Hacker News Daily — 2026-05-18 Daily Best
Scraped at 08:33, May 19, 2026 (PDT)
(1) Andrej Karpathy joins Anthropic
Andrej Karpathy joins Anthropic, signaling talent moves and emphasis on safety-aligned AI; his expertise could boost research and engineering for scalable, aligned AI.
Andrej Karpathy 加入 Anthropic，显示该公司在安全对齐 AI 方面的持续人才投入；他的专长可能提升研究与工程，推动可扩展、对齐的 AI 能力。
OpenBSD 7.9 ships with improved security hardening, updated drivers, and new ports, continuing the project's emphasis on correctness and minimalism. The release also expands platform support and performance improvements.
OpenBSD 7.9 发布，带来更强的安全强化、更新的驱动以及新端口，延续对正确性与简约设计的坚持。此次版本还扩展了平台支持并带来性能提升。
(3) New accessibility features powered by Apple Intelligence
Apple introduces a suite of accessibility features powered by on-device AI, expanding supports across vision, hearing, motor, and cognitive accessibility. The updates emphasize private, fast AI-assisted tools and broader inclusion across Apple devices.
苹果推出一系列由本地 AI 提供支持的新无障碍功能，覆盖视觉、听力、运动和认知等方面。更新强调隐私友好的高性能 AI 助手与跨设备的更广泛无障碍覆盖。
(4) Gaussian Splat of a Strawberry
The project showcases a Gaussian splat rendering of a strawberry, producing soft volumetric shading from a point-based approach. It illustrates how advanced rendering techniques are being used in art and experiments, signaling future directions for real-time graphics pipelines.
该作品以草莓为例，展示高斯斑点渲染技术，利用点云实现柔和体积光照与纹理效果。它体现了前沿渲染技术在艺术创作中的应用趋势，以及对实时图形管线的潜在影响。
(5) The last six months in LLMs in five minutes
A rapid recap of the most consequential developments in large language models over the last six months. It highlights breakthroughs in open-source models, ongoing safety and alignment debates, and the expanding tooling and deployment ecosystem that are shaping how practitioners build and scale applications.
在过去六个月里，LLM 领域出现了一系列关键进展。重点包括开源模型的突破、关于安全与对齐的持续讨论，以及推动应用落地的工具链与部署生态的扩展。
(6) Elon Musk has lost his lawsuit against Sam Altman and OpenAI
A US court ruled against Elon Musk in his lawsuit against Sam Altman and OpenAI. The decision halts his bid to challenge OpenAI’s conduct in court and underscores the uphill road for high-profile disputes in the AI leadership circle.
美国法院裁定马斯克在针对萨姆·奥特曼与 OpenAI 的诉讼中败诉。裁决终结了他提出的对 OpenAI 行为的诉讼，凸显了 AI 领域高知名度对立中的法律挑战。
(7) Mini Shai-Hulud Strikes Again: 314 npm Packages Compromised
A supply-chain intrusion hit the npm ecosystem, compromising 314 packages as part of the Mini Shai-Hulud campaign. The incident underscores the fragility of dependency-heavy tooling and reinforces best practices like lockfiles, trusted maintainers, and regular audits to detect tampering.
小型 Shai-Hulud 攻击再次蔓延，314个 npm 包被入侵篡改。这一事件暴露了依赖密集型工具的脆弱性，强调使用 lockfile、信任的维护者审计以及定期安全检查的重要性。
Peter G. Neumann, a longtime observer and author on computer security risks, has died. His work on computer-related risks shaped how engineers approach safety, reliability, and policy in software and networks. His writings and talks continue to influence risk management practices in software and cybersecurity.
彼得·诺伊曼是一位长期关注并著作 Computer Related Risks 的计算机科学家，其关于计算机安全风险的工作影响了工程师对软件与网络安全、可靠性与政策的思考。他的著作和演讲继续影响软件与网络安全领域的风险管理实践。
(9) Click (2016)
A 2016 work centered on click interactions as a lens into human-computer interfaces.
