.6：Sol、Terra、Luna 和 Thinking Effort</title>
    <link href="https://canace.site/gpt-5-6-sol-terra-luna/"/>
    <id>https://canace.site/gpt-5-6-sol-terra-luna/</id>
    <published>2026-07-10T02:49:12.000Z</published>
    <updated>2026-07-12T04:03:04.272Z</updated>
    
    
    <summary type="html">OpenAI 从 GPT-5.6 起用 Sol / Terra / Luna 表示能力层级，Thinking Effort 则是另一维的思考时长参数。</summary>
    
    
    
    <category term="AI探索" scheme="https://canace.site/categories/intelligent/"/>
    
    
    <category term="AI原理" scheme="https://canace.site/tags/AI%E5%8E%9F%E7%90%86/"/>
    
  </entry>
  
  <entry>
    <title>在 CLI 里用 Claude Design 做原型设计</title>
    <link href="https://canace.site/claude-design-cli-prototyping/"/>
    <id>https://canace.site/claude-design-cli-prototyping/</id>
    <published>2026-07-09T06:56:25.000Z</published>
    <updated>2026-07-12T04:03:07.763Z</updated>
    
    
    <summary type="html">配置 Claude Design MCP 后，可在终端用自然语言生成设计系统、导出交互原型，并同步成前端组件代码。</summary>
    
    
    
    <category term="AI工程化" scheme="https://canace.site/categories/AI%E5%B7%A5%E7%A8%8B%E5%8C%96/"/>
    
    
    <category term="AI编程" scheme="https://canace.site/tags/AI%E7%BC%96%E7%A8%8B/"/>
    
  </entry>
  
  <entry>
    <title>Vitest 笔记</title>
    <link href="https://canace.site/vitest-vs-jest-advantages/"/>
    <id>https://canace.site/vitest-vs-jest-advantages/</id>
    <published>2026-06-29T02:00:00.000Z</published>
    <updated>2026-06-29T03:22:09.517Z</updated>
    
    
    <summary type="html">Vitest 相对 Jest 的配置复用、ESM 支持与开发体验对比，附 vi/beforeEach 用法、版本要求及 it/test 语法来源</summary>
    
    
    
    <category term="工程化与运维" scheme="https://canace.site/categories/cli/"/>
    
    
    <category term="代码质量" scheme="https://canace.site/tags/code-quality/"/>
    
  </entry>
  
  <entry>
    <title>原来我一直用错了 Cowork</title>
    <link href="https://canace.site/use-cowork/"/>
    <id>https://canace.site/use-cowork/</id>
    <published>2026-06-23T02:00:00.000Z</published>
    <updated>2026-06-23T03:01:40.568Z</updated>
    
    
    <summary type="html">Cowork、Claude Code 与 Cursor 定位不同，token 消耗差异大</summary>
    
    
    
    <category term="AI工程化" scheme="https://canace.site/categories/AI%E5%B7%A5%E7%A8%8B%E5%8C%96/"/>
    
    
    <category term="AI编程" scheme="https://canace.site/tags/AI%E7%BC%96%E7%A8%8B/"/>
    
  </entry>
  
  <entry>
    <title>AI Native 不是接个 API：我们踩坑总结的方法论</title>
    <link href="https://canace.site/ai-native/"/>
    <id>https://canace.site/ai-native/</id>
    <published>2026-06-20T02:00:00.000Z</published>
    <updated>2026-06-21T05:07:33.522Z</updated>
    
    
    <summary type="html">为游戏研发团队做 AI Native 应用的一年实践，总结 AI 化与 AI Native 的差别，以及围绕业务实体重构生产体系的方法</summary>
    
    
    
    <category term="AI工程化" scheme="https://canace.site/categories/AI%E5%B7%A5%E7%A8%8B%E5%8C%96/"/>
    
