<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
<channel>
  <title>AI 武林：Sonnet</title>
  <link>https://aiwulin.itsmygo.uk/tools/sonnet/</link>
  <description>AI 武林收錄的內容裡，最新提到 Sonnet 的 20 筆，每筆附中文摘要。</description>
  <language>zh-TW</language>
  <lastBuildDate>Thu, 08 Oct 2026 17:22:41 GMT</lastBuildDate>
  <atom:link href="https://aiwulin.itsmygo.uk/tools/sonnet/rss.xml" rel="self" type="application/rss+xml"/>
  <item>
    <title>Using Claude Opus 5.5 as your daily driver</title>
    <link>https://aiwulin.itsmygo.uk/c/296c77e93e/</link>
    <guid isPermaLink="false">aiwulin-296c77e93e</guid>
    <pubDate>Thu, 24 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>Claude</dc:creator>
    <description>&lt;p&gt;展示 Claude Opus 5.5 相比前代 Opus 5 在速度、準確度與成本上的提升，透過實際程式碼修復案例對比兩者表現。觀眾能了解如何設定努力程度與使用子代理來最佳化效能，並掌握 API 審計命令的用法。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=jKRl_CSVxyI&quot;&gt;https://www.youtube.com/watch?v=jKRl_CSVxyI&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>模型發布與實測</category>
  </item>
  <item>
    <title>Designing How AI Grows — Tom McGrath</title>
    <link>https://aiwulin.itsmygo.uk/c/e8029b042b/</link>
    <guid isPermaLink="false">aiwulin-e8029b042b</guid>
    <pubDate>Thu, 03 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>Machine Learning Street Talk</dc:creator>
    <description>&lt;p&gt;Tom McGrath 探討神經網路如何像自然科學般學習概念，並提出將可解釋性納入訓練迴路以實現有意義的設計。他分析 AlphaZero 的知識收斂性、模型如何發現錯誤並產生誤對齊，以及稀疏自編碼器在理解高維結構上的侷限。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://podcasters.spotify.com/pod/show/machinelearningstreettalk/episodes/Designing-How-AI-Grows--Tom-McGrath-e3o8p04&quot;&gt;https://podcasters.spotify.com/pod/show/machinelearningstreettalk/episodes/Designing-How-AI-Grows--Tom-McGrath-e3o8p04&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>可解釋性</category>
    <category>AI 安全與治理</category>
    <category>AI 與科學研究</category>
  </item>
  <item>
    <title>高效电商 AI 智能体解剖指南 | 选自 Anthropic 的 Claude 博客</title>
    <link>https://aiwulin.itsmygo.uk/c/0a920e0082/</link>
    <guid isPermaLink="false">aiwulin-0a920e0082</guid>
    <pubDate>Wed, 02 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>寶玉</dc:creator>
    <description>&lt;p&gt;深入解析如何構建高效且低延遲的電商 AI 智慧體，涵蓋架構設計、成本最佳化與生產環境部署。讀者可學習如何透過技能模組替代子智慧體以降低延遲，並掌握評估體系與安全策略，讓 AI 系統能穩定執行於實際商業場景。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://baoyu.io/translations/2026-09-02/the-anatomy-of-effective-commerce-agents&quot;&gt;https://baoyu.io/translations/2026-09-02/the-anatomy-of-effective-commerce-agents&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>AI Agent 基礎</category>
  </item>
  <item>
    <title>Anthropic Is &quot;Increasing&quot; Your Limits</title>
    <link>https://aiwulin.itsmygo.uk/c/f914137b01/</link>
    <guid isPermaLink="false">aiwulin-f914137b01</guid>
    <pubDate>Mon, 31 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>Theo - t3․gg</dc:creator>
    <description>&lt;p&gt;討論 Anthropic 將 Claude Code 周限永久調降 17%，卻用「增加 25%」的誤導性說法，並批評其溝通方式與對 Fable 模型的嚴格限制。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=Q7n0PGbMW_U&quot;&gt;https://www.youtube.com/watch?v=Q7n0PGbMW_U&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>Claude Code</category>
    <category>Coding Agent</category>
  </item>
  <item>
    <title>Cisco &amp; Stanford on Why Skills Are the New Code</title>
    <link>https://aiwulin.itsmygo.uk/c/e94dc37a24/</link>
    <guid isPermaLink="false">aiwulin-e94dc37a24</guid>
    <pubDate>Wed, 26 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>AI Native Dev</dc:creator>