一部聚焦点击交互、以此观察人机界面的2016年作品。
(10) Show HN: Files.md – Open-source alternative to Obsidian
Files.md is an open-source knowledge management app pitched as an Obsidian alternative, focusing on local Markdown files and a lightweight footprint. It appeals to users seeking privacy and ownership of their data.
Files.md 是一款开源知识管理应用，被定位为 Obsidian 的替代方案，注重本地 Markdown 文件与轻量化。它吸引那些寻求隐私和对数据拥有权的用户。
(11) Anthropic acquires Stainless
Anthropic has acquired Stainless, a startup focused on safety tooling and governance capabilities. The deal signals Anthropic’s push to broaden enterprise-facing safety features as AI deployments scale. Integration could accelerate productization for customers and strengthen governance tooling.
Anthropic 收购 Stainless，一家专注于安全工具与治理能力的初创公司。此举昭示 Anthropic 将进一步扩展面向企业的安全功能，以应对大规模 AI 部署的合规与风险管理需求。并购有望加速对外提供的产品落地，并强化治理能力。
(12) Garry Tan, the CEO of YC, accused me of unethical reporting
Radley Balko recounts Garry Tan's accusation that his reporting was unethical, setting up a rare public clash between a startup founder and a journalist. The exchange highlights how power dynamics shape tech coverage and where accountability boundaries lie. It argues for transparent standards and robust verification in tech journalism.
Radley Balko 描述了 YC CEO Garry Tan 指控其报道不道德的对峙，揭示创始人与记者之间罕见的公开冲突。此事凸显权力关系如何影响科技报道的边界与问责。强调在科技新闻中坚持透明标准与充分核实的重要性。
(13) Pope Leo XIV’s first encyclical Magnifica humanitas to be published May 25
Announces the forthcoming publication of Pope Leo XIV's first encyclical Magnifica humanitas, outlining expected themes around human dignity and social charity.
梵蒂冈宣布教皇利奥十四世的首部通谕 Magnifica humanitas 将于5月25日公布，预计聚焦人类尊严与社会慈善等主题。
(14) We stopped AI bot spam in our GitHub repo using Git's –author flag
By enforcing commit authorship with Git's --author flag, the team significantly cut spam from automated bots in their repo. The approach blocks impersonation at the source and complements existing security controls. It shows how simple version-control metadata can curb abuse without heavy ML defenses.
通过在提交时强制使用 Git 的 --author 标志来标识作者，该团队显著降低了仓库中的机器人垃圾信息。此做法在源头阻断冒充行为，辅以其他安全控制。该案例证明了版本控制元数据在反滪用方面的实用性，不必依赖复杂的机器学习防护。
(15) Peter Salus has died
Obituary for Peter Neumann, a renowned computer scientist known for security research and contributions to information safety and policy.
纪念在计算机安全与信息风险研究领域具影响力的彼得·纽曼，他的离去被视为对安全科学和社区的一大损失。
(16) The FBI Wants to Buy Nationwide Access to License Plate Readers
Covers a government-wide push by the FBI to acquire nationwide access to license plate reader data. It also touches implications for vendors building apps that rely on LPR data.
报道 FBI 求购全国范围的车牌识别数据访问权，讨论相关隐私与公民自由风险，以及广泛数据访问可能带来的监控扩张。并点出对依赖车牌识别数据的应用开发者的潜在影响。
(17) Haiku OS runs on M1 Macs now
Haiku OS, a BeOS-inspired open-source system, now demonstrates running on Apple Silicon M1 Macs, marking progress in ARM64 support for this project. The update highlights the ongoing porting efforts and the challenges of booting modern Macs with niche operating systems.
