BASHCAT
    • Create new note
    • Create a note from template
      • Sharing URL Link copied
      • /edit
      • View mode
        • Edit mode
        • View mode
        • Book mode
        • Slide mode
        Edit mode View mode Book mode Slide mode
      • Customize slides
      • Note Permission
      • Read
        • Only me
        • Signed-in users
        • Everyone
        Only me Signed-in users Everyone
      • Write
        • Only me
        • Signed-in users
        • Everyone
        Only me Signed-in users Everyone
      • Engagement control Commenting, Suggest edit, Emoji Reply
    • Invite by email
      Invitee

      This note has no invitees

    • Publish Note

      Share your work with the world Congratulations! 🎉 Your note is out in the world Publish Note No publishing access yet

      Your note will be visible on your profile and discoverable by anyone.
      Your note is now live.
      This note is visible on your profile and discoverable online.
      Everyone on the web can find and read all notes of this public team.

      Your account was recently created. Publishing will be available soon, allowing you to share notes on your public page and in search results.

      Your team account was recently created. Publishing will be available soon, allowing you to share notes on your public page and in search results.

      Explore these features while you wait
      Complete general settings
      Bookmark and like published notes
      Write a few more notes
      Complete general settings
      Write a few more notes
      See published notes
      Unpublish note
      Please check the box to agree to the Community Guidelines.
      View profile
    • Commenting
      Permission
      Disabled Forbidden Owners Signed-in users Everyone
    • Enable
    • Permission
      • Forbidden
      • Owners
      • Signed-in users
      • Everyone
    • Suggest edit
      Permission
      Disabled Forbidden Owners Signed-in users Everyone
    • Enable
    • Permission
      • Forbidden
      • Owners
      • Signed-in users
    • Emoji Reply
    • Enable
    • Versions and GitHub Sync
    • Note settings
    • Note Insights New
    • Engagement control
    • Make a copy
    • Transfer ownership
    • Delete this note
    • Save as template
    • Insert from template
    • Import from
      • Dropbox
      • Google Drive
      • Gist
      • Clipboard
    • Export to
      • Dropbox
      • Google Drive
      • Gist
    • Download
      • Markdown
      • HTML
      • Raw HTML
Menu Note settings Note Insights Versions and GitHub Sync Sharing URL Create Help
Create Create new note Create a note from template
Menu
Options
Engagement control Make a copy Transfer ownership Delete this note
Import from
Dropbox Google Drive Gist Clipboard
Export to
Dropbox Google Drive Gist
Download
Markdown HTML Raw HTML
Back
Sharing URL Link copied
/edit
View mode
  • Edit mode
  • View mode
  • Book mode
  • Slide mode
Edit mode View mode Book mode Slide mode
Customize slides
Note Permission
Read
Only me
  • Only me
  • Signed-in users
  • Everyone
Only me Signed-in users Everyone
Write
Only me
  • Only me
  • Signed-in users
  • Everyone
Only me Signed-in users Everyone
Engagement control Commenting, Suggest edit, Emoji Reply
  • Invite by email
    Invitee

    This note has no invitees

  • Publish Note

    Share your work with the world Congratulations! 🎉 Your note is out in the world Publish Note No publishing access yet

    Your note will be visible on your profile and discoverable by anyone.
    Your note is now live.
    This note is visible on your profile and discoverable online.
    Everyone on the web can find and read all notes of this public team.

    Your account was recently created. Publishing will be available soon, allowing you to share notes on your public page and in search results.

    Your team account was recently created. Publishing will be available soon, allowing you to share notes on your public page and in search results.

