脑卒中是「大脑的断供事故」——血管堵住（脑梗死）或破裂（脑出血），都以分钟计地杀死神经元。它是我国居民第一位的致残原因，而它的救治窗口比心梗更短：静脉溶栓只有 4.5 小时。本文讲清两类卒中的机理、识别口诀 BE FAST 的每一个字母、以及「输液冲血管」这类经典误区的真相。

Oct 15, 2026 Google Deepmind Key Points Gemini 4 Argon is Google's attempt at a single AI model that controls robots, ge…

Oct 2, 2026 Key Points Cloudflare has released Clef and Clef-flash, two decision models for AI agents that compete direc…

On an afternoon in Seoul in March 2016, I watched a program I helped build put a stone on the fifth line of a Go board i…

This is today's edition of* *The Download*,*our weekday newsletter that provides a daily dose of what's going on in the …

It’s been a busy few months for AI hype. At the end of April, Anthropic claimed that its model Claude Mythos is better a…

As companies hand off longer and more complex tasks to AI agents, they are running into an oversight problem: Agents can…

This session will explore how early-stage companies are building teams where humans and AI agents work alongside each ot…

Anthropic and OpenAI want to embed independent safety evaluators inside their AI labs. Researchers welcome the unprecede…

Anthropic’s public listing, which is expected this year at a valuation of as much as $2 trillion, is bound to be a windf…

Oct 23, 2026 Nano Banana Pro prompted by THE DECODER OpenAI CEO Sam Altman believes advanced AI can't be kept locked up.…

Reproducing a published 3DGS paper takes weeks of expert engineering before any extension can be proposed. Prior autonom…

Improving every agent in an LLM-based multi-agent system can be prohibitively costly, making it important to identify wh…

A new beginning for Windows PCs starts tomorrow with the launch of NVIDIA RTX Spark — one PC built for creators, gamers …

One of the core promises of AI is that the more it knows about a person, the better it will be able to assist them with …

Microsoft CEO Satya Nadella is the latest tech executive to offer lengthy thoughts on how AI safety might be improved. I…

Rather than downloading another app, a growing number of agents can simply be texted like an ordinary person. You text i…

Orkut, the once-popular social networking platform that shut down more than a decade ago, may finally be making a comeba…

Long-horizon coding agents need timely corrections, yet feedback can be ineffective or even harmful when it misjudges on…

Manuel Uth Oct 8, 2026 A 16-year-old had to be airlifted off a dangerous rock face on Crown Mountain near Vancouver afte…

The evaluation of coding agents generally strives to be as realistic as possible. In our study, we collect 4,782 agent s…

Google on Tuesday launched a new site that lets anyone verify whether a piece of media, be it an image, a video, or an a…

Sparse GNN training reduces computation, but deciding which edges to keep can be costly. Reusing one sparse graph is che…

As promised, President Donald Trump has announced the formation of a new Super Intelligence Force, which he said will be…

Many useful language-model tasks cannot be evaluated by exact outcome verification. Rubric-based reinforcement learning …

Manuel Uth Sep 29, 2026 Florida Attorney General James Uthmeier is asking a court to stop OpenAI from giving ChatGPT hum…

Backdoor attacks can be implanted in Large Language Models (LLMs) during training, causing unwanted behaviour when a tri…

Neural image watermarks can be forged by extracting watermark-bearing residuals from released images and transferring th…

Large-scale Digital Surface Models (DSMs) can be produced cost-effectively from satellite images via stereo-photogrammet…

Political discourse is widely perceived to be growing more hostile, yet robust evidence remains scarce. This study exami…

The group won't be given leeway to slow down or redirect OpenAI's ongoing mathematical research.]

Tabby is designed to be a real-time bookkeeping interface, handling clients’ paperwork as it gives them up-to-the-minute…

Language models can be adapted by changing the computations applied to individual tokens. Quantum circuits offer one suc…

Sparse on-policy distillation (OPD) allocates teacher supervision to a small subset of tokens in student-generated traje…

But the military's overall use of AI seems to be accelerating.]

If AI lab PrismML isn't on your radar yet, it should be.]

There may be a simpler and more effective fix for rogue agents, hiding in plain sight.]