    
    <category term="AI编程" scheme="https://canace.site/tags/AI%E7%BC%96%E7%A8%8B/"/>
    
  </entry>
  
  <entry>
    <title>跟 AI 说 Ok，它为什么有时不继续</title>
    <link href="https://canace.site/go-ahead-vs-continue-ai-chat/"/>
    <id>https://canace.site/go-ahead-vs-continue-ai-chat/</id>
    <published>2026-06-17T02:00:00.000Z</published>
    <updated>2026-06-21T05:08:52.229Z</updated>
    
    
    <summary type="html">go ahead 和 continue 在跟 AI 对话时该怎么选，以及为什么单独回 ok 常常拦不住 AI 停下来</summary>
    
    
    
    <category term="AI工程化" scheme="https://canace.site/categories/AI%E5%B7%A5%E7%A8%8B%E5%8C%96/"/>
    
    
    <category term="AI编程" scheme="https://canace.site/tags/AI%E7%BC%96%E7%A8%8B/"/>
    
  </entry>
  
  <entry>
    <title>程序员愿意为 Claude 写文档，但不愿为同事写</title>
    <link href="https://canace.site/claude-handoff-doc-to-repo/"/>
    <id>https://canace.site/claude-handoff-doc-to-repo/</id>
    <published>2026-06-13T02:00:00.000Z</published>
    <updated>2026-06-21T05:07:33.513Z</updated>
    
    
    <summary type="html">Mark Dominus 把 Claude 交接文档和项目总结审阅后提交进仓库，AI 写、人来审，解决没人愿意写文档的老问题。</summary>
    
    
    
    <category term="AI工程化" scheme="https://canace.site/categories/AI%E5%B7%A5%E7%A8%8B%E5%8C%96/"/>
    
    
    <category term="AI编程" scheme="https://canace.site/tags/AI%E7%BC%96%E7%A8%8B/"/>
    
  </entry>
  
  <entry>
    <title>RAG vs Graph RAG</title>
    <link href="https://canace.site/rag-vs-graph-rag/"/>
    <id>https://canace.site/rag-vs-graph-rag/</id>
    <published>2026-06-02T23:41:25.000Z</published>
    <updated>2026-06-21T05:07:42.452Z</updated>
    
    
    <summary type="html">对比传统 RAG 与 Graph RAG 的工作原理、适用场景与成本差异，帮助选择合适的知识库检索方案。</summary>
    
    
    
    <category term="AI工程化" scheme="https://canace.site/categories/AI%E5%B7%A5%E7%A8%8B%E5%8C%96/"/>
    
    
    <category term="AI编程" scheme="https://canace.site/tags/AI%E7%BC%96%E7%A8%8B/"/>
    
  </entry>
  
  <entry>
    <title>LLM 训练的真相：Andrej Karpathy 的分享笔记</title>
    <link href="https://canace.site/llm-training-truth/"/>
    <id>https://canace.site/llm-training-truth/</id>
    <published>2026-05-10T02:00:00.000Z</published>
    <updated>2026-05-10T03:52:23.049Z</updated>
    
    
    <summary type="html">结合 Andrej Karpathy 的分享，把 LLM 的诞生拆解成三个核心步骤：爬取、压缩、预测。</summary>
    
    
    
    <category term="AI探索" scheme="https://canace.site/categories/intelligent/"/>
    
    
    <category term="AI原理" scheme="https://canace.site/tags/AI%E5%8E%9F%E7%90%86/"/>
    
  </entry>
  
  <entry>
    <title>AI 如何让我们躺平</title>
    <link href="https://canace.site/how-ai-lets-us-lie-flat/"/>
    <id>https://canace.site/how-ai-lets-us-lie-flat/</id>
    <published>2026-05-03T02:00:00.000Z</published>
    <updated>2026-05-03T02:14:01.731Z</updated>
    
    
    <summary type="html">借「方向要对、提效后要少做工时」讨论躺平：AI 若边界不清，容易加长屏幕时间与无效产出；更合理的用法是在圈定范围内提效，避免越界堆出 vibe coding 却挤占生活与深耕。</summary>
    
    
    