    <description>&lt;p&gt;探討技能工程如何取代模型選擇，成為決定 AI 代理效能的關鍵。透過實測資料揭示技能濫用、版本滯後及無視性問題，並展示中階模型搭配精準上下文即可達成高價值。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=PabKgxoegGg&quot;&gt;https://www.youtube.com/watch?v=PabKgxoegGg&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>Agent Skills 與 Harness</category>
    <category>上下文工程</category>
    <category>AI Agent 基礎</category>
  </item>
  <item>
    <title>Telling your agent “no” is a moat now, rearward deployed engineers, and harnessing the context for your SDLC</title>
    <link>https://aiwulin.itsmygo.uk/c/e66e65a261/</link>
    <guid isPermaLink="false">aiwulin-e66e65a261</guid>
    <pubDate>Sat, 15 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>Dev Interrupted</dc:creator>
    <description>&lt;p&gt;探討 Uber 將工程師外派至非技術部門（如法律、行銷）以推廣 AI 工作流的「後向部署」策略，並分析 Claude Code 預設開啟安全自動模式、Meta 推出本地執行模型 Muse Glimmer 的動態，以及 Tim O'…&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.buzzsprout.com/1422892/episodes/19648099-telling-your-agent-no-is-a-moat-now-rearward-deployed-engineers-and-harnessing-the-context-for-your-sdlc&quot;&gt;https://www.buzzsprout.com/1422892/episodes/19648099-telling-your-agent-no-is-a-moat-now-rearward-deployed-engineers-and-harnessing-the-context-for-your-sdlc&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>AI Agent 基礎</category>
    <category>Coding Agent</category>
    <category>Claude Code</category>
  </item>
  <item>
    <title>Omnigent: Composition, Control, and Collaboration for AI Agents</title>
    <link>https://aiwulin.itsmygo.uk/c/ae35caef7a/</link>
    <guid isPermaLink="false">aiwulin-ae35caef7a</guid>
    <pubDate>Sun, 02 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>AAIF Live</dc:creator>
    <description>&lt;p&gt;分享如何利用 OmniGent 等工具，透過讓多個 AI 模型互相辯論來提升程式碼品質與研究效率。講者分享實戰經驗，包括用不同模型測試、處理繁體中文語料以及建立 AI 代理的工作流程。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=MQqV-v5HqaU&quot;&gt;https://www.youtube.com/watch?v=MQqV-v5HqaU&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>AI Agent 基礎</category>
    <category>Coding Agent</category>
  </item>
  <item>
    <title>The Tessl Agent: Build Your Software Factory on Autopilot</title>
    <link>https://aiwulin.itsmygo.uk/c/d2271aaf27/</link>
    <guid isPermaLink="false">aiwulin-d2271aaf27</guid>
    <pubDate>Tue, 30 Jun 2026 00:00:00 +0800</pubDate>
    <dc:creator>The AI Native Dev</dc:creator>
    <description>&lt;p&gt;Tessl 產品負責人 Simon Maple 在播客中介紹 Tessl Agent，這是一款用於建立自動化軟體工廠的介面。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.buzzsprout.com/2375985/episodes/19421665-the-tessl-agent-build-your-software-factory-on-autopilot&quot;&gt;https://www.buzzsprout.com/2375985/episodes/19421665-the-tessl-agent-build-your-software-factory-on-autopilot&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>AI Agent 基礎</category>
    <category>Coding Agent</category>
    <category>工作流自動化</category>
  </item>
  <item>
    <title>The Flat Curve Society</title>
    <link>https://aiwulin.itsmygo.uk/c/476ce10324/</link>
    <guid isPermaLink="false">aiwulin-476ce10324</guid>
    <pubDate>Fri, 19 Jun 2026 00:00:00 +0800</pubDate>
    <dc:creator>Steve Yegge（部落格）</dc:creator>
    <description>&lt;p&gt;Steve Yegge 探討 AI 能力曲線看似平坦化的現象，指出因安全限制與人類辨識度天花板，多數人將無法接觸到超級智慧，只能使用現有模型。文章強調企業需重視員工的 AI 素養培訓，並重新評估 SaaS 與自建的成本效益。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://steve-yegge.medium.com/the-flat-curve-society-36c8b01eb33b&quot;&gt;https://steve-yegge.medium.com/the-flat-curve-society-36c8b01eb33b&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>AGI 與長期趨勢</category>
    <category>AI 安全與治理</category>
  </item>
  <item>
    <title>The Age of Async Agents — Cognition's Walden Yan &amp; OpenInspect's Cole Murray</title>
