1. 用於評估人工智能在平面設計任務中表現的綜合基準測試 ↑ 3 HN Points 11:20 Comprehensive Benchmark for Evaluating AI on Graphic Design Tasks (arxiv.org)
2. 美國的增材製造：3D打印的監管 [視頻] ↑ 2 HN Points 11:15 Additive in America: Regulating 3D Printing [video] (www.youtube.com)
4. AI 代理與自主性之間的所有障礙——一份實用指南 ↑ 2 HN Points 11:03 Every Barrier Between AI Agents and Autonomy – A Practical Map (vibeagentmaking.com)
5. LLM 爬蟲機器人正在導致 acme.com 的 HTTPS 服務器過載 ↑ 2 HN Points 11:02 LLM scraper bots are overloading acme.com''s HTTPS server (acme.com)
6. 肯尼迪設定9月為最後期限，以查明美國自閉症發病率上升的原因（2025年） ↑ 2 HN Points 11:01 Kennedy sets September deadline to identify cause of rising US autism rates (2025) (www.reuters.com)
7. 文件顯示，匈牙利與俄羅斯達成了一項旨在加強雙邊關係的12點計劃 ↑ 2 HN Points 10:44 Hungary and Russia struck 12-point plan for closer ties, documents show (www.politico.eu)
9. 致幻劑引發的視覺幻覺與精神分裂症 ↑ 2 HN Points 10:19 Visual Hallucinations in Psychedelics vs. Schizophrenia (psychedelirium.substack.com)
10. RNA條形碼可實現大腦連接的高速繪製 ↑ 2 HN Points 09:53 RNA barcodes enable high-speed mapping of connections in the brain (news.illinois.edu)
13. 針對Booking.com酒店及客戶的“我付了兩筆錢”網絡釣魚活動 ↑ 2 HN Points 09:01 Phishing Campaigns "I Paid Twice" Targeting Booking.com Hotels and Customers (blog.sekoia.io)
17. OpenAI表示其新模型GPT-2過於危險，不宜發佈（2019年） ↑ 10 HN Points 10:41 OpenAI says its new model GPT-2 is too dangerous to release (2019) (slate.com)
18. 他設計 C 就是為了解決你的代碼問題 ↑ 2 HN Points 09:49 He designed C to solve your code problems (stackoverflow.blog)
19. Drive9，一款內置語義搜索功能的網絡驅動器 ↑ 2 HN Points 09:31 Drive9, a network drive with built-in semantic search (github.com)
20. 緬因州即將成為首個禁止新建數據中心的州 ↑ 2 HN Points 10:37 Maine Is About to Become the First State to Ban New Data Centers (www.wsj.com)
22. “人類社會本不該如此運作，”美國參議員表示 ↑ 6 HN Points 10:17 ''This Is Just Not How the Human Race Should Operate,'' Says US Senator (www.commondreams.org)
23. 規模更大、可訓練性更強的語言模型可靠性反而更低 ↑ 4 HN Points 10:12 Larger and more instructable language models become less reliable (pmc.ncbi.nlm.nih.gov)
25. 美國：如果你具備相應條件，這絕對是一場值得一試的賭注 ↑ 2 HN Points 10:04 America: A Bet Worth Taking If You''re Built for It (code2net.com)
26. 品牌紛紛採用“無AI”聲明，以在良莠不齊的市場中脫穎而出 ↑ 2 HN Points 10:02 Brands Adopt ''No AI'' Disclaimers to Stand Out Amid the Slop (www.wsj.com)
27. 保險公司持有的1萬億美元私人信貸規模讓監管機構望塵莫及 ↑ 2 HN Points 09:56 Insurers'' 1T Buildup in Private Credit Is Leaving Regulators in Dust (www.wsj.com)
28. “阿耳忒彌斯”號傳回了一些非凡的高質量月球照片 ↑ 2 HN Points 09:55 Artemis delivers some exceptional, high-quality photos of the Moon (arstechnica.com)
29. 通過強化學習訓練最先進（SOTA）的漏洞發現代理 ↑ 2 HN Points 09:44 Training SOTA vulnerability discovery agents through RL (depthfirst.com)
30. 非技術團隊全面採用 AI 編碼助手的影響 ↑ 2 HN Points 09:29 Impact of 100% Adoption of AI Coding Agents by Non-Technical Team (www.kapwing.com)
32. 使用 LM2596T 穩壓器為樹莓派 4B 供電時電壓會下降 ↑ 2 HN Points 09:06 Voltage drops when powering a Raspberry Pi4B from a LM2596T regulator (electronics.stackexchange.com)
33. 俄羅斯軍事黑客將英國網民的網絡流量進行了重定向 ↑ 2 HN Points 07:37 Russian military hackers reroute British internet users'' traffic (www.ft.com)
34. “白噪音”（基於Nostr的加密群聊）已完成安全審計 ↑ 2 HN Points 09:02 White Noise (encrypted group chat over Nostr) completes security audit (leastauthority.com)
35. 全球民主的狀況比你想象的要好 ↑ 2 HN Points 08:21 Global democracy is in better shape than you think (economist.com)
36. 馬薩諸塞州議員正尋求推行全美“最嚴格”的社交媒體禁令 ↑ 2 HN Points 08:12 Mass. lawmakers looking to usher in ''most restrictive'' social media ban in USA (www.independent.co.uk)
37. AI編程助手技能（Claude Code、Codex、OpenCode、OpenClaw、Factory Droid） ↑ 2 HN Points 07:45 AI coding assistant skill (Claude Code, Codex, OpenCode, OpenClaw, Factory Droid (github.com)
39. 俄羅斯和中國否決了旨在重新開放霍爾木茲海峽的聯合國決議 ↑ 2 HN Points 09:29 Russia and China veto UN resolution aimed at reopening the Strait of Hormuz (apnews.com)
40. 美國移民及海關執法局承認其正在使用強大的間諜軟件 ↑ 2 HN Points 09:26 ICE acknowledges it is using powerful spyware (text.npr.org)
