单词学习：![[2026031388986345.mp3]]

新闻播报：![[2026031344429348.mp3]]

移动端AI入口竞争白热化，阿里云正式上线JVSClaw“龙虾”应用

简报：

2026年3月13日，阿里云手机版OpenClaw应用“JVSClaw”正式登陆各大应用商店，标志着主流云厂商对AI移动端入口的争夺进入新阶段。
该应用延续了前期的免费内测优惠，用户在注册后可创建1个自定义Bot，并获赠8000Credits的模型额度。
市场竞争对手动作频繁，腾讯旗下的“小龙虾”版本WorkBuddy已于3月12日完成更新，通过打通微信一键直连提升了生态连接稳定性。
自Peter Steinberger于2025年11月发布OpenClaw以来，该开源AI代理工具在中国引发“养虾”热潮，字节跳动（ArkClaw）及智谱AI等厂商也已相继推出相关产品。
尽管OpenClaw带动了Token消耗量创下新高，但安全机构也发出预警，近期出现了伪装成安装包的GhostClaw木马等恶意程序。

相关链接：

阿里云上线手机版 OpenClaw“龙虾”JVSClaw

Competition for AI Entry Points on Mobile Heats Up as Alibaba Cloud Officially Launches the JVSClaw “Lobster” App 📱🦞

Briefing:

On March 13, 2026, Alibaba Cloud’s mobile OpenClaw app, JVSClaw, officially launched on major app stores, marking a new phase in the battle among leading cloud providers for AI entry points on mobile devices.
The app continues its earlier free beta promotion. After registering, users can create one custom Bot and receive 8,000 Credits in model usage quota.
Competitors have also been moving quickly. Tencent’s “Little Lobster” version, WorkBuddy, completed an update on March 12, improving ecosystem connectivity stability through direct one-click integration with WeChat.
Since Peter Steinberger released OpenClaw in November 2025, the open-source AI agent tool has sparked a “raising lobsters” craze in China. Companies including ByteDance (ArkClaw) and Zhipu AI have also rolled out related products.
Although OpenClaw has driven token consumption to record highs, security agencies have also issued warnings, noting the recent emergence of malicious programs such as the GhostClaw trojan disguised as installation packages. ⚠️

Related link:

Alibaba Cloud launches the mobile OpenClaw “Lobster” JVSClaw

Emergence /ɪˈmɜːrdʒəns/

n. 出现；兴起

▶ "The emergence of AI has transformed many industries."

[例句] 人工智能的兴起已经改变了许多行业。

◼ 词根分析

e-(ex-)

向外

merg-

沉没

-ence

名词后缀

◼ 衍生词

emerge (v.) 出现

emergent (adj.) 新兴的

emergency (n.) 紧急情况

苹果首款折叠屏手机CAD图流出：采用独特“方圆”机身，定价或上探2000美元

简报：

知名爆料达人 Sonny Dickson 曝光了苹果首款折叠手机（暂称 iPhone Fold 或 Ultra）的 3D CAD 设计图，揭示了其融合复古与未来的工业设计。
硬件设计上，该机型背部弃用三摄，转而采用类似 iPhone Air 风格的双镜头相机平台，机身四角呈现“两圆两方”的非对称结构，旨在完美容纳折叠铰链机构。
屏幕配置显示其拥有外侧副屏幕及展开后的 7.8 英寸巨型主屏幕，主屏左上角设有微小相机开孔，整体比例接近 iPad Mini 或首代 Pixel Fold 的矮胖造型。
市场传闻该产品将定位为超越 Pro Max 的全新“Ultra”级超旗舰，单机售价可能落在 2,000 美元至 2,400 美元之间（约合人民币 1.4 万至 1.7 万元）。
供应链消息称，三星显示（Samsung Display）已被选为该设备 OLED 折叠面板的独家供应商，量产预计于 2026 年中期开始，有望在 2026 年秋季与 iPhone 18 系列一同发布。

相关链接：

Apple’s First Foldable iPhone CAD Images Leaked: Featuring a Distinctive “Square-and-Round” Body, Price May Exceed $2,000 📱✨

Brief:

Well-known leaker Sonny Dickson has revealed 3D CAD design files of Apple’s first foldable phone (tentatively called the iPhone Fold or Ultra), showcasing an industrial design that blends retro and futuristic elements.
In terms of hardware design, the device reportedly drops the triple-camera setup in favor of a dual-camera platform similar to the iPhone Air style. Its four corners feature an asymmetric “two round, two square” structure, intended to perfectly accommodate the folding hinge mechanism.
Screen specifications indicate that it will feature an outer secondary display and a large 7.8-inch main screen when unfolded. A small camera cutout is located in the upper-left corner of the main display, and the overall aspect ratio appears closer to the shorter, wider form factor of the iPad mini or the first-generation Pixel Fold.
Market rumors suggest the product will be positioned as a brand-new “Ultra”-level flagship above the Pro Max, with a starting price likely ranging from $2,000 to$ 2,400 (approximately RMB 14,000 to RMB 17,000). 💰
Supply chain sources say Samsung Display has been selected as the exclusive supplier of the device’s OLED foldable panels. Mass production is expected to begin in mid-2026, with a potential launch alongside the iPhone 18 series in the fall of 2026. 🚀

