单词学习：![[4be5287d-e129-4640-9ed7-293260f256da.mp3]]

新闻播报：![[cfd3b077-08b0-4f37-9163-6d7bcbd98f8b.mp3]]

昆仑元AI发布昇腾跨模态模型及AI硬件

简报：

• 在2025世界计算大会上，湖南昆仑元人工智能应用软件有限公司正式发布了基于昇腾（Ascend）的跨模态融合模型BaiZe-Omni-14b-a2b。
• 同时，该公司与AMD联合发布了GPT-Factory Mini AI工作站，并与超聚变签署战略合作协议，共同推出FusionXpark硬件平台。

相关链接：

• 昆仑元AI发布基于昇腾（Ascend）的跨模态融合模型BaiZe-Omni-14b-a2b

Kunlunyuan AI Unveils Ascend Cross-Modal Model and AI Hardware 🚀

Brief:

• At the 2025 World Computing Conference, Hunan Kunlunyuan Artificial Intelligence Application Software Co., Ltd. officially released its Ascend-based cross-modal fusion model, BaiZe-Omni-14b-a2b. ✨
• Concurrently, the company, in collaboration with AMD, launched the GPT-Factory Mini AI workstation. It also signed a strategic cooperation agreement with FusionXpark to jointly introduce the FusionXpark hardware platform. 🤝

Related Links:

• Kunlunyuan AI Releases Ascend-Based Cross-Modal Fusion Model BaiZe-Omni-14b-a2b

AMD携手IBM发布ZAYA1，首个纯AMD硬件训练大模型亮相

简报：

• AMD、IBM及AI初创公司Zyphra于11月24日联合宣布，成功训练出首个完全基于AMD硬件生态系统的大型混合专家（MoE）基础模型ZAYA1。
• 训练过程在IBM云上进行，使用了由1024块AMD Instinct MI300X GPU、Pensando网络技术和ROCm软件平台组成的大规模集群。
• ZAYA1模型在预训练阶段使用了14万亿tokens的数据，并采用了新的CCA注意力机制和改进的混合专家模型路由结构。
• 基准测试结果显示，ZAYA1的综合性能与业界领先的Qwen3系列模型相当，并优于SmolLM3、Phi4等主流开源模型。
• 目前发布的是ZAYA1基础模型预览版，未来团队将发布经过完整后训练的版本。

相关链接：

• 挑战英伟达：首个纯 AMD 硬件训练的 AI 大模型 ZAYA1 登场

AMD and IBM Unveil ZAYA1: The First Large AI Model Trained Exclusively on AMD Hardware 🚀

Briefing:

• AMD, IBM, and AI startup Zyphra jointly announced on November 24th the successful training of ZAYA1, the first large Mixture-of-Experts (MoE) foundational model built entirely on the AMD hardware ecosystem. 💡
• The training was conducted on IBM Cloud, utilizing a massive cluster comprising 1024 AMD Instinct MI300X GPUs, Pensando networking technology, and the ROCm software platform.
• The ZAYA1 model leveraged 14 trillion tokens of data during its pre-training phase and incorporates a novel CCA attention mechanism along with an enhanced Mixture-of-Experts routing structure.
• Benchmark test results indicate that ZAYA1's overall performance is comparable to industry-leading Qwen3 series models and surpasses mainstream open-source models like SmolLM3 and Phi4.
• The currently released version is a preview of the ZAYA1 foundational model; the team plans to release a fully post-trained version in the future. 🤖

Related Links:

• Challenging NVIDIA: The First AI Large Model ZAYA1 Trained Exclusively on AMD Hardware Debuts

阿里千问取代Meta，成新加坡国家AI计划新选择

简报：据外媒11月25日报道，新加坡国家人工智能计划（AISG）在其最新的东南亚语言大模型项目中进行战略调整，决定放弃Meta模型，转向采用阿里巴巴的通义千问Qwen开源架构。

相关链接：

• 新加坡国家AI计划放弃Meta模型，转向阿里千问

Alibaba's Tongyi Qianwen Replaces Meta, Becomes New Choice for Singapore's National AI Plan

Brief: According to foreign media reports on November 25, Singapore's National AI Programme (AISG) has made a strategic adjustment in its latest Southeast Asian language large model project, deciding to abandon Meta's models and turn to adopt Alibaba's Tongyi Qianwen Qwen open-source architecture. 🇸🇬✨

Related Link:

• Singapore's National AI Plan Dumps Meta Models, Pivots to Alibaba's Tongyi Qianwen

OpenAI与Jony Ive联手，打造无屏AI硬件终结数字干扰

简报：

• OpenAI CEO山姆·奥特曼与前苹果设计师乔纳森·艾维正合作开发一款“无屏幕”的袖珍AI硬件设备，旨在终结现代科技带来的“数字喧嚣”和干扰。
• 奥特曼将当前设备体验比作嘈杂的“纽约时代广场”，而新设备将提供如“湖边木屋”般宁静祥和的体验。
• 该设备的核心理念是极致简洁、“几乎不用思考就能使用”，并基于用户信任和情境感知能力来为用户过滤和处理信息。
• 这款AI硬件设备预计将在两年内上市。

相关链接：

• OpenAI携手Jony Ive:无屏AI设备将重新定义用户体验

OpenAI & Jony Ive Partner to Build Screenless AI Hardware, Aiming to End Digital Distraction

Briefing:

• OpenAI CEO Sam Altman and former Apple designer Jony Ive are collaborating on a "screenless" pocket-sized AI hardware device, aiming to put an end to the "digital noise" and distractions brought by modern technology. 📵
• Altman likened the current device experience to a noisy "Times Square in New York," while the new device promises a tranquil and peaceful experience akin to a "lakeside cabin." 🏞️
• The core philosophy behind the device is extreme simplicity, "almost thoughtless to use," and it will filter and process information for users based on trust and situational awareness.
• This AI hardware device is expected to be available on the market within two years. 🚀