    <link>https://aiwulin.itsmygo.uk/c/cf994e29e8/</link>
    <guid isPermaLink="false">aiwulin-cf994e29e8</guid>
    <pubDate>Fri, 29 May 2026 00:00:00 +0800</pubDate>
    <dc:creator>Latent Space</dc:creator>
    <description>&lt;p&gt;訪談 Cognition 創辦人 Walden Yan 與 OpenInspect 創辦人 Cole Murray，探討背景代理（Background Agents）如何從概念走向實用，以及 Devin 等工具在 2025 年底後的演變。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.latent.space/p/cognition&quot;&gt;https://www.latent.space/p/cognition&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>AI Agent 基礎</category>
    <category>Coding Agent</category>
  </item>
  <item>
    <title>Agents that remember</title>
    <link>https://aiwulin.itsmygo.uk/c/e938824ce1/</link>
    <guid isPermaLink="false">aiwulin-e938824ce1</guid>
    <pubDate>Sat, 23 May 2026 00:00:00 +0800</pubDate>
    <dc:creator>Claude</dc:creator>
    <description>&lt;p&gt;本工作坊示範如何為 Claude 代理建立持久記憶庫，讓其能在不同會話間保留資訊。透過「Dreaming」功能，可自動整理、去重並增強記憶內容，解決記憶庫無限制膨脹與資訊過時的問題。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=geUv4CjPpxI&quot;&gt;https://www.youtube.com/watch?v=geUv4CjPpxI&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>AI Agent 基礎</category>
  </item>
  <item>
    <title>The prompting playbook</title>
    <link>https://aiwulin.itsmygo.uk/c/41b73cb131/</link>
    <guid isPermaLink="false">aiwulin-41b73cb131</guid>
    <pubDate>Fri, 22 May 2026 00:00:00 +0800</pubDate>
    <dc:creator>Claude</dc:creator>
    <description>&lt;p&gt;分享如何透過評估套件與結構化提示最佳化 AI 代理系統，涵蓋從維護舊提示到從零建構新用例的實戰案例。觀眾可學習如何針對失敗模式逐一修復，並利用工具與代理迴圈提升效能。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=G2B0YWuJUgI&quot;&gt;https://www.youtube.com/watch?v=G2B0YWuJUgI&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>提示工程</category>
    <category>AI Agent 基礎</category>
    <category>AI 評測</category>
  </item>
  <item>
    <title>Picking the right model</title>
    <link>https://aiwulin.itsmygo.uk/c/b494252084/</link>
    <guid isPermaLink="false">aiwulin-b494252084</guid>
    <pubDate>Fri, 22 May 2026 00:00:00 +0800</pubDate>
    <dc:creator>Claude</dc:creator>
    <description>&lt;p&gt;分享如何建立專屬評估流程，透過實際測試而非公開榜單來選擇最適合業務場景的 AI 模型。重點在於定義任務成功標準、區分基礎設施失敗與模型表現，並利用思考模式與努力參數調整成本與品質的平衡點。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=P0uMXS6emHA&quot;&gt;https://www.youtube.com/watch?v=P0uMXS6emHA&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>AI 評測</category>
  </item>
  <item>
    <title>Anthropic's Boris Cherny: Coding's Printing Press Moment</title>
    <link>https://aiwulin.itsmygo.uk/c/6ffccdb879/</link>
    <guid isPermaLink="false">aiwulin-6ffccdb879</guid>
    <pubDate>Tue, 05 May 2026 00:00:00 +0800</pubDate>
    <dc:creator>Training Data</dc:creator>
    <description>&lt;p&gt;Boris Cherny 分享 Claude Code 從實驗到爆發的歷程，解釋為何他已完全依賴 AI 寫程式，並介紹 Loop 與 Routines 等自動化工作流。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI2772021536.mp3&quot;&gt;https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI2772021536.mp3&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>Coding Agent</category>
    <category>Claude Code</category>
    <category>工作流自動化</category>
  </item>
  <item>
    <title>Claude Code 省 Token 指南：慎用 1M 上下文，不开新会话或者总是开新会话都不对</title>
    <link>https://aiwulin.itsmygo.uk/c/41cf8b0751/</link>
    <guid isPermaLink="false">aiwulin-41cf8b0751</guid>
    <pubDate>Mon, 06 Apr 2026 00:00:00 +0800</pubDate>
    <dc:creator>寶玉</dc:creator>
    <description>&lt;p&gt;深入解析 Claude Code 的提示快取機制，說明為何頻繁開啟新會話反而更耗錢。作者提供具體策略，包括在活躍會話中繼續工作以維持快取、善用子智慧體隔離上下文、以及設定自動壓縮閾值來控制成本。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://baoyu.io/blog/2026-04-06/claude-code-token-optimization&quot;&gt;https://baoyu.io/blog/2026-04-06/claude-code-token-optimization&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>Claude Code</category>
