目录
单词学习：![[fd605ad7-8b78-43d3-83cf-d1ea7678ffe7.mp3]]
新闻播报：![[e8176990-0280-47c0-acb7-f87e3d0c60c8.mp3]]
谷歌以 Gemini3 驱动搜索变革，打造 AI 对话新体验
简报：
- 谷歌宣布为其 AI Overviews (AI 概览) 功能引入最新的 Gemini3 模型，旨在将搜索体验从传统的链接跳转转变为 AI 对话模式。
- 用户可在搜索结果中进行无缝追问，或一键切换至全新的 AI 模式 (AI Mode)，在搜索页面内完成深度交互，以解决复杂问题。
- 该变革由 Gemini 模型的技术进步支撑，例如新发布的 “智能体视觉” (Agentic Vision) 功能，该功能让 AI 能够模仿人类“思考-行动-观察”的循环来主动分析图像，从而提升视觉理解的准确性。
相关链接：
Google Drives Search Transformation with Gemini 3, Creating a New AI Conversation Experience
Brief:
- Google announced the integration of its latest Gemini 3 model into its AI Overviews feature, aiming to transform the search experience from traditional link navigation to an AI conversation model. 🚀
- Users can now follow up seamlessly within search results or switch with a single click to the new AI Mode, enabling deep interaction directly on the search page to solve complex problems. 💬
- This transformation is underpinned by technological advancements in the Gemini model, such as the newly released “Agentic Vision” feature, which allows AI to proactively analyze images by mimicking the human 'think-act-observe' loop, thereby enhancing the accuracy of visual understanding. ✨
Related Links:
软银洽谈巨额增资，OpenAI估值或达8300亿
简报：
- 据知情人士透露，软银集团正洽谈向OpenAI追加至多300亿美元的投资。
- 此举是OpenAI寻求募集至多1000亿美元新资金计划的一部分，若融资完成，其估值最高可达8300亿美元。
- 目前相关交易洽谈仍在进行中，具体条款仍有可能发生变动。
相关链接：
SoftBank in Talks for Mega Funding Boost, OpenAI Valuation Could Hit $830 Billion 🚀
Briefing:
- SoftBank Group is reportedly in talks to inject up to $30 billion into OpenAI, according to sources familiar with the matter.
- This potential move is part of OpenAI's broader plan to raise up to dJyH6kDkfC5RlM16.answerText00 billion in new capital, which could push its valuation as high as $830 billion upon completion. 💰
- Discussions for the deal are still ongoing, and specific terms remain subject to change. ⏳
Related Link:
OpenAI发布科研协作平台Prism，集成多工具简化研究流程
简报：
- OpenAI正式推出专为科研人员打造的在线协作平台Prism，由最新的GPT-5.2模型驱动。
- 该平台整合了文本编辑器、LaTeX编译器、PDF阅读器、参考文献管理及AI聊天界面，旨在解决科研写作中多软件频繁切换的问题。
- Prism的核心功能包括协助推理复杂科学问题、重构数学公式、润色论文，并支持手绘草图转LaTeX图示、语音编辑等。
- 平台目前已对ChatGPT个人免费用户开放，并将在未来几周内陆续向Business和Enterprise付费方案用户推出。
相关链接：
- OpenAI 发布 Prism：基于 GPT-5.2 的科学家专属“AI 原生”工作空间
- 连 OpenAI CEO 都没忍住：奥尔特曼称 AI 太好用，两小时就交出了电脑权限
- IT之家报道：奥尔特曼分享自己两小时就给AI交出电脑权限的经历
OpenAI Releases Research Collaboration Platform Prism, Integrating Multiple Tools to Streamline Research Workflow
Brief:
- OpenAI has officially launched Prism, an online collaboration platform specifically designed for researchers, powered by the latest GPT-5.2 model. 🚀
- The platform integrates a text editor, LaTeX compiler, PDF reader, reference manager, and an AI chat interface, aiming to solve the frequent switching between multiple software applications during scientific writing. 💡
- Prism's core features include assisting with complex scientific reasoning, reconstructing mathematical formulas, refining papers, and supporting handwritten sketch-to-LaTeX diagram conversion, as well as voice editing. ✍️
- The platform is currently available to ChatGPT individual free users and will be rolled out to Business and Enterprise paid plan users in the coming weeks.