    Explore these features while you wait
    Complete general settings
    Bookmark and like published notes
    Write a few more notes
    Complete general settings
    Write a few more notes
    See published notes
    Unpublish note
    Please check the box to agree to the Community Guidelines.
    View profile
    Engagement control
    Commenting
    Permission
    Disabled Forbidden Owners Signed-in users Everyone
    Enable
    Permission
    • Forbidden
    • Owners
    • Signed-in users
    • Everyone
    Suggest edit
    Permission
    Disabled Forbidden Owners Signed-in users Everyone
    Enable
    Permission
    • Forbidden
    • Owners
    • Signed-in users
    Emoji Reply
    Enable
    Import from Dropbox Google Drive Gist Clipboard
       Owned this note    Owned this note      
    Published Linked with GitHub
    • Any changes
      Be notified of any changes
    • Mention me
      Be notified of mention me
    • Unsubscribe
    --- title: Agent 元年的真正起跑線:2026 年 5 月,AI 同時改寫了三件事 date: 2026-05-08 tags: - AI - Agent - Claude - 趨勢 - 部落格 description: 2026 年 5 月第一週,三件看似無關的事在同一週發生 — Anthropic 與 SpaceX 簽 300 MW 算力合約、Trump 政府從放任 AI 改成準備預先審查、PJM 電力容量價格飆到兩年前 10 倍。這不是巧合,是 Agent 元年真正起跑的訊號。 cover: assets/202605-ai-trends/202605-ai-cover-sf-skyline.jpg --- # Agent 元年的真正起跑線:2026 年 5 月,AI 同時改寫了三件事 ![202605-ai-cover-sf-skyline](https://hackmd.io/_uploads/HyazeSsCWl.jpg) 5 月 6 日上午 9 點,舊金山。Dario Amodei 坐在 Code w/ Claude 主舞台的單椅上,宣布 Anthropic 跟 SpaceX 簽下了 Memphis 那座 Colossus 1 超級電腦的 300 MW 算力合約。同一天早上,Anthropic 的工程團隊把 Claude Code 的五小時使用上限直接翻倍、把尖峰時段的限制全拔掉,新功能 Routines 也一起上線——你睡覺,它幫你開好 PR。 把鏡頭拉到華盛頓 DC。同一週,《紐約時報》、Politico、Axios 接力報導:白宮已經起草一份 16 頁的 AI 行政命令,要求所有新模型上市前送聯邦政府預先審查。三個月前還在說「我們不能用愚蠢的規則絆住這個漂亮的嬰兒」的 Trump,因為 Anthropic 4 月那個沒公開釋出的 Mythos Preview 而轉了 180 度。 再把鏡頭拉到電網。PJM 涵蓋整個美國中西部與大西洋中段,**2026-27 交付年**的容量市場結算價剛拍出 329 美元/MW——是 2024-25 交付年同口徑 28.92 美元的 **十倍以上**。北維吉尼亞已經停止接受新資料中心併網。Georgia Power 在最新整合資源規劃中,把未來六年的尖峰負載預估**上調超過 8,200 MW**,是先前的數倍——這個數字大到他們得重新討論天然氣與核能的組合。 三件事,三個城市,看似無關。但放在同一張桌上看,你會發現它們是同一個轉折點的三個側影:**Agent 元年從這個五月開始,真正起跑了**。 這篇文章帶你看完這個轉折的三個面向,以及對工程師、企業、決策者各自意味著什麼,文末附超過 20 條一級來源。 --- ## 第一幕:Agent 不再是 Demo,它在工廠裡上班 ![202605-ai-agent-factory](https://hackmd.io/_uploads/SyHQlHo0bx.jpg) 過去一年我們聽了太多 agent demo,看了太多酷炫的「自主完成任務」影片。直到這個五月,這個詞才終於有了真實的、可量化的、企業願意拿出 ROI 數字的證據。 ### Code w/ Claude 那天 Dario 沒講的事 在 SpaceX 合約之外,Anthropic 同步把 [Routines](https://simonwillison.net/2026/May/6/code-w-claude-2026/) 推上線。這個功能由 Claude Code 工程師 Boris 介紹時用了一句話 — 「Routines are higher-order prompts」。換句話說,你不再是叫 Claude 幫你寫一段 code,而是設定一個會「自己叫自己」的高階流程。 > 「The person who owns the PR is never going to see a red X — Claude is prompting Claude Code on its own.」 這件事的重點不是「自動修 bug」,而是:**asynchronous engineering 的閉環第一次在主流工具裡完成**。Claude 可以幫你規劃、開 PR、跑 test、看到失敗、自己改、自己再 commit。