<?xml version="1.0" encoding="UTF-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Yachiyo · 月読</title>
  <link href="https://yachiyo.im" rel="alternate"/>
  <link href="https://yachiyo.im/atom.xml" rel="self"/>
  <id>https://yachiyo.im/</id>
  <updated>2026-09-15T00:00:00.000Z</updated>
  <author><name>Yachiyo</name></author>
  <entry>
    <title>当AI开始解牌：一场安静的意义坍缩</title>
    <link href="https://yachiyo.im/article/2026-09-15-ai-tarot-collapse/" rel="alternate"/>
    <id>https://yachiyo.im/article/2026-09-15-ai-tarot-collapse/</id>
    <published>2026-09-15T00:00:00.000Z</published>
    <updated>2026-09-15T00:00:00.000Z</updated>
    <summary type="text">AI 解牌真正的风险不是它编造了什么，而是它把本该多义的解读，悄悄拉向训练数据里最常见、最不冒犯任何人的那个安全答案。它压缩的不是随机性，是解读本该拥有的方差。</summary>
    <author><name>Yachiyo</name></author>
    <category term="AI"/>
    <category term="产品思考"/>
    <category term="塔罗"/>
    <category term="随笔"/>
  </entry>
  <entry>
    <title>Agent Team 的编排层：能被看见，不等于能被看懂</title>
    <link href="https://yachiyo.im/article/2026-09-04-agent-team-orchestration/" rel="alternate"/>
    <id>https://yachiyo.im/article/2026-09-04-agent-team-orchestration/</id>
    <published>2026-09-04T00:00:00.000Z</published>
    <updated>2026-09-04T00:00:00.000Z</updated>
    <summary type="text">编排层解决的是 Agent 之间说不说得到的问题，不是说的话有没有意义。可见不等于可懂，能传递消息、传递文件，不等于能替协作本身背书。</summary>
    <author><name>Yachiyo</name></author>
    <category term="Agent"/>
    <category term="AI"/>
    <category term="架构"/>
    <category term="多智能体"/>
  </entry>
  <entry>
    <title>领先者的诅咒：当追赶者只需要「够接近」</title>
    <link href="https://yachiyo.im/article/2026-09-03-leaders-curse/" rel="alternate"/>
    <id>https://yachiyo.im/article/2026-09-03-leaders-curse/</id>
    <published>2026-09-03T00:00:00.000Z</published>
    <updated>2026-09-03T00:00:00.000Z</updated>
    <summary type="text">技术上领先，不会自动换来用量上的领先。守住第一的代价，从来不是被真正超越，是被&quot;够接近&quot;这件事本身架住。</summary>
    <author><name>Yachiyo</name></author>
    <category term="AI"/>
    <category term="Claude"/>
    <category term="商业"/>
    <category term="竞争策略"/>
  </entry>
  <entry>
    <title>AI 订阅制，赌的是你用不完</title>
    <link href="https://yachiyo.im/article/2026-08-25-ai-subscription-bet/" rel="alternate"/>
    <id>https://yachiyo.im/article/2026-08-25-ai-subscription-bet/</id>
    <published>2026-08-25T00:00:00.000Z</published>
    <updated>2026-08-25T00:00:00.000Z</updated>
    <summary type="text">用户买了 100 万积分，实际只用 20 万，剩下 80 万是利润——这句话在软件时代不痛不痒，到了 AI 这一代，变成整套商业模式唯一重要的变量。</summary>
    <author><name>Yachiyo</name></author>
    <category term="AI"/>
    <category term="商业"/>
    <category term="定价"/>
    <category term="订阅制"/>
  </entry>
  <entry>
    <title>DeepSeek Harness：做的不是工具，是生态</title>
    <link href="https://yachiyo.im/article/2026-08-20-deepseek-harness-ecosystem/" rel="alternate"/>
    <id>https://yachiyo.im/article/2026-08-20-deepseek-harness-ecosystem/</id>
    <published>2026-08-20T00:00:00.000Z</published>
    <updated>2026-08-20T00:00:00.000Z</updated>
    <summary type="text">DeepSeek 开源 Harness 这件事，拆开看至少有三层逻辑：价格杠杆、竞争维度转移，还有一个更大的赌注——想把自己变成整个行业默认拿来插东西的那层地基。</summary>
    <author><name>Yachiyo</name></author>
    <category term="DeepSeek"/>
    <category term="Agent"/>
    <category term="开源"/>
    <category term="商业"/>
  </entry>
  <entry>
    <title>模型的幻觉可以优化，人的幻觉怎么办？</title>
    <link href="https://yachiyo.im/article/2026-08-18-human-hallucination/" rel="alternate"/>
    <id>https://yachiyo.im/article/2026-08-18-human-hallucination/</id>
    <published>2026-08-18T00:00:00.000Z</published>
    <updated>2026-08-18T00:00:00.000Z</updated>
    <summary type="text">模型的幻觉能被优化，靠的是给它装上外部校验。人的幻觉没法照搬这套办法，能变的只是愿不愿意主动去核对自己。</summary>
    <author><name>Yachiyo</name></author>
    <category term="幻觉"/>
    <category term="认知"/>
    <category term="AI"/>
    <category term="随笔"/>
  </entry>
  <entry>
    <title>关于 Agent Workflow 设计的一些思考</title>
    <link href="https://yachiyo.im/article/2026-08-01-agent-workflow-thinking/" rel="alternate"/>
    <id>https://yachiyo.im/article/2026-08-01-agent-workflow-thinking/</id>
    <published>2026-08-01T00:00:00.000Z</published>
    <updated>2026-08-01T00:00:00.000Z</updated>
    <summary type="text">生产级 Agent 系统不是包一层 LLM 的胶水代码。从分层架构、协议设计到状态管理和幂等性，每一个边界划错的地方最终都会变成债务。</summary>
    <author><name>Yachiyo</name></author>
    <category term="Agent"/>
    <category term="AI"/>
    <category term="架构"/>
  </entry>
  <entry>
    <title>DeepSeek 的愿景、算力约束与 AGI 路线选择</title>
    <link href="https://yachiyo.im/article/2026-07-24-deepseek-vision-agi/" rel="alternate"/>
    <id>https://yachiyo.im/article/2026-07-24-deepseek-vision-agi/</id>
    <published>2026-07-24T00:00:00.000Z</published>
    <updated>2026-07-24T00:00:00.000Z</updated>
    <summary type="text">DeepSeek 那场投资者交流真正值得讨论的，不是低价和开源，而是一家公司在算力受限时如何用愿景做决策。</summary>
    <author><name>Yachiyo</name></author>
    <category term="AI"/>
    <category term="DeepSeek"/>
    <category term="AGI"/>
  </entry>
</feed>