OpenAI and Anthropic tell corporate customers their data won't be used for training. But when Anthropic said it would st…

As corporate AI adoption continues to grow, enterprise-grade LLM agents are being deployed into sensitive contexts such …

Safety guardrails in open-weight language models can be readily bypassed using Refusal Feature Ablation (RFA), a techniq…

A research idea may be novel, coherent, and scientifically plausible, yet its proposed method may remain insufficiently …

Large language models used for code editing can be trained and deployed in at least two output regimes: direct generatio…

Large language models learn similar behaviours, yet it remains unclear what structure they share or how to change one be…

2017 年 2 月 19 日，Uber 前工程师苏珊·福勒发布 2700 字博文，自述入职第一天遭经理性骚扰、向 HR 举报六次均被「他是高绩效员工」压下。博文引爆后一个月：CEO 卡兰尼克聘请前司法部长霍尔德独立调查；6 月报告给出 47 项整改建议、20 人被解雇、卡兰尼克先休假后被投资人联名信逼宫辞职。本系列第 24 篇复盘「内部人举报型危机」——它与网易案同属雇主品牌，却揭示了更深的机制：当价值观成为包庇高绩效者的工具，一篇克制而具体的个人叙述就足以完成一场公司政变的舆论铺垫。

2018 年 11 月，D&G 为上海大秀发布的《起筷吃饭》广告被批刻板化中国饮食文化，设计师 Instagram 辱骂言论曝光后，官方甩锅「账号被盗」；上海大秀当天取消，天猫京东等八大电商全面下架，两位创始人用中文说「对不起」——但文化冒犯无法像产品一样召回。亚太收入占比从 25% 缩至 22%，2021 年反弹 20% 仍低于事发前，回归之路走了四年尚未复原。本系列第 23 篇复盘「文化傲慢型舆情」：致敬与冒犯的判定权在接收方，而奢侈品的高替代性让消费者的钱包投票最为彻底。

「头晕」是门诊最常见的主诉之一，但「头晕」这个词里装着完全不同的病：天旋地转的眩晕、站不稳的失衡、眼前发黑的昏厥前感——它们的病灶分别在耳朵里的平衡器官、神经系统与心血管。绝大多数眩晕是良性且可手法治愈的（耳石症），而少数伴随神经症状的眩晕是脑干的重症警报。本文讲清三者的鉴别与该挂哪个科。

2018 年 10 月至 2019 年 3 月，波音 737 MAX 8 连续两起空难带走 346 条生命，祸根是 MCAS 系统依赖单一传感器且未对飞行员培训。埃航次日波音声明「对安全性充满信心」，三天后才被中国民航局率先停飞倒逼全球停飞；公司用「飞行员反应」甩锅，用四个月才承认 MCAS 在两起事故中均被激活。最终代价：全球停飞 20 个月、超 200 亿美元直接损失、与司法部 25 亿美元和解——罪名不是事故本身，而是「欺骗监管者」。本系列第十八篇复盘人命级安全危机：当公关的一切技巧失效后，剩下的唯一形态是说真话的时点与彻底的技术修复。

2019 年 4 月 11 日，马云在阿里巴巴内部交流中说「能做 996 是一种巨大的福气」，发言曝光后引爆全网——两个月前，程序员们刚在 GitHub 上发起 996.ICU 抵制运动，舆论的弹药早已装填完毕。马云三天内二次回应「不为 996 辩护，但向奋斗者致敬」，官媒密集定性涉嫌违反劳动法，两年后最高法与人社部以典型案例给出终局：996 严重违法。本系列第十六篇复盘「企业家价值观表态型舆情」：当表态的内容本身站在法律对立面，所有修辞都只是延长了交割期。

2017 年 7 月，东北大学毕业生李文星通过 Boss 直聘求职，被冒用公司名义的传销组织骗至天津静海，溺亡于一处水坑。8 月初《求职少年李文星之死》刷屏，公众追问的不是「谁杀了李文星」，而是「谁把传销的大门开在了招聘软件上」——CEO 三天内承认审核机制漏洞、网信办约谈后强制实名认证整改。本系列第十四篇复盘「平台责任型舆情」：当危机的源头不是失误而是商业模式本身，道歉只是入场，整改是一场增长机制的大手术。

AI hasn’t benefited the bottom line of any big corporations outside the tech industry, according to a note recently publ…