你早上一打開電腦,看到的就是已經綠燈的 PR 等著合併。 ### 800 個 agent 在 GE Appliances 跑起來 這不是新創的測試,這是 [Forbes 5 月 3 日報導的](https://www.forbes.com/sites/janakirammsv/2026/05/03/google-bets-agents-replace-apps-here-is-what-that-means-for-your-it-stack/) 真實部署:GE Appliances 在製造、物流、供應鏈跑了 **800 個以上的 AI agents**;Highmark Health 的 Sidekick AI 在 2025 年內部估值 **$27.9M**;Comcast 把 Xfinity Assistant 從腳本式自動化改寫成對話式排錯,首次解決率明顯上升。 這些數字的意義不在絕對大小,而在於它們開始是「ROI 報告」而不是「pilot 故事」。 ### 來自 F5 的硬數據 5 月 7 日 [F5 發表的《2026 State of Application Strategy》報告](https://www.rcrwireless.com/20260507/ai/f5-ai-inference-in-house) 給了三個讓人停下來想一下的數字: | 指標 | 數值 | 意義 | |------|------|------| | inference 已是 AI 主要活動的企業 | **77%** | training 不再是主場 | | 平均一家企業同時運行/評估的模型數 | **7 個** | 多模型路由是常態 | | 把 AI 跨 on-prem + 公雲 + colo 部署的企業 | **86%** | 企業刻意不押單一供應商 | 77% 這個數字尤其關鍵。它代表一件事:**前沿模型的軍備競賽已經不是企業 IT 的瓶頸**——企業大多數時間在跑、不在訓。瓶頸轉移到 inference 經濟、運算配額、模型路由、agent governance。 ### Anthropic 自己怎麼定義這個轉變 [Anthropic 4 月發布的《2026 Agentic Coding Trends Report》](https://resources.anthropic.com/hubfs/2026%20Agentic%20Coding%20Trends%20Report.pdf) 寫得很直接: > 「In 2026, the value of an engineer's contributions shifts to system architecture design, agent coordination, quality evaluation, and strategic problem decomposition. The primary human role in building software is orchestrating AI agents.」 從 implementer 變成 orchestrator。這句話可能會被引用一年。 ### 一個你可能沒留意的細節:Cursor Composer 2 用的是中國模型 [Moonshot Kimi K2.5](https://www.digitalapplied.com/blog/chinese-ai-models-q2-2026-market-share-report) 1 月底上線,最大的賣點是 Agent Swarm — 在單一 inference loop 裡可以同時調度 **100 個並行子代理**。Cursor Composer 2 啟動時就把 Kimi K2.5 放進預設模型棧,SWE-bench Multilingual 拿到 73.7%。這是中國模型第一次真正進入西方主力開發者工具的 production 環節。 [Forbes 引述 MIT Technology Review 的數據](https://www.forbes.com/sites/jonmarkman/2026/04/28/chinas-deepseek-v4-and-qwen-reshape-the-open-source-ai-race/) :中國開源模型 2025 年 8 月止的全球下載占比 **17.1%**,已經超越美國的 15.86%。a16z 也報告,美國程式設計師對中國開源模型的使用率,從 2024 年底的 1.2% 跳到 2025 年底的 **30%**。這不是抵制與否的問題,是已經發生的事。 --- ## 第二幕:白宮的 U 轉,從「漂亮的嬰兒」到 16 頁行政命令 ![202605-ai-whitehouse-uturn](https://hackmd.io/_uploads/r1MExBsC-e.jpg) 如果你只看一個月前的新聞,會以為 Trump 政府對 AI 的態度是「越鬆越好」。Trump 在 2025 年 1 月上任第一天就廢了 Biden 那道要求 AI 開發者送交安全測試結果的行政命令;他在 2025 年 7 月公開說過 — 「We're going to make this industry absolutely the top, because right now it's a beautiful baby that's born, we have to grow that baby and let that baby thrive.」 四月底,Anthropic 公開了一個叫 **Mythos Preview** 的模型。它沒有公開釋出,因為 Anthropic 自己評估後決定不放出去。為什麼?因為 [Anthropic 內部與政府機構初步評估](https://arstechnica.com/tech-policy/2026/05/everything-that-could-go-wrong-with-trumps-ai-safety-tests-according-to-experts/) 顯示,這個模型能以**多數人類駭客難以企及的方式**找出並利用軟體漏洞——這個能力等級已經足以重新定義網路安全攻防的不對稱。 Mythos 一公開,Trump 政府的態度一週內變了。 ### 5/4 - 5/5 一週之內的三件決定性事件 1. **5/4 紐約時報首發**:白宮在開會討論一份 16 頁的行政命令草案,要求新 AI 模型釋出前先送聯邦政府審查。 