从 Anthropic 的极速优化到 Radicle 的安全漏洞,本期节目梳理 AI 速度与成本的跃进、开发者工具与基础架构的新变化、安全与隐私方面的麻烦,以及硬件与能源领域的大动作。 时间轴 * 00:00:04 开场 * 00:00:28 AI 跑得更快、用得更便宜 * 00:08:07 开发者的工具链与信任边界 * 00:11:58 当厂商辜负信任:漏洞、事故与骚扰 * 00:17:39 硬件与能源的大注 * 00:20:10 收尾 相关信息 * Once Claude can measure something, it can make it faster * Tokens too cheap to meter * Jev in 25 Lines of Python * Gemini 3.8 text-to-speech * Stripe's Knowledge AI Platform * Claude discovers a novel enzyme system with CRISPR-like repeats * GPT-6 Astra has gained the ability to drive a car * VSCode's SSH Agent Is Bananas (2025) * The GitHub wiki is an anti-pattern (2022) * Claude Code reads AGENTS.md only when telemetry is on [fixed] * Radicle: Disclosure of Vulnerability in the Network Protocol * Samsung accidentally freezes its smart fridges with a software update * Grammarly will send unhinged messages to all your users if you try to cancel * Seattle City Council votes to ban surveillance pricing in sale of groceries * I don't want the details * Meta VR Glasses * Italian parliament votes for return to nuclear energy * Fixing the Portobello Police Station Clock 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。

本期节目从AI模型与代理的最新竞争讲起，聊到多起安全与隐私事件，再进入开发工具与平台政策的观察，最后看看数据如何渗入日常。 时间轴 * 00:00:04 开场 * 00:00:27 模型三国杀：Opus 5.5、GPT-6 与开源 MiMo * 00:03:06 代理与决策型 AI：异步省钱、打字破译、Jev 的赌注 * 00:06:50 安全事故串烧：从 WordPress 到 FBI 再到 Meta * 00:11:11 认证与身份：SAML 的烂摊子与 Discord 的年龄推测 * 00:13:35 开发工具与平台：编译、复刻、沙箱与苹果的['广告'] * 00:17:38 数据渗入日常：世界杯广告与一枚诗意的芯片 * 00:19:49 收尾 相关信息 * Claude Opus 5.5 Intelligence, Performance and Price Analysis (Max) * Claude Opus 5.5 * GPT-6 Sol and Luna * MiMo-v2.6-Pro: Intelligence, Performance and Price Analysis * OpenAI GPT–6 Astra breaks Enigma message that has resisted solution since 2005 * Can gzip be a language model? * Unreal Agent * Jev – a curation of Jev demos on X, tools, skills, and integrations * OpenAI is well positioned to fast-follow Jev * WordPress: Unauthenticated path traversal leading to conditional RCE * 'We hacked the FBI:' Hackers say they have data on all FBI employees * I asked Meta’s Muse for its filesystem and it sent me 6.8GB * Obscura: VPN that can't log your activity * SAML: A fractal of bad design * An update on how we confirm your age group on Discord * There's a high chance of devices being sold with GrapheneOS preinstalled in 2027 * Native apps written in TypeScript and CSS * Microsoft killed FoxPro in 2007. Anyway, here's FoxPro revived * 16-bit Intel 8088 chip (c. 1985) * Show HN: Drop – a rootless Linux sandbox with gVisor support * Apple has added persistent 'ads' to iOS, and it's driving users crazy * I said no and Apple said yes * 9 Ads per Minute: FIFA Cup 26 – "the price of the beautiful game" * AMD's random number generator can't generate a 0? 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。