自研算力突围：Meta 发布新一代 MTIA 芯片，推理性能超越英伟达 H100

简报：

Meta 正式发布新一代自研 AI 芯片 MTIA 3（Meta 训练与推理加速器），主要用于优化 Facebook、Instagram 的推荐算法及 Llama 大模型的实时推理。
官方基准测试显示，MTIA 3 在特定工作负载下的推理效能已超越英伟达的主力 H100 芯片，具备更高的能效比和算力密度。
该芯片通过与 PyTorch 框架深度底层耦合，可显著提升 AI 算法的部署速度，并逐年摊薄 Meta 在 AI 基础设施上的巨额支出。
Meta 计划到 2027 年底前部署四代自研芯片（包括 MTIA 300、400、450 和 500），旨在减少对外部供应链的过度依赖并确保供应链安全。
尽管推进自研芯片进程，Meta 目前仍维持与英伟达、AMD 的巨额硬件采购协议，以支撑其快速扩张的 AI 工作负载。

相关链接：

Breaking Through with In-House Compute: Meta Unveils Next-Generation MTIA Chip, Delivering Inference Performance Beyond Nvidia H100 🚀

Brief:

Meta has officially launched its next-generation in-house AI chip, MTIA 3 (Meta Training and Inference Accelerator), primarily designed to optimize recommendation algorithms for Facebook and Instagram, as well as real-time inference for the Llama family of large models.
According to official benchmark results, MTIA 3 has surpassed Nvidia’s flagship H100 chip in inference efficiency under specific workloads, offering higher energy efficiency and greater compute density ⚡.
Through deep low-level integration with the PyTorch framework, the chip can significantly accelerate AI model deployment while helping Meta gradually amortize its massive spending on AI infrastructure year by year.
Meta plans to deploy four generations of self-developed chips by the end of 2027, including the MTIA 300, 400, 450, and 500, with the goal of reducing excessive reliance on external supply chains and strengthening supply chain security.
Despite advancing its in-house chip roadmap, Meta is still maintaining major hardware procurement agreements with Nvidia and AMD to support its rapidly expanding AI workloads.

英伟达发布开源大模型 Nemotron 3 Super，推理速度提升 3 倍并主打智能体应用

简报：

英伟达（NVIDIA）于 2026 年 3 月 12 日正式发布专为 AI 智能体（Agent）设计的开源大模型 Nemotron 3 Super。
该模型采用创新的 Mamba-MoE 混合架构，参数总量 1200 亿，激活参数仅为 120 亿，推理速度提升 300%，吞吐量增长 5 倍。
Nemotron 3 Super 支持高达 1M 的超长上下文，在 OpenClaw 智能体任务中的成功率达 85.6%，性能表现直逼 GPT-5.4 和 Claude Opus 4.6。
模型针对英伟达最新的 Blackwell 硬件平台进行了优化，支持 NVFP4 训练特性，可显著提升算力利用率并降低训练成本。
目前该模型已获得 Perplexity、西门子、戴尔等科技巨头集成，并同步上线 AWS、Azure 和 Google Cloud 等主流云平台。

相关链接：

NVIDIA发布开源大模型Nemotron 3 Super：性能直逼GPT-5.4

NVIDIA Unveils Open-Source Large Model Nemotron 3 Super, Delivering 3x Faster Inference and Targeting AI Agent Applications 🚀🤖

Brief:

On March 12, 2026, NVIDIA officially released Nemotron 3 Super, an open-source large model purpose-built for AI agents.
The model adopts an innovative Mamba-MoE hybrid architecture, with 120 billion total parameters and only 12 billion activated parameters, delivering a 300% increase in inference speed and 5x higher throughput.
Nemotron 3 Super supports an ultra-long context window of up to 1M tokens, achieves an 85.6% success rate on OpenClaw agent tasks, and delivers performance close to GPT-5.4 and Claude Opus 4.6.
The model is optimized for NVIDIA’s latest Blackwell hardware platform and supports NVFP4 training features, significantly improving compute utilization while reducing training costs.
The model has already been integrated by major technology companies including Perplexity, Siemens, and Dell, and is now available on leading cloud platforms such as AWS, Azure, and Google Cloud. ☁️