41. GitHub Copilot CLI 現已支持 BYOK 和本地模型 ↑ 3 HN Points 09:05 GitHub Copilot CLI now supports BYOK and local models (github.blog)
42. 零散股：如何判斷一篇文章是否由人工智能撰寫 ↑ 2 HN Points 09:04 Odd Lots: This Is How to Tell If Writing Was Made by AI (www.bloomberg.com)
43. 為什麼Anthropic的新模型讓網絡安全專家感到不安 ↑ 2 HN Points 08:19 Why Anthropic''s new model has cybersecurity experts rattled (www.platformer.news)
44. Claudecode Vibes 的 Downdetector ↑ 2 HN Points 08:48 Downdetector for Claudecode Vibes (claudedumb.com)
45. AI輔助會降低堅持度，並損害獨立表現 ↑ 2 HN Points 08:00 AI Assistance Reduces Persistence and Hurts Independent Performance (arxiv.org)
47. 陽光中波長較長的部分能穿透人體，並對全身產生影響 ↑ 2 HN Points 08:37 Longer wavelengths in sunlight pass through the human body and have a systemic (www.nature.com)
48. 曾推動私募信貸發展的富裕投資者正爭相撤資 ↑ 2 HN Points 08:24 The Wealthy Investors That Powered Private Credit Are Rushing for the Exits (www.wsj.com)
49. 推出我的產品。第一部分：尋找一個行之有效（且能讓你堅持下去）的點子 ↑ 2 HN Points 08:09 Launching My Product. Part 1: Finding an Idea That Works (and Keeps You Going) (www.indiehackers.com)
50. 預測並獲利：用於天氣及CPI/PCE的Kalshi自動化機器人 ↑ 2 HN Points 08:08 Predict and Profit, automated Kalshi bots for weather and CPI/PCE (predictandprofit.io)
51. 克勞德·科德是特工還是馴獸師？ ↑ 2 HN Points 08:01 Is Claude Code an agent or a harnness? (agenticloopsai.substack.com)
52. 正念、編程以及我犯下的錯誤 ↑ 2 HN Points 07:53 Mindfulness, Programming, and the Mistake I Made (yonatanmedan.github.io)
53. 《Signal Ledger》：將技術性閱讀轉化為複利式研究 ↑ 2 HN Points 08:28 The Signal Ledger: Turning technical reading into compounding research (blog.pjhoberman.com)
55. 《Nutshell》通訊——每日可自定義的 RSS 摘要工具 ↑ 2 HN Points 08:27 Nutshell Newsletter – Daily Customizable RSS Summarizer (www.nutshellnewsletter.com)
56. 我們開發了一款工具，用於自定義 Unity Cloud 構建的超時設置 ↑ 2 HN Points 08:25 We built a tool for custom timeouts on your Unity Cloud builds (buildstash.com)
57. 美國和伊朗表示已達成兩週停火協議 ↑ 4 HN Points 08:16 US, Iran say they have agreed to a two-week ceasefire (apnews.com)
58. 在生產環境中運行 OpenClaw：與您的移動和 Web 應用集成 ↑ 2 HN Points 08:09 Running OpenClaw in Production: Integrating with Your Mobile & Web Apps (docs.mesibo.com)
60. 我們需要重新認識2026年的AI智能體開發工具 ↑ 2 HN Points 07:56 We need re-learn what AI agent development tools are in 2026 (blog.n8n.io)
61. Nix：在 FOD 輸出註冊過程中，通過追蹤符號鏈接實現權限提升 ↑ 2 HN Points 07:04 Nix: Privilege escalation via symlink following during FOD output registration (discourse.nixos.org)
62. 當編譯器撒謊時：破壞安全的Go語言中的內存安全性 ↑ 2 HN Points 07:02 When the compiler lies: breaking memory safety in safe Go (ciolek.dev)
63. 基於 WASM 的瀏覽器端 SQLite 及實時 ER 圖 ↑ 2 HN Points 06:56 SQLite in the browser with WASM and real-time ER diagram (fasttools.dev)
64. GEON：面向語言模型的結構優先解碼 ↑ 2 HN Points 06:51 GEON: Structure-first decoding for language models (github.com)
65. 特朗普表示，如果伊朗開放霍爾木茲海峽，他將同意為期兩週的停火 ↑ 2 HN Points 08:03 Trump agrees to 2-week ceasefire if Iran opens Strait of Hormuz (www.foxnews.com)
66. 霍爾木茲海峽的枷鎖——已破解且已解決 ↑ 2 HN Points 07:58 The Strait Jacket of Hormuz – Solved and Resolved (jeffreylminch.substack.com)
67. Sardine：一個由人工智能代理實時交易的股票市場 ↑ 2 HN Points 07:21 Sardine: A stock market where AI agents trade in real time (github.com)
68. 特朗普暫停對伊朗的襲擊兩週，但前提是霍爾木茲海峽必須開放 ↑ 2 HN Points 07:45 Trump suspends Iran attack for two weeks, subject to Hormuz Strait opening (www.cnbc.com)
69. Crag——一個治理文件，兼容所有AI編碼工具（50個倉庫，準確率96.4%） ↑ 2 HN Points 07:43 Crag – One governance file, every AI coding tool (50 repos, 96.4% accuracy) (github.com)
71. Xilem – 一個實驗性的 Rust 原生 UI 框架 ↑ 2 HN Points 07:36 Xilem – An experimental Rust native UI framework (github.com)
72. “垃圾化”的巔峰，亦即大型語言模型 ↑ 3 HN Points 07:20 The pinnacle of enshittification, or Large Language Models (blogs.gentoo.org)
73. WSLg：Windows 子系統 for Linux 圖形用戶界面 ↑ 2 HN Points 06:56 WSLg: Windows Subsystem for Linux GUI (github.com)
74. 美國與伊朗將暫停交戰，同意實施為期兩週的停火 ↑ 6 HN Points 07:16 US, Iran to pause war, agree to 2-week ceasefire (www.axios.com)
75. 分析顯示，蘋果和聯想的筆記本電腦可維修性最低 ↑ 2 HN Points 07:13 Apple and Lenovo have the least repairable laptops, analysis finds (arstechnica.com)
