单词学习：![[984bfb74-840d-4fee-9d7b-a4616f37ed92.mp3]]

新闻播报：![[6370a773-9901-48ce-97bf-cfafbc88e70b.mp3]]

火山引擎：AI市场将迎十倍增长，竞争核心在于共同做大蛋糕

简报：

• 火山引擎总裁谭待在2025 FORCE原动力大会上预测，到2026年大模型市场规模有望实现十倍增长。
• 谭待认为，当前AI竞争的核心不是零和博弈，而是通过合作共同将市场“蛋糕做大”。
• 他将2025年定义为“智能体（Agent）年”，并预计2026年将迎来全面爆发，企业级智能体部署正从广度和深度上实现跨越。
• 谭待强调，衡量AI产业落地的核心指标是Tokens调用量，模型的价值在真实调用中才能释放。
• 对于AI手机等终端，他认为行业仍处早期，关键是能否以更低成本便利地满足用户需求，而非载体形式。

相关链接：

• 中信建投：继续看好AI算力板块
• 火山引擎谭待：2026年大模型市场将迎十倍增长，AI竞争不是零和博弈而是共同做大蛋糕
• 国内首个国标 VLA 大模型开源!北京人形 XR-1让机器人真正“会干活”
• 蓝色光标官宣深度接入豆包大模型，全面融合大语言与视频生成能力
• 豆包上线 Seedance 1.5 Pro，可直接生成有声视频

Volcengine: AI Market to See Tenfold Growth, Competition Core Lies in Expanding the Pie Together 📈

Brief:

• Tan Dai, President of Volcengine, predicted at the 2025 FORCE Original Power Conference that the large model market size is expected to achieve a tenfold growth by 2026.
• Tan Dai believes that the current core of AI competition is not a zero-sum game, but rather about collaborating to "make the market pie bigger" together. 🤝
• He defined 2025 as the "Agent Year" and anticipates a full-scale boom in 2026, with enterprise-level intelligent agent deployment achieving significant leaps in both breadth and depth.
• Tan Dai emphasized that the key metric for measuring AI industry adoption is the volume of Token calls, as the value of models is only unleashed through real-world usage.
• Regarding AI phones and other terminals, he believes the industry is still in its early stages. The key is whether user needs can be conveniently met at a lower cost, rather than the specific form factor of the carrier device. 📱

Related Links:

• Citic Securities: Continues to Be Optimistic About the AI Computing Power Sector
• Volcengine Tan Dai: Large Model Market to See Tenfold Growth by 2026, AI Competition is Not a Zero-Sum Game But About Expanding the Pie Together
• China's First National Standard VLA Large Model Open-Sourced! Beijing Humanoid XR-1 Makes Robots Truly "Capable"
• BlueFocus Officially Announces Deep Integration with Doubao Large Model, Fully Fusing Large Language and Video Generation Capabilities
• Doubao Launches Seedance 1.5 Pro, Can Directly Generate Videos with Sound

首个国标VLA模型XR-1开源，助人形机器人迈入应用阶段

简报：

• 北京人形机器人创新中心于2025年12月18日宣布，正式开源国内首个通过具身智能国家标准测试的视觉-语言-动作（VLA）大模型XR-1。
• XR-1被定位为具身智能的“小脑”，负责将上层大模型的指令转化为机器人精确的物理动作，使其能完成取放、推拉等复杂双臂技能。
• 该模型具备跨机器人本体的通用操作能力，可在天工2.0、UR、Franka等不同平台上进行知识迁移和应用。
• 随模型一同开源的还包括数据底座RoboMIND2.0及高保真数字资产数据集ArtVIP，旨在降低行业开发门槛，推动机器人产业进入大规模应用阶段。

相关链接：

• 国内首个国标 VLA 大模型开源!北京人形 XR-1让机器人真正“会干活”

China's First National Standard VLA Model, XR-1, Open-Sourced to Propel Humanoid Robots into Application Phase 🤖

Briefing:

• On December 18, 2025, the Beijing Humanoid Robot Innovation Center announced the official open-sourcing of XR-1, the first domestic Vision-Language-Action (VLA) large model to pass the national standard test for embodied intelligence.
• XR-1 is positioned as the "cerebellum" of embodied intelligence, responsible for translating instructions from high-level large models into precise physical actions for robots, enabling them to perform complex dual-arm skills such as picking, placing, pushing, and pulling.
• The model boasts universal operational capabilities across different robot platforms, allowing for knowledge transfer and application on platforms like Tiangong 2.0, UR, and Franka. 🦾
• Open-sourced alongside the model are the data foundation RoboMIND2.0 and the high-fidelity digital asset dataset ArtVIP. These initiatives aim to lower industry development barriers and drive the robot industry into a phase of large-scale application. 🚀