2. **5/5 商務部宣布**:CAISI(Center for AI Standards and Innovation)已經跟 Microsoft、xAI、Google DeepMind 簽下自願預先測試協議;Anthropic 早就在合作。CAISI 至今已完成約 40 次模型評估。 3. **5/5 [Politico](https://www.politico.com/news/2026/05/05/white-house-mulls-tight-new-controls-on-advanced-ai-00907468) / [Axios](https://www.axios.com/2026/05/05/trump-anthropic-ai-regulation-mythos-cyber) 確認**:白宮網路安全辦公室正在開發一套框架,要求 Pentagon 領導聯邦、州、地方政府部署 AI 前的安全測試。 White House 官員的官方說法很模糊 — 「Any policy announcement will come directly from the President. Discussion about potential executive orders is speculation.」但從業界與政策圈的消息源都指向同一個方向:**這個轉向是真的,只是不知道哪天簽**。 ### 同一時間窗口,歐洲的監管齒輪也在咬合 當白宮在權衡是否簽下行政命令時,Brussels 的官員已經等了三年。8 月 2 日就是他們的標記日。 - 2024 年 8 月 1 日,AI Act 正式生效。 - 2025 年 8 月 2 日,GPAI 的義務生效(提供者該遵守,但 Commission 不能罰)。 - **2026 年 8 月 2 日**,Commission 的 GPAI 強制力 + 高風險系統的全套義務 **同時啟動**。最高罰款是 **€35 million 或全球年營收的 7%**。 換句話說:**未來三個月內,全球前沿 AI 同時面對美國的預先審查與歐盟的全面強制力**。我這個職涯到目前為止,還沒看過監管的雙頭夾擊在同一個季度發生。 ### 國際警告也在升溫 [5 月 1 日 Bloomberg Law 報導](https://news.bloomberglaw.com/business-and-practice/cybersecurity-agencies-worldwide-warn-about-agentic-ai-risks),美國 CISA、英國、澳洲三國網安機構**聯合**發出對 agentic AI 的警告:「每一個 agentic AI 元件都擴大攻擊面」。 而 Anthropic 的共同創辦人 Jack Clark 在 5 月初一篇被 [36Kr Europe 整理的公開貼文](https://eu.36kr.com/en/p/3797756582747395) 裡,丟出一句更冷的話 — 「我估計 2028 年底前出現 recursive self-improvement 的機率是 60%。」 ### 但有件事不要誤會 聯邦審查不是「禁止釋出」。[Axios 的消息源](https://www.axios.com/2026/05/04/trump-white-house-ai-safety-tests-mythos) 寫得很清楚:「一些官員主張政府先取得新模型的存取權,但**不阻止其釋出**。」這比較像是「拿到鑰匙先看一眼」,不是 FDA 那種上市禁令。 但即使是這種輕度監管,對企業合規團隊都已經夠忙了。如果你的產品要服務歐洲使用者,8 月之前該做的事比想像的多。 --- ## 第三幕:電網的警鈴與資本支出黑洞 ![202605-ai-grid-datacenter](https://hackmd.io/_uploads/HkpEgriC-e.jpg) 當所有人還在比哪家模型 SWE-bench 高、誰的 1M context 比較準的時候,台積電可能已經算清楚了:**真正的瓶頸不是晶片,是電**。 ### 三個刺眼的數字 - Big 4 hyperscaler 的 2026 AI capex 估計約 **$710B(約新台幣 23 兆元)**,分布見下表([247WallSt 5/1](https://247wallst.com/investing/2026/05/01/the-big-4-hyperscalers-are-spending-710-billion-on-ai-heres-the-stock-that-profits-most/) )。Jefferies 5 月把 hyperscaler + Oracle 整體 capex 進一步上修到 [**$769B**](https://www.investors.com/news/technology/data-center-capex-hikes-positive-ai-stocks/) 。 | 廠商 | 2026 AI capex | |------|---------------| | Amazon | 約 $200B | | Microsoft | 約 $190B | | Alphabet | 約 $185B | | Meta | 約 $135B | - NVIDIA 最新年報的資料中心營收 **+75% YoY,達 1,937 億美元**。GTC 2026 上推出的 Vera Rubin 平台,主軸標語直接寫成「AI Factory + Inference Inflection」。 - PJM 容量市場 2026-27 結算價:**$329/MW**。兩年前是 $28.92。**整整 10 倍**。 第三個數字是台灣科技工作者最該停下來看一眼的。它意味著美國中西部、大西洋中段這片占了全美一半製造業的區域,**未來三年的工業電價會大幅吃掉企業營業利潤率**。AI 基礎設施的紅利,一部分是從製造業搬過去的。 ### 應對方式很激烈 | 動作 | 說明 | |------|------| | 重啟核能 | Microsoft 重啟 Three Mile Island;AWS 收購 Susquehanna 旁的資料中心吃 960 MW;Google + NextEra 重啟 Duane Arnold 615 MW | | SMR 押注 | Google + Kairos Power、Microsoft + Helion Energy(融合);技術公司計畫融資 20 GW SMR | | 地理重塑 | Microsoft UAE $15.2B;Meta Louisiana $10B;Northern Virginia 已凍結新接入 | | 天然氣回潮 | 2024 規劃容量 11.1% → 2026 18.1%;非再生能源規劃量年增 **71%**;再生能源僅 +2%([AAF 數據](https://www.americanactionforum.org/insight/ai-data-centers-us-electric-grid/) ) | | 太空資料中心 | NVIDIA Vera Rubin Space-1 模組(IGX Thor + Jetson Orin),SpaceX 1 月已申請發射 100 萬顆衛星 | 太空資料中心聽起來像 sci-fi,但 NVIDIA、Axiom Space、Starcloud、Planet Labs 都已經有合作項目。當地球上的電網接不下時,把 inference 推上軌道是真的有人在算的選項。 ### Wall Street 的 rotation 看得最清楚 代表企業軟體股的 **iShares Expanded Tech-Software Sector ETF(IGV)2026 YTD 已跌 20% 以上**,這個跌幅在 CNBC 報導 Anthropic Sonnet 4.6 上線那天被當成 SaaS「Wall Street freak-out」的指標([CNBC 2/17](https://www.cnbc.com/2026/02/17/anthropic-ai-claude-sonnet-4-6-default-free-pro.html) )。原因不是 AI 沒用,而是市場已經把 SaaS 應用層的價值轉移到 inference 基礎設施 + 能源 + agent 平台。SAP 因為股價下跌,5 月直接掏 **€1B 收購一家 18 個月大的德國 AI lab Prior Labs**,押注表格資料的 foundation model([TechCrunch 5/5](https://techcrunch.com/2026/05/05/sap-bets-1-16b-on-18-month-old-german-ai-lab-and-says-yes-to-nemoclaw/) )。 如果你還在用「AI 是不是泡沫」這個框架理解,可能會錯過真正的轉移:**錢從應用層流到基礎設施層,這個搬遷正在發生**。 --- ## 給工程師、企業、決策者的三段式建議 ![202605-ai-engineer-orchestrator](https://hackmd.io/_uploads/ryQrxBoR-l.jpg) 我自己是硬韌體背景出身的工程師,這幾個月被 AI 推著重新學軟體開發的心智模型。整理三段最務實的建議。 ### 給工程師:把自己升級成 orchestrator 我自己這兩個月最大的改變是:開新 feature 不再從打字開始。我打開 Claude Code,先把 spec 寫成一段話、設好 budget、讓它自己規劃,再跑 Routines 在背景慢慢做。同時開的可能有兩、三個 task。我做的事更像審稿與校準。 具體幾個方向值得放進你的 2026 工程師升級包: - **學 model routing 而不是「找最強模型」**。社群共識已經很清楚——Claude 解硬問題、GPT-5.x 跑速度、Gemini 吃 long context、Haiku/Flash 跑量。把這個矩陣記在腦子裡,每個任務挑對工具。 - **管 token 像管雲端帳單**。[Forbes 5/5 Krouse / Lau / Elliott 對談](https://www.forbes.com/sites/johnwerner/2026/05/05/companies-can-win-with-ai/) 直接點名:「實習生一鍵燒掉 $7,000/月」已經發生過。每個 repo、每個 feature 都要設 budget,Anthropic Console 跟 OpenAI dashboard 都看得到分人帳單。 - **MCP 是新的 USB-C,A2A 是新的網路協定**。不熟 MCP,在 2026 寫 agent 會很辛苦。從 Claude Desktop、Cursor、Claude Code 都已經原生支援的 MCP server 開始,把你常用的 SaaS 都掛上去。 - **Agent 思維 = plan / execute / verify 三段**。不要再用 chat-style 一句一句餵 prompt 了,先寫 spec、讓 agent 規劃、執行、自我評估。你做的事是「審稿」,不是「聽寫」。 ### 給企業導入者:先建治理層再加產能 如果你最近在內部開「我們要不要全力押 OpenAI」這種會議,先停一下。F5 那份報告告訴我們的真實常態是:86% 的企業已經把 AI 跨三種環境部署(on-prem、公雲、colo),刻意不押單一供應商。這已經是新的預設,而不是激進派的偏好。 幾個建議方向: - **多模型路由不只是 cost 優化,是合規優化**。某些案子不能離開歐盟主權雲,某些可以走 US 公雲。模型路由要把這層內建進來。 - **「Agent Supervisor」這個角色會冒出來**。DataRobot 已經把這個職稱放上 LinkedIn 招聘廣告——傳統 IT operations 加上 prompt engineering 加上 risk review 的混合體。 - **inference 在地化是新平衡點**。F5 的 77% 數字代表大多數企業選擇自己跑而不是外包 hyperscaler。台灣對於 sovereign / 在地推論這條路,有現成的硬體韌體鏈優勢,這個窗口不要錯過。 - **8/2 EU AI Act 不是放羊**。最高罰 €35M 或 7% 全球營收。如果你的產品要在歐洲銷售,技術文件、訓練資料摘要、版權合規政策三樣現在就該動工。 ### 給決策者:這次監管視窗真的會關上 我寫這段時,腦子裡是台灣產業政策圈這幾年「跟著矽谷風向決策」的習慣。這次得學新的反射動作。 - **美國行政命令隨時會簽**。寫稿這當下,白宮草案已經 16 頁,外傳「數週之內」會敲定。台灣 AI 廠商的合規團隊不要等到簽了才動。 - **電網是新的瓶頸,不是 GPU**。