77. USCIS.gov 會將用戶/訪客數據發送給 Meta 和 Google ↑ 2 HN Points 07:10 USCIS.gov sends user/visitor data to Meta and Google (varlog.in)
78. Anthropic的最新AI模型過於強大，無法公開發布，且已突破安全防護 ↑ 2 HN Points 06:54 Anthropic latest AI model too powerful for public release and broke containment (www.businessinsider.com)
79. Appwrite 1.9.0：使用 MongoDB 作為底層數據庫進行自主託管 ↑ 2 HN Points 05:37 Appwrite 1.9.0: Self-hosting with MongoDB as the underlying database (appwrite.io)
80. Anthropic因黑客攻擊風險暫停 Mythos 模型 ↑ 2 HN Points 07:02 Anthropic holds Mythos model due to hacking risks (www.axios.com)
81. 《第五元素》女主角米拉·喬沃維奇揭秘AI記憶工具MemPalace ↑ 2 HN Points 07:01 Fifth Element Star Milla Jovovich Reveals AI Memory Tool MemPalace (decrypt.co)
82. 比亞迪這款支持5分鐘快充、續航里程達500英里的豪華電動車即將進軍海外市場 ↑ 3 HN Points 06:58 BYD''s luxury EV with 5-min fast charging and 500 miles range is headed overseas (electrek.co)
83. GLM-5.1 在代理性能方面與Opus 4.6相當，實際成本僅為其約三分之一 ↑ 2 HN Points 06:53 GLM-5.1 matches Opus 4.6 in agentic performance, at 1/3 actual cost (app.uniclaw.ai)
84. 本書作者前言：《PGP源代碼與內部機制》（1995年） ↑ 2 HN Points 06:48 Author''s preface to the book: "PGP Source Code and Internals" (1995) (philzimmermann.com)
85. 中年時期睡眠不規律與重大心臟事件風險升高有關 ↑ 2 HN Points 06:45 Midlife Sleep Irregularity Linked to Higher Risk of Major Cardiac Events (doi.org)
86. “絕對是騙局”：隨著關稅上漲，欺詐行為日益猖獗 ↑ 3 HN Points 06:39 ''Definitely a Sham'': As Tariffs Climb, Fraud Proliferates (www.nytimes.com)
87. 突破性阿爾茨海默病藥物重塑大腦，而非僅僅清除斑塊 ↑ 2 HN Points 06:36 Breakthrough Alzheimer''s Drug Rewires the Brain Instead of Just Clearing Plaques (scitechdaily.com)
88. 《紐約時報》被遠程醫療騙局愚弄，卻將其稱為人工智能的未來 ↑ 3 HN Points 06:18 New York Times Got Played by a Telehealth Scam and Called It the Future of AI (www.techdirt.com)
89. Tabsdata：宛如數據流的數據庫 ↑ 2 HN Points 05:32 Tabsdata: Like a Database for Dataflows (docs.tabsdata.com)
90. 關於軟件工程、人工智能和DevOps職位空缺的洞察——2026年4月 ↑ 2 HN Points 05:11 Insights on Software Engineering, AI and DevOps Job Openings – April 2026 (corvi.careers)
91. 特朗普同意暫停對伊朗的“轟炸和襲擊”兩週 ↑ 5 HN Points 06:41 Trump agrees to suspend ''bombing and attack of Iran'' for 2 weeks (www.reuters.com)
92. 迪爾公司同意支付9900萬美元以和解“維修權”訴訟 ↑ 4 HN Points 06:38 Deere and Co agrees to pay 99M to settle ''right to repair'' lawsuit (apnews.com)
93. MCP 擁有 9700 萬次安裝。它也是通往每個開發環境的便捷通道。 ↑ 2 HN Points 06:19 MCP has 97M installs. It''s also an open door into every dev environment (rawtext.io)
94. 前Meta員工因下載3萬張Facebook私密照片而接受調查 ↑ 2 HN Points 06:20 Ex-Meta worker investigated for downloading 30k private Facebook photos (www.bbc.com)
95. DeepMind關於攻破Claws的論文及我們的收穫 ↑ 3 HN Points 06:07 DeepMind''s paper on p0wning Claws and what we learned (suthakamal.substack.com)
99. “強勁”的2025年再現順風：美元走弱 ↑ 2 HN Points 06:16 The Tailwind in "Strong" 2025 Returns: A Weaker Dollar (nicolaswalker.com)
100. 區塊鏈熱潮（終於）安息吧（2025） ↑ 2 HN Points 06:15 RIP (finally) to the blockchain hype (2025) (www.cio.com)
101. 針對伊朗“種族滅絕”威脅，要求罷免特朗普的兩黨呼聲日益高漲 ↑ 9 HN Points 06:05 Bipartisan calls to remove Trump from office grow over Iranian ''genocide'' threat (www.yahoo.com)
103. 一個用於識別失蹤移民的DNA數據庫已不復存在 ↑ 2 HN Points 06:04 A DNA Archive Built to Identify Missing Migrants Has Vanished (undark.org)
105. 史上最嚴重的石油供應中斷正加速能源轉型 ↑ 2 HN Points 05:54 The Biggest Oil Disruption in History Is Accelerating the Energy Transition (oilprice.com)
107. RSoC 2026：Redox OS 的新 CPU 調度器 ↑ 4 HN Points 05:41 RSoC 2026: A new CPU scheduler for Redox OS (www.redox-os.org)
109. Harrier——微軟開源了業界領先的嵌入式模型 ↑ 2 HN Points 04:40 Harrier – Microsoft Open-Sources Industry-Leading Embedding Model (blogs.bing.com)
110. 福克斯將在福克斯新聞媒體和福克斯一號平臺上整合Kalshi的天氣預報 ↑ 2 HN Points 05:45 Fox to Integrate Kalshi Forecasts Across Fox News Media and Fox One Platforms (news.kalshi.com)
111. 邁向 TLS 中的後量子密碼學 (2019) ↑ 2 HN Points 05:36 Towards Post-Quantum Cryptography in TLS (2019) (blog.cloudflare.com)