谷歌 Gemini 进化为“超级代理”：手机端任务自动化功能开启 Beta 测试

简报：

谷歌正式推出基于 Gemini 的“任务自动化”（或称“屏幕自动化”）功能，目前已在三星 Galaxy S26 系列手机上开启 Beta 测试。
该功能标志着 AI 助手从信息查询向“数字代办”转型，能够直接模拟人类在 UI 界面上的操作，实现跨应用执行打车、订餐及点咖啡等复杂流程。
目前首批适配的应用包括 Uber、Uber Eats、Lyft、Starbucks、Grubhub 和 Doordash，未来计划在 Pixel 10 系列等更多机型上推广。
在安全机制上，系统设置了人工复核逻辑：AI 会在后台完成加购或填写目的地等操作，但在最终支付或提交订单前，会强制停留并等待用户手动确认。
实际测试显示，AI 能自主处理如选择星巴克具体单品、确认航站楼等细节，但在处理复杂滚动选择时偶尔仍显笨拙，且在 Beta 阶段可能出现预览模式锁定等系统稳定性问题。

相关链接：

Google Gemini Evolves into a “Super Agent”: Beta Testing Begins for Mobile Task Automation 🤖📱

Brief:

Google has officially launched Gemini-powered “Task Automation” (also referred to as “screen automation”), which is now in Beta testing on the Samsung Galaxy S26 series.
This feature marks a shift for AI assistants from information lookup to becoming a true “digital concierge,” capable of directly simulating human actions within UI interfaces to complete complex cross-app tasks such as booking rides, ordering food, and buying coffee.
The first wave of supported apps includes Uber, Uber Eats, Lyft, Starbucks, Grubhub, and DoorDash, with plans to expand to more devices in the future, including the Pixel 10 series.
On the security side, the system includes a human review mechanism: the AI can complete actions such as adding items to a cart or entering a destination in the background, but it will always pause before final payment or order submission and require manual user confirmation.
Early hands-on testing shows that the AI can independently handle details such as selecting specific Starbucks menu items or confirming the correct airport terminal. However, it can still appear clumsy when dealing with complex scrolling selections, and Beta-stage stability issues—such as getting stuck in preview mode—may still occur. ☕✨

Anthropic 拨 1 亿美元启动合作伙伴网络，加速企业级 AI 落地

简报：

Anthropic 正式启动“Claude 合作伙伴网络”并投入 1 亿美元初始资金，旨在资助能帮助企业将 AI 落地到业务流程的服务商，并同步推出支持技术认证的项目及代码现代化工具。
该公司同步推出“Claude Marketplace”应用市场，允许企业客户直接使用现有的 API 预算购买基于 Claude 构建的第三方软件（如 Snowflake、GitLab 等），且启动初期不收取佣金。
此举旨在发挥 Claude 作为唯一在 AWS、谷歌云和微软云上均能直接调用的模型优势，通过构建商业生态系统降低企业部署门槛，与 OpenAI 争夺企业级 AI 市场主导权。

相关链接：

Anthropic Commits dJyH6kDkfC5RlM16.answerText00 Million to Launch Partner Network, Accelerating Enterprise AI Adoption 🚀

Brief:

Anthropic has officially launched the “Claude Partner Network” with an initial dJyH6kDkfC5RlM16.answerText00 million investment, aimed at funding service providers that can help enterprises integrate AI into their business workflows. At the same time, the company is rolling out programs for technical certification support and code modernization tools.
The company has also introduced the “Claude Marketplace,” an app marketplace that allows enterprise customers to directly use their existing API budgets to purchase third-party software built on Claude, including offerings from Snowflake, GitLab, and others. No commission will be charged during the initial launch phase 💼.
This move is designed to capitalize on Claude’s unique advantage as the only model directly accessible across AWS, Google Cloud, and Microsoft Cloud. By building a broader commercial ecosystem, Anthropic aims to lower deployment barriers for enterprises and compete with OpenAI for leadership in the enterprise AI market 🤖.

腾讯 WorkBuddy 升级全量开放：手机微信可远程“遥控”桌面 AI 助理

简报：

腾讯自研 AI 原生桌面智能体工作台 WorkBuddy（原 OpenClaw 腾讯版）宣布全量开放，并实现与微信的深度直连。
用户通过微信发送文字或语音指令即可远程操控办公室电脑，由 AI 助理完成资料查询、文案撰写或报表处理，结果实时回传手机。
新增“自动化打工”模式，支持设定定时任务（如每日抓取热点、每周整理纪要），并自动推送 Markdown 或 PDF 格式的成品。
升级底层架构以支持企业微信断网自动重连，并推出包含云端隔离、沙箱及漏洞扫描在内的全链路安全防御体系。
针对全量开放推出福利，即日起至 2026 年 3 月 31 日，新老用户注册或更新即可无门槛领取 5000 积分体验补贴。

