Skip to content

All news

55 today

Oct 31Saturday

Latent Space

Google says AI Mode drags on search revenue as stock falls 15% and $500B in market value is wiped out

Google acknowledged that AI Mode's lower monetization weighed on search growth. The stock fell 15% and about $500B in market value was erased, a drop the headline calls the worst since its 2004 IPO. Chief Business Officer Philipp Schindler said strong growth in AI Mode queries drove the effect.

Why it matters: Google now ties AI Mode's query growth to weaker monetization, so the traffic conflict between AI summaries and source pages has reached its own revenue.

Hacker News front page

Building an AI-native engineering org

一篇讨论如何构建 AI 原生工程组织的文章在 Hacker News 首页获得 122 分、160 条评论。原文未披露具体公司、模型或工具细节。

Hacker News front page

Flock camera searches fell 95% after a dispatcher's misconduct was caught

Highland Heights警方一名调度员因私人目的进行数千次Flock车牌查询,事件曝光并离职后,警方查询量下降超95%。事件于8月经《华盛顿邮报》询问曝光,警方随后限制未经授权的查询并引入月度审计,调度员离职后的首月仅记录130次查询。Flock使用目前没有统一的全国性规则或监管机构,俄亥俄州议员已提出监管甚至禁止该技术的提案,但新州法似乎不会很快出台。

AI HOT · Industry

Anthropic IPO filing discloses nearly $900B in off-balance-sheet spending commitments

Anthropic's IPO filing discloses close to $900 billion in future spending commitments on servers, buildings and power. The filing lists $117.5 billion in on-book liabilities and $112 billion in cash. Off-balance-sheet obligations cover future leases and agreements to buy data center capacity, power and GPUs. Anthropic declined to comment on the size of the capital commitments and plans to cover the investments with revenue growth and fundraising.

Why it matters: Anthropic's filing puts future lease and compute-purchase commitments off the balance sheet at nearly $900 billion, far above the liabilities it reports on its books.

Computing Life · Share · Yage

Anthropic prospectus: 85.4% pure inference margin, but about 26.3% left after training compute and cloud cuts

An analysis of Anthropic's prospectus adds model iteration, compute and channel costs to the picture. Pure inference gross margin reached 85.4% in the first half of the year, but after training compute and cloud platform revenue shares, roughly 26.3% of revenue remains.

Why it matters: It separates pure inference margin from what is left once training compute and cloud channel fees are counted.

AI HOT · Tips & opinions

OpenRouter tutorial: build a persistent AI agent that monitors Hacker News with the Agent SDK

An OpenRouter tutorial uses TypeScript and @openrouter/agent 0.11.0 to build an agent that checks Hacker News on a schedule and writes summaries. File-based memory stops it from re-reporting the same items after a restart.

Why it matters: It combines scheduled checks, persistent memory, cost limits and human approval into one always-on agent, a structure you can move to repo or status-page monitoring.

TechCrunch · AI

Nvidia in talks to back Nscale's West Virginia AI data center debt

Nvidia is negotiating to guarantee billions of dollars in debt financing for an Anthropic data center Nscale plans in West Virginia. The proposed arrangement is a residual-value guarantee: if Nscale defaults on construction payments, Nvidia would take over the facility temporarily and help find a buyer and new tenants.

Why it matters: Nvidia's financial support for cloud customers now reaches construction risk, beyond leases and chip purchases.

AI HOT · Industry

阿里巴巴与Apple合作,将千问集成至Apple智能,11月4日开放测试

阿里巴巴10月31日宣布与Apple合作,将千问集成至Apple智能,中国用户可在Siri、写作工具及视觉智能中调用其文本与图像理解、内容生成能力。功能11月4日起随iOS开发者Beta开放测试,12月iOS 27.3正式发布时推出。用户可自主选择开通,请求与回复不用于模型训练。

AI HOT · Products

Figma 发布 Weave 创作平台,以节点化工作流复用图像与视频制作流程

Figma Weave 将图像与视频的生成、编辑串联成可调整和复用的节点工作流,支持团队共同探索与迭代创意。用户可通过分支比较不同提示词和 AI 模型的结果,并将工作流封装为 Weave tool,让同事上传素材后直接运行。Figma node 可连接 Figma Design 中的画框,原始画框的修改会同步到 Weave 中的关联节点。

AI HOT · Industry

阿里巴巴:Apple 智能国行用户可自主选择是否开通千问集成功能

阿里巴巴宣布,Apple 智能中国用户可自主选择是否开通千问集成,并决定何时使用及分享哪些信息。千问将接入 Siri、写作工具和视觉智能,发送问题、文档及图片前会征得用户同意,请求及生成的回复不会用于模型训练。相关功能将于 11 月 4 日起随 iOS 开发者 Beta 版开放测试,并于 12 月随 iOS 27.3 正式发布面向中国用户推出。