Related Link:

• 国内首个国标 VLA 大模型开源!北京人形 XR-1让机器人真正“会干活”

Luma AI推出新工具，视频编辑可换脸换景不换表演

简报：

• 3D与AI视频公司Luma AI发布全新模型Ray3 Modify，用于高保真实拍视频的AI修改。
• 该工具的核心突破在于，能在更换角色外观、服装或场景的同时，完整保留原始演员的动作节奏、眼神和情感表达等表演细节。
• 用户通过提供角色参考图和设定起止关键帧，即可实现“表演不变、表现可变”的精准视频编辑。
• Luma AI CEO表示，该工具将真实世界与AI创造力融合，创作者用普通相机拍摄一次表演后，即可通过AI将其置于任何想象的场景中。
• Ray3 Modify模型现已集成至Luma的Dream Machine平台，面向专业创作者开放。

相关链接：

• Luma发布Ray3 Modify：AI视频编辑新突破，保留演员表演细节，一键换装、换场景

Luma AI Launches New Tool for Video Editing: Swap Faces and Scenes, Keep the Performance Intact 🤩

Brief:

• 3D and AI video company Luma AI has released a new model, Ray3 Modify, for AI modification of high-fidelity live-action videos.
• The core breakthrough of this tool is its ability to fully preserve the original actor's performance details, such as movement rhythm, eye gaze, and emotional expression, while simultaneously changing the character's appearance, costume, or scene.
• Users can achieve precise video editing with "unchanged performance, variable presentation" by providing character reference images and setting start and end keyframes.
• Luma AI CEO stated that the tool merges the real world with AI creativity. Creators can film a performance once with a regular camera and then use AI to place it in any imaginative scene. 🎬
• The Ray3 Modify model is now integrated into Luma's Dream Machine platform and is available to professional creators. ✨

Related Links:

• Luma发布Ray3 Modify：AI视频编辑新突破，保留演员表演细节，一键换装、换场景

微软集成Kimi模型，推出Office自动化Agent

简报：

• 微软在Azure云服务接入月之暗面（Moonshot AI）的Kimi模型后，将合作拓展至应用层。
• 微软计划于2025年12月上线全新的Agent功能，旨在实现Office产品的全面自动化。
• 该Agent将深度利用Kimi模型的长程推理与工具调用能力。
• 此举是微软为应对本土市场竞争，通过引入外部顶尖模型来补强其第一梯队AI能力并完善其AI云业务闭环的策略。

相关链接：

• 微软联手 Kimi 打造 Office 自动化 Agent，AI 云巨头开启本土化“自我修正”

Microsoft Integrates Kimi Model, Launches Office Automation Agent 🚀

Brief:

• Following the integration of Moonshot AI's Kimi model into its Azure cloud services, Microsoft is extending its collaboration to the application layer. ✨
• Microsoft plans to launch a new Agent feature by December 2025, aiming for comprehensive automation of Office products.
• This Agent will deeply leverage the Kimi model's long-context reasoning and tool-calling capabilities. 🤖
• This move is a strategic step by Microsoft to address local market competition, strengthen its top-tier AI capabilities by integrating external leading models, and complete its AI cloud business closed loop.

Related Link:

• Microsoft Teams Up with Kimi to Create Office Automation Agent, AI Cloud Giant Begins Localization "Self-Correction"

AI芯片公司Mythic获1.25亿美元融资，主打模拟计算与NVIDIA互补

简报：

• AI芯片公司Mythic成功筹集1.25亿美元资金，由深科技风险投资公司DCVC领投，本田汽车、洛克希德・马丁等也参与其中。
• 资金将用于开发其模拟处理单元（APU），目标是实现比现有GPU高出100倍的能效，并完善其软件开发工具包。
• Mythic的APU采用模拟内存计算技术，旨在作为GPU的补充加速器而非替代品，其现行架构据称每瓦可提供120万亿次操作。
• 公司CEO、前NVIDIA高管坦尔・奥兹切利表示，Mythic的芯片能够在不依赖高速互连的情况下运行万亿级参数的大型语言模型。
• Mythic的目标市场包括数据中心、汽车系统、机器人及国防领域，并已推出集成了其芯片的“星光”（Starlight）亚瓦特传感平台。

