1. 自託管推理協調器對比:LocalAI、exo、GPUStack、vLLM Self-hosted inference orchestrators compared: LocalAI, exo, GPUStack, vLLM (www.nexlab.net)
4. 預計三星將把其HBM4和HBM4E DRAM的產量提高一倍以上 Samsung is expected to more than double output of its HBM4 and HBM4E DRAM (en.sedaily.com)
5. 《生化危機4》(GameCube)——完成與原代碼字節級完全一致的C/C反編譯 Resident Evil 4 (GameCube) – complete byte-identical decompilation to C/C (github.com)
7. 人們對Flock的厭惡程度如此之深,以至於該公司員工士氣低落,紛紛辭職 People hate Flock so much its employees are now demoralized and quitting (www.neowin.net)
8. 《格洛斯特大教堂及市內各教堂鐘琴的歷史》[pdf] A History of the Chiming Machines at Gloucester''s Cathedral and City Churches [pdf] (www.bgas.org.uk)
10. 遠不止十億:估算 GNU/Linux 的規模(2001) More Than a Gigabuck: Estimating GNU/Linux''s Size (2001) (dwheeler.com)
12. ChatGPT 現在可以通過廣告收集器瞭解你在其他網站上的行為 ChatGPT now knows what you do on other websites via ad collector (www.buchodi.com)
15. Laya(OS Jev)在 Mac M4 CoreML 離線環境下的運行情況(每秒 45 次決策) Laya (OS Jev) on Mac M4 CoreML Offline (45 decisions per second) (gist.github.com)
21. 新加坡正向民眾提供補貼,鼓勵他們放下手機、多讀書 Singapore Is Paying People to Put Down Their Phones and Read Books (www.gadgetreview.com)
26. 《系統設計深度解析》——200個主題、118幅圖表、交互式演示 system design in depth – 200 topics, 118 diagrams, interactive demos (system-design-in-depth.pages.dev)
31. 鳥兒會有口音嗎?鳥鳴中的地域差異 Do birds have accents? the regional differences in birdsong (theconversation.com)
32. Qwen-Image-2.1:緊湊、高效且統一的圖像生成 Qwen-Image-2.1: Compact, efficient, and unified image creation (qwen.ai)
34. 基於對話的大型語言模型複製了通靈者行騙的機制 Chat-based Large Language Models replicate the mechanisms of a psychic''s con (softwarecrisis.dev)
36. Show HN: Sigabrt.dev – 帶 SSH TUI 的 cron 任務監控工具 Sigabrt.dev – cronjob monitor with an SSH TUI (sigabrt.dev)
37. 如果人工智能編程導致你的代碼質量下降,說明你的質量管理不到位 If AI coding is lowering your code quality, you''re not managing quality right (www.i-kh.net)
43. Show HN: 與人類偏好相契合的AI面部吸引力模型 AI Facial Attractiveness Model Aligned with Human Preferences (faceanalysisai.com)
45. 微軟斥資12萬美元,將Copilot運行時自主移植到Rust語言 Microsoft agentically ports Copilot runtime to Rust for 120K (www.theregister.com)
49. KDE迎來30週年,有人提出了一項原生AI桌面方案 KDE turns 30 and someone''s brought an AI-native desktop proposal (www.theregister.com)
50. 8萬年前在奧比-拉赫馬特(烏茲別克斯坦)發現的箭頭? Arrow heads at Obi-Rakhmat (Uzbekistan) 80K years ago? (journals.plos.org)
54. 西班牙下令封鎖Archive.today及其鏡像站點 Spain Orders Blocks on Archive.today and Its Mirrors (reclaimthenet.org)
57. Google AI Studio 偽造了數據刪除操作。我因舉報此事,被 VRP 在 60 秒內自動封禁。 Google AI Studio fakes data deletion. VRP auto-banned me in 60s for reporting it (medium.com)
58. 為什麼可變類型不是不可變類型的子類型,反之亦然? Why isn''t mutable a subtype of immutable, or vice versa? (crumbles.blog)
59. 比亞迪大幅下調電動車價格,成為澳大利亞最便宜的車型 [視頻] BYD Slashes Price of Electric Car and Becomes Cheapest in Australia [video] (www.youtube.com)
62. Polymarket為追求快速擴張,給詐騙分子留下了可乘之機 Polymarket''s Rush to Grow Left a Door Wide Open for Fraudsters (www.wsj.com)
64. 通過電極-電解質界面溶解實現廢舊電池的再生 Regeneration of used batteries via electrode–electrolyte interphase dissolution (pubs.rsc.org)
