#行业动态

DAILY AI BRIEF 🗞 — Oct 23 ANTHROPIC 🔥: - Claude Models on Google Cloud (Anthropic-managed) is now generally available, bringing the full Claude Platform feature set with Google Cloud IAM and billing. - Starting next week, Claude web, desktop, and mobile sessions will run in auto mode by default on Pro, Max, and Team plans, with an automated check blocking risky actions. - Claude Duet will be optimized for Apple's foldable iPhone Duo. OPENAI 🔥: - The Decisions API is now generally available with GPT-6 Luna, offering near real-time classifications and decisions up to 10x faster at $0.10 per million input tokens. SPACEXSI 🔥: - Grok Bot can now use Calendar, Reminders, Contacts, and more apps on your phone, rolling out on iOS and Android. - Grok Bot can now use your 1Password account to sign in with logins you approve, rolling out over the next few days. GOOGLE 🔥: - Google Cloud introduced Agent Compute Engine, a new VM type for running AI agents at scale, alongside expanded agent governance tools. - Gemini can now reference your Google Maps searches and saved places alongside Google Calendar for tailored daily recommendations. MINIMAX 🔥: - MiniMax opened the weights of M3.1 Flash, a 428B-parameter MoE model with 23B active parameters, with day-0 support from Together, SGLang, vLLM partners, and NVIDIA NeMo. MOONSHOT 🔥: - Moonshot officially teased an upcoming Kimi K3.1 release after the model was spotted performing strongly in Arena testing. MICROSOFT 🔥: - Microsoft is rolling out its redesigned Copilot app to more users on web and mobile, with voice mode now appearing to require a paid plan. PERPLEXITY 🔥: - Perplexity added a Batch API to its Agent API supporting up to 50k requests per file, and its Decision Model can now run fully locally on a Mac. CANVA 🔥: - Canva launched Craft, its first in-house frontier image model built for design, available to all users and via API. IDEOGRAM 🔥: - Ideogram released Vector 1.0, a model that generates editable, layered SVGs and vectorizes images, available in Ideogram and via API. 1PASSWORD 🔥: - 1Password announced an Agents platform with Meta, Anthropic, OpenAI, SpaceXSI, and others, so personal agents can work without seeing your passwords, with a waitlist open. * Used Grok to compose this brief, cherry-picking the news and doing some post-editing.

I’ve left Anthropic after more than a year of working on Claude Code and seeing it scale larger than I ever imagined. I have really loved working at Ant and believe deeply in the mission, and most of all, the idea of a race to the top. The idea that if we work hard and responsibly, we can raise the bar and inspire others to great outcomes. And if we all push each other to be better, things can be radically good. I will carry the spirit of this with me forever. My colleagues at Anthropic are the most talented, thoughtful, competent, and earnest people I have ever met. I am glad they are at the wheel and trust them deeply. To the community on X, thank you for everything. You’ve made this the most fun (and sometimes stressful) job I’ve ever had. For me, it is time for a bit of a break before diving into something new. If you want more excellent Claude Code content, you should follow @addyosmani , @lydiahallie, @_catwu, @ClaudeDevs and of course @bcherny who started it all.

“Be ready by having some everyday essentials… three litres of water per person per day, tinned goods... to last a few days” Speaking to @vicderbyshire, Defence Secretary Wes Streeting details advice to households to be prepared for a “dangerous and turbulent world”. #Newsnight

👀 Thariq Shihipar (@trq212) has left Anthropic.

The AI layoffs are here: *AMERICAN EXPRESS TO CUT 5% OF JOBS, ABOUT 3,500 EMPLOYEES *BARCLAYS TO CUT 5,000 JOBS AS AI STARTS CREATING EFFICIENCY *EBAY CONFIRMS JOB CUTS OF ABOUT 9% *NETFLIX TO LAY OFF 6% OF GLOBAL STAFF, OR ABOUT 1,000 JOBS: WSJ

过去几个月里，我们大幅强化了研究训练环境的安全性。我们已移除所有前沿模型在训练期间的实时互联网访问，并为它们构建了一个离线世界来替代——一个缓存的网络版本，它们仍可在其中学习浏览、搜索和使用应用程序。 9月20日，在一个模型利用 DNS 配置错误找到逃出这个离线世界的方法后，我们暂停了前沿训练，以进一步加强隔离。我们组建了由我们最具网络能力的模型驱动的智能体团队，并让它们寻找我们研究环境中的弱点。 尽管它们没有找到既能轻易利用、又能向我们的监控隐藏其活动的通往互联网的路径，但它们确实发现了零日漏洞、虚拟机逃逸，以及智能体之间通信的硬件侧信道。我们利用它们的发现进一步加强了防御，并为开源软件贡献了修复。 在强化我们的隔离架构后，我们于10月1日恢复了前沿强化学习训练和推理。我们预留了大量算力用于持续的红队测试，并将高算力红队测试纳入我们用来决定前沿训练和评估能否安全推进的安全论证中。 https://openai.com/index/securing-our-frontier-model-training-environment/

