单词学习：![[3277a8b0-f4f6-4f65-b2e7-5c5c1abeaa65.mp3]]

新闻播报：![[62debfac-e676-49ee-bc77-7b704f81fbd4.mp3]]

谷歌Gemini升级为自主研究代理，可自行规划分析任务

简报：

• 谷歌发布基于Gemini 3 Pro的全新深度研究助手，将其从报告撰写工具转变为一个能自主规划研究步骤、执行长时间推理和复杂分析的自主代理。
• 新系统旨在减少AI“幻觉”的发生，能分析并浓缩大量复杂数据，提供更准确的报告，并已应用于金融服务、生物技术等领域。
• 谷歌将为开发者提供新工具，以便将该研究能力嵌入自有应用，并计划将该代理整合进谷歌搜索、财经及NotebookLM等服务。
• 谷歌搜索已开始测试“+”按钮，允许部分用户上传图片和文档，利用Gemini AI进行深度分析。
• 在谷歌发布此更新的同一天，OpenAI也推出了GPT-5.2模型，加剧了AI领域的竞争。

相关链接：

• 谷歌发布全新 Gemini 深度研究助手，AI 竞争再升级
• 谷歌搜索测试“+”按钮:允许用户上传图片文档，利用 Gemini AI 进行深度分析

Google Gemini Upgrades to Autonomous Research Agent, Capable of Self-Planning Analysis Tasks

Brief:

• Google has launched a new deep research assistant powered by Gemini 3 Pro, transforming it from a report-writing tool into an autonomous agent capable of independently planning research steps, conducting prolonged reasoning, and performing complex analysis. 🚀
• The new system aims to mitigate AI "hallucinations," analyze and synthesize vast amounts of complex data, provide more accurate reports, and has already been applied in sectors like financial services and biotechnology. 💡
• Google will provide developers with new tools to embed this research capability into their own applications and plans to integrate this agent into services such as Google Search, Google Finance, and NotebookLM.
• Google Search has begun testing a "+" button, allowing some users to upload images and documents for in-depth analysis using Gemini AI.
• On the same day Google released this update, OpenAI also launched its GPT-5.2 model, intensifying competition in the AI sector. 🥊

Related Links:

• Google Launches New Gemini Deep Research Assistant, AI Competition Escalates
• Google Search Testing "+" Button: Allows Users to Upload Images and Documents for Deep Analysis with Gemini AI

苹果启动自研AI服务器芯片，旨在摆脱对英伟达的依赖

简报：

• 苹果公司已启动代号为“Baltra”的首款自研AI服务器芯片项目，旨在深化其“垂直整合”战略，摆脱对英伟达芯片的依赖。
• 该芯片将专注于“AI推理”（Inference）而非模型训练，以处理“Apple Intelligence”云端服务的用户请求，强调低延迟与高并发吞吐量。
• 苹果选择博通（Broadcom）作为网络技术合作伙伴，芯片预计将采用台积电先进的3nm “N3E”工艺。
• “Baltra”芯片项目设计工作预计在未来12个月内完成，最早可能在2027年正式投入使用。

相关链接：

• 摆脱英伟达依赖!苹果启动首款自研 AI 服务器芯片“Baltra”项目

Apple Kicks Off In-House AI Server Chip Development to Break Free from Nvidia 🚀

Brief:

• Apple Inc. has initiated its first in-house AI server chip project, codenamed "Baltra," aiming to deepen its "vertical integration" strategy and reduce reliance on Nvidia chips.
• The chip will focus on "AI Inference" rather than model training, processing user requests for "Apple Intelligence" cloud services, emphasizing low latency and high concurrent throughput.
• Apple has chosen Broadcom as its networking technology partner, and the chip is expected to utilize TSMC's advanced 3nm "N3E" process.
• Design work for the "Baltra" chip project is anticipated to be completed within the next 12 months, with formal deployment potentially as early as 2027. 🗓️💡

Related Link:

• 摆脱英伟达依赖!苹果启动首款自研 AI 服务器芯片“Baltra”项目

OpenAI以股权换IP，Sora获迪士尼海量角色授权

简报：

• OpenAI与迪士尼达成合作，通过提供认股权证（warrants）而非现金，获得了迪士尼旗下超过200个经典角色的使用权。
• 此次授权涵盖米老鼠、灰姑娘、小美人鱼Ariel、狮子王辛巴等标志性IP，将用于OpenAI的文生视频模型Sora。
• 该合作模式被视为“以IP换未来”，迪士尼押注OpenAI的长期股权价值，而OpenAI则降低了前期成本，并获得了可落地的商业内容生态。
• 此举旨在帮助Sora在与谷歌Gemini等产品的竞争中，通过绑定顶级IP获得优势，推动AI生成内容进入专业影视制作流程。

相关链接：

• OpenAI与迪士尼达成“股权换IP”重磅合作：Sora获米老鼠等200+角色授权，押注AI视频未来
• OpenAI 通信一把手 Hannah Wong 明年交棒，曾主导化解“奥尔特曼被罢免”公关危机
• 用 AI 从零设计抗体，OpenAI 支持的 Chai Discovery 获 1.3 亿美元融资

OpenAI Trades Equity for IP, Sora Secures Massive Character Licensing from Disney 🎬

Brief:

• OpenAI has partnered with Disney, obtaining usage rights for over 200 classic characters from the Disney portfolio through warrants rather than cash.
• This authorization includes iconic IPs such as Mickey Mouse, Cinderella, Ariel the Little Mermaid, and Simba from The Lion King, and will be utilized by OpenAI's text-to-video model, Sora.
• This cooperation model is viewed as "exchanging IP for the future," with Disney betting on OpenAI's long-term equity value, while OpenAI reduces upfront costs and gains access to a practical commercial content ecosystem.
• The move aims to give Sora an edge against competitors like Google's Gemini by integrating top-tier IP, pushing AI-generated content into professional film and television production workflows. ✨

Related Links:

• OpenAI and Disney Reach Major "Equity-for-IP" Deal: Sora Licensed 200+ Characters Including Mickey Mouse, Betting on AI Video Future
• OpenAI Communications Head Hannah Wong to Step Down Next Year, Once Led Crisis Management During "Altman's Ouster"
• Designing Antibodies from Scratch with AI, OpenAI-Backed Chai Discovery Raises dJyH6kDkfC5RlM16.answerText30 Million 🚀

ChatGPT移动端迎来对话分支功能

简报：OpenAI已将“对话分支”（Branch Conversations）功能扩展至ChatGPT的iOS和Android移动客户端，用户现可在移动设备上从对话的任意节点创建多个独立的并行分支，以探索不同思路而不影响主对话流程。

相关链接：

• ​ChatGPT移动端上线“对话分支”功能，多线程思考终于自由切换

ChatGPT Mobile App Gains Conversation Branching Feature 💬

Brief: OpenAI has extended the "Branch Conversations" feature to ChatGPT's iOS and Android mobile clients. Users can now create multiple independent, parallel branches from any point in a conversation on their mobile devices, allowing them to explore different ideas without disrupting the main conversation flow. 🚀✨

Related Link:

• ChatGPT mobile launches "conversation branching" feature, multi-threaded thinking finally free to switch

台积电熊本二厂为AI需求或升级4奈米，施工暂停

简报：

• 据日经亚洲报道，为应对人工智能（AI）相关需求的激增，台积电正考虑将其在日本熊本的第二座工厂生产制程从原计划的6奈米与7奈米，升级为更先进的4奈米技术。
• 此举可能导致该厂延后原定于2027年的量产时程并需要重新设计，工地现场照片显示，熊本二厂的施工已于12月初暂停。
• 知情人士称，台积电也考虑将其先进的晶片封装技术引入日本，但所有计划尚未最终确定。
• 台积电官方回应称，其日本计划正在推进中，公司正与合作伙伴商讨详细的施工执行计划。

相关链接：

• 台積電熊本二廠傳考慮產「更先進」4奈米晶片！　因應AI需求
• 張忠謀領6億股息不是最多！台積電今除息　「他」爽入袋82.68億元登榜首

TSMC Kumamoto Fab 2 May Upgrade to 4nm for AI Demand, Construction Halted 🚧

Briefing:

• According to Nikkei Asia, TSMC is considering upgrading the production process at its second factory in Kumamoto, Japan, from the originally planned 6nm and 7nm to the more advanced 4nm technology, in response to surging artificial intelligence (AI)-related demand. 🚀
• This move could delay the factory's mass production schedule, originally set for 2027, and necessitate a redesign. On-site photos show that construction at Kumamoto Fab 2 was halted in early December.
• Sources indicate that TSMC is also considering introducing its advanced chip packaging technology to Japan, though all plans are not yet finalized.
• TSMC officially responded that its Japan plans are moving forward, and the company is discussing detailed construction execution plans with its partners. 🤝

Related Links:

• TSMC Kumamoto Fab 2 Reportedly Considering Production of "More Advanced" 4nm Chips! To Meet AI Demand
• Morris Chang's NT$600 Million Dividend is Not the Most! TSMC Ex-dividend Today, "He" Rakes in NT$8.268 Billion to Top the List

阿里发布万相2.6视频模型，首创角色扮演功能

简报：

• 12月16日，阿里巴巴发布新一代万相2.6系列模型，官方称其为“全球功能最全的视频生成模型”。
• 该模型为国内首个支持“角色扮演”功能的视频模型，能够参考输入视频中角色的外观和音色，按照提示词生成新视频。
• 新增专业级分镜控制功能，可将简单提示词转换为多分镜脚本；单次视频生成时长提升至国内最高的15秒。
• 万相2.6已上线阿里云百炼和万相官网，个人用户可直接体验，企业用户可通过API调用。

相关链接：

• 阿里万相2.6发布:支持“角色扮演”与多分镜控制，单次视频时长达15秒

Alibaba Unveils Wanxiang 2.6 Video Model, Debuts First-Ever Role-Playing Feature

Briefing:

• On December 16, Alibaba released its next-generation Wanxiang 2.6 series model, officially touting it as the "world's most feature-rich video generation model." 🌍
• This model is the first in China to support a "role-playing" function, capable of generating new videos based on prompts while referencing the appearance and vocal characteristics of a character from an input video. 🎭
• New professional-grade storyboard control features allow simple prompts to be converted into multi-scene scripts; single video generation length has been extended to a domestic high of 15 seconds. ⏱️
• Wanxiang 2.6 is now available on Alibaba Cloud Bailian and the Wanxiang official website, allowing individual users direct access and enterprise users API calls.

Related Link:

• Alibaba Wanxiang 2.6 Released: Supports "Role-Playing" and Multi-Scene Control, Single Video Length Up to 15 Seconds