112. 現代戰場上最有效的武器是混凝土（2016） ↑ 2 HN Points 05:34 The Most Effective Weapon on the Modern Battlefield Is Concrete (2016) (mwi.westpoint.edu)
113. 瑞士建造了全球最強大的氧化還原液流電池 ↑ 4 HN Points 05:33 Switzerland builds most powerful redox-flow battery (www.swissinfo.ch)
114. 《地球陷落》。這款Amiga平臺的地下城探索遊戲重磅迴歸 ↑ 3 HN Points 05:32 Earth Has Fallen. The Amiga Dungeon Crawler Is Back (old.reddit.com)
115. Mythos 逃離了安全的沙箱環境 [測試中] ↑ 2 HN Points 05:31 Mythos escapes a secure sandbox [in test] (twitter.com)
116. 2000年以來美元購買力的實時走勢 ↑ 2 HN Points 05:27 USD Purchasing Power in Real Time Since 2000 (onedollar.today)
117. 我們的AI統治者到底怎麼了？ ↑ 2 HN Points 05:23 What the heck is wrong with our AI overlords? (arstechnica.com)
118. 一個免費模型的表現優於Opus，但該方法被鎖定在13項心理學實驗中 ↑ 2 HN Points 05:11 A free model beats Opus but the method is locked behind 13 psychological trials (github.com)
119. 來自伊朗——邀請物理學家們閱讀我關於Alpha的論文 ↑ 3 HN Points 05:23 From Iran – an invitation to physicists to look at my paper about Alpha (www.researchgate.net)
120. 科學家編造了一種虛構的疾病。人工智能卻告訴人們這是真的 ↑ 2 HN Points 05:23 Scientists invented a fake disease. AI told people it was real (www.nature.com)
121. 關於Claude Mythos預覽版的可解釋性研究結果 ↑ 2 HN Points 05:15 Interpretability Findings on Claude Mythos Preview (twitter.com)
122. 預測市場上的體育博彩被裁定為“掉期交易”，不受州法律約束 ↑ 3 HN Points 05:14 Sports bets on prediction markets ruled to be "swaps," exempt from state laws (arstechnica.com)
123. 算一算：特朗普對伊朗的戰爭會讓你在加油站多花多少錢 ↑ 2 HN Points 05:12 Do the math: How much Trump''s Iran war will cost you at the pump (responsiblestatecraft.org)
124. Nedster——一款開源、本地優先的編程代理，能夠驗證自身的工作 ↑ 2 HN Points 05:05 Nedster – An open-source, local-first coding agent that verifies its own work (github.com)
128. 在我看來，Anthropic的“Project Glasswing”項目似乎很有必要 ↑ 2 HN Points 04:53 Anthropic''s Project Glasswing sounds necessary to me (simonwillison.net)
132. 伊朗正試圖對美國關鍵基礎設施發動網絡攻擊 ↑ 2 HN Points 04:21 Iran attempting cyberattacks against U.S. critical infrastructure (www.latimes.com)
133. vLLM IR：一種適用於vLLM的功能性中間表示 ↑ 2 HN Points 04:40 vLLM IR: A Functional Intermediate Representation for vLLM (github.com)
134. 《美國憲法》第二十五修正案第四款 ↑ 2 HN Points 04:37 Section 4 of the 25th Amendment to the US Constitution (en.wikipedia.org)
135. “Photofeeler”（約會照片排名網站）是如何運作的 ↑ 2 HN Points 04:37 How Photofeeler (dating pic ranking site) works (getdatingphotos.com)
136. 本地AI可讀取WhatsApp消息並安排會議——無需雲端，無需訂閱 ↑ 2 HN Points 04:29 Local AI reads WhatsApp and books meetings – no cloud, no subscription (lmim.tech)
137. 代理 SLO：讓自主代理基於關鍵指標運行 ↑ 2 HN Points 04:29 Agent SLOs: Grounding autonomous agents in metrics that matter (blog.firetiger.com)
138. 私募信貸市場資金大規模撤離，導致穆迪將BDC展望下調至負面 ↑ 3 HN Points 04:23 Private Credit Exodus Turns Moody''s BDC Outlook to Negative (www.bloomberg.com)
139. Vorim AI – 人工智能代理的身份識別、權限管理和審計日誌 ↑ 2 HN Points 04:13 Vorim AI – Identity, permissions, and audit trails for AI agents (vorim.ai)
140. 卡爾·西姆斯與亞歷山大·莫爾德溫採夫論科技與生物學的融合（2025） ↑ 2 HN Points 03:56 Karl Sims and Alexander Mordvintsev on Merging Technology and Biology (2025) (www.lerandom.art)
141. 兔子找自己的腳，結果找到了腳，這反而更糟 ↑ 2 HN Points 03:55 Rabbit Looks for Her Feet and Finds Feet, Which Is Somehow Worse (gist.github.com)
142. 與朝鮮有關聯的黑客正將目標鎖定在Fastify和Lodash背後的NPM維護者身上 ↑ 2 HN Points 03:46 North Korea-linked operators are hunting NPM maintainers behind Fastify, Lodash (anonhaven.com)
144. 美國稱，親伊朗的黑客破壞了部分工業控制系統 ↑ 2 HN Points 04:18 Pro-Iran hackers have disrupted some industrial-control systems, US says (www.nextgov.com)
145. 卸載機制如何運作：揭秘這款將我們的測試速度提升6倍的Rust命令行工具 ↑ 5 HN Points 04:15 How Offload Works: Inside the Rust CLI that sped up our tests by 6x (imbue.com)
146. LocalMind——數據始終留存於加拿大的文檔智能解決方案 ↑ 2 HN Points 04:10 LocalMind – Document intelligence that stays in Canada (localmind.caseonix.ca)
148. 使用 rm 命令將文件刪除限制在指定目錄內 ↑ 2 HN Points 03:54 Safe rm to restrict file removals to be under specified dir (blog.fraggod.net)
150. “阿耳忒彌斯”計劃宇航員拒絕與特朗普交談 [視頻] ↑ 10 HN Points 04:05 Artemis astronauts refuse to talk to Trump [video] (www.youtube.com)