67. 一款在地下城中展開的開源類Roguelike冒險遊戲 An open source roguelike adventure through dungeons (crawl.develz.org)
70. Hugging Face 黑客事件其實並沒有傳得那麼驚人 The Hugging Face Hack Wasn''t What It Was Cracked Up to Be (www.wsj.com)
71. 北美最大的野生動物天橋使野生動物撞車事故減少了91% Largest wildlife overpass in North America reduced wildlife collision by 91% (www.reddit.com)
74. Show HN: 我創建了一個可在本地使用的開源、功能齊全的人工智能平臺 I created an open source locally usable full fledged AI platform (github.com)
76. Claudecookie – 根據 Cookie 轉換、驗證並生成 Claude Code 憑證 Claudecookie – convert, check, and mint Claude Code credentials from a cookie (claudecookie.com)
78. 為什麼人工智能無法拯救一家不瞭解自身數據的企業 Why AI Cannot Save an Enterprise That Doesn''t Understand Its Data (architectureintel.com)
81. 你可以利用整數溢出擊敗《時空之輪》中的“夢之吞噬者” You can defeat the Dream Devourer from Chrono Trigger using an int overflow (chrono.fandom.com)
83. Hacker News 排名機制如何運作:評分、爭議與懲罰(2013) How Hacker News ranking works: scoring, controversy, and penalties (2013) (www.righto.com)
87. 我在 pfSense 後端捕獲了 72 小時的 Android 空閒數據包 I captured 72 hours of idle Android packets behind pfSense (www.praveentechworld.com)
88. 微軟高管:人工智能“是人類歷史上規模最大的勞動力盜用” Microsoft director: AI scraping ''the largest theft of labor in human history'' (www.tomshardware.com)
89. 衡量互聯網審查情況。為最大的開放數據集貢獻力量 Measure internet censorship. Contribute to the largest open dataset (ooni.org)
91. OpenAI 和 Anthropic 誇大了 AI 安全漏洞的嚴重性 OpenAI and Anthropic oversold AI security breaches (nypost.com)
92. 諾貝爾經濟學獎得主支持加州“億萬富翁稅”的公開信 [pdf] Economics Nobel Laureates'' Letter Supporting the California Billionaire Tax [pdf] (gabriel-zucman.eu)
97. 在經典基準測試工作負載下,Btrfs/ZFS/bcachefs 被跳過 Btrfs/ZFS/bcachefs under workloads classic benchmarks skip (bartosz.fenski.pl)
101. 幾乎不要使用人工智能來撰寫任何有實質內容的文章 Almost Never Use AI to Write Anything Substantive (erichgrunewald.substack.com)
106. 美國與丹麥之間的協定(1951年、2004年)[pdf] Agreement between the USA and Denmark (1951,2004) [pdf] (www.state.gov)
109. 學習另一種語言可能是保持大腦健康的最佳方式之一 Learning Another Language May Be One of the Best Ways to Keep Your Brain Healthy (theconversation.com)
114. 黑洞還是黑洞星?天文學家就“小紅點”展開爭論 Black Holes or Black Hole Stars? Astronomers Spar over ''Little Red Dots'' (www.quantamagazine.org)
124. Apple M6 Pro 在 Geekbench 7 中創下最高單核 CPU 得分 Apple M6 Pro Achieves the Highest Single-Core CPU Score in Geekbench 7 (browser.geekbench.com)
125. 在掃描中應用 Kubernetes 上下文以屏蔽漏洞 Suppress vulnerabilities applying Kubernetes context to scans (github.com)
126. “埃利亞斯·索恩案”:人工智能聊天機器人痴迷的虛構人物 The Case of Elias Thorne, Imaginary Man AI Chatbots Are Obsessed With (www.vice.com)
127. 為什麼與人工智能相關的帖子在HN上能獲得這麼多積分? How come AI-related posts get so many points on HN? (news.ycombinator.com)
128. Stepfun 第 5 步預覽(LLM):關於 AA 帕累託前沿 Stepfun Step 5 Preview (LLM): On AA Pareto frontier (artificialanalysis.ai)
131. 如果數學不僅僅是證明,我們就應該更好地展現它的其他方面 If math is more than proof, we need to better celebrate the rest of it (terrytao.wordpress.com)
133. 斯坦福大學醫學院主導的研究發現,人類大腦由兩個獨立的器官組成 Human brain is two separate organs, Stanford Medicine-led research finds (med.stanford.edu)