    <category>上下文工程</category>
  </item>
  <item>
    <title>My Pi Agent Teams. Claude Code Leak SIGNAL. Harness Engineering</title>
    <link>https://aiwulin.itsmygo.uk/c/ae47c95955/</link>
    <guid isPermaLink="false">aiwulin-ae47c95955</guid>
    <pubDate>Mon, 06 Apr 2026 00:00:00 +0800</pubDate>
    <dc:creator>IndyDevDan</dc:creator>
    <description>&lt;p&gt;解析 Claude Code 洩漏後的核心訊號，指出「Agent Harness」是建構自主代理的關鍵基礎設施，而非僅依賴特定模型。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=RairMJflUSA&quot;&gt;https://www.youtube.com/watch?v=RairMJflUSA&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>Agent Skills 與 Harness</category>
    <category>Claude Code</category>
    <category>AI Agent 基礎</category>
  </item>
  <item>
    <title>我们是如何定义 OpenClaw for Teams 新产品形态的｜对谈 Kuse&amp;Junior 联创兼 CTO 宇豪</title>
    <link>https://aiwulin.itsmygo.uk/c/a480bc2ee7/</link>
    <guid isPermaLink="false">aiwulin-a480bc2ee7</guid>
    <pubDate>Thu, 19 Mar 2026 00:00:00 +0800</pubDate>
    <dc:creator>42章经</dc:creator>
    <description>&lt;p&gt;訪談 Kuse.ai 與 Junior.so 聯創兼 CTO 宇豪，探討如何從零融資做到千萬美元 ARR，以及 OpenClaw 類產品的演進。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.xiaoyuzhoufm.com/episode/69bbc8ea3c625cc5ae21b461&quot;&gt;https://www.xiaoyuzhoufm.com/episode/69bbc8ea3c625cc5ae21b461&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>AI 工作術</category>
    <category>AI Agent 基礎</category>
    <category>AI 產品設計</category>
  </item>
  <item>
    <title>The End of Fragmented Agent Context</title>
    <link>https://aiwulin.itsmygo.uk/c/67b821501b/</link>
    <guid isPermaLink="false">aiwulin-67b821501b</guid>
    <pubDate>Wed, 04 Feb 2026 00:00:00 +0800</pubDate>
    <dc:creator>The AI Native Dev</dc:creator>
    <description>&lt;p&gt;Podcast 訪談 Guy Podjarny 與 Simon Maple，探討 AI Agent 技能的現狀、評估與最佳實踐。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.buzzsprout.com/2375985/episodes/18618689-the-end-of-fragmented-agent-context&quot;&gt;https://www.buzzsprout.com/2375985/episodes/18618689-the-end-of-fragmented-agent-context&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>AI Agent 基礎</category>
    <category>Agent Skills 與 Harness</category>
  </item>
  <item>
    <title>Episode 8</title>
    <link>https://aiwulin.itsmygo.uk/c/adca2de9d8/</link>
    <guid isPermaLink="false">aiwulin-adca2de9d8</guid>
    <pubDate>Fri, 22 Aug 2025 00:00:00 +0800</pubDate>
    <dc:creator>Raising An Agent</dc:creator>
    <description>&lt;p&gt;探討 Amp 團隊如何評估與選擇不同大型語言模型用於程式碼代理。內容涵蓋為何工具呼叫能力是關鍵差異，Gemini Pro 在複雜任務中的不足，以及開放模型如 K2 和 Qwen 的潛力。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://1e80eb49-3264-4e2c-b53c-c7308a102a9a.libsyn.com/episode-8&quot;&gt;https://1e80eb49-3264-4e2c-b53c-c7308a102a9a.libsyn.com/episode-8&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>模型發布與實測</category>
    <category>開源模型</category>
    <category>Coding Agent</category>
  </item>
  <item>
    <title>【生成式AI時代下的機器學習(2025)】第八講：大型語言模型的推理過程不用太長、夠用就好</title>
    <link>https://aiwulin.itsmygo.uk/c/e5146a6322/</link>
    <guid isPermaLink="false">aiwulin-e5146a6322</guid>
    <pubDate>Sat, 03 May 2025 00:00:00 +0800</pubDate>
    <dc:creator>Hung-yi Lee</dc:creator>
    <description>&lt;p&gt;探討大型語言模型推理過程過長是否必要，指出推理長度與正確率未必正相關，反而消耗額外算力。透過控制提示詞、調整參數、知識蒸餾及強化學習獎勵機制，可訓練出推理短且準確的模型，達到資源效率與效能的平衡。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=ip3XnTpcxoA&quot;&gt;https://www.youtube.com/watch?v=ip3XnTpcxoA&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>大型語言模型入門</category>
  </item>
</channel>
</rss>
