单词学习：![[8581f9b4-ce8e-42eb-81ac-38ee6022eba7.mp3]]

新闻播报：![[9b4064f5-15f4-4b8e-bcfd-ab41f538400b.mp3]]

百川发布医疗大模型M3 Plus，宣称幻觉率全球最低

简报：

• 1月22日，百川智能发布医疗大模型Baichuan-M3 Plus，据称其在严肃医疗场景的问答准确性与可靠性刷新了世界纪录。
• 该模型凭借独创的“六源循证”技术，将幻觉率降低至2.6%，官方称其为全球最低水平。
• M3 Plus首创“证据锚定”技术，能将模型生成的每一句医学结论，精确锚定至原始论文中的对应证据段落。

相关链接：

• 百川推出最低幻觉循证增强医疗大模型M3 Plus

Baichuan Launches Medical LLM M3 Plus, Claims Lowest Hallucination Rate Globally

Brief:

• On January 22, Baichuan Intelligent launched its medical large language model (LLM) Baichuan-M3 Plus, claiming it sets a new world record for Q&A accuracy and reliability in serious medical scenarios.
• Leveraging its proprietary 'Six-Source Evidence-Based' technology, the model has reduced its hallucination rate to an industry-low 2.6%, officially touted as the lowest globally. 🧠
• M3 Plus also introduces an innovative 'Evidence Anchoring' technology, which precisely links every medical conclusion generated by the model to its corresponding evidence paragraph in original research papers. 🔗

Related Links:

• Check out more details! 👇 Baichuan Launches M3 Plus, a Medical LLM with Lowest Hallucination Rate and Evidence-Based Enhancement

苹果AI布局曝光：内部上线Enchanté和Enterprise Assistant双系统

据报道，苹果公司自2025年11月起已在内部上线两套生成式AI工具，分别名为Enchanté和Enterprise Assistant，前者用于辅助代码编写和文案创作，后者则作为公司内部知识库，解答有关公司规章的提问。

相关链接：

• 苹果内部AI双系统首度曝光
• 知情人士透露苹果AI工具细节

Apple's AI Strategy Revealed: Internal Launch of Enchanté and Enterprise Assistant Dual Systems 🍏

According to reports, Apple has internally launched two generative AI tools since November 2025, named Enchanté and Enterprise Assistant, respectively. The former is designed to assist with code writing and content creation, while the latter serves as an internal knowledge base, answering questions about company regulations. 🤖✨

Related Links:

• Apple's Internal AI Dual Systems Revealed for the First Time
• Insiders Disclose Details of Apple's AI Tools

苹果研发AirTag大小AI胸针，探索独立穿戴新形态

简报：

• 苹果公司正研发一款全新可穿戴AI胸针设备，其定位是独立的AI硬件，而非手机配件。
• 该设备外形与AirTag接近，采用轻薄圆润的设计，机身材质预计为铝合金与玻璃。
• 设备内置高精度摄像头和麦克风，用于通过AI实时捕捉和分析用户周边环境信息，充当随身智能助理。

相关链接：

• 苹果秘密研发“AI胸针”：大小如AirTag
• 挑战下一代穿戴设备新形态

Apple Developing AirTag-Sized AI Pin, Exploring New Standalone Wearable Form Factor 📍

Brief:

• Apple Inc. is reportedly developing a new wearable AI pin device, positioned as standalone AI hardware rather than a mere phone accessory. ✨
• The device's form factor is said to be similar to an AirTag, featuring a thin and rounded design, with an anticipated body material of aluminum alloy and glass.
• It will feature a built-in high-precision camera and microphone to capture and analyze the user's surrounding environment information in real-time via AI, acting as a personal smart assistant. 💡

Related Links:

• Apple Secretly Developing 'AI Pin': AirTag-Sized
• Challenging the Next Generation of Wearable Device Form Factors

微软发布Rho-alpha模型，推动机器人进入“物理AI”新纪元

简报：

• 微软于2026年1月21日发布了专为机器人领域设计的全新AI模型“Rho-alpha”，旨在让机器人能够在复杂和不可预测的真实世界中执行任务。
• 该模型由微软研究院基于Phi视觉-语言模型家族开发，是微软“物理AI”战略的一部分，强调智能体与物理世界的直接交互。
• Rho-alpha的核心能力是将自然语言指令直接转换成机器人控制信号，支持复杂的双手协同操作，无需预设脚本。
• 模型不仅融合了视觉和语言，还引入了触觉感知能力，使机器人能根据触觉反馈实时调整动作。
• Rho-alpha具备适应性，能够在运行中学习，人类操作员可以通过3D输入设备等工具进行干预和纠正，帮助系统持续优化。
• 为了解决训练数据不足的问题，该模型采用真实机器人演示、仿真任务和大规模视觉问答数据相结合的混合训练方式。
• 微软计划首先通过研究型早期访问计划向外部提供Rho-alpha。

相关链接：

• 微软发布Rho-alpha模型，让AI机器人听懂语言指令
• 微软发布 Rho-alpha 模型，让 AI 机器人听懂语言指令并完成复杂操作

Microsoft Releases Rho-alpha Model, Propelling Robotics into a New Era of 'Physical AI' ✨

Briefing:

• On January 21, 2026, Microsoft unveiled 'Rho-alpha,' a new AI model specifically designed for robotics, aimed at enabling robots to perform tasks in complex and unpredictable real-world environments.
• Developed by Microsoft Research based on the Phi family of vision-language models, Rho-alpha is part of Microsoft's 'Physical AI' strategy, emphasizing direct interaction between intelligent agents and the physical world.
• Rho-alpha's core capability is transforming natural language instructions directly into robot control signals, supporting complex dual-arm collaborative operations without the need for pre-set scripts. 🤖
• The model integrates not only vision and language but also introduces haptic perception capabilities, allowing robots to adjust movements in real-time based on tactile feedback.
• Rho-alpha is adaptive, capable of learning on the fly. Human operators can intervene and correct using tools like 3D input devices, helping the system continuously optimize.
• To address the challenge of insufficient training data, the model employs a hybrid training approach, combining real robot demonstrations, simulated tasks, and large-scale visual question-answering data.
• Microsoft plans to initially offer Rho-alpha externally through a research-oriented early access program. 🚀

Related Links:

• Microsoft Releases Rho-alpha Model, Enabling AI Robots to Understand Language Commands
• Microsoft Releases Rho-alpha Model, Enabling AI Robots to Understand Language Commands and Perform Complex Operations