Haiku OS 作为 BeOS 启发的开源系统，现已在 Apple Silicon 的 M1 Mac 上运行，标志着其 ARM64 支持的进展。该进展凸显了持续的移植工作以及在现代 Mac 硬件上引导小众操作系统所面临的挑战。
(18) Who will buy your services if you fire us all?
The piece argues that aggressive outsourcing or mass layoffs undermine long-term customer value, as vendors risk losing the relationship, trust, and capability to deliver. It stresses sustainable staffing and human-centric service as a competitive differentiator.
文章指出，过度裁员或大规模外包会削弱长期客户价值，因为失去关系、信任与交付能力。强调以人本、可持续的人员配置提供服务，避免因削减成本而损害客户信任。
(19) We let AIs run radio stations
Andon Labs demonstrates AI-operated radio stations, where AI curates playlists and hosts content with automated narration. The project probes how far automation can go in media, and what remains uniquely human in storytelling and curation.
Andon Labs 展示了让 AI 运营广播电台的尝试，AI 负责选曲并进行自动化主持。该项目探索媒体领域自动化的边界，以及在叙事与选曲方面仍需人工介入的部分。
(20) Iran starts Bitcoin-backed ship insurance for Hormuz strait
Iran launches shipping insurance backed by Bitcoin for vessels through the Strait of Hormuz. The crypto-backed instrument offers an alternative risk transfer tool amid sanctions and regional instability, signaling crypto's growing role in traditional finance and risk management.
伊朗为通过霍尔木兹海峡的船舶推出比特币背书的保险，这一以加密货币为担保的保险工具为制裁与区域不稳局势下的风险转移提供了替代方案，体现加密货币在传统金融与风险管理中的日益嵌入。
(21) Project Glasswing: what Mythos showed us
Mythos demonstrates a new approach to cybersecurity tooling with Glasswing, showcasing how frontier models can assist threat hunting and incident response. The project illustrates how AI models integrated with domain-specific data can accelerate cognitive tasks in security operations, while highlighting limits like data sensitivity and model hallucination. It signals a trend: domain-specific AI tooling for cybersecurity is moving from concept to practice.
Mythos 的 Glasswing 展示了将前沿模型应用于网络安全工具的新思路，覆盖从威胁狩猎到事件响应的场景。该项目凸显了将 AI 模型与领域数据结合以提升安全运维认知任务的潜力，同时也提醒数据敏感性与模型可信性等局限。预示着以安全领域为目标的 AI 工具正从概念走向落地。
(22) Actually, democracy dies in H.R.
The piece argues that HR processes can quietly undermine democratic norms within organizations, through centralized control, surveillance, and opaque decision-making. It connects these patterns to broader concerns about democracy and calls for greater employee participation and governance safeguards.
文章指出人力资源流程可能通过集权、监控和隐性决策等方式侵蚀组织内部的民主原则，与更广泛的民主健康相关。呼吁加强员工参与、治理保障与透明度提升。
(23) Eric Schmidt speech about AI booed during graduation
Eric Schmidt’s AI-themed graduation remarks were booed, signaling skepticism about the AI hype among students. The moment reflects broader tensions between industry optimism and concerns about automation's impact on jobs and society.
施密特在毕业典礼上就 AI 的讲话遭遇嘘声，显示学生群体对 AI 热潮的怀疑情绪。此刻折射出行业乐观情绪与对自动化对就业及社会影响的担忧之间的紧张。
(24) Cursor Introduces Composer 2.5
Cursor released Composer 2.5, adding improvements to the AI agent toolchain, such as better task orchestration, reliability, and debugging. The update helps developers build more capable agents with fewer edge-case issues.
Cursor 发布了 Composer 2.5，带来改进的 AI 代理工具链，如更好的任务编排、稳定性与调试能力。更新帮助开发者构建更强大的代理，减少边缘情况问题。
(25) Qwen 3.7 Preview
Alibaba's Qwen 3.7 preview reveals improvements in reasoning, multilingual support, and safety, continuing to advance open AI copilots. The release hints at how enterprise AI stacks will evolve with stronger safety features and integration.