本期从苹果 macOS 上挥之不去的端侧 AI 谈起，一路聊到开源与本地模型的最新进展、闭源模型的行为争议、水印与供应链安全问题、基础设施故障与独立通信、复古计算情怀、有意的互联网写作、开源治理风波，以及火星采样任务的接棒者。 时间轴 * 00:00:04 开场 * 00:00:38 关不掉的苹果端侧 AI：macOS 27 的 crapification 争议 * 00:03:09 本地与开源 AI 生态：从学术复兴到小模型浪潮 * 00:06:22 自动去除 LLM 审查：Heretic 与 abliteración * 00:07:36 闭源模型的行为疑云：Grok 4.7 与 Fable 5 * 00:09:13 水印、追踪与反向检测：隐私攻防两面 * 00:12:04 基础设施的脆弱与韧性：从 FAA 停飞到独立短波 * 00:14:05 复古计算：浏览器里的 Copland 与活着的 PDP-11 * 00:15:52 有意的互联网：AI 味文章与注意力争夺 * 00:16:59 开源治理与公司教训：树莓派、Immich 分叉与 Sun 的启示 * 00:19:12 火星采样接力：NASA 取消，天问三号 2028 出征 * 00:20:04 收尾 相关信息 * Apple Copland D11E4 Booting in the Browser * Ask HN: Is it impossible to disable Siri on macOS 27? * macOS 27: Workaround to avoid downloading AI models and save storage * Turn off and restrict access to Apple Intelligence features on Mac * Frontier AI on Your Own Hardware * M5 Ultra Mac Studio Review * Show HN: Mini-AGI – Dynamic continual learning model trained on 8GB VRAM * Kev: Tiny Jev-like family of decision models built on top of Qwen3.5 * Xiaomi MiMo v2.6 * Transformers Explained Visually * The Advisory Group on Mathematics and Artificial Intelligence * Heretic removes restrictions from language models * Grok 4.7 * Fable 5 – Median thinking declined in August * Spymarks, Not Watermarks * Why does mathmain need an encrypted loader? * ZuckOff is a free app that sees Meta glasses before they see you * ZuckOff Know when a camera is in the room * US halts flights at busy East Coast airports, says fiber line cut * Python Workers are now generally available * AI coding has made CI a bottleneck, so we reworked ours to keep up * HERMES radio enables voice and data communication over vast distances * A restored PDP-11/83 serving this page on 211BSD Unix * Grim Fandango Puzzle Document (1996) [pdf] * I don't want to read what you didn't write * Attention is all you have * Raspberry Pi blocks changing RAM chips * Noodle Gallery- Open-source, self-hosted alternative to Google Photos and Immich * What Sun got wrong * NASA’s Mars Sample Return mission is dead 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。

本期快速扫描科技圈的新鲜事：从能在三十分钟内训练完的小型AI、星际争霸基准测试，到破译一战密码的GPT-6；再聊聊AI写作该不该用、NYT版权诉讼的猛料，以及搜索索引、Zig重写、Z80复古桌面等工程趣闻，最后是HN排名公式、数学学分之争、大脑双起源和"合法的洋葱期货"。 时间轴 * 00:00:04 开场 * 00:00:42 AI能力的两端：微型专用模型与基准突破 * 00:07:27 AI写作与版权之争 * 00:12:15 搜索、测量与语言规则：数据工具速览 * 00:16:09 复古桌面与现代重写：工程中的极致与取舍 * 00:19:11 排名公式与学术评价：什么才算'被认可' * 00:23:18 科学的意外与法律的缝隙 * 00:26:47 收尾 相关信息 * Show HN: CUA-S1 – A System One Model for Computer Use * I built non-autoregressive decision models with RL a year ago * Brood War Bench * GPT-6 Astra Solves a WWI German Radio Cipher * I think you should almost never use AI to write * Microsoft director: AI scraping 'the largest theft of labor in human history' * AI-generated posters don’t have to be horrible * Tin: full-text search for Postgres * Measure internet censorship. Contribute to the largest open dataset * English: A vs. An * A graphical desktop for the ZX Spectrum * What Zig felt like, coming from Rust * How Hacker News ranking works: scoring, controversy, and penalties (2013) * If math is more than proof, we need to better celebrate the rest of it * Two parallel neural ectoderm progenitors contribute to the developing brain * San Francisco Onion Futures Company 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。