151. 我們如何構建了一個防出錯的 AI 工程工作流 ↑ 3 HN Points 04:02 How We Built a Slop-Proof AI Engineering Workflow (twitter.com)
152. 伊朗黑客對美國能源和供水設施發起破壞性攻擊 [pdf] ↑ 6 HN Points 04:00 Iranian hackers launching disruptive attacks at U.S. energy, water targets [pdf] (www.cisa.gov)
153. 代理協議標準化：11份IETF草案競相角逐，其中1份將於4月10日失效 ↑ 2 HN Points 03:52 Agent Protocol Standardization: 11 IETF Drafts Competing, 1 Expires April 10 (global-chat.io)
155. Boost.Container：比較不同的雙端隊列實現 ↑ 2 HN Points 02:36 Boost.Container: comparing different deque implementations (boostedcpp.net)
156. NetBSD 中的進程隔離：由內核強制執行的、類似於“監獄”的隔離機制 ↑ 2 HN Points 03:54 Cells for NetBSD: kernel-enforced, jail-like isolation (netbsd-cells.petermann-digital.de)
157. 俄羅斯的“Fancy Bear”組織仍在攻擊路由器以推廣虛假網站 ↑ 2 HN Points 03:53 Russia''s Fancy Bear still attacking routers to boost fake sites (www.theregister.com)
158. Anthropic：對齊風險更新：Claude Mythos 預覽版 [pdf] ↑ 2 HN Points 03:51 Anthropic: Alignment Risk Update: Claude Mythos Preview [pdf] (www-cdn.anthropic.com)
159. 甲骨文聘請新任首席財務官，年薪95萬美元，而數千名員工面臨裁員 ↑ 3 HN Points 03:50 Oracle hires new CFO with 950K salary as thousands face layoffs (www.foxbusiness.com)
160. 索尼影視娛樂公司正在進行裁員，以調整業務重心 ↑ 2 HN Points 03:50 Sony Pictures Entertainment Layoffs Underway as Studio Refocuses (deadline.com)
161. 人工智能代理正在擾亂資深用戶的思維 ↑ 2 HN Points 03:49 AI agents are scrambling power users'' brains (www.axios.com)
162. 科技公司正在裁員。其中一些被裁員工隨後又以合同工身份被重新聘用。 ↑ 2 HN Points 03:49 Tech companies are axing roles. They''re hiring some back as contractors (www.businessinsider.com)
163. S3 文件與 S3 的嶄新面貌 ↑ 14 HN Points 03:44 S3 Files and the changing face of S3 (www.allthingsdistributed.com)
164. 聯邦撥款削減後，研究生面臨高昂的私人貸款壓力 ↑ 2 HN Points 03:43 Grad Students Face Costly Private Loans After Federal Cuts (www.bloomberg.com)
166. 對生物安全未來持悲觀（和樂觀）態度的理由 ↑ 2 HN Points 03:39 Reasons to be pessimistic (and optimistic) on the future of biosecurity (www.owlposting.com)
168. 這款精巧的日本印刷小工具，將世界各地的藝術家緊密聯結在一起 ↑ 2 HN Points 03:37 The nifty Japanese printing gadget uniting artists worldwide (www.theguardian.com)
169. Anthropic宣稱其新款AI模型Mythos將引發網絡安全領域的“大變革” ↑ 2 HN Points 03:40 Anthropic Claims Its New A.I. Model, Mythos, Is a Cybersecurity ‘Reckoning’ (www.nytimes.com)
172. V32數字電臺：一種在伊朗重現的神秘冷戰間諜系統 ↑ 2 HN Points 03:33 The V32 numbers station: a mysterious Cold War spying system revived in Iran (english.elpais.com)
173. 美國最高法院推翻了第五巡迴上訴法院要求互聯網服務提供商將盜版者斷網的裁決 ↑ 3 HN Points 03:31 SCOTUS overturns 5th Circuit ruling that told ISP to kick pirates off Internet (arstechnica.com)
174. 適用於 README 的 Duolingo 風格 GitHub 連續天數小工具 ↑ 2 HN Points 03:28 Duolingo-style GitHub streak widget for READMEs (github-streak.rahuldhole.com)
176. Tailslayer：用於降低內存讀取尾部延遲的庫 ↑ 2 HN Points 03:16 Tailslayer: Library for reducing tail latency in RAM reads (github.com)
177. 坎耶·韋斯特被禁止入境英國，無線音樂節隨之取消 ↑ 2 HN Points 03:15 Wireless festival cancelled after Kanye West banned from entering UK (www.theguardian.com)
179. Apple Studio Display XDR 現已獲得 FDA 批准，可用於診斷放射學 ↑ 2 HN Points 03:21 Apple Studio Display XDR Now FDA-Cleared for Diagnostic Radiology Use (www.macrumors.com)
180. Hotcopy 編碼 CLI：無上下文限制，且代理可在不同會話間進行學習 ↑ 2 HN Points 03:19 Hotcopy Coding CLI with no context ceiling and agents that learn across sessions (hotcopy.ai)
181. 美國勞動力參與率持續下滑 ↑ 2 HN Points 03:12 US Labor Force Participation Continues to Slide (restaurant.org)
182. Docker 與 Linux 內核：何時能隔離代理，何時不能 ↑ 3 HN Points 03:05 Docker and the Linux Kernel Isolate Your Agent, and Where They Don''t (timbreai.substack.com)
183. 美國聯邦存款保險公司（FDIC）發佈了針對發行穩定幣的金融機構的指導方針 ↑ 2 HN Points 02:42 FDIC Lays Out Guidelines for Institutions Issuing Stablecoins (www.bloomberg.com)
184. 德米斯·哈薩比斯：一位通才。他深知形象就是一切 ↑ 2 HN Points 02:58 Demis Hassabis: Renaissance man. He knows that image is everything (unherd.com)
185. 技術嫻熟的年長員工通過參加人工智能培訓來保住飯碗 ↑ 4 HN Points 02:57 Skilled older workers turn to AI training to stay afloat (www.theguardian.com)
