![[facdb6dc-11ea-4a5a-8544-0bfcfe7fb451.mp3]]

OpenAI或将下周首度开源语言模型，或引发与微软分歧

简报：

• 据The Verge等媒体2025年7月10日报道，OpenAI最快下周将推出开源的大语言AI模型，提供类似o3 mini的推理能力，这将是其自2019年以来首次开源语言模型。
• OpenAI此前仅发布过Whisper系列的语音转文本模型权重，其余模型皆未公开，此举有望打破闭源AI的格局。
• OpenAI已邀请开发者和研究者测试该模型，并征求AI社区反馈。
• 当前OpenAI正与微软重新谈判合作协议，微软作为主要云服务合作伙伴，拥有OpenAI多数模型的独家商业访问权，开源模型可能加剧双方分歧。

相关链接：

• OpenAI计划下周发布开源语言模型，或加剧美微软分歧
• OpenAI酝酿推出开源AI模型

OpenAI to Open-Source Language Model Next Week for First Time, Potentially Causing Rift with Microsoft

Briefing:

• According to reports from The Verge and other media outlets on July 10, 2025, OpenAI is set to launch an open-source large language AI model as early as next week, offering inference capabilities similar to o3 mini. This would mark its first open-source language model release since 2019. 🚀
• Previously, OpenAI had only released weights for its Whisper series speech-to-text models, with all other models remaining closed-source. This move is expected to disrupt the current landscape of closed-source AI.
• OpenAI has already invited developers and researchers to test the model and is soliciting feedback from the AI community. 🤝
• Currently, OpenAI is renegotiating its partnership agreement with Microsoft. As the primary cloud service partner, Microsoft holds exclusive commercial access to most of OpenAI's models. An open-source model could potentially intensify disagreements between the two parties. 😬

Related Links:

• OpenAI Plans to Release Open-Source Language Model Next Week, Potentially Intensifying Disagreement with Microsoft
• OpenAI Brewing Open-Source AI Model Launch

Hugging Face推出3B参数的小型开源模型SmolLM3

简报：

• Hugging Face于近日开源了名为SmolLM3的小型语言模型，拥有30亿（3B）参数。
• SmolLM3支持高达128K的上下文窗口，适合处理长文本任务。
• 该模型面向多种下游应用优化，强调高性能与灵活开源许可，旨在为开发者提供更高效的模型选项。

相关链接：

• Hugging Face开源SmolLM3：3B参数小模型，128K上下文

Hugging Face Launches SmolLM3: A 3B-Parameter Open-Source Small Model

Briefing:

• Hugging Face recently open-sourced SmolLM3, a small language model featuring 3 billion (3B) parameters. 🤖
• SmolLM3 supports a context window of up to 128K, making it suitable for long-text processing tasks. 📚
• Optimized for various downstream applications, this model emphasizes high performance and flexible open-source licensing, aiming to provide developers with more efficient model options. ✨

Related Links:

• Hugging Face开源SmolLM3：3B参数小模型，128K上下文

英伟达CEO将再访中国，专供中国市场AI芯片预计9月上市

简报：

• 英伟达CEO黄仁勋计划下周访问中国，参加在北京举行的中国国际供应链促进博览会，并将会见中国高级官员，重申公司对中国市场的承诺。
• 英伟达计划最快于今年9月推出专为中国市场设计的新型AI芯片，该芯片为现有Blackwell RTX Pro 6000处理器的降级版，删减了高带宽内存和NVLink等部分先进技术，以符合美国出口管制规定。
• 新芯片在正式开售前需获得美国政府的保证确保不违反出口限制，上市时间及具体规格仍可能根据美方政策变化调整。
• 中国是英伟达第四大市场，截至2025财年营收达171亿美元，占总销售额的13%。

相关链接：

• 英伟达CEO黄仁勋将于下周访华，特供版AI芯片最快9月推出
• 英伟达CEO黄仁勋下周访华，中国特供AI芯片或9月推出
• 黄仁勋拟访华，预计9月推中国专供AI芯片

NVIDIA CEO to Revisit China, AI Chips Exclusively for Chinese Market Expected in September 🤝

Brief:

• NVIDIA CEO Jensen Huang is scheduled to visit China next week to attend the China International Supply Chain Expo in Beijing and meet with senior Chinese officials, reaffirming the company's commitment to the Chinese market.
• NVIDIA plans to launch a new AI chip designed specifically for the Chinese market as early as September this year. This chip is a downgraded version of the existing Blackwell RTX Pro 6000 processor, with advanced technologies like high-bandwidth memory and NVLink removed to comply with US export control regulations. 📉
• Before its official launch, the new chip requires assurance from the US government to ensure it does not violate export restrictions. The launch date and specific specifications may still be adjusted based on changes in US policy. ⏳
• China is NVIDIA's fourth-largest market, with revenue reaching $17.1 billion as of fiscal year 2025, accounting for 13% of total sales.