阿里巴巴的 Qwen 3.7 预览在推理能力、多语言支持和安全性方面有所提升，继续推动开放式 AI 助手的发展。版本透露，企业级 AI 架构将通过更强的安全特性与集成进一步演进。
(26) Show HN: Auto-identity-remove – Automated data broker opt-out runner for macOS
Auto-identity-remove is an open-source macOS tool that automates opt-out requests to data brokers, aiming to reduce personal data exposure. It demonstrates a practical approach to privacy hygiene with automation.
Auto-identity-remove 是一个开源的 macOS 工具，自动化向数据经纪人提交退出请求，旨在降低个人数据暴露。它展示了通过自动化实现隐私卫生的实用方法。
(27) Shutterstock to pay $35M over hard-to-cancel subscriptions
Shutterstock has agreed to pay $35 million to settle FTC allegations that its subscription practices made cancellations difficult and misled consumers. The case highlights regulatory scrutiny of online subscriptions and the need for clear, user-friendly cancellation flows.
Shutterstock 同意支付 3500 万美元以解决 FTC 对其难以取消订阅做法的指控。此案凸显对在线订阅的监管关注以及提升取消流程透明度与易用性的必要性。
(28) Hyperpolyglot Lisp: Common Lisp, Racket, Clojure, Emacs Lisp
A survey of the main Lisp dialects—Common Lisp, Racket, Clojure, and Emacs Lisp—highlighting where each shines and where they clash. It covers language design, tooling, ecosystems, and practical tips for choosing or bridging between dialects.
文章梳理了常用的 Lisp 方言：Common Lisp、Racket、Clojure 与 Emacs Lisp，比较各自的优势与局限。它讨论了语言设计、开发工具和生态系统，并给出在项目中选择或实现跨方言互操作的实用思路。
(29) AI eats the world (Spring 26) [pdf]
A 2026 AI landscape report surveying adoption, risks, and policy implications across industries, offering a snapshot of how AI investments are evolving globally.
一份2026年 AI 现状报告，梳理各行业的普及、风险与政策含义，提供全球 AI 投资演变的快照。
(30) No more JetBrains products for me
The author explains leaving JetBrains tools due to licensing, performance, or policy concerns, detailing impact on workflow and suggesting alternative tools and adjustments.
作者解释放弃 JetBrains 工具的原因，涉及许可证、性能和政策等因素。文章还讨论对工作流程的影响，并提出可替代工具和工作流调整的建议。
(31) New York to tax luxury second homes in NYC
New York proposes taxing luxury second homes in NYC, broadening the tax base beyond primary residences. The move could influence real estate demand, residency choices, and city revenue.
纽约市拟对豪华第二套住宅征税，扩大税基至非主要居所。此举可能影响房地产需求、居住选择与城市财政收入。
(32) 'We mould trees to grow into the shape of chairs'
The BBC feature explores designers growing trees into chair shapes, blending biology with furniture design for sustainable, living materials. It highlights how living processes can redefine manufacturing, aesthetics, and material sourcing.
这篇 BBC 报道介绍设计师让树木按椅子形状成长，运用生物学实现可持续、有机的家具设计。它展现了活体材料如何重新定义制造、美学与材料来源。
(33) Where Are the Vibecoded Photoshops?
The piece investigates the status of Vibecoded Photoshop works and what it means for image provenance and authenticity amid evolving manipulation tech.
本文探究 Vibecoded Photoshop 图像的现状，以及在日益成熟的图像操控技术背景下，对图像出处与真实性可能带来的影响。
(34) GenCAD
GenCAD appears to be a project centered on automated CAD generation and design workflows.