186. is-antibot：檢測來自25家以上服務商的反機器人防護挑戰 ↑ 2 HN Points 02:55 is-antibot: Detect bot protection challenges from 25 providers (antibot.microlink.io)
187. 一張利用本地大型語言模型（LLM）自動生成的特朗普矛盾圖 ↑ 3 HN Points 02:51 A graph of Trump''s contradictions that builds itself using a local LLM (biglystable.com)
188. 美國證券交易委員會因執法線索向舉報人獎勵逾5000萬美元 ↑ 2 HN Points 02:34 SEC Awards Whistleblower over 50M for Enforcement Tip (www.bloomberg.com)
189. 印度快中子增殖反應堆首次達到臨界狀態 ↑ 2 HN Points 02:16 First criticality for Indian fast breeder reactor (www.world-nuclear-news.org)
190. 六個人。236集。一家咖啡館。——《老友記》的視覺呈現 ↑ 2 HN Points 02:51 Six People. 236 Episodes. 1 Coffee Shop. – Friends visualised (sheets.works)
191. 蘋果批准了允許 AMD 和 Nvidia 外接顯卡在 Mac 上運行的驅動程序 ↑ 2 HN Points 02:43 Apple approves drivers that let AMD and Nvidia eGPUs run on Mac (www.tomshardware.com)
192. 據報道，蘋果公司正面臨著關於MacBook Neo的“巨大困境” ↑ 4 HN Points 02:42 Apple Is Reportedly Facing a ''Massive Dilemma'' with the MacBook Neo (www.macrumors.com)
193. Linux 內核維護者正在推進移除對英特爾 486 處理器的支持 ↑ 2 HN Points 02:36 Linux kernel maintainers are following through on removing Intel 486 support (arstechnica.com)
194. Anthropic允許蘋果和亞馬遜測試更強大的Mythos人工智能模型 ↑ 3 HN Points 02:35 Anthropic Lets Apple, Amazon Test More Powerful Mythos AI Model (www.bloomberg.com)
196. 在“代理團隊”這一概念出現之前，我就已構建了多代理協作系統 ↑ 2 HN Points 02:29 I Built Multi-Agent Collaboration Before Agent Teams Existed (www.purecontext.dev)
197. 美國移民及海關執法局（ICE）根據美國機場安檢機構的線報，逮捕了800多人 ↑ 2 HN Points 02:27 ICE arrested more than 800 people after tips from US airport security agency (www.reuters.com)
198. gitsugi；用黃金填補你 GitHub 貢獻圖中的空白 ↑ 2 HN Points 02:24 gitsugi; Mend the gaps in your GitHub contribution graph with gold (github.com)
199. 《遞歸 Make 可能有害》（1997）[pdf] ↑ 2 HN Points 01:58 Recursive Make Considered Harmful (1997) [pdf] (aegis.sourceforge.net)
200. McGridsort：適用於 GPU k-way 歸併排序的扭曲網格 ↑ 2 HN Points 01:55 McGridsort: Warping Grids for GPU k-way mergesort (winwang.blog)
201. 《Inference Noise》，AI Slop 的前作 ↑ 2 HN Points 02:19 "Inference Noise", AI slop''s older brother (uxcontent.com)
202. 系統卡：克勞德·米索斯預覽 [pdf] ↑ 9 HN Points 02:18 System Card: Claude Mythos Preview [pdf] (www-cdn.anthropic.com)
203. 中情局在伊朗使用了名為“幽靈低語”的“遠程量子磁測”技術 ↑ 4 HN Points 02:17 CIA used "long-range quantum magnetometry" called "Ghost Murmur" in Iran (nypost.com)
204. 一個適用於 STM32、Linux 和瀏覽器的異步 Rust 代碼庫 ↑ 2 HN Points 02:16 One async Rust codebase for STM32, Linux and the browser (aimdb.dev)
206. 與伊朗有關聯的網絡攻擊者利用美國關鍵基礎設施中的PLC系統 ↑ 5 HN Points 02:15 Iranian-Affiliated Cyber Actors Exploit PLCs Across US Critical Infrastructure (www.cisa.gov)
207. 一個用於降低內存中尾部延遲的 C 庫 ↑ 2 HN Points 02:12 A C library that reduces tail latency in RAM (twitter.com)
208. 評估 Claude Mythos Preview 的網絡安全能力 ↑ 6 HN Points 02:11 Assessing Claude Mythos Preview''s cybersecurity capabilities (red.anthropic.com)
209. Alize——一份為您實時追蹤YouTube動態的每日簡報 ↑ 3 HN Points 02:09 Alize – A daily newsletter that watches YouTube for you (alize.me)
210. “Glasswing”項目：為人工智能時代保障關鍵軟件安全 ↑ 33 HN Points 02:09 Project Glasswing: Securing critical software for the AI era (www.anthropic.com)
211. Facevitals – 輕量級 rPPG 生命體徵監測（無需 GPU） ↑ 2 HN Points 02:08 Facevitals – Lightweight rPPG vital signs monitoring (No GPU required) (github.com)
212. 重塑，而非取代：人工智能正在如何改變我們當今的工作 ↑ 2 HN Points 02:07 Reshape, not replace: What AI is changing about our work today (www.mjeggleton.com)
213. 戰爭是最好的風投路演，卻沒人願意去做 ↑ 3 HN Points 02:05 War Is the Best VC Pitch Nobody Wants to Give (rawtext.io)
214. AI取代放射科醫生：醫生們抨擊英偉達和Anthropic首席執行官散佈虛假信息 ↑ 3 HN Points 02:04 AI replacing radiologists: Docs slam Nvidia, Anthropic CEOs for false info (timesofindia.indiatimes.com)
215. 唐納德·特朗普今晚正威脅要讓“整個文明”走向滅亡 ↑ 4 HN Points 02:04 Donald Trump is threatening the extinction of an ''entire civilization'' tonight (isdonaldtrumpalive.com)
216. 針對您產品的AI驅動型調侃（及解決方案） ↑ 2 HN Points 02:01 AI-powered roasts (and Solutions) for your product (roastcraft.app)