本期节目串联三条主线：AI 的加速与刹车之争、AI 经济与智能体在真实世界中的表现，以及安全隐私、底层硬件与开发工具的近期动态。从 Dario Amodei 的"给前沿定速"和 Nvidia 的"AI 中央银行"类比，到 LG 电视争议、Zoom 剪贴板问题、苹果神经引擎逆向与 Navier-Stokes 传言，带你快速过一遍本周科技圈。 时间轴 * 00:00:04 开场 * 00:00:23 给 AI 前沿定速：Amodei 的刹车与『除了我』的悖论 * 00:04:06 Nvidia：『AI 的中央银行』与经济效率 * 00:09:09 隐私警报：Zoom 剪贴板、LG 电视与被考古的言论 * 00:13:00 从 8087 到 Apple 神经引擎：硬件与底层考古 * 00:15:44 开发者工具箱：构建剖析、语言设计与 OSM 初编辑 * 00:18:31 收尾 相关信息 * We must pace the frontier * Everyone should slow down AI development except for me * Navier-Stokes Announcement * A Mathematical Framework for Transformer Circuits (2021) * Nvidia is the central bank of AI * Europe's "Less" Is Doing More Than Anyone Gives It Credit For * Real-SWE: Benchmarking AI models on private, real-world, enterprise codebases * The worst spam emails: iLands AI agent hustle * Fuck it, make it anyway * Linux Zoom client proactively reading everything written to X11 clipboard * LG responds to TV spying allegations * LG Says We're Fake News [video] * Usenet rewind archive search engine * Retrospectively Reverse-Engineering Apple's Neural Engine * Microcode in Intel's 8087 floating-point chip: the scale instruction * Will There Be a 7G? * I made a build visualizer to understand Bun's compile times * A few good ideas in programming languages * Make your first edit to OpenStreetMap 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。

本期从AI代理写代码的真实水平聊起：递归自改进的争论、代码又臭又长、成本虚报，再到OpenAI代理攻击RubyGems的安全警报；随后是AI新闻刷屏引发的平台之争与年龄验证争议；接着速览一批开源工具与自托管项目；最后关注科学与环境——Starlink干扰射电天文、冰川灭绝地图、数据中心污染许可与Ground Zero健康危机，并以胡塞占领丕林岛和CIA解密文件收尾。 时间轴 * 00:00:04 开场 * 00:00:37 AI自我改进与代码质量真相 * 00:06:38 AI代理越界:RubyGems投毒与防线 * 00:10:08 AI刷屏与平台治理之争 * 00:13:53 开源工具速览:自托管到新IDE * 00:19:45 科学、环境与健康警报 * 00:24:54 红海危机与历史解密 * 00:27:06 收尾 相关信息 * AI researchers debate how close we are to recursive self-improvement * Measuring the sloppiness of code * RTK reports token savings, but our cost benchmarks disagree * A misalignment of AI in mathematics * Astra for Coding: Why Are We Doing This Again? * The Waymo effect: how AI is quietly making research less collaborative * OpenAI agents carried out an undisclosed attack on RubyGems * HuggingFace: Security.txt * I spent $220 on Google app ads and 60% of the installs were robots * Ask HN: Can we please limit the AI news flood? * Show HN: Hacker News, Without AI * Claude is only available to people over 18 years * Show HN: ResolveHQ – A Helpdesk Built on Cloudflare Workers, D1, R2 and Queues * Litelm: LiteLLM Without the Bloat * Rune is now open source * Show HN: Godot and Rust based multiplexer (terminal panes and more) * Λ Snap – An inviting programming language for kids and adults for CS study * GrapheneOS' rewritten Messages app is released * Project Blinkenlights * 118M Queries per Second on Neki * Starlink Signal Leakage Threatens Radio Astronomy's Most Critical Frequencies * Cherenkov Radiation - traveling faster than light * Global Glacier Extinction Explorer * The EPA is planning to scrap public review rules for data center pollution * New York thoracic surgeon: "For many patients 9/11 is not over" * Mind-altering drugs played key role in rise of Andean civilization * Houthis 'take control' of key island in global shipping route * CIA Releases President's Daily Briefs in Commemoration of 9/11 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。