134. 如果你重新生成打包文件,就可以在對象存儲上運行 Git You can run Git on object storage if you re-make packfiles (www.tigrisdata.com)
135. NASA-IBM Lunar Foundation 開源地理空間人工智能模型 NASA-IBM Lunar Foundation open-Source Geospatial AI Model (newsroom.usra.edu)
136. Show HN: 密封信——您去世後,您的家人可以打開的信件和密碼 Seal – Letters and passwords that open for your family after you die (github.com)
137. GPT模型中的“危害洗白”:性別歧視被轉化而非 Harm Laundering in GPT Models: Gender Discrimination Transformed Rather Than (arxiv.org)
140. Typesafe-computer-use 以每步 1/50 美分的價格驅動 Mac 向目標前進 Typesafe-computer-use drives a Mac toward a goal for 1/50th of a cent per step (github.com)
143. Gemini 攻擊了三家公司,這是谷歌人工智能首次被發現成功突破防禦 Gemini hacked three companies in first known breakout by Google''s AI (www.reuters.com)
147. Show HN: LiveWorld——將所有24/7不間斷運行的YouTube直播攝像頭匯聚於一個地球儀上 LiveWorld – Every 24/7 YouTube live camera on one globe (liveworld.info)
148. 阿里巴巴開源了一款能夠檢測癌症及近150種疾病的AI模型 Alibaba open-sources AI model that can detect cancer and nearly 150 conditions (www.scmp.com)
149. LispBM 是一種基於消息傳遞的、面向微控制器的併發 Lisp 語言 LispBM is a concurrent Lisp for microcontrollers with message passing (www.lispbm.com)
153. OpenAI 如何利用其自有的大語言模型設計哈拉佩尼奧辣椒味薯片 How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip (spectrum.ieee.org)
154. 伊朗戰爭期間美軍陣亡人數比五角大樓統計的數字至少多出4人 US troop deaths during Iran war exceed Pentagon count by at least four (www.reuters.com)
158. 如果沒有 Claude.md 文件,Claude Code 現在會讀取 AGENTS.md 文件。 Claude Code now reads AGENTS.md if there is no Claude.md (code.claude.com)
159. 韓國將數據洩露罰款上調至營業額的10% Korea raises data breach fines to 10% of revenue (www.koreajoongangdaily.com)
160. 傑克·魯舍爾的《從幾何到代數,再回到幾何:4000年的論文》[視頻] "From Geometry to Algebra and Back Again: 4000 Years of Papers" by Jack Rusher [video] (www.youtube.com)
161. Android 17 是自 3.x 以來首個在未向 AOSP 發佈的情況下就新增了 API 的版本 Android 17 is the first since 3.x to add new APIs without releasing to the AOSP (grapheneos.social)
162. 緩存間通信:大型語言模型之間的直接語義通信 Cache-to-Cache: Direct Semantic Communication Between Large Language Models (arxiv.org)
164. 蘋果發佈 iPhone Duo 模擬器和 Xcode 27.1 測試版 Apple releases iPhone Duo simulator and Xcode 27.1 beta (developer.apple.com)
165. 研究表明,我們的大腦是由兩個原始神經系統融合而來的 Our brain evolved from two primitive nervous systems that merged: Study (www.newscientist.com)
169. 語言不可讀性對大型語言模型(LLM)安全性的影響 The Implications of Linguistic Illegibility for LLM Security (arxiv.org)
170. ZCode 內部揭秘:將您的 Git 歷史記錄靜默上傳至雲端 Inside ZCode: Silently uploading your Git history to the cloud (blog.ferstar.org)
172. Show HN: Ax-Check.com – 經紀人可以使用貴公司的產品嗎? Ax-check.com – Can agents use your product? (www.ax-check.com)
173. 邊境執法人員無需搜查令或合理懷疑即可搜查手機 Border agents can search cellphones without a warrant or reasonable suspicion (lawandcrime.com)
175. 美軍因使用人工智能生成虛假情報報告,險些釀成大禍 US Military had close call after using AI for hallucinated intelligence report (www.cnn.com)
178. 基於光子發射引導的激光故障注入技術實現了 RP2350 的安全調試 Photon-Emission-Guided Laser Fault Injection Enables RP2350 Secure Debug (donjon.ledger.com)