相关链接：

腾讯 WorkBuddy 迎来重大更新：直连微信，让“小龙虾”全天候为你打工

Tencent WorkBuddy Upgrade Fully Rolls Out: Remotely “Control” Your Desktop AI Assistant via WeChat 🤖📱

Brief:

Tencent’s self-developed AI-native desktop agent workspace, WorkBuddy (formerly OpenClaw Tencent Edition), has officially become fully available and now features deep integration with WeChat.
Users can remotely control their office computers by sending text or voice commands through WeChat, allowing the AI assistant to handle information searches, copywriting, or report processing, with results sent back to the phone in real time.
A new “automated work” mode has been added, supporting scheduled tasks such as fetching trending topics daily or organizing weekly meeting notes, and automatically delivering finished files in Markdown or PDF format.
The underlying architecture has been upgraded to support automatic reconnection for WeCom after network interruptions, while also introducing an end-to-end security defense system that includes cloud isolation, sandboxing, and vulnerability scanning 🔒.
To celebrate the full rollout, Tencent is offering a special benefit: from now until March 31, 2026, both new and existing users can register or update to claim a 5,000-point experience subsidy with no eligibility threshold.

GitHub Copilot 学生版取消高级模型手动选择权，被指变相“降本”

简报：

微软旗下 GitHub 宣布自 2026 年 3 月 12 日起调整 Copilot 学生免费套餐，将其纳入新的管理计划以确保长期访问的可持续性。
此次调整的核心变动是学生用户将无法再手动选择 GPT-5.4、Claude Opus 和 Sonnet 等高级 AI 模型。
GitHub 官方将通过“自动模式”为学生分配来自 OpenAI、Anthropic 和谷歌的模型，现有的高级请求单位（PRU）权益则保持不变。
这一举措引发了大量学生用户的失望与不满，用户认为限制高品质推理模型的使用将显著影响学习效率和代码理解能力。

相关链接：

微软 GitHub Copilot 学生免费套餐“降本”：无法手动选择高级 AI 模型，引发众怒

GitHub Copilot Student Plan Loses Manual Access to Advanced Models, Criticized as a Disguised Cost-Cutting Move 😕💸

Brief:

GitHub, a Microsoft subsidiary, announced that starting March 12, 2026, it will adjust the Copilot free plan for students by moving it into a new management program to ensure the sustainability of long-term access.
The core change is that student users will no longer be able to manually select advanced AI models such as GPT-5.4, Claude Opus, and Sonnet.
GitHub said it will assign models from OpenAI, Anthropic, and Google to students through an “automatic mode,” while existing Premium Request Unit (PRU) benefits will remain unchanged.
The move has triggered widespread disappointment and dissatisfaction among student users, who argue that restricting access to high-quality reasoning models will significantly reduce learning efficiency and weaken code comprehension capabilities. 📉

目录

移动端AI入口竞争白热化，阿里云正式上线JVSClaw“龙虾”应用

Competition for AI Entry Points on Mobile Heats Up as Alibaba Cloud Officially Launches the JVSClaw “Lobster” App 📱🦞

苹果首款折叠屏手机CAD图流出：采用独特“方圆”机身，定价或上探2000美元

Apple’s First Foldable iPhone CAD Images Leaked: Featuring a Distinctive “Square-and-Round” Body, Price May Exceed $2,000 📱✨

自研算力突围：Meta 发布新一代 MTIA 芯片，推理性能超越英伟达 H100

Breaking Through with In-House Compute: Meta Unveils Next-Generation MTIA Chip, Delivering Inference Performance Beyond Nvidia H100 🚀

英伟达发布开源大模型 Nemotron 3 Super，推理速度提升 3 倍并主打智能体应用

NVIDIA Unveils Open-Source Large Model Nemotron 3 Super, Delivering 3x Faster Inference and Targeting AI Agent Applications 🚀🤖

谷歌 Gemini 进化为“超级代理”：手机端任务自动化功能开启 Beta 测试

Google Gemini Evolves into a “Super Agent”: Beta Testing Begins for Mobile Task Automation 🤖📱

Anthropic 拨 1 亿美元启动合作伙伴网络，加速企业级 AI 落地

Anthropic Commits dJyH6kDkfC5RlM16.answerText00 Million to Launch Partner Network, Accelerating Enterprise AI Adoption 🚀

腾讯 WorkBuddy 升级全量开放：手机微信可远程“遥控”桌面 AI 助理

Tencent WorkBuddy Upgrade Fully Rolls Out: Remotely “Control” Your Desktop AI Assistant via WeChat 🤖📱

GitHub Copilot 学生版取消高级模型手动选择权，被指变相“降本”

GitHub Copilot Student Plan Loses Manual Access to Advanced Models, Criticized as a Disguised Cost-Cutting Move 😕💸