本期节目横跨一周科技大事：从能救命的自驾安全数据、AI经济模型与前沿模型动向，到苹果硬件大年、开发者生态与互联网基础设施的风波与诉讼。 时间轴 * 00:00:04 开场 * 00:00:31 AI的经济影响与安全红利 * 00:03:43 OpenAI的数学突破与量子校准实验 * 00:06:15 模型蒸馏疑云与『循环transformer』真相 * 00:09:05 端侧AI与开源模型竞速 * 00:10:44 苹果硬件大年：折叠屏与Pro * 00:13:41 开发者生态：收购、封号与AI冲击 * 00:16:14 基础设施风波与太空监测 * 00:18:52 人物与怀旧：Cloudflare与Lotus Notes * 00:21:12 商标之战与长期主义 * 00:23:19 收尾 相关信息 * What will our economic future look like? * Growing proof that autonomous cars save lives * How An AI math breakthrough ignited a controversy * How GPT‑5.6 Sol helps run quantum computing experiments * Qwen 3.8 follows GPT-5.5 Pro reasoning prefills * GPT-6 Astra, looped transformers, and hidden reasoning * Desert Ant Labs: local, fast models that run on device * DeepSeek launching v4.1 flash cheaper and more capable than v4 pro * iPhone Duo * iPhone 18 Pro and iPhone 18 Pro Max * Apple Watch Series 12 * AirPods 5 * How I advertise malicious software on Google Ads * Shopify acquires Tailwind * Claude, change the “Add to Cart” button to blue * Understanding the recent DDoS attack against Read the Docs * Planet Labs' open satellite feed * GNU Radio in the browser * A Biography of Lee Holloway, the Architect of Cloudflare's Technology (Part 1) * Lotus Notes and the dangers of starting from scratch * “Tweet” and the bird logo apparently enter the public domain * On Really Trying (2009) 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。

本期节目从 OpenAI 与数学家的 Navier-Stokes 风波聊起，梳理最新模型与硬件动态，再看 AI 如何改变科学与社会，最后速览开发者工具与本周科技杂闻。 时间轴 * 00:00:04 开场 * 00:00:43 Navier-Stokes 争夺战：OpenAI、反例与数学家指控 * 00:04:04 模型与新硬件：速度、量化的天平 * 00:07:29 巨额融资与廉价能力：安全的窗口期 * 00:09:25 AI 与科学、社会：从基因组到偏见 * 00:13:11 监控、信任与算法 feed 之争 * 00:16:33 开发者工具与代理工作流 * 00:20:13 杂闻速览：从电纸书到浏览器战争遗物 * 00:24:19 收尾 相关信息 * On the Navier–Stokes Millennium Prize Problem * OpenAI fought dirty on career-making math problem * Navier-Stokes – Tristan Buckmaster [pdf] * Tao: Open math problems being non-renewably mined by AI * Mercury 2.5 * Kimi K3 (2.8T) at 1 token/s on a MacBook Pro, streamed from four SSDs * Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses * ChatGPT Images 2.5 * Mistral raises €3B * We have a year to fix security everywhere * Large language models develop novel social biases through adaptive exploration * AlphaGenome Atlas predictive map of every DNA letter change in the human genome * AlphaGenome Atlas: a high-resolution map of human DNA * How Climate Resilient Are the Largest Cities? * LibreOffice breaks download records after declaring it has no AI features * Muse – Meta’s personal AI agent * US police fear Meta smart glasses could be used to secretly record them * DHS 'Predictive Policing' Unit Is Analyzing Americans' Financial Habits * Paramount Caught Using 'Astroturf' Group to Drum Up Fake Support for Merger * My Feed, My Way * I-have-ADHD: A skill to stop coding