GenCAD 似乎是一个专注于自动化 CAD 生成与设计工作流的项目。
(35) Linux security mailing list 'almost unmanageable'
Linus Torvalds says AI-powered bug hunters have made the Linux security mailing list almost unmanageable due to volume and noise. The note highlights the tension between openness and signal quality in security discourse.
托瓦尔兹表示 AI 驱动的漏洞猎人让 Linux 安全邮件列表几乎难以管理，原因是信息量大且噪声较多。这体现了开放性与信息质量之间的张力。
(36) 1024000^2 Blocks, 2B2T Minecraft Server World Download Project, and Discoveries
A project attempting to download the entire 2B2T world with a staggering 1.0e12 blocks reveals the scale of Minecraft's procedurally generated terrain and the challenges of data capture. The effort yields insights into storage, chunking, and network transfer, along with surprising finds from the sprawling map. It underscores the tensions between archival ambition and practical limits.
一次企图以约 1 万亿方块规模下载 2B2T 世界的工程暴露了 Minecraft 程序化生成地形的规模与数据捕获的挑战。该工作带来了存储、分块和传输方面的见解，并在庞大地图中发现了有趣的结构与遗迹。它凸显了档案化野心与现实极限之间的张力。
(37) At least 25 Flock cameras have been destroyed in five states since April 2025
At least 25 Flock cameras have been destroyed across five states since April 2025. The incidents highlight tensions around public surveillance, civil liberties, and the politics of security infrastructure in urban areas. The pattern suggests that security deployments can become flashpoints in local policy debates.
自2025年4月以来，五个州至少有25台 Flock 摄像头被毁，反映出公共监控、公民自由与城市安全基础设施之间的紧张态势。此类事件模式表明安防部署可能成为地方政策辩论中的焦点。
(38) It is time to give up the dualism introduced by the debate on consciousness
Philosophical piece urging readers to drop the traditional mind–body dualism invoked by debates about consciousness. It argues for a monist or non-dual framework that treats consciousness as emergent from physical processes, challenging long-standing dichotomies.
本文呼吁放弃在意识辩论中常见的二元论思维，主张以更贴近物理过程的单一框架来理解意识。通过将意识视为物理过程的涌现，挑战长期存在的二分法。
(39) Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep
Semble is a code search tool designed for agents that uses far fewer tokens than grep, enabling more efficient code discovery in AI agent workflows. It demonstrates how smarter indexing can reduce prompt sizes.
Semble 是一个为代理优化的代码搜索工具，其所需 token 远少于 grep，提升了 AI 代理工作流中的代码发现效率。展示了更智能的索引如何减小提示尺寸。
(40) Enough with the AI FOMO, go slow-mo, says Domo CDO
Domo's Chief Data Officer urges a slower, more deliberate approach to AI adoption, warning against AI hype and stressing data quality, governance, and measurable ROI.
Domo 首席数据官呼吁放慢对 AI 的追逐，抵制盲目炒作，强调数据质量、治理和可衡量投资回报的重要性。
(41) Learn Harness Engineering
Learn Harness Engineering is a new educational resource from Walking Labs, introducing Harness engineering concepts and practical tutorials. It provides a structured pathway for developers to explore the Harness framework.
Walking Labs 推出名为《学习 Harness 工程》的新教育资源，系统介绍 Harness 的工程概念与实用教程。它为开发者提供了探索该框架的清晰学习路径。
(42) Multiple commencement speakers booed for AI comments during graduation speeches
Several commencement speakers faced boos after bringing up AI, highlighting a divide over AI's role in society and education. The reactions illustrate how the AI conversation can polarize audiences, even in celebratory settings, as institutions weigh automation's implications.
多名毕业典礼发言人在提及人工智能后遭遇观众嘘声，凸显社会与教育领域对 AI 角色的分歧。此类反应显示人工智能话题已成为公众关注的焦点，即使在庆祝场合也会引发两极化讨论。
(43) Voice AI Systems Are Vulnerable to Hidden Audio Attacks
Researchers show that voice AI systems can be manipulated by hidden audio attacks—commands played at inaudible frequencies. The vulnerability matters for consumer devices and enterprise systems alike, highlighting a need for robust microphone filtering, anomaly detection, and adversarial audio defenses.