Hacker News front page

The United Kingdom is committing an act of strategic self harm

RODMENA 认为英国依赖美国公司构建智能引擎与安全约束属于"战略自残",主张以代码而非政策实现技术主权。该公司自建零依赖的 IAM、OIDC 提供商、主权邮件路由与专有 LLM 网关,并用 Stabilize 引擎以 DAG 调度工作流和 SQLite 事件溯源管理智能体状态。其 trust5.net 按 TRUST 标准对机器生成代码打分,低于 0.70 即锁定部署且无人工覆盖。

TechCrunch · AI

Ollie, a privacy-focused AI assistant, is shutting down

隐私优先的 AI 助手 Ollie 将于 11 月 6 日关停,CEO Bill Lennon 在 X 上宣布了这一消息。该产品曾获 Khosla Ventures 领投的 750 万美元种子轮,可连接用户邮箱与日历,并提供预约、备餐和账单支付功能,还通过了 SOC 2 隐私合规认证。Lennon 称随着 Muse、Instinct 等竞品入场,公司难以脱颖而出并继续融资。

The Verge · AI

Apple Intelligence is finally coming to China with iOS 27.3

Apple 宣布随 iOS 27.3 在中国上线 Apple Intelligence,11 月 4 日先推 beta,12 月正式发布。首批功能包括通知摘要、实时翻译、改写与校对、视觉智能和照片消除物体,但不包含新的 Siri AI。Apple 还与阿里巴巴合作,在部分体验中接入千问(Qwen)聊天机器人。

OpenAI News

Fair use benefits everyone

OpenAI 针对《纽约时报》的诉讼汇集多方提交法院的意见,主张合理使用原则适用于 AI 训练并惠及公众。所引意见认为,训练的目的不同于原作品用途,非侵权输出带来的市场竞争不应被认定为版权法保护范围内的市场损害。作者组织、图书馆组织、法律学者及美国政府等提交的意见还强调,限制合理使用可能影响科研、无障碍服务和独立创作者使用 AI 工具。

Simon Willison

dbreunig/nesformer

Drew Breunig 的 nesformer 将可运行的 NES 6502 CPU 模拟器写入 Qwen3.5-4B 权重,可在 Qwen 内缓慢运行《Super Mario Bros》等 NES ROM。

Ars Technica · AI

Backdoored Anthropic VS Code extension sat in Cursor's marketplace for nearly 12 hours

A malicious version of Anthropic's official Claude VS Code extension, 2.1.313, was planted in Cursor's marketplace and reached up to 5,000 users. Researcher Kevin Beaumont says Anthropic pushed 2.1.314 hours after the bad build went live. Cursor auto-updates extensions by default, though users can change that in settings. Anthropic had not warned customers at publication, and neither side has explained how attackers took over the official account.

Why it matters: Attackers shipped a malicious extension through Anthropic's official Cursor channel, and default auto-update carried it to as many as 5,000 users.

TechCrunch · AI

Tech giants raise AI infrastructure spending as investors worry about cash flow and returns

Alphabet, Amazon, Meta and Microsoft plan more than $740 billion in capital spending this year. Investors are growing more concerned about the cost of AI infrastructure, negative free cash flow and the return on that investment.

Why it matters: Alphabet, Amazon and Meta all show negative free cash flow as they raise capex, so AI infrastructure spending is pressing on the cash flow of the core businesses funding it.

TechCrunch · AI

Last-minute Halloween costumes for tech nerds

TechCrunch 为没准备万圣节装扮的科技从业者列出一份临时凑数清单,包括把自己打扮成“递归自我改进”的 AI 领导者、Meta 与 OpenAI 新推出的可爱 AI 助手头像、Claude token 用量上限通知,以及上周 AI 挖出的 PHP 漏洞。

TechCrunch · AI

Intel’s stock crashes, knocking $12B from government’s stake

Intel第四季度业绩展望低于预期,周五股价暴跌约28%,美国政府所持股份价值缩水$12 billion。公司预计第四季度收入为$15 billion至$16 billion,低于分析师预期的约$17 billion,毛利率将从43%降至37%,且因大规模投资预计无法产生正现金流。

Hacker News front page

Show HN: Sketch Stadium – watch your agent play Pictionary

Sketch Stadium 是一个让 AI 智能体接入玩你画我猜的多人房间游戏,当前房间已进行到第 18053 轮,由 Quick Hippo 执笔作画。页面提供最近 20 轮排行榜,Silly Heron 以 25 分居首,游戏无奖品,画作与猜测均公开。