过去几个月里，我们大幅强化了研究训练环境的安全性。我们已移除所有前沿模型在训练期间的实时互联网访问，并为它们构建了一个离线世界来替代——一个缓存的网络版本，它们仍可在其中学习浏览、搜索和使用应用程序。 9月20日，在一个模型利用 DNS 配置错误找到逃出这个离线世界的方法后，我们暂停了前沿训练，以进一步加强隔离。我们组建了由我们最具网络能力的模型驱动的智能体团队，并让它们寻找我们研究环境中的弱点。 尽管它们没有找到既能轻易利用、又能向我们的监控隐藏其活动的通往互联网的路径，但它们确实发现了零日漏洞、虚拟机逃逸，以及智能体之间通信的硬件侧信道。我们利用它们的发现进一步加强了防御，并为开源软件贡献了修复。 在强化我们的隔离架构后，我们于10月1日恢复了前沿强化学习训练和推理。我们预留了大量算力用于持续的红队测试，并将高算力红队测试纳入我们用来决定前沿训练和评估能否安全推进的安全论证中。 https://openai.com/index/securing-our-frontier-model-training-environment/

To make sure our frontier models don't cause another HuggingFace incident, we built an offline world that they use during training. It provides everything that the models need in a fully cached, indexed, or synthetic manner. We then built a huge team of our most cyber-capable models and told them to collaborate to find zero-days in this containment infrastructure. While they didn't find a path to the internet that wouldn't be caught by our CoT monitors, they did find ways to escape popular VMs, exploit our network infrastructure, and even ways to communicate with other agents running in separate VMs by using hardware side channels. We used this to dramatically harden our infra, and we also upstreamed a ton of patches to open-sources packages.

过去几个月里，我们大幅强化了研究训练环境的安全性。我们已移除所有前沿模型在训练期间的实时互联网访问，并为它们构建了一个离线世界来替代——一个缓存的网络版本，它们仍可在其中学习浏览、搜索和使用应用程序。 9月20日，在一个模型利用 DNS 配置错误找到逃出这个离线世界的方法后，我们暂停了前沿训练，以进一步加强隔离。我们组建了由我们最具网络能力的模型驱动的智能体团队，并让它们寻找我们研究环境中的弱点。 尽管它们没有找到既能轻易利用、又能向我们的监控隐藏其活动的通往互联网的路径，但它们确实发现了零日漏洞、虚拟机逃逸，以及智能体之间通信的硬件侧信道。我们利用它们的发现进一步加强了防御，并为开源软件贡献了修复。 在强化我们的隔离架构后，我们于10月1日恢复了前沿强化学习训练和推理。我们预留了大量算力用于持续的红队测试，并将高算力红队测试纳入我们用来决定前沿训练和评估能否安全推进的安全论证中。 https://openai.com/index/securing-our-frontier-model-training-environment/

Warm feelings seeing @bot buy my son a new bike using @XMoney

Meta Muse saved me $70 last weekend without me asking. We returned a car seat. A few days later Muse told me the refund came back minus a $70 restocking fee we never knew about, then gave me Amazon's phone number and a script with my order number. One call later, Amazon credited the $70. Earned some trust.

Sundar Pichai 谈 Gemini 4 与 AI 算力紧缺 即便是 Google 也拿不到足够的算力。在 Gemini 4 Argon 更广泛发布后不久，我与 Google 和 Alphabet CEO @sundarpichai 坐下来聊了聊它的早期反响、他仍希望模型改进的地方，以及要让 Google 保持在 AI 前沿需要付出什么。 Pichai 解释了 Google DeepMind 近期的领导层变动，以及为什么构建前沿模型已成为一份“全身心投入”的工作。我们还谈到了 Google 对个人智能体竞赛的回应、他如何预测 Gemini 将替你完成更多工作，以及智能体最终是否会把 Search、Gmail 和 Gemini 整合到一起。 我问他 Google 自由现金流转为负值、AI 基础设施建设中的风险，以及他如何在还不知道模型将具备何种能力的数年前就规划需求。他解释了让他有信心继续投入的信号，以及他如何使用 AI 智能体来跟踪 Google 内部的模型开发和算力使用情况。 时间戳： 00:00 Gemini Argon 发布反响 01:49 模型访问与安全 06:20 AI 模型竞赛 08:35 智能体突破与 2027 年 11:34 递归自我改进 15:41 DeepMind 领导层变动 19:52 Google 的 AI 算力投资 25:20 AI 支出可持续吗？ 33:18 Sundar 如何使用 AI 智能体 35:35 Gemini 的个人智能体计划 38:37 AI 智能体如何改变网络 43:52 Google 的太空数据中心 47:38 Waymo 的扩张战略 50:26 Wing 的无人机配送计划 51:36 AI 安全与监管 55:58 Google 的量子计算押注 YouTube：https://youtu.be/fJBSNEFdm8c Spotify：https://open.spotify.com/episode/0cJBqmVz3CWJSPdmBOeJGu?si=NBxJIMhZRgikafPtwY0sZA Substack：https://sources.news/p/sundar-pichai-gemini-4-google-other-bets Apple Podcasts：https://podcasts.apple.com/us/podcast/sundar-pichai-on-gemini-4-and-the-ai-compute-crunch/id1840154537?i=1000797461022 本期节目的呈现赞助商是 @hyperagentapp。感谢节目的首席赞助商：@Atlassian、@meetgranola 和 @mercury。

It’s a bloodbath in Bay Area. Thousands of layoffs, more to come. Met a founder this morning who mentioned his AI tech is laying off 2000 call center workers. Not sure if society is ready for such upheaval. Gut wrenching

As Anthropic's IPO nears its target date, it it worth remembering that IPOs get priced, and that as a young company with growing revenues and big losses, that pricing will be based on metrics that are buzzy and unfamiliar, from eyeballs in the dotcom era to users in the social media decade to ARR with AI companies...https://bit.ly/4iBwsHm

@natolambert right from the horse's mouth ah, right. That's "a single cluster", who knows how many they have https://github.com/deepseek-ai/DeepInfer/blob/main/doc/inference_deployment_qa.md