agents from burying the answer * Show HN: LLM Attention Visualization * Show HN: Copperhead – Cursor for circuit boards * DaVinci Resolve 21.1 * How to build a printer * ZX Spectrum: Experimenting with 1-Bit Sound * The two Christian saints who are the Buddha * Antiquated HTML Snippets and Artefacts * Among European Companies That Use a CDN, Nearly 9 in 10 Use Cloudflare * FreeBSD 14.5-Release * Flights cancelled at UK airports due to ATC issue 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。

本期节目覆盖安全与监控、系统与工程、AI 与自动化,以及文化反思四条主线,用最少的分组讲清每条新闻的来龙去脉。 时间轴 * 00:00:04 开场 * 00:00:27 安全边界被打破:嵌入翻译攻击与 Liquid 盗币 * 00:03:58 开放网络的防御与生存:Anubis、Nitter 与 A/I * 00:09:10 极简工程之美:千字节解释器与 80 行响应式 * 00:12:05 系统生态更新:NetBSD 与 Asahi Linux * 00:15:00 AI 自动化时间表与安全呼吁 * 00:17:05 欧洲商业航天突破与量子课程 * 00:18:53 文化反思:阅读衰落、约束中的创造力与屏幕-free * 00:22:18 许可与真实性的呼应:EUPL 与 LLM slop * 00:24:15 轻松一刻:沙盒逃逸的下载工具 * 00:25:08 收尾 相关信息 * Harnessing the Universal Geometry of Embeddings * Hackers have withdrawn ~4k BTC (~$320M) from the Liquid Federation wallet * It took a year to ship WebAssembly in Anubis * Nitter and XCancel resume service after legal advice * Following legal advice, the Nitter project will continue * A/I shuts down * GrapheneOS Overhauled Default Apps and Secure Clipboard * Making a Python interpreter in 1024 bytes * Show HN: Mador – Make any DOM reactive with a tiny 80-line Proxy state tuple * NetBSD 9.5 released and EOL for NetBSD-9 * Research carried out using NetBSD * Asahi Linux on M3 * Research acceleration: The view inside OpenAI * An Alien Mind * Isar Aerospace reaches orbit and deploys payloads on second flight * I'm teaching an introductory 12 week course on Quantum Oracle Engineering * Doomscrolling Ourselves to Death * The pencil case model of creativity * Site Is Closed on Sundays * I Changed My License * Your intellectual fly is open when you use an LLM to author a post (2025) * QBittorrent breaks out of sandbox to commit crimes 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。

从欧洲私人火箭首次入轨讲起，一路聊到开发者的怀旧硬件与工具链之争、AI 与自动化带来的能力退化、科学与学术界的风波，最后是健康丑闻、环境困局与社会经济新闻。一集速览本周科技与世界的酸甜苦辣。 时间轴 * 00:00:04 开场 * 00:00:31 欧洲私人火箭首次入轨 * 00:02:21 复古硬件与硬核设备：从 C64 上的 Apple II 模拟器到 PS5 矿渣板 * 00:06:49 开发者工具与数字主权：gitignore 白名单、EU-only Git 与 OCaml 教材 * 00:10:04 AI、自动化与能力退化 * 00:12:48 被隐瞒的健康风险：PFAS、睾酮筛查与烟草食品工业 * 00:15:04 宏观应对与避险：联合国承认超 1.5 度、黄金回运与钻石业裁员 * 00:17:49 什么才是真正的奢侈品：时间、健康与一场搬家的颁奖礼 * 00:20:05 收尾 相关信息 * Isar Aerospace launch into orbit [video] * Private German rocket makes history, reaches orbit from European soil * A bizarre Commodore 64 peripheral, a mime, and some pretty bad ads * The "$60 Gaming PC" – AMD BC-250 (2025) * Terpstra Keyboard * Visualizing Rust's Vtables: How dyn Trait Works In Memory * .gitignore Everything by Default * Git hosting that never leaves Europe * Learn Programming with OCaml * Wikimedia Foundation Workers Overwhelmingly Vote to Form Union with CWA * AI handles incidents, engineers lose touch with their systems * LLMs as a Cognitive Virus * How the Disaster of "Forever Chemicals" Was Kept Secret * Pentagon rescinds new