    <category term="软技能与思考" scheme="https://canace.site/categories/soft/"/>
    
    
    <category term="随笔" scheme="https://canace.site/tags/essays/"/>
    
  </entry>
  
  <entry>
    <title>Openclaw定时任务</title>
    <link href="https://canace.site/scheduled-tasks/"/>
    <id>https://canace.site/scheduled-tasks/</id>
    <published>2026-04-23T02:00:00.000Z</published>
    <updated>2026-04-25T08:03:29.453Z</updated>
    
    
    <summary type="html">系统 crontab 与 openclaw 工作空间任务的分工，以及 Mac 上查看编辑与五段式表达式写法备忘。</summary>
    
    
    
    <category term="工程化与运维" scheme="https://canace.site/categories/cli/"/>
    
    
    <category term="开发工具" scheme="https://canace.site/tags/dev-tools/"/>
    
  </entry>
  
  <entry>
    <title>节奏心理学：叙事中的“情绪推手”</title>
    <link href="https://canace.site/rhythm-psychology/"/>
    <id>https://canace.site/rhythm-psychology/</id>
    <published>2026-04-22T04:00:00.000Z</published>
    <updated>2026-04-22T02:39:02.243Z</updated>
    
    
    <summary type="html">“爽点节奏”不仅是网文的套路，更是大脑分配注意力与管理多巴胺的科学。本文将从心理学维度深度拆解叙事节奏的底层逻辑</summary>
    
    
    
    <category term="软技能与思考" scheme="https://canace.site/categories/soft/"/>
    
    
    <category term="笔记" scheme="https://canace.site/tags/notes/"/>
    
  </entry>
  
  <entry>
    <title>叙事研究</title>
    <link href="https://canace.site/narrative-research/"/>
    <id>https://canace.site/narrative-research/</id>
    <published>2026-04-22T02:00:00.000Z</published>
    <updated>2026-04-22T02:39:28.668Z</updated>
    
    
      
      
        
        
    <summary type="html"></summary>
        
      
    
    
    
    <category term="软技能与思考" scheme="https://canace.site/categories/soft/"/>
    
    
    <category term="笔记" scheme="https://canace.site/tags/notes/"/>
    
  </entry>
  
  <entry>
    <title>OpenClaw 的一些使用体验</title>
    <link href="https://canace.site/openclaw-usage-experience/"/>
    <id>https://canace.site/openclaw-usage-experience/</id>
    <published>2026-04-18T02:00:00.000Z</published>
    <updated>2026-06-21T05:07:33.533Z</updated>
    
    
    <summary type="html">围绕 OpenClaw 的 memory 与自建 wiki 同步 Notion、RSS 筛选写入数据库、条件盯盘与自研 Prompt 工具整合等场景的实践记录，并顺带写了对模型是否遵守记忆规则、QQ 推送限制与 SOUL 设定的一点感受。</summary>
    
    
    
    <category term="AI工程化" scheme="https://canace.site/categories/AI%E5%B7%A5%E7%A8%8B%E5%8C%96/"/>
    
    
    <category term="AI编程" scheme="https://canace.site/tags/AI%E7%BC%96%E7%A8%8B/"/>
    
  </entry>
  
  <entry>
    <title>产品更新、RSS订阅以及文件diff类需求实现模型</title>
    <link href="https://canace.site/product-updates-rss-file-diff-mental-model/"/>
    <id>https://canace.site/product-updates-rss-file-diff-mental-model/</id>
    <published>2026-04-14T02:00:00.000Z</published>
    <updated>2026-04-14T05:59:41.094Z</updated>
    
    
    <summary type="html">把 Web 版本更新、RSS、应用更新与文件 diff 抽象为「发现更新—判断差异—处理差异」与五层模型，先定差异粒度再选发现机制与状态描述；附场景对照表、落地四问，并设想 checklist、prompt 与 skill，尚未实践。</summary>
    
    
    
    <category term="软技能与思考" scheme="https://canace.site/categories/soft/"/>
    
    
    <category term="随笔" scheme="https://canace.site/tags/essays/"/>
    
  </entry>
  
  <entry>
    <title>RSS 推送到 TG 的可行性探索</title>
    <link href="https://canace.site/rss-telegram-selfhost/"/>
    <id>https://canace.site/rss-telegram-selfhost/</id>
    <published>2026-04-12T02:00:00.000Z</published>
    <updated>2026-04-12T08:47:30.375Z</updated>
    