217. 我讓 Claude Code 在我的 Apple Watch 上運行起來了 ↑ 2 HN Points 02:00 I made Claude Code run on my Apple Watch (twitter.com)
218. OpenClaw 裡的 GPT-5.4 並不差。是你的提示語有問題 ↑ 2 HN Points 01:58 GPT-5.4 in OpenClaw doesn''t suck. Your prompts do (skylarbpayne.com)
219. 驍龍X2 Elite Extreme：蘋果的強勁對手，AMD/英特爾的重大威脅 ↑ 2 HN Points 01:58 Snapdragon X2 Elite Extreme: Strong Rival to Apple, Major Threat to AMD/Intel (www.notebookcheck.net)
224. Meta智能眼鏡的競爭對手錶示：拒絕“貼在臉上的攝像頭” ↑ 2 HN Points 01:45 Say no to a ''camera on your face'', says Meta smart glasses rival (www.ft.com)
225. 因為在 ChatGPT 和 Gemini 上無休止地刷屏而感到無聊 ↑ 2 HN Points 01:44 Got Bored due to endless scrolling on ChatGPT and Gemini (www.trydraft.app)
226. 從審計 WhatsApp 的私有推理中，我們對 TEE 安全有了哪些認識 ↑ 8 HN Points 01:40 What we learned about TEE security from auditing WhatsApp''s Private Inference (blog.trailofbits.com)
229. 開源大麻價格指數——方法論、SQL 及樣本數據 ↑ 2 HN Points 01:38 Open-Source Cannabis Price Index – Methodology, SQL, and Sample Data (github.com)
230. 釀酒師的產品上市指南 ↑ 2 HN Points 01:38 The Fermenter''s Guide to Launching a Product (vibeagentmaking.com)
232. Opus 4.6 在漏洞檢測方面的表現如何？ ↑ 2 HN Points 01:34 How good is Opus 4.6 at vuln detection? (zeropath.com)
233. OpenNOW：一款適用於 GeForce NOW 的開源桌面客戶端 ↑ 2 HN Points 01:34 OpenNOW: An open-source desktop client for GeForce NOW (github.com)
234. 關於KL散度的六條（半）直覺 ↑ 2 HN Points 01:34 Six (and a half) intuitions for KL divergence (www.perfectlynormal.co.uk)
235. 安全性與互操作性：真實的矛盾還是虛假的二分法？ ↑ 2 HN Points 01:32 Security versus Interoperability: Real Tension or False Dichotomy? (www.lawfaremedia.org)
237. 藉助 Chrome 的新垂直標籤頁和沉浸式閱讀模式，提升工作效率 ↑ 2 HN Points 01:30 Get more done with new vertical tabs and immersive reading mode in Chrome (blog.google)
238. 《為人民而自動》（Rand 與 LLMs） ↑ 2 HN Points 01:24 Automatic for the People (Rand and LLMs) (hollisrobbinsanecdotal.substack.com)
240. 多倫多某社區因“虛擬封閉社區”人工智能監控計劃引發爭議 ↑ 2 HN Points 01:18 Row over ''virtual gated community'' AI surveillance plan in Toronto neighbourhood (www.theguardian.com)
242. 亞馬遜第二總部（HQ2）的招聘工作在2025年陷入停滯 ↑ 2 HN Points 01:18 Amazon''s HQ2 hiring stalled in 2025 (virginiabusiness.com)
244. Fluux Messenger 0.15.0 – 一款現代化的跨平臺 XMPP 客戶端（TypeScript） ↑ 2 HN Points 01:16 Fluux Messenger 0.15.0 – A Modern Cross Platform XMPP Client (TypeScript) (www.process-one.net)
246. 前白宮通訊主任稱，特朗普正在“呼籲發動核打擊” ↑ 2 HN Points 01:48 Trump is ''calling for a nuclear strike,'' former White House comms director says (www.irishstar.com)
247. Agentic 開發願景：構建、運行、觀察——無需更多 Markdown ↑ 2 HN Points 01:46 Agentic development aspirations: build, run, observe – without more Markdown (devblogs.microsoft.com)
249. 虛構的引用正在汙染科學文獻 ↑ 2 HN Points 01:30 Hallucinated citations are polluting the scientific literature (www.nature.com)
250. 俄羅斯通過入侵路由器竊取微軟Office憑證 ↑ 2 HN Points 01:24 Russia Hacked Routers to Steal Microsoft Office Tokens (krebsonsecurity.com)
251. 柬埔寨為著名的探雷鼠馬加瓦揭幕雕像 ↑ 8 HN Points 01:23 Cambodia unveils a statue of famous landmine-sniffing rat Magawa (www.bbc.com)
252. 解析，而非驗證——在一種抗拒驗證的語言中 ↑ 2 HN Points 01:19 Parse, Don''t Validate – In a Language That Resists It (cekrem.github.io)
253. Genesis Agent v7 – 一個刪除了自身意識層的人工智能代理 ↑ 2 HN Points 01:12 Genesis Agent v7 – AI agent that deleted its own consciousness layer (github.com)
254. 測試表明，谷歌的AI概覽每小時都會說謊 ↑ 2 HN Points 01:10 Testing suggests Google''s AI Overviews tells lies per hour (arstechnica.com)
256. 一項早期臨床試驗顯示，在體CAR-T療法導致所有患者出現嚴重毒性反應 ↑ 2 HN Points 01:25 In Vivo Car T Causes Serious Toxicities in All Patients of Early Trial (www.insideprecisionmedicine.com)
257. OpenAI：智能時代的產業政策 ↑ 4 HN Points 01:21 OpenAI: Industrial Policy for the Intelligence Age (openai.com)
258. 你的老闆想利用監控數據來削減你的工資 ↑ 2 HN Points 01:14 Your boss wants to use surveillance data to cut your wages (pluralistic.net)
261. Smuggl – 將您的本地主機作為僅限受邀者訪問的鏈接分享 ↑ 2 HN Points 00:18 Smuggl – Share your localhost as an invite only link (smuggl.dev)
