单词学习：![[5d604604-785a-4c35-9e86-878ad45a399d.mp3]]

新闻播报：![[449f2032-165a-4c96-a998-c487c09d0c78.mp3]]

谷歌强化AI搜索：AI概览上线追问功能并标配Gemini 3

简报：谷歌于1月27日宣布升级其搜索功能，“AI概览”（AI Overviews）现已支持用户进行上下文追问以实现更自然的对话式搜索，并已将Gemini 3模型作为全球范围内的默认配置，旨在提升AI响应质量。

相关链接：

• 谷歌搜索升级：AI概览支持追问，Gemini 3 成为全球默认模型
• 谷歌 AI Plus 订阅版图再扩张：登陆全球 35 个市场，月费 7.99 美元
• 谷歌发布 Agentic Vision：模仿人类处理图像，让 AI 学会“放大看”
• 可标记网页 AI 生成内容，谷歌拟为 Chrome 浏览器引入新功能

Google Boosts AI Search: AI Overviews Launch Follow-up Questions and Gemini 3 as Default

Brief: Google announced on January 27 an upgrade to its search capabilities 🚀. "AI Overviews" now support users asking contextual follow-up questions for a more natural conversational search experience 💬. Furthermore, the Gemini 3 model has been made the default configuration globally, aiming to enhance AI response quality 💡.

Related Links:

• Google Search Upgrades: AI Overviews Support Follow-up Questions, Gemini 3 Becomes Global Default Model
• Google AI Plus Subscription Expands Again: Launches in 35 Global Markets, Monthly Fee $7.99
• Google Releases Agentic Vision: Mimics Human Image Processing, Teaches AI to 'Zoom In'
• Google Plans New Feature for Chrome Browser: Ability to Mark AI-Generated Web Content

谷歌推出智能体视觉，AI可主动调查图像细节

简报：

• 谷歌DeepMind团队在Gemini 3 Flash模型基础上，推出了“智能体视觉”（Agentic Vision）新功能。
• 该功能改变了传统AI单次静态扫描图像的方式，通过模仿人类“思考、行动、观察”的循环，将视觉理解转变为一个主动调查的过程。
• 模型能生成并执行Python代码来主动操作图像（如裁剪、旋转），以聚焦并分析图像中的细微细节，为最终答案确立视觉证据。
• 引入该功能后，模型在多数视觉基准测试中质量提升了5-10%，并能有效减少在处理视觉数学问题时出现的“幻觉”。

相关链接：

• 谷歌发布 Agentic Vision：模仿人类处理图像，让 AI 学会“放大看”

Google Launches Agentic Vision: AI Can Actively Investigate Image Details

Brief:

• The Google DeepMind team has launched a new feature called “Agentic Vision,” built upon the Gemini 3 Flash model. 🚀
• This feature transforms the traditional method of AI performing a single, static scan of an image, turning visual understanding into an active investigative process by mimicking the human "think, act, observe" loop.
• The model can generate and execute Python code to actively manipulate images (e.g., cropping, rotating) to focus on and analyze subtle details within the image, establishing visual evidence for the final answer. 🔍
• With the introduction of this feature, the model has seen a 5-10% quality improvement in most visual benchmarks and can effectively reduce "hallucinations" encountered when processing visual math problems. ✨

Related Links:

• Google Releases Agentic Vision: Mimicking Human Image Processing, Teaching AI to 'Zoom In'

OpenAI发布Prism，为科研人员提供免费AI协作空间

简报：

• OpenAI于1月27日宣布推出由GPT-5.2驱动的AI原生科研协作空间Prism，旨在解决日常科研工作的碎片化问题。
• 该工具将初稿撰写、润色修订、团队协作和投稿准备整合到统一的云端LaTeX原生工作空间中。
• Prism即日起向所有拥有ChatGPT个人账户的用户免费开放，项目数量与协作人数不设上限，并将很快向组织版用户开放。

相关链接：

• OpenAI 推出 Prism：GPT-5.2 驱动的免费 AI 原生科研协作空间
• OpenAI 推出 Prism：GPT-5.2 驱动的免费 AI 原生科研协作空间
• OpenAI 推出 Prism：GPT-5.2 驱动的免费 AI 原生科研协作空间

OpenAI Launches Prism: Free AI Collaboration Space for Researchers

Brief:

• OpenAI announced on January 27th the launch of Prism, an AI-native research collaboration space powered by GPT-5.2, designed to address the fragmentation issues in daily research work 🚀.
• This tool integrates initial draft writing, refinement and revision, team collaboration, and submission preparation into a unified, cloud-based LaTeX-native workspace 💡.
• Prism is now freely available to all users with a ChatGPT personal account, with no limits on the number of projects or collaborators. It will soon be accessible to enterprise users 😊.