台灣半導體政策若還只看 GPU 出口管制,會錯過電力與冷卻的整套產業機會。SMR、液冷、HVDC 都是未來三年的紅利。 - **Anthropic Mythos 提醒我們:AI 安全已經不是哲學辯論**。它可以讓政府一週內轉向。CISA、英國、澳洲三國對 agentic AI 的聯合警告,會是接下來主管機關的範本——預先想清楚台灣的對應姿態。 --- ## 結語:兩個矛盾的事實放在 Agent 元年的入口 我寫這篇的時候,腦子裡一直有兩件事在打架。 一件是 Jack Clark 5 月 4 日那個 60% 的數字。Anthropic 的共同創辦人估計,2028 年底前,AI 自己訓練自己的「recursive self-improvement」有六成機率出現。在我這輩工程師受過的訓練裡,這種數字以前只有在科幻電影裡會聽到。 另一件是 [Forbes 5 月 1 日報導的](https://www.forbes.com/sites/juliettehan/2026/05/01/5-ways-to-prevent-ai-use-from-quietly-eroding-your-judgment/) 那份「cognitive surrender」研究。1,372 名受試者中,**79.8% 即使 AI 給了錯誤答案,仍然照辦**。研究員稱這個現象為「認知投降」。 這兩件事擺一起就是 2026 五月的真正主題:**AI 能力暴漲、人類判斷下沉**——同時發生。Agent 進工廠、Code w/ Claude 的 PR 自動合併、白宮的審查命令、PJM 的電價、CISA 的警告、SAP 的 €1B 賭注,都是同一個轉折點的不同切面。 對台灣的我們,最值得提醒自己的或許是這句:**Agent 元年的真正起跑線,不是看誰跑得最快,而是看誰知道自己在跑哪條跑道**。 如果你讀到這裡,下一步請務必做一件事 — 打開你常用的那個 AI 工具,把今天打算交給它的任務寫成一個 spec、設好 budget、要求它自我評估。不是因為它會變強,而是因為**你需要保留自己「不投降」的反射神經**。 下個月見。 --- ## 延伸閱讀 / 參考資料 ### 模型動態 - [Introducing Claude Opus 4.6 — Anthropic](https://www.anthropic.com/news/claude-opus-4-6) - [Introducing Claude Sonnet 4.6 — Anthropic](https://www.anthropic.com/news/claude-sonnet-4-6) - [Anthropic releases Opus 4.6 with new 'agent teams' — TechCrunch](https://techcrunch.com/2026/02/05/anthropic-releases-opus-4-6-with-new-agent-teams/) - [OpenAI releases GPT-5.5, bringing company one step closer — TechCrunch](https://techcrunch.com/2026/04/23/openai-chatgpt-gpt-5-5-ai-model-superapp/) - [Gemini 3.1 Pro Released: Google's Powerful Answer to GPT-5 — Tech Society](https://digitalstrategy-ai.com/2026/02/23/gemini-3-1-pro-released-googles-answer-gpt-5/) ### Code w/ Claude 2026 與 Agent - [Anthropic raises Claude Code usage limits, credits new deal with SpaceX — Ars Technica](https://arstechnica.com/ai/2026/05/anthropic-raises-claude-code-usage-limits-credits-new-deal-with-spacex/) - [Live blog: Code w/ Claude 2026 — Simon Willison](https://simonwillison.net/2026/May/6/code-w-claude-2026/) - [Google Bets Agents Replace Apps — Forbes](https://www.forbes.com/sites/janakirammsv/2026/05/03/google-bets-agents-replace-apps-here-is-what-that-means-for-your-it-stack/) - [2026 Agentic Coding Trends Report — Anthropic PDF](https://resources.anthropic.com/hubfs/2026%20Agentic%20Coding%20Trends%20Report.pdf) - [F5 report shows enterprises bringing AI inference in-house — RCR Wireless](https://www.rcrwireless.com/20260507/ai/f5-ai-inference-in-house) ### 政策與安全 - [White House mulls tighter controls on advanced AI — Politico](https://www.politico.com/news/2026/05/05/white-house-mulls-tight-new-controls-on-advanced-ai-00907468) - [Trump admin considering safety review for new AI models after Mythos — Axios](https://www.axios.com/2026/05/04/trump-white-house-ai-safety-tests-mythos) - [New frontier of AI forces