研究人员表明，语音 AI 系统可能被听不见的隐藏音频攻击所利用，发出不可听见的指令即可执行。此漏洞对消费级设备和企业系统都具有现实意义，强调需要加强麦克风滤波、异常检测以及对抗性音频防护。
(44) Two EA-18 fighter jets collide at Mountain Home airshow, pilots ejected safely
Two EA-18 fighter jets collided during an air show at Mountain Home AFB, with both pilots ejecting safely. The incident will trigger an official investigation and could affect future safety procedures at civilian-military air shows.
在 Mountain Home 空军基地的航空表演中，两架 EA-18 战斗机相撞，飞行员均安全弹出。事件将启动官方调查，可能影响未来军民航空表演的安全措施。
(45) WriteUp: 16 Bytes of x86 that turn Matrix rain into sound
A write-up demonstrates how 16 bytes of x86 code can synthesize audible sound from the Matrix rain visualization, illustrating extreme code-golf techniques and low-level audio synthesis. The result highlights how tiny snippets can produce multimedia.
这篇写作展示了用仅仅16字节的 x86 代码就能把矩阵雨可转成声音，体现了极简编码与多媒体合成的极限。对于低级编程与代码高球爱好者而言，这是一个有趣的演示，展示了紧凑代码如何实现音频合成。
(46) Prolog Coding Horror
Prolog has a reputation for a steep learning curve and quirky behavior that makes debugging a headache. This piece highlights Prolog's backtracking and logic-first mindset, arguing many developers find it a painful fit for real-world projects.
本文直陈 Prolog 的学习曲线陡峭、行为怪异且回溯机制复杂，调试常常令人头疼。文章聚焦其逻辑驱动的工作方式，认为在实际项目中，许多开发者会因此感到深刻的“编码恐惧”。
(47) Meta deletes popular 1M follower account after Kuwaiti request
Meta deleted a popular account with about a million followers following a Kuwaiti government request, illustrating platform compliance with state demands. The move spotlights censorship, geopolitics, and platform accountability in a global social web.
Meta 在科威特政府请求后删除了一个拥有约一百万粉丝的热门账号，凸显平台对政府要求的合规性。这一举动引发对全球社交网络中文章审查、地缘政治与平台问责的讨论。
(48) VoIP brings back old-fashioned pay phones to rural Vermont (2025)
A pilot uses VoIP-based payphones to fill gaps in rural Vermont where mobile coverage is spotty. Internet-connected kiosks offer emergency calling and everyday use, reviving a classic public amenity with modern networking.
VoIP 技术在佛蒙特州农村重新部署电话亭，弥补移动覆盖薄弱带来的通信盲区。借助互联网连接，这些亭子既能提供紧急呼叫，也能完成日常通话，展示用现代网络改造旧有基础设施的尝试。
(49) Most Americans don't trust AI – or the people in charge of it (2025)
Pew/Gallup data show broad American distrust of AI and the people shaping it in 2025. The sentiment spans technology, governance, and corporate motives, posing legitimacy challenges for AI initiatives.
皮尤/盖洛普数据表明，2025 年多数美国人对 AI 及其掌控者缺乏信任，涉及科技、治理与企业动机，给 AI 项目带来公信力挑战。
(50) EU weighs restricting use of US cloud platforms to process government data
EU policymakers weigh restricting US cloud platforms for processing government data due to sovereignty and security concerns. The move could spur local cloud alternatives and reshape global cloud dynamics.
欧盟在主权与安全担忧驱动下，权衡限制美国云平台处理政府数据。此举可能推动本地云方案的兴起并重塑全球云格局。