Related Links:

• OpenAI 推出 Prism：GPT-5.2 驱动的免费 AI 原生科研协作空间
• OpenAI 推出 Prism：GPT-5.2 驱动的免费 AI 原生科研协作空间
• OpenAI 推出 Prism：GPT-5.2 驱动的免费 AI 原生科研协作空间

Meta推出AI付费订阅，整合Manus与Vibes工具

简报：Meta于1月27日宣布将测试一项全新订阅服务，计划在Instagram、Facebook和WhatsApp上通过付费模式，为用户提供整合了AI智能体Manus和短视频工具Vibes的增强功能与独家生产力工具。

相关链接：

• Meta官宣高级订阅服务，开启AI视频增值时代
• Meta官宣高级订阅服务，开启AI视频增值时代

Meta Launches AI Paid Subscription, Integrating Manus and Vibes Tools 💰

Brief: Meta announced on January 27 that it will test a new subscription service, planning to offer users enhanced features and exclusive productivity tools, integrating the AI agent Manus and short video tool Vibes, via a paid model on Instagram, Facebook, and WhatsApp. 🚀💡

Related Links:

• Meta Officially Announces Premium Subscription Service, Ushering in an Era of AI Video Value-Added Services
• Meta Officially Announces Premium Subscription Service, Ushering in an Era of AI Video Value-Added Services

Claude模型深入集成Slack与Figma

简报：人工智能公司Anthropic宣布，旗下的大型语言模型Claude现已深度集成到Slack和Figma等多款主流工作应用中。

相关链接：

• Anthropic 宣布 Claude 已深度集成到 Slack 和 Figma 等应用中
• Claude与协作工具集成详情

Claude Model Deeply Integrated with Slack & Figma 🚀

Brief: AI company Anthropic announced that its large language model Claude is now deeply integrated into several mainstream work applications, including Slack and Figma. This move aims to enhance user productivity and collaboration! 🤖✨

Related Links:

• Anthropic Announces Deep Integration of Claude into Slack, Figma, and Other Apps
• Details of Claude's Integration with Collaboration Tools

微软Copilot功能升级，新增长期记忆并放宽输入限制

简报：

• 微软为Copilot AI引入“长期记忆”功能，使其能够跨越不同对话记住用户提供的信息，用户可在设置中查看或删除记忆。
• Copilot的单次输入字符上限提升至超过10240个字符，并支持将长文本自动转换为文件进行处理。
• 新增“对话置顶”功能，允许用户将重要的聊天固定在列表顶部，以解决“找对话难”的问题。
• macOS版Copilot功能更新后与Windows 11版基本对齐，同时为iOS用户推出了新的桌面小组件。

相关链接：

• 追赶 ChatGPT：微软 Copilot AI 新增“长期记忆”，输入上限破 1 万字符

Microsoft Copilot Feature Upgrade: Long-Term Memory Added, Input Limits Relaxed

Brief:

• Microsoft introduces a 'long-term memory' feature for Copilot AI, enabling it to remember user-provided information across different conversations. Users can view or delete these memories in the settings. 🧠
• Copilot's single input character limit has been increased to over 10,240 characters. It also supports automatic conversion of long texts into files for processing. 🚀
• A new 'pin conversation' feature allows users to fix important chats to the top of the list, addressing the difficulty of finding specific conversations. 📌
• The macOS version of Copilot has been updated to largely align with the Windows 11 version. Additionally, new desktop widgets have been launched for iOS users.

Related Link:

• 追赶 ChatGPT：微软 Copilot AI 新增“长期记忆”，输入上限破 1 万字符