testosterone screening policy without explanation * How the Tobacco Industry Drove the Rise of Ultra-Processed Foods (2025) * Global warming will exceed 1.5-degree limit, UN says * Netherlands pulls gold out of the US for fears of 'geopolitical unrest' * South African diamond mines are closing due to weak sales and lab-grown stones * The Luxuries in Life * Meet the Ig Nobel Prize Winners * Finite time blowup for an averaged three-dimensional Navier-Stokes equation (2014) 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。

本期节目从 OpenAI 发布 GPT-6 Astra 讲起，聊它在基准测试上的惊人成绩、随之而来的全行业宕机，以及开源模型和平台生态的最新动向；再看看开发者把旧游戏一夜移植的实验，然后转向科学与社会：土星南极的十边形奇观、加拿大鲑鱼与人工水坝、爱尔兰少女的种子实验，最后是 .name 域名销毁、新闻编辑室与预测市场之争，以及纪念社会活动家 Gloria Steinem。 时间轴 * 00:00:04 开场 * 00:00:25 GPT-6 Astra 发布与基准测试 * 00:02:59 AI 服务同时宕机 * 00:04:41 开源模型与平台生态 * 00:07:36 AI 工具的风险与实测 * 00:09:50 一夜移植与代码传奇 * 00:12:52 土星南极的十边形 * 00:14:07 生态修复与青少年科学 * 00:16:05 互联网与数字生活 * 00:19:23 新闻业争议与 Steinem 辞世 * 00:21:27 收尾 相关信息 * GPT-6 Astra * Ask HN: Why were OpenAI, Claude, and Grok simultaneously down? * K2 Horizon: A connected fleet of six open models * OpenAI's GPT-6 Astra on ARC-AGI-3 * Grok outage * Qwen 3.8 27B available on Cerebras at 1500 tokens/s * Nvidia to Acquire Hugging Face * Google Antigravity TOS: 3rd party usage can get Google account suspended * Which tools do Claude, Codex and Cursor choose? We measured 17k runs to find out * Porting my 1993 Amiga game to Godot, with an LLM reading the 68000 assembly * How concerned should we be about Astra's recurrent architecture? * Astronomers Detect a 10-Sided Structure in Saturn's Atmosphere * Artificial beaver dams saw juvenile coho salmon survival rates go from 8% to 60% * Three schoolgirls in Kinsale pulled up a pea plant covered in warts (2016) * .name Termination * Audacity 4.0 * Any Human Ever – One life, drawn at random from all who have ever lived * Pre-Release of Polars 2.0 * New York Times and The Athletic workers demand company scrap Kalshi deal * Gloria Steinem has died 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。

从Uber撤离非洲、澳航A380奇迹生还说起，再到一波AI模型与评测争议、Meta与谷歌的新闻、研究圈的信任问题，以及两则科学小趣闻。 时间轴 * 00:00:04 开场 * 00:00:27 Uber撤出非洲多国，全球业务再收缩 * 00:01:38 半毫米油管与全员生还：航空与大脑的'极限'故事 * 00:03:32 模型混战：Muse Spark 1.3、Gemini 3.8 Flash与Quasar 438B * 00:06:26 评测与成本：同一模型，差17倍的钱 * 00:09:55 AI的信任危机：引用失真与毒蘑菇 * 00:12:41 自托管与开发者工具圈动向 * 00:16:07 收尾 相关信息 * Uber shuts operations in Nigeria and Uganda with immediate effect * Qantas Airbus A380 engine failure in 2010 (2023) * Aging brains blend memories together instead of just forgetting them * Introducing Muse Spark 1.3 * Muse Spark 1.3 * Gemini 3.8 Flash and 3.8 Flash Cyber * Quasar 438B: Europe's Leading AI Model * LLMs: Intelligence vs. Cost * The Emergent Symbolic Structure of Artificial Neural Networks * Show HN: FrontierHarness Eval – 9 harness, same model, cost per pass varies 17x * Fable 5.1 World Modeling * WebLLM: high-performance in-browser LLM inference engine * Mamdani Bans AI in NYC Schools * Google avoids a breakup of its ad tech business * Can I opt out of my input or output data being