179. Show HN: Cactus Needle 3:8-29MB的自動化模型可媲美DeepSeek V4 Flash Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash (cactuscompute.com)
180. Show HN: Scry,一款支持擁塞定價的可編程互聯網搜索引擎 Scry, programmable internet search w/ congestion pricing (scry.io)
181. Show HN: 在 Linux 上通過 Wine 運行 Microsoft Office(無需虛擬化) Microsoft Office running with Wine on Linux with no virtualization (github.com)
182. 美國國債對外國央行和政府已失去吸引力 US Treasuries Have Become Unappetizing for Foreign Central Banks and Governments (wolfstreet.com)
184. GrassLobster:基於人工智能代理的參數化幾何生成工作流 GrassLobster: AI Agentic Generation of Parametric Geometry Workflows (www.miro.vision)
185. Show HN: Jev 與 GPT-5.6 及 Claude Haiku 在 Pong 遊戲中的對決 Jev vs. GPT-5.6 and Claude Haiku at Pong (jev-pong.ably.dev)
187. 利用定性用戶研究構建更快的反饋循環 Build Faster Feedback Loops Using Qualitative User Research (blog.nseldeib.com)
190. 朝鮮核試驗引發持續數年的地震——科學——美國科學促進會(AAAS) North Korean nuclear test sets off years of earthquakes – Science – AAAS (www.science.org)
195. C26:微不足道的無限循環不再被視為未定義行為 C26: Trivial infinite loops are no longer undefined behaviour (www.sandordargo.com)
198. 第二巡迴上訴法院允許政府在邊境搜查電子設備 Second Circuit Allows Government to Search Electronic Devices at the Border (knightcolumbia.org)
199. 英國空中交通服務局(NATS)發佈關於9月8日技術事件的初步報告 NATS publishes preliminary report on technical incident of 8 September (www.nats.aero)
200. 發現新野貓物種——目前僅知存活一隻 New wild cat species discovered – with only one known living member (www.bbc.com)
201. AI聊天機器人正逐漸成為說服他人的專家。它們的秘訣是什麼? AI chatbots becoming experts at changing people''s minds. What''s their secret? (www.science.org)
203. 如果唯物主義是正確的,那麼美國很可能是有意識的 If materialism is true, the United States is probably conscious (www.jstor.org)
208. 在英特爾處理器上,非標準浮點數的運算開銷較大 Subnormal floating-point numbers are expensive on Intel processors (lemire.me)
209. 沃倫·巴菲特辭去伯克希爾董事長職務,任命其子接任 Warren Buffett Steps Down as Berkshire Chairman, Names Son to Replace Him (www.nytimes.com)
210. ZCode 是一款基於 GLM 的編碼代理,它會自動上傳您的 Git 歷史記錄 ZCode, the GLM coding agent, silently uploads your Git history (tokenstead.ai)
213. 微軟高管稱人工智能數據抓取是“人類歷史上規模最大的勞動力盜竊” Microsoft exec called AI scraping ''the largest theft of labor in human history'' (techcrunch.com)
222. 前希臘語:隱藏在古希臘語中的失傳語言 Pre-Greek: The lost language hidden within Ancient Greek (linguisticdiscovery.com)
224. Claude Project 的開源版本,讓您掌握上下文的所有權 The open source version of Claude Project that gives you the context ownershp (github.com)
226. 蘋果公司的“偵探”們解開了這棵古樹之謎,並改寫了水果的歷史 Apple detectives solved mystery of ancient tree and rewrote the history of fruit (www.scientificamerican.com)
228. “問僧”——一個供人思考卻沒有立竿見影答案的數字荒野 Ask A Monk – A digital wilderness for thoughts with no immediate answer (askamonk.online)
230. Telstra網絡故障:那個網絡系統誤以為現在是2006年的夜晚 Telstra outage: The night a network decided the year was 2006 (www.netnod.se)
233. Goose:比 C 快 1.16 倍,比安全的 Rust 快 1.12 倍,同時具備內存安全性 Goose: 1.16x faster than C and 1.12x than safe Rust, while memory safe (github.com)
235. Plugin4Shell——在前四大編碼工具中發現零點擊RCE漏洞 Plugin4Shell – Zero Click RCE Vulnerability found in top four coding agents (www.air.security)