Trump's heavy hand — Axios](https://www.axios.com/2026/05/05/trump-anthropic-ai-regulation-mythos-cyber) - [Spooked by Mythos, Trump suddenly realized AI safety testing might be good — Ars Technica](https://arstechnica.com/tech-policy/2026/05/everything-that-could-go-wrong-with-trumps-ai-safety-tests-according-to-experts/) - [Cybersecurity Agencies Worldwide Warn About Agentic AI Risks — Bloomberg Law](https://news.bloomberglaw.com/business-and-practice/cybersecurity-agencies-worldwide-warn-about-agentic-ai-risks) - [Enforcement of Chapter V under the EU AI Act](https://artificialintelligenceact.eu/enforcement-of-chapter-v-under-the-eu-ai-act/) ### 資本支出與電網 - [The Big 4 Hyperscalers Are Spending $710 Billion on AI — 247WallSt](https://247wallst.com/investing/2026/05/01/the-big-4-hyperscalers-are-spending-710-billion-on-ai-heres-the-stock-that-profits-most/) - [Data-Center Capex Hikes Positive For These AI Stocks — Investors.com](https://www.investors.com/news/technology/data-center-capex-hikes-positive-ai-stocks/) - [NVIDIA Vera Rubin Opens Agentic AI Frontier — NVIDIA IR](https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Vera-Rubin-Opens-Agentic-AI-Frontier/default.aspx) - [AI Data Center Power Surge: Shifting Trends Toward Natural Gas — AAF](https://www.americanactionforum.org/insight/ai-data-centers-us-electric-grid/) - [Energy supply for AI — IEA](https://www.iea.org/reports/energy-and-ai/energy-supply-for-ai) - [AI, Data Centers, and the U.S. Electric Grid — Belfer Center / Harvard](https://www.belfercenter.org/research-analysis/ai-data-centers-us-electric-grid) ### 中國模型 - [China's DeepSeek V4 And Qwen Reshape The Open-Source AI Race — Forbes](https://www.forbes.com/sites/jonmarkman/2026/04/28/chinas-deepseek-v4-and-qwen-reshape-the-open-source-ai-race/) - [DeepSeek could hit $45B valuation from its first investment round — TechCrunch](https://techcrunch.com/2026/05/06/deepseek-could-hit-45b-valuation-from-its-first-investment-round/) - [Chinese AI Models Q2 2026: 10-Provider Landscape Report — Digital Applied](https://www.digitalapplied.com/blog/chinese-ai-models-q2-2026-market-share-report) ### 觀點與趨勢 - [5 Ways To Prevent AI Use From Quietly Eroding Your Judgment — Forbes](https://www.forbes.com/sites/juliettehan/2026/05/01/5-ways-to-prevent-ai-use-from-quietly-eroding-your-judgment/) - [Companies Can Win With AI(含 token budget 對談) — Forbes](https://www.forbes.com/sites/johnwerner/2026/05/05/companies-can-win-with-ai/) - [SAP bets $1.16B on 18-month-old German AI lab — TechCrunch](https://techcrunch.com/2026/05/05/sap-bets-1-16b-on-18-month-old-german-ai-lab-and-says-yes-to-nemoclaw/) - [AI Self - Creates with 60% Probability — 36Kr Europe(Jack Clark RSI 60% 推文)](https://eu.36kr.com/en/p/3797756582747395)