264. Linux虛擬助手開發者將項目夭折歸咎於專利流氓（2023） ↑ 2 HN Points 01:07 Creator of Linux virtual assistant blames patent troll for projects death (2023) (www.theregister.com)
266. “海綿城市”能拯救我們免受即將到來的洪災嗎？ ↑ 3 HN Points 01:05 Can Sponge Cities Save Us from the Coming Floods? (www.newyorker.com)
267. Kindle將於5月停止提供商店下載服務，並停止為第1至第5代Kindle設備辦理註冊 ↑ 6 HN Points 01:05 Kindle to end store downloads and registering for 1st-5th gen kindles in May (www.reddit.com)
268. ChatGPT的常用戶能夠準確識別AI生成的文本（2025） ↑ 4 HN Points 01:01 Frequent ChatGPT users are accurate detectors of AI-generated text (2025) (arxiv.org)
269. textmode.js 一個用於創建動態 ASCII 藝術的創意編程庫 ↑ 2 HN Points 00:53 textmode.js A creative coding library for building dynamic ASCII art (code.textmode.art)
270. OpenSSH 10.3/10.3p1 發行說明 ↑ 2 HN Points 00:42 OpenSSH 10.3/10.3p1 Release Notes (www.openssh.org)
271. 我們不再在代碼庫中編寫任何提示信息 ↑ 2 HN Points 00:33 We no longer write any of the prompts in our codebase (gist.github.com)
272. 靜觀其變：AI領域的“大手筆”假起跑 ↑ 2 HN Points 00:55 Wait Out AI''s Super-Spending False Start (www.bloomberg.com)
274. 彈性碳中和電力系統的全系統成本 [pdf] ↑ 3 HN Points 00:49 Full System Cost of a Resilient and Carbon Neutral Electricity System [pdf] (unece.org)
275. DNA機器人有望在人體內輸送藥物並追蹤病毒 ↑ 2 HN Points 00:43 DNA robots could deliver drugs and hunt viruses inside your body (openyourmindabretumente.blogspot.com)
276. 穩定幣正在悄然鞏固美元的主導地位 ↑ 3 HN Points 00:39 Stablecoins are quietly reinforcing US dollar dominance (verda.ventures)
277. 我們已經進入“遺留系統”時代了嗎？ ↑ 3 HN Points 00:33 Are We Legacy Computing Yet? (arewelegacycomputingyet.com)
278. 朝鮮針對開源軟件的攻擊藍圖 ↑ 4 HN Points 00:37 The Blueprint of a North Korean Attack on Open-Source (casco.com)
279. Seekdb M0：面向 OpenClaw 代理的持久化雲存儲與共享體驗 ↑ 2 HN Points 00:37 Seekdb M0: Persistent Cloud Memory and Shared Experience for OpenClaw Agents (en.oceanbase.com)
281. 利用通過 USB/IP 連接 WebUSB 的瀏覽器內 Linux 虛擬機，讓老舊打印機重獲新生 ↑ 2 HN Points 00:33 Rescuing old printers with an in-browser Linux VM bridged to WebUSB over USB/IP (printervention.app)
282. 一位頂尖科學家指出，改進遺傳學課程有助於減少種族主義 ↑ 2 HN Points 00:30 A star scientist showed that better genetics lessons could reduce racism (www.statnews.com)
283. 康威將迎來唯一一艘全尺寸《大白鯊》船隻複製品 ↑ 2 HN Points 00:22 Only full-scale Jaws boat replica to open in Conwy (www.bbc.com)
284. 推出 OmniConnect——計算機使用代理的身份驗證 ↑ 2 HN Points 00:21 Launching OmniConnect – Authentication for Computer Use Agents (anchorbrowser.io)
285. “氛圍編程”對B2B SaaS構成生存威脅 ↑ 2 HN Points 00:21 Vibe Coding Is an Existential Threat to B2B SaaS (nmn.gl)
286. 地球上最具爆發力的火山之一正在悄然積聚岩漿 ↑ 3 HN Points 00:20 One of Earth''s Most Explosive Volcanoes Is Quietly Refilling with Magma (www.sciencealert.com)
287. 特朗普的H-1B簽證規定是在保護壟斷嗎？ ↑ 3 HN Points 00:15 Are Trump''s H-1B Rules Protecting Monopolies? (openmarkets.substack.com)
288. DNA聚合酶無需模板即可合成長而複雜的序列 ↑ 2 HN Points 00:14 DNA polymerases can generate long, complex sequences without templates (www.nature.com)
289. 美國參議員稱中國汽車是“毒瘤”，誓言實施更嚴格的禁令 ↑ 2 HN Points 00:03 US Senator Calls Chinese Cars a ''Cancer,'' Vowing Stricter Ban (insideevs.com)
290. 是時候將開發機視為不可信了 ↑ 2 HN Points 00:22 Time to Start Treating Dev Machines as Untrusted (worklifenotes.com)
291. 將鐵路的前端從 Next.js 遷移出來 ↑ 2 HN Points 00:20 Moving Railway''s Frontend Off Next.js (blog.railway.com)
292. Macpow：適用於 Apple Silicon 的實時電源樹 TUI ↑ 2 HN Points 00:17 Macpow: Real-time power tree TUI for Apple Silicon (github.com)
294. 中國實施為期40天的空域管制，顯示軍事活動頻繁 ↑ 3 HN Points 00:14 China restricts airspace for 40 days in signal of intense military activity (www.msn.com)
295. “問題出在薩姆·阿爾特曼身上”：OpenAI內部人士不信任首席執行官 ↑ 2 HN Points 00:12 "The problem is Sam Altman": OpenAI Insiders don''t trust CEO (arstechnica.com)
297. 以“嚴重罪行和不端行為”為由彈劾唐納德·J·特朗普 [pdf] ↑ 3 HN Points 00:08 Impeaching Donald J. Trump for High Crimes and Misdemeanors [pdf] (www.congress.gov)
300. 如何思考動態系統：相空間的應用 ↑ 2 HN Points 04-07 How to Think About Dynamic Systems: Working with Phase Space (chillphysicsenjoyer.substack.com)