    
    <summary type="html">从 Kindle 推送停摆到想在 Telegram 看 RSS：试了一圈机器人后自建 flowerss-bot，顺带记下 Token 与 Docker 里走代理的几件事。</summary>
    
    
    
    <category term="软技能与思考" scheme="https://canace.site/categories/soft/"/>
    
    
    <category term="随笔" scheme="https://canace.site/tags/essays/"/>
    
  </entry>
  
  <entry>
    <title>为什么不要让LLM帮我们写文档</title>
    <link href="https://canace.site/why-not-let-ai-write-for-us/"/>
    <id>https://canace.site/why-not-let-ai-write-for-us/</id>
    <published>2026-04-10T02:00:00.000Z</published>
    <updated>2026-04-10T08:24:54.027Z</updated>
    
    
    <summary type="html">写作是自问自答与建立信任的过程；代写像付费健身，舒适却练不到自己。讨论 LLM 代写对公信力与思考真实性的影响，以及调研、转录与创意场景下如何善用模型。</summary>
    
    
    
    <category term="软技能与思考" scheme="https://canace.site/categories/soft/"/>
    
    
    <category term="随笔" scheme="https://canace.site/tags/essays/"/>
    
  </entry>
  
  <entry>
    <title>使用大模型来维护知识库</title>
    <link href="https://canace.site/llm-wiki/"/>
    <id>https://canace.site/llm-wiki/</id>
    <published>2026-04-08T01:45:45.000Z</published>
    <updated>2026-06-21T05:07:42.448Z</updated>
    
    
    <summary type="html">介绍「原始资料 + 规则文件 + wiki」分工：人只负责输入，大模型维护持久化知识库（索引、对话总结、报告等），作为跨 agent 的长期上下文；说明三层架构、wiki 目录设计与落地步骤，并关联 Karpathy 的 LLM Wiki 思路。</summary>
    
    
    
    <category term="AI工程化" scheme="https://canace.site/categories/AI%E5%B7%A5%E7%A8%8B%E5%8C%96/"/>
    
    
    <category term="AI编程" scheme="https://canace.site/tags/AI%E7%BC%96%E7%A8%8B/"/>
    
  </entry>
  
  <entry>
    <title>Gemini 在打开新会话时，是如何有记忆的</title>
    <link href="https://canace.site/gemini-new-session-memory/"/>
    <id>https://canace.site/gemini-new-session-memory/</id>
    <published>2026-04-02T07:00:00.000Z</published>
    <updated>2026-04-02T03:53:17.674Z</updated>
    
    
    <summary type="html">探讨 Google Gemini 在新对话中仍能呈现「记得你」的体验背后，可能涉及的账户级记忆、Saved info、跨会话摘要等机制与边界。</summary>
    
    
    
    <category term="AI探索" scheme="https://canace.site/categories/intelligent/"/>
    
    
    <category term="AI原理" scheme="https://canace.site/tags/AI%E5%8E%9F%E7%90%86/"/>
    
  </entry>
  
  <entry>
    <title>AI 让人变得无趣</title>
    <link href="https://canace.site/ai-make-you-boring/"/>
    <id>https://canace.site/ai-make-you-boring/</id>
    <published>2026-04-02T02:40:43.000Z</published>
    <updated>2026-04-02T02:47:10.972Z</updated>
    
    
    <summary type="html">由 Hacker News 上关于 Show HN 质量下滑的讨论展开，谈 AI 辅助下的「无趣」项目、思维外包如何削弱原创与沉浸，以及为何提示词替代不了真正的构思过程。</summary>
    
    
    
    <category term="软技能与思考" scheme="https://canace.site/categories/soft/"/>
    
    
    <category term="翻译" scheme="https://canace.site/tags/%E7%BF%BB%E8%AF%91/"/>
    
  </entry>
  
</feed>