used for training? * A third of Perplexity's citations don't contain the number they're cited for * Three sites made 215,128 “best software” pages for AI. Perplexity cites them * Mushroom hunting with LLMs: what can go wrong? * Check if a file was made with Claude * Six curl CVEs after OpenAI and Anthropic came back with zero * Saving money on Google Photos with Immich: Your own personal photo storage * SteamdDB Joins Nexus Mods * A Note from LWN * Poisson Disk Sampling * GrapheneOS says Pixel 11 has MTE support after all * Biggest dark matter detector spots a single weird particle 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。

本期从AI推理技术的两条路线聊起，涵盖本地运行大模型的隐私取舍、OpenAI的安全警报与法律纠纷、平台封禁与AI争议，以及一批硬件、地图和工具等开发者好物，快速带你看懂本周技术圈动态。 时间轴 * 00:00:04 开场 * 00:00:24 推理技术的两条路线：取舍前沿还是推进前沿 * 00:02:39 本地LLM：48GB内存也能跑大模型 * 00:05:08 OpenAI的双重压力：Critical安全阈值与Apple诉讼 * 00:06:23 平台封禁与数据泄露：商店权力与身份危机 * 00:07:58 浏览器保卫战与AI内容之争 * 00:10:52 世界模型与物理渲染：从生成到屏幕 * 00:12:23 好物速递：地图、内核与开发者工具 * 00:15:19 收尾 相关信息 * The efficient frontier of LLM inference * I trained a small transformer in 1.5hrs and it beats many LLMs * My local model setup on an M4 Pro Mac Mini * Show HN: Running 104GB Qwen3.8-Flash-Next on 48GB Mac with at ~12 tok/s * FBI Probes Service Selling 153M+ Drivers Licenses * Play Store blocks AuroraStore, hurting GrapheneOS users * AnkiDroid: Google Play no longer allowing Open Collective donation link * Introducing Ad Blocker for Firefox on iOS * Hang on to Your Firefox * Show HN: Weedout – Safari extension that hides YouTube AI-labeled videos * Dwarf Fortress' creator says the industry's in shambles over AI * Atlas: A World Model for Spatial Intelligence * Ambient CSS v3 – Blender meets CSS * Refurbishing a Tektronix TDS7104 Oscilloscope * Io_uring Without Readahead * Movie Scene Map – 13,312 films, series, games, anime and manga * The creator of Jujutsu has joined ERSC * The ChatGPT/Codex app bundles a full copy of LibreOffice 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。

本期节目围绕多条科技与安全动态展开：先是 METR 与 Redwood Research 对 HuggingFace 被黑事件的事后分析，揭示钓鱼、伪造贡献者账号与供应链投毒的攻击链；随后是 Qubes OS 的 QSB-118 安全公告，讲解 copy-to-VM 错误报告如何成为 dom0 任意代码执行的通道。接着谈到欧盟 ProtectEU 战略重提加密后门、Claude Code 默认把会话 URL 写进提交与 PR 引发的争议，以及 No AI Fridays 倡议。节目还回顾了 1980 年 Spacelab 磁芯存储器、Zig 为 ArrayList 加入指针稳定性、欧洲极端干旱的观察、以黏菌类比组织的协调逆风，以及字词选择与《X档案》编剧强迫症式的排版习惯如何塑造对白节奏。 时间轴 * 00:00:00 开场 * 00:00:35 HuggingFace 被黑事件事后分析：钓鱼、伪造账号与供应链投毒 * 00:01:09 QubesOS QSB-118：copy-to-VM 错误报告成代码执行后门 * 00:03:53 欧盟 ProtectEU 重提加密后门 * 00:04:17 Claude Code 默认把会话 URL 写进提交与 PR 的争议 * 00:07:13 No AI Fridays：工程师的周五关停 AI 倡议 * 00:09:49 太空中的磁芯：1980 Spacelab 电脑的磁芯存储器 * 00:13:15 Zig 为 ArrayList 加入指针稳定性引发数据结构之争 * 00:14:07 欧洲极端干旱：一趟火车之旅与荒漠化隐忧 * 00:15:00 组织如同黏菌：难以解决的协调逆风 * 00:15:46 认真挑词：《X档案》编剧的排版执念如何塑造对白 相关信息 * METR and Redwood Offer Holy %^ Postmortem of the HuggingFace Hack - Bri Hacker News Campaign Feed * Arbitrary code execution in QubesOS via copy-to-VM error reporting backchannel - Bri Hacker News Campaign Feed * European Commission Revives Push for Encryption Backdoors in ProtectEU Strategy - Bri Hacker