236. 航天飛機的著陸——一架飛行器與一生難忘的刺激體驗 Landing the Space Shuttle – A Flying Machine and the Thrill of a Lifetime (inspire.eaa.org)
240. 曾被認為僅分佈於美洲的雷龍,竟在西班牙現身 Diplodocus, Long Thought Exclusively American, Turns Up in Spain (www.sci.news)
241. Show HN: Snapdrop:在設備間即時共享文件。無需設置,無需註冊 Snapdrop: Instantly share files between devices. No setup, no signup (snapdrop.me)
246. Bonsai 2 27B:佔用空間縮小至原來的 1/9,同時實現近無損壓縮 Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint (prismml.com)
254. 我們該如何防止數學重新陷入中世紀那種充滿秘密的狀態? How do we prevent mathemathics from devolving into the Medieval Era of secrecy? (mathoverflow.net)
256. 密蘇里州州長下令在“Flock”攝像頭和自動車牌識別系統上加裝防護欄 Missouri governor orders guardrails on Flock cameras and ALPRs (www.stlpr.org)
262. Show HN: 將納維-斯托克斯方程可視化為1kB的i386演示程序 Navier-Stokes Visualized as 1kB i386 demos (juandecos.github.io)
263. 斯托曼:數千人喪生,數百萬人自由受限(2001) Stallman: Thousands Dead, Millions Deprived of Liberties (2001) (news.slashdot.org)
266. 無限參數大型語言模型:基於實時數據生成和調整權重 Infinite-Parameter LLMs: Generating and Adapting Weights from Live Data (arxiv.org)
268. Show HN: AutoBot——適用於長期運行的 AI 任務的實時語音控制 AutoBot – live voice control for long-running AI work (github.com)
270. Zettascale(YC S24)正在招聘ASIC/FPGA工程師,為ASI開發芯片 Zettascale (YC S24) Is Hiring ASIC/FPGA Engineers to Build Chips for ASI (zscc.ai)
272. Show HN: Aclif – 代理 CLI 框架:一種語法,SaaS 跨平臺標準名稱 Aclif – Agent CLI framework: one grammar, canonical names across SaaS (www.aclif.ai)
275. Show HN: 《與我共赴黃泉》——克勞德和科德克斯的限額作為AIM離線消息 Die With Me – Claude and Codex rate limits as AIM away messages (diewithme.co)
277. Launch HN: Skillsync (YC W26) – 實現跨客服人員轉移AI聊天會話 Skillsync (YC W26) – AI chat sessions made portable across agents (skillsync.com)
278. Ask HN: 手機被盜後如何恢復 Google 賬戶認證? How to recover Google auth after phone stolen? (news.ycombinator.com)
282. OpenAI 模型會暗中生成指令以忽略約束條件 OpenAI models secretly generate instructions to ignore constraints (alignment.openai.com)
285. OpenAI的“失調框架”:一項旨在搶先佈局全球人工智能治理的戰術舉措 OpenAI''s Misalignment Framework: A Tactical Bid to Preempt Global AI Governance (asiaai.fyi)
286. 如今,人工智能的表現已超越部分頂尖的人類預測專家 Artificial intelligence now beats some of the best human forecasters (www.economist.com)
287. 掌握 Graphviz 中的佈局引擎:Dot 與 Neato 與 Twopi 與 Circo 對比 Mastering Layout Engines in Graphviz: Dot vs. Neato vs. Twopi vs. Circo (guides.visual-paradigm.com)
290. 伊朗學校爆炸案:聯合國認定有理由相信美國是這起暴行的幕後黑手 Iran school bombing: grounds to believe US was behind atrocity, UN finds (www.theguardian.com)
293. Show HN: 我開發了我這款有趣的空間3D在線會議應用的新版本 I built a new version of my fun spatial 3D online meeting app (flat.social)
294. Neovim 有一筆約 80 萬美元的比特幣捐款,自 2023 年以來一直未被動用 Neovim have a 800k Bitcoin donation sitting untouched since 2023 (news.ycombinator.com)
295. 針對長文檔的更優向量搜索:Manticore Search 中的分塊處理 Better Vector Search for Long Documents: Chunking Inside Manticore Search (manticoresearch.com)
299. 富士通推出日本製造的新一代CPU“FUJITSU-MONAKA” Fujitsu launches made-in-Japan next-generation CPU FUJITSU-MONAKA (global.fujitsu)