相关链接：

• Mythic 融资 1.25 亿美元，力图以超高能效 AI 芯片挑战 NVIDIA
• Mythic获新融资，研发模拟处理单元APU
• 可玩转千余款游戏，英伟达联合斯坦福等研发 NitroGen 开源模型

AI Chip Company Mythic Secures dJyH6kDkfC5RlM16.answerText25 Million Funding, Focusing on Analog Compute to Complement NVIDIA 🚀

Brief:

• AI chip company Mythic has successfully raised dJyH6kDkfC5RlM16.answerText25 million in funding, led by deep-tech venture capital firm DCVC, with participation from Honda Motor, Lockheed Martin, and others. 💰
• The funds will be utilized to develop its Analog Processing Unit (APU), aiming to achieve 100 times greater energy efficiency than existing GPUs, and to refine its software development kit.
• Mythic's APU leverages analog in-memory computing technology, designed to serve as a complementary accelerator to GPUs rather than a replacement. Its current architecture reportedly delivers 120 trillion operations per watt.
• Taner Ozcelik, CEO of the company and a former NVIDIA executive, stated that Mythic's chips are capable of running large language models with trillions of parameters without relying on high-speed interconnects.
• Mythic targets markets including data centers, automotive systems, robotics, and defense sectors, and has already launched its "Starlight" sub-watt sensing platform, which integrates its chips. ✨

Related Links:

• Mythic 融资 1.25 亿美元，力图以超高能效 AI 芯片挑战 NVIDIA
• Mythic获新融资，研发模拟处理单元APU
• 可玩转千余款游戏，英伟达联合斯坦福等研发 NitroGen 开源模型

AI巨头联手：用AI预测年龄保护青少年

简报：为防范未成年人使用风险，OpenAI与Anthropic宣布将推出AI年龄预测功能，其中OpenAI计划在识别出用户可能未满18岁时自动启用保护机制，而Anthropic则旨在通过分析对话特征来识别并关停未成年人账户。

相关链接：

• 防范未成年人风险
  与Anthropic将推出AI年龄预测功能
• ChatGPT 文风，原产地肯尼亚
• 为防范未成年人使用风险，OpenAI 与 Anthropic 将推出 AI 年龄预测功能

AI Giants Join Forces: Using AI to Predict Age to Protect Minors 🛡️

Brief: To mitigate risks associated with underage users, OpenAI and Anthropic have announced the upcoming launch of an AI age prediction feature. OpenAI plans to automatically activate protective mechanisms when a user is identified as potentially being under 18, while Anthropic aims to identify and deactivate minor accounts by analyzing conversation characteristics. 🤖

Related Links:

• 防范未成年人风险
  与Anthropic将推出AI年龄预测功能
• ChatGPT 文风，原产地肯尼亚
• 为防范未成年人使用风险，OpenAI 与 Anthropic 将推出 AI 年龄预测功能

纽约签署AI安全法案，违规最高罚300万美元

简报：

• 纽约州州长凯西・霍楚尔已签署《RAISE 法案》，使纽约成为继加州之后美国第二个出台重大人工智能安全法案的州。
• 法案要求大型人工智能开发商需公开其安全协议信息，并在发生安全事故后72小时内向州政府报备。
• 对于未按要求提交报告或存在虚假申报的企业，最高可被处以100万美元罚款，屡犯者罚款上限将升至300万美元。
• 该法案旨在与加州已确立的监管框架为基础，推动两大科技前沿州建立统一的行业标准。

相关链接：

• 纽约出台 AI 安全法案：要求大企业公示安全协议，违规最高罚 300 万美元

New York Signs AI Safety Bill, Violators Face Up to $3 Million Fine

Brief:

• New York Governor Kathy Hochul has signed the RAISE Act, making New York the second U.S. state, after California, to enact a significant AI safety bill. 🗽
• The bill mandates that large AI developers disclose their security protocol information and report security incidents to the state government within 72 hours of their occurrence. 🛡️
• Companies that fail to submit required reports or make false declarations can face fines of up to dJyH6kDkfC5RlM16.answerText million, with the maximum penalty for repeat offenders increasing to $3 million. 💸
• This legislation aims to build upon California's established regulatory framework, pushing for the creation of unified industry standards between these two tech-forward states.

Related Link:

• New York Enacts AI Safety Bill: Requires Large Companies to Disclose Security Protocols, Violators Face Up to $3 Million Fine