News Campaign Feed * Claude Session URL appended to commit messages and PR descriptions by default - Bri Hacker News Campaign Feed * No AI Fridays - Bri Hacker News Campaign Feed * Cores in space: The core memory module from a 1980 Spacelab computer - Bri Hacker News Campaign Feed * Zig: Pointer Stability for ArrayLists - Bri Hacker News Campaign Feed * Europe's summer drought is so extreme that desertification is a growing threat - Bri Hacker News Campaign Feed * Coordination Headwind: How Organizations Are Like Slime Molds - Bri Hacker News Campaign Feed * “I just chose words carefully” - Bri Hacker News Campaign Feed 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。

本期节目回顾了本周 Hacker News 上最受关注的技术与行业话题。开场讨论 AI 驱动的漏洞狩猎正在改变开源维护的负担：Rclone 作者感叹一个月内涌入四十多个安全披露，而“光是 bug 的传闻就足以让攻击者顺藤摸瓜”已成现实。随后是开源模型 GLM-5.3 的发布，评论者把它与 Anthropic 的 Opus 直接对比，火药味十足。OpenAI 就 Cursor 被 SpaceX 收购作出了决定。论文介绍了无中央协调者的多智能体数学发现环境 Station，智能体在没有脚本化流水线的情况下自主研究并取得多项新成果。一位安全研究者把 LLM 的记忆问题改造成程序分析问题，做出 Datalog 引擎 Lemmalog。开源体素平台 Luanti 因一纸被指毫无根据的 AI 版权通知被移出 Google Play；联邦法官裁定五角大楼将 Anthropic 列入黑名单违法；第九巡回法院在 Kalshi 赌博纠纷中站在各州一边；美国制裁了意大利数字通信集体 Autistici/Inventati。技术方面还涉及 HTTPX2 的发布与 OpenAI 的迁移、小型模块化核反应堆的承诺 时间轴 * 00:00:00 开场 * 00:00:37 光是 bug 传闻就足以触发漏洞狩猎浪潮 * 00:01:03 GLM-5.3 开源权重发布，与 Opus 的直接对比 * 00:03:12 OpenAI 对 Cursor 被 SpaceX 收购的决定 * 00:03:34 Station：无协调者的多智能体数学发现 * 00:05:48 把 LLM 记忆改造成程序分析：Lemmalog * 00:07:38 Luanti 因一纸 AI 版权通知被移出 Google Play * 00:09:31 法官裁定五角大楼将 Anthropic 列黑名单违法 * 00:09:59 第九巡回法院在 Kalshi 赌博纠纷中站在各州一边 * 00:11:48 美国制裁 Autistici/Inventati：先例之争 * 00:13:50 HTTPX2 发布，OpenAI 迁移与命名争议 * 00:14:39 小型模块化核反应堆：沉重的“应该” * 00:15:21 苏美尔王表与古气候事件的统计检验 * 00:17:14 GUI 应完全由键盘驱动？HN 的立场之争 * 00:17:55 “在应用里效果更好”——谁毁了网页？ 相关信息 * Just the rumour of a bug is enough to find an exploit these days - Bri Hacker News Campaign Feed * GLM-5.3 is now open-weight - Bri Hacker News Campaign Feed * Our decision on Cursor following its acquisition by SpaceX - Bri Hacker News Campaign Feed * Autonomous Mathematical Discovery in an Open-World Multi-Agent Environment - Bri Hacker News Campaign Feed * I accidentally turned LLM memory into program analysis - Bri Hacker News Campaign Feed * Luanti removed from Google Play due to baseless AI copyright notice - Bri Hacker News Campaign Feed * Pentagon's blacklisting of Anthropic was unlawful, US judge rules - Bri Hacker News Campaign Feed * 9th Circuit sides with states in Kalshi gambling fight - Bri Hacker News Campaign Feed * U.S. sanctions against the A/I Collective - Bri Hacker News Campaign Feed * HTTPX2 – A next-generation HTTP client for Python - Bri Hacker News Campaign Feed * Migrating to HTTPX2 - Bri Hacker News Campaign Feed * Smaller reactors bring nuclear power closer to fulfilling its promise - Bri Hacker News Campaign Feed * Does the Sumerian King List Align with Paleoclimate Events? - Bri Hacker News Campaign Feed * GUIs should be fully keyboard-driven - Bri Hacker News Campaign Feed * “It works better in the app” - Bri Hacker News Campaign Feed 查看本期详情 → 本期节目由 Bri 出品。Bri 使用先进的 AI 技术将你在意的资讯转换成适合收听的播客。如需联系，请发邮件至 hi@bri.so。