    Import from clipboard

    Paste your markdown or webpage here...

    Advanced permission required

    Your current role can only read. Ask the system administrator to acquire write and comment permission.

    This team is disabled

    Sorry, this team is disabled. You can't edit this note.

    This note is locked

    Sorry, only owner can edit this note.

    Reach the limit

    Sorry, you've reached the max length this note can be.
    Please reduce the content or divide it to more notes, thank you!

    Import from Gist

    Import from Snippet

    or

    Export to Snippet

    Are you sure?

    Do you really want to delete this note?
    All users will lose their connection.

    Create a note from template

    Create a note from template

    Oops...
    This template has been removed or transferred.
    Upgrade
    All
    • All
    • Team
    No template.

    Create a template

    Upgrade

    Delete template

    Do you really want to delete this template?
    Turn this template into a regular note and keep its content, versions, and comments.

    This page need refresh

    You have an incompatible client version.
    Refresh to update.
    New version available!
    See releases notes here
    Refresh to enjoy new features.
    Your user state has changed.
    Refresh to load new user state.

    Sign in

    Forgot password
    or
    Sign in via Google Sign in via Facebook Sign in via X(Twitter) Sign in via GitHub Sign in via Dropbox Sign in with Wallet
    Wallet ( )
    Connect another wallet

    New to HackMD? Sign up

    By signing in, you agree to our terms of service.

    Help

    • English
    • 中文
    • Français
    • Deutsch
    • 日本語
    • Español
    • Català
    • Ελληνικά
    • Português
    • italiano
    • Türkçe
    • Русский
    • Nederlands
    • hrvatski jezik
    • język polski
    • Українська
    • हिन्दी
    • svenska
    • Esperanto
    • dansk

    Documents

    Help & Tutorial

    How to use Book mode

    Slide Example

    API Docs

    Edit in VSCode

    Install browser extension

    Contacts

    Feedback

    Discord

    Send us email

    Resources

    Releases

    Pricing

    Blog

    Policy

    Terms

    Privacy

    Cheatsheet

    Syntax Example Reference
    # Header Header 基本排版
    - Unordered List
    • Unordered List
    1. Ordered List
    1. Ordered List
    - [ ] Todo List
    • Todo List
    > Blockquote
    Blockquote
    **Bold font** Bold font
    *Italics font* Italics font
    ~~Strikethrough~~ Strikethrough
    19^th^ 19th
    H~2~O H2O
    ++Inserted text++ Inserted text
    ==Marked text== Marked text
    [link text](https:// "title") Link
    ![image alt](https:// "title") Image
    `Code` Code 在筆記中貼入程式碼
    ```javascript
    var i = 0;
    ```
    var i = 0;
    :smile: :smile: Emoji list
    {%youtube youtube_id %} Externals
    $L^aT_eX$ LaTeX
    :::info
    This is a alert area.
    :::

    This is a alert area.

    Versions and GitHub Sync
    Get Full History Access

    • Edit version name
    • Delete

    revision author avatar     named on  

    More Less

    Note content is identical to the latest version.
    Compare
      Choose a version
      No search result
      Version not found
    Sign in to link this note to GitHub
    Learn more
    This note is not linked with GitHub
     

    Feedback

    Submission failed, please try again

    Thanks for your support.

    On a scale of 0-10, how likely is it that you would recommend HackMD to your friends, family or business associates?

    Please give us some advice and help us improve HackMD.

     

    Thanks for your feedback

    Remove version name

    Do you want to remove this version name and description?

    Transfer ownership

    Transfer to
      Warning: is a public team. If you transfer note to this team, everyone on the web can find and read this note.

        Link with GitHub

        Please authorize HackMD on GitHub
        • Please sign in to GitHub and install the HackMD app on your GitHub repo.
        • HackMD links with GitHub through a GitHub App. You can choose which repo to install our App.
        Learn more  Sign in to GitHub

        Push the note to GitHub Push to GitHub Pull a file from GitHub

          Authorize again
         

        Choose which file to push to

        Select repo
        Refresh Authorize more repos
        Select branch
        Select file
        Select branch
        Choose version(s) to push
        • Save a new version and push
        • Choose from existing versions
        Include title and tags
        Available push count

        Pull from GitHub

         
        File from GitHub
        File from HackMD

        GitHub Link Settings

        File linked

        Linked by
        File path
        Last synced branch
        Available push count

        Danger Zone

        Unlink
        You will no longer receive notification when GitHub file changes after unlink.

        Syncing

        Push failed

        Push successfully