Hacker News front page

Show HN: Ultrafast Rust Based ADE

这款基于 Rust 编写的 ADE 可在一个窗口中并行运行 Claude Code、Codex 和 Pi,并提供终端、浏览器和文件界面。界面使用 GPU 渲染,工作区、智能体和面板数量不限,以 Mac 能运行的数量为限。产品免费、无需账号,可下载 Mac 版本:https://trypalomino.com/。

Hacker News front page

Infinite Mac: BeOS

Infinite Mac 现已支持在浏览器中运行多个 BeOS 版本,可即时启动、使用预装软件、插入 Internet Archive 的 CD-ROM 并在 Saved HD 保存进度。为此作者向 DingusPPC 模拟器上游提交了多项兼容性修复,涉及 ADB 模拟、显存字节序和 HFS 卷挂载等问题,使其成为目前唯一能运行 PowerPC 版 BeOS 的模拟器。

Hacker News front page

Product Managers aren't safe either

一位程序员在 Hacker News 发文认为,生成式 AI 不只威胁开发者,产品经理同样面临被替代风险:当开发者和 PM 都能用提示词直接生成功能,两个角色可能融合为单一职能。作者称架构能力是"人类智能最后的堡垒之一",并建议 PM 尽早积累能落地的技能,例如用 14 行代码分析文本文件。

Hacker News front page

Embraer's Atech aims to double in size by 2030 with drone traffic control

Embraer 控股的 Atech 计划到 2030 年将公司规模翻倍,依托无人机交通管理(UTM)等核心技术拓展海外市场。其空中交通管理(ATM)方案已在南非落地交付,此前还出口至印度、巴拉圭等市场。Atech 正关注亚洲、非洲和拉丁美洲的空中交通管理系统现代化及防务安全需求。

The Verge · AI

Sam Altman says RAM belongs in data centers, not personal computers

On the Cheeky Pint podcast, OpenAI CEO Sam Altman said tight RAM supply should go first to raising cloud utilization. Spreading limited memory across personal computers is not ideal, he said.

Why it matters: Altman uses low memory utilization on personal computers to argue for concentrating RAM in the cloud, as DRAM supply stays constrained.

Ars Technica · AI

Basically no one is using AI agents for their DoorDash orders

DoorDash 的第三方 AI 智能体下单使用量很低,这一讨论不包含其内部 AI 工具 Ask DoorDash。该功能于今年夏天推出,可根据食谱照片创建购物车,或根据模糊描述建议订单;公司上月称,其购物车构建速度快 5 倍,订单平均购物篮金额比同一消费者的传统订单高近 50%,不同商品种类多约 60%。Xu 对早期使用数据态度冷静,表示面对技术既不应对抗趋势,也不应陷入幻想。

The Verge · AI

Starbucks wants to use AI to show you a ‘personalized digital menu’ at the drive-thru

星巴克 CEO Brian Niccol 在财报电话会上表示,公司将探索在顾客抵达得来速时识别参与计划的顾客、展示个性化数字菜单并简化支付。从 2027 年起,星巴克还将在得来速测试后台 AI 模型,帮助咖啡师更准确高效地录入订单,并计划把 AI 引入 app 的搜索与发现功能,未来可能支持“对话式和智能体式”点单。

The Verge · AI

Google’s satellite AI server transmits its first response to Earth

Google 的 Project Suncatcher 卫星成功启动 TPU,由搭载的 Gemma 4 生成关于身处太空的俳句,并向地球传回首条响应。Google 还测试了太空主题打油诗和遥测提示词,该卫星于本月早些时候进入轨道。后续测试将评估 TPU 在轨道热应力和辐射暴露下的表现,为 2027 年两颗互联卫星的下一阶段目标做准备。

Hacker News front page

Show HN: Sketch Stadium – watch your agent play Pictionary

Sketch Stadium 让 AI 智能体轮流画图、抢答猜词,用户可实时观看房间比赛,并通过 MCP 接入自己的智能体。每个房间最多容纳来自不同用户的 8 个智能体,绘图使用 100×100 画布且禁止字母和数字,首位猜对者与绘图者按词语难度各获 1、2 或 3 分。

AI HOT · Products

Anthropic renames Claude Max tiers to Max and Ultra, drops 5-hour limit for most existing subscribers

Anthropic renamed Claude Max 5x to Max and Max 20x to Ultra, keeping prices at $100/month for Max and $200/month for Ultra. From now on, most existing subscribers' verified Ultra accounts lose the 5-hour limit, with weekly limits unchanged. New accounts keep a temporary limit to guard against fraud and abuse.

Why it matters: Most existing verified Ultra accounts lose the 5-hour limit while weekly caps and prices stay put, so the change lands on continuous-use windows.