Related Links:
- OpenAI Releases Prism: A Scientist-Exclusive 'AI-Native' Workspace Based on GPT-5.2
- Even OpenAI CEO Couldn't Resist: Altman Says AI is So Good He Handed Over Computer Control in Two Hours
- IT Home Reports: Altman Shares His Experience of Giving AI Control of His Computer in Two Hours
腾讯开源800亿参数混元图像3.0模型
简报：
- 腾讯宣布其混元图像3.0（Hunyuan-DiT 3.0）模型正式开源，该模型是一个拥有800亿参数的图生图（image-to-image）模型，旨在助力AI创作。
相关链接：
Tencent Open-Sources 80 Billion Parameter Hunyuan Image 3.0 Model
Brief: 🚀 Tencent announced that its Hunyuan Image 3.0 (Hunyuan-DiT 3.0) model is now officially open source. This model is an 80-billion-parameter image-to-image (DiT) model, designed to empower AI creation. ✨
Related Links:
蚂蚁开源具身智能与空间感知两大模型
简报：
- 蚂蚁集团旗下具身智能公司蚂蚁灵波科技宣布，全面开源LingBot-VLA具身大模型及后训练代码，和LingBot-Depth空间感知模型。
- LingBot-VLA的后训练工具链在8卡GPU配置下，训练效率达到主流框架的1.5至2.8倍，并已适配星海图、松灵、乐聚等机器人厂商。
- 在上海交通大学开源的GM-100评测基准中，引入深度信息后，LingBot-VLA在3个不同真实机器人平台上的平均成功率达到17.3%。
- LingBot-Depth空间感知模型旨在为机器人、自动驾驶汽车等智能终端提供更精准的三维视觉，其在深度精度与像素覆盖率两项核心指标上均超越业界顶级工业级深度相机。
相关链接：
Ant Open-Sources Two Major Models for Embodied AI and Spatial Perception
Briefing:
- Ant Lingbo Technology, an embodied AI company under Ant Group, announced the full open-sourcing of its LingBot-VLA embodied large model and its post-training code, along with the LingBot-Depth spatial perception model. 🤖
- LingBot-VLA's post-training toolchain achieves 1.5 to 2.8 times the training efficiency of mainstream frameworks under an 8-GPU configuration and has been adapted for robot manufacturers such as Xinghaitu, Songling, and Leju. 🚀
- In the GM-100 evaluation benchmark, open-sourced by Shanghai Jiao Tong University, LingBot-VLA achieved an average success rate of 17.3% on 3 different real robot platforms after incorporating depth information.
- The LingBot-Depth spatial perception model aims to provide more precise 3D vision for intelligent terminals like robots and autonomous vehicles. It surpasses industry-leading industrial-grade depth cameras in both its core metrics: depth accuracy and pixel coverage. ✨
Related Links:
Hugging Face坚守中立，婉拒英伟达巨额投资
简报： 据报道，Hugging Face拒绝英伟达潜在的巨额投资，主要是为了保持公司的独立性和作为AI社区中立平台的定位。公司CEO Clément Delangue强调，Hugging Face致力于成为一个开放且硬件无关（hardware-agnostic）的平台，服务于整个机器学习生态系统。接受来自单一硬件巨头的大量资金可能会损害其中立性，并使其与使用其他硬件（如AMD、Google TPU等）的开发者和公司产生利益冲突。因此，Hugging Face在后续的D轮融资中，选择了包括谷歌、亚马逊、英伟达、Salesforce在内的多元化战略投资者，以确保其开放和多平台合作的战略。
相关链接：
Hugging Face Upholds Neutrality, Politely Declines Nvidia's Hefty Investment 🛡️
Brief: According to reports, Hugging Face rejected a potential massive investment from Nvidia, primarily to maintain its independence and its position as a neutral platform for the AI community. Company CEO Clément Delangue emphasized that Hugging Face is committed to being an open and hardware-agnostic platform, serving the entire machine learning ecosystem. 💡 Accepting substantial funding from a single hardware giant could compromise its neutrality and create conflicts of interest with developers and companies using other hardware (e.g., AMD, Google TPUs). Therefore, in its subsequent Series D funding round, Hugging Face chose a diverse group of strategic investors including Google, Amazon, Nvidia, and Salesforce, to ensure its strategy of openness and multi-platform collaboration. 🚀
Related Link:
QuestMobile报告：闲鱼AI应用用户规模登顶
简报：
- 根据第三方数据机构QuestMobile于1月27日发布的《2025-2026年度核心趋势报告》，闲鱼登上AI赛道用户规模第一的应用榜单。
- 报告显示，闲鱼在AI文案写作赛道中排名第一，其AI发布产品的月活跃用户数（MAU）已达1807.8万。
- 此外，闲鱼也持续保持其在闲置交易赛道用户规模第一的地位。
相关链接： QuestMobile：闲鱼登上AI赛道用户规模NO.1应用榜
QuestMobile Report: Xianyu AI Application Tops User Scale Rankings 💡
Brief:
- According to the "2025-2026 Core Trend Report" released by third-party data agency QuestMobile on January 27, Xianyu has topped the list of applications by user scale in the AI sector. 🚀
- The report indicates that Xianyu ranks first in the AI copywriting sector, with its AI publishing product reaching 18.078 million Monthly Active Users (MAU).
- Furthermore, Xianyu consistently maintains its leading position in user scale within the idle goods trading sector. 🥇
Related Link: QuestMobile: Xianyu Tops AI Sector User Scale Application Ranking
本文作者:topwind
本文链接:
版权声明:本博客所有文章除特别声明外，均采用 BY-NC-SA 许可协议。转载请注明出处！