Related Links:

• OpenAI Joins Forces with Jony Ive: Screenless AI Device to Redefine User Experience

ChatGPT上线购物研究功能，提供无广告比价报告

简报：

• OpenAI为ChatGPT Plus、Pro及Team订阅用户推出“Shopping Research”（购物研究）新功能，可在网页端与App端使用。
• 该功能可实时抓取电商、论坛、评测视频及价格追踪网站的数据，并自动过滤刷单评价。
• 它能生成结构化的购物报告，内容包括性价比排行、产品优缺点列表、多平台实时比价（含历史低价）以及个性化推荐。
• OpenAI表示该功能采用“零商业化”策略，不接受品牌付费排位，以确保报告的客观性。
• 首批支持数码、家电、美妆、家居和服饰五大品类，并计划未来扩展至汽车和二手房。

相关链接：

• OpenAI携手Jony Ive:无屏AI设备将重新定义用户体验，告别数字干扰
• 库克明年上半年退休？古尔曼直言“假消息”，OpenAI狂挖苹果40+硬件人才成焦点
• ChatGPT“购物研究”上线：全网实时比价+过滤刷单， 3 分钟输出零广告导购报告

ChatGPT Launches Shopping Research Feature, Offering Ad-Free Price Comparison Reports

Briefing:

• OpenAI has rolled out a new "Shopping Research" feature for ChatGPT Plus, Pro, and Team subscribers, available on both web and app platforms. 🛍️
• This function can gather real-time data from e-commerce sites, forums, review videos, and price tracking websites, automatically filtering out fake reviews. 🔎
• It generates structured shopping reports, including cost-effectiveness rankings, product pros and cons lists, real-time multi-platform price comparisons (featuring historical low prices), and personalized recommendations.
• OpenAI states that this feature adopts a "zero commercialization" strategy, not accepting paid placements from brands to ensure the objectivity of its reports. ✅
• Initially, it supports five major categories: digital products, home appliances, beauty, home furnishings, and apparel, with plans to expand to cars and second-hand homes in the future.

Related Links:

• OpenAI Teams Up with Jony Ive: Screenless AI Devices to Redefine User Experience, Bidding Farewell to Digital Disturbance
• Cook to Retire by H1 Next Year? Gurman Calls it "Fake News," OpenAI Poaching 40+ Apple Hardware Talents Becomes a Focus
• ChatGPT "Shopping Research" Goes Live: Real-time Price Comparison Across the Web + Fake Review Filtering, 3-Minute Ad-Free Buying Guide Report Output

Gemini 3.0发布引爆流量，日访问量突破5400万

简报：

• 根据Similarweb的最新数据，在Gemini 3.0版本发布后，其日访问量创下历史新高。
• 发布当天的访问量激增至约5400万次，显著高于此前4400万次的峰值。
• 访问量的增长主要归因于新版本引入了多项备受用户期待的新功能。

相关链接：

• Gemini 3 发布后日访问量创历史新高，用户热情高涨
• 谷歌Nano Bana Pro模型首秀
  新增幻灯片和信息图功能
• 谷歌牵手Accel推出“AI Futures × Atoms”计划：单项目共投 200 万美元，瞄准印度AI种子轮

Gemini 3.0 Launch Explodes Traffic 🚀, Daily Visits Surpass 54 Million

Briefing:

• According to Similarweb's latest data, following the release of Gemini 3.0, its daily visits reached an all-time high.
• On the day of its launch, visits surged to approximately 54 million, significantly surpassing the previous peak of 44 million.
• The growth in visits is primarily attributed to the introduction of several highly anticipated new features in the latest version. ✨

Related Links:

• Gemini 3 Reaches All-Time High Daily Visits After Launch, User Enthusiasm Soars
• Google's Nano Bana Pro Model Debuts: NotebookLM Adds Slides and Infographic Features
• Google Partners with Accel to Launch "AI Futures × Atoms" Program: Investing $2 Million Per Project, Targeting Indian AI Seed Rounds

台积电应对AI芯片需求，拟加码9000亿新台币建三座2纳米新厂

简报：

• 供应链消息透露，因应AI芯片订单激增及先进产能供不应求，台积电计划在台湾加码投资兴建3座2纳米芯片厂。
• 新厂总投资额估计高达9000亿新台币，地点可能选在南部科學園區附近的特定区。
• 台积电已与政府单位会商资源需求，若用地取得与环评顺利，最快可能在明年动土。
• 目前台积电在竹科及高雄已有7座2纳米厂的规划。

相关链接：

• 晶片供不應求！台積電加碼投9千億元「建3座2奈米廠」　落腳地點曝光

TSMC Responds to Soaring AI Chip Demand, Plans NT$900 Billion Investment for Three New 2nm Fabs 🚀

Brief:

• Supply chain sources indicate that TSMC plans to significantly invest in building three additional 2nm chip fabs in Taiwan, driven by a surge in AI chip orders and an overwhelming demand for advanced manufacturing capacity.
• The total investment for these new plants is estimated to reach an astonishing NT$900 billion (approximately US$27.7 billion), with potential locations eyed in specific zones near the Southern Taiwan Science Park.
• TSMC has reportedly engaged with government agencies to discuss resource requirements. If land acquisition and environmental assessments proceed smoothly, groundbreaking could commence as early as next year.
• Currently, TSMC already has plans for seven 2nm fabs in the Hsinchu Science Park and Kaohsiung. 🏭✨

Related links:

• 晶片供不應求！台積電加碼投9千億元「建3座2奈米廠」　落腳地點曝光