Related Links:

• NVIDIA CEO Jensen Huang to visit China next week, special AI chips for China expected as early as September
• NVIDIA CEO Jensen Huang to visit China next week, China-specific AI chips may launch in September
• Jensen Huang plans to visit China, special AI chips for China expected in September

谷歌AI助手Gemini即将取代Google Assistant，登陆多品牌Wear OS智能手表

简报：

• 谷歌宣布，Gemini人工智能助手将在未来几周内全面登陆包括Pixel、三星、OPPO、一加、小米等品牌的Wear OS 4及以上系统智能手表，并计划取代原有的Google Assistant。
• 用户可通过“Hey Google”或长按手表按钮、点击Gemini图标等方式唤醒Gemini助手，实现语音查询、信息摘要、日历管理、信息记录等多种功能，提升操作便捷性和智能化体验。
• 随着Gemini在Wear OS平台部署，用户将可在手表上快捷管理日常任务、应用操作和生活琐事，无需频繁使用手机。

相关链接：

• IT之家报道
• 三立新闻网报道
• AIBase新闻

Google AI Assistant Gemini to Replace Google Assistant, Arriving on Multi-Brand Wear OS Smartwatches

Briefing:

• Google has announced that its Gemini AI assistant will be fully rolling out to Wear OS 4 and above smartwatches from various brands, including Pixel, Samsung, OPPO, OnePlus, and Xiaomi, in the coming weeks. It is planned to replace the existing Google Assistant. 🚀
• Users will be able to activate the Gemini assistant via “Hey Google,” by long-pressing the watch button, or by tapping the Gemini icon. This will enable various functions such as voice queries, information summaries, calendar management, and note-taking, enhancing operational convenience and intelligent experience. ✨
• With Gemini's deployment on the Wear OS platform, users will be able to quickly manage daily tasks, app operations, and daily errands directly on their watch, reducing the need to frequently use their phone. ⌚

Related Links:

• IT Home Report
• Sanli News Network Report
• AIBase News

AI手术机器人SRT-H首次完全自主完成复杂胆囊切除，准确率达100%

简报：

• 美国约翰斯·霍普金斯大学研究团队开发的AI手术机器人SRT-H，成功实现了完整自主的复杂胆囊切除手术，在八组猪胆囊和肝脏分离测试中达到100%准确率。
• 该系统采用“语言引导模仿学习”人工智能框架训练，具备自主决策和应变能力，能够在不同解剖结构下执行抓取、夹持、切割等环节，模拟人类手术中遇到的各种复杂情况。
• 目前SRT-H手术速度略慢于人类外科医生，研究团队计划继续优化其效率，并推进机器人在更复杂手术场景的应用。
• 与当前广泛应用需医生远程操控的达芬奇手术系统不同，SRT-H可实现全自主操作，有望缓解外科医生短缺，提升医疗服务质量。

相关链接：

• IT之家报道
• aibase新闻

SRT-H AI Surgical Robot Achieves First Fully Autonomous Complex Gallbladder Removal with 100% Accuracy

Briefing:

• The AI surgical robot SRT-H, developed by a research team at Johns Hopkins University in the United States, has successfully performed a fully autonomous complex gallbladder removal. It achieved 100% accuracy in eight sets of pig gallbladder and liver separation tests. 🤖
• The system is trained using a "language-guided imitation learning" AI framework, endowing it with autonomous decision-making and adaptive capabilities. It can perform grasping, clamping, and cutting in various anatomical structures, simulating complex situations encountered in human surgery. 🧠
• Currently, SRT-H's surgical speed is slightly slower than that of human surgeons. The research team plans to continue optimizing its efficiency and advance the robot's application in even more complex surgical scenarios.
• Unlike the widely used da Vinci Surgical System, which requires remote control by a surgeon, SRT-H can operate fully autonomously. This development holds promise for alleviating the shortage of surgeons and improving the quality of medical services. 💖

Related Links:

• IT之家报道
• aibase新闻