![[d19d3c09-ead5-4e61-b26f-0998e7007573.mp3]]

智元机器人获LG电子、未来资产联合战略投资

简报：

• 智元机器人宣布获得新的国际集团战略入股。
• 本轮投资由LG电子（LG Electronics）与韩国未来资产集团（Mirae Asset）联合领投。
• 根据公开信息，此次投资是LG电子在全球具身智能领域的首次对外布局。

相关链接：

• 智元机器人获国际集团战略入股

Zhipu Robot Secures Joint Strategic Investment from LG Electronics and Mirae Asset 🤖

Briefing:

• Zhipu Robot has announced a new strategic equity investment from an international consortium.
• This investment round was co-led by LG Electronics and South Korea's Mirae Asset Group. 🤝
• According to publicly available information, this marks LG Electronics' first external strategic move into the global embodied AI sector. 💰

Related Link:

• Zhipu Robot Secures Strategic Investment from International Group

DeepSeek公布大模型部署新专利，旨在提升硬件利用率与效率

简报：

• DeepSeek关联公司“杭州深度求索人工智能基础技术研究有限公司”近日公布了一项名为“一种大语言模型的部署方法及系统”的专利。
• 该专利的核心方法是将大语言模型的预填充（prefill）阶段和解码（decoding）阶段分别部署在高性能计算能力的机器和拥有大内存的机器上。
• 此举旨在均衡负载任务，从而最大化硬件利用率、减少闲置算力、降低整体延迟、提高吞吐量，并增强系统的扩展性和容错性。

DeepSeek Unveils New Patent for Large Model Deployment, Aiming to Boost Hardware Utilization and Efficiency 🚀

Briefing:

• Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd., an affiliate of DeepSeek, recently announced a patent titled "A Deployment Method and System for Large Language Models."
• The core method of this patent involves deploying the prefill phase and the decoding phase of large language models separately on machines with high-performance computing capabilities and machines with large memory.
• This initiative aims to balance workload tasks, thereby maximizing hardware utilization, reducing idle compute power, lowering overall latency, increasing throughput, and enhancing system scalability and fault tolerance. 💡⚙️

通义千问推出Qwen3-Coder-Flash，专为代码平台与函数调用优化

简报：

• 阿里巴巴通义千问发布了新的Qwen3-Coder-Flash模型。
• 该模型针对Qwen Code、Cline、Roo Code、Kilo Code等平台进行了优化。
• 模型具备特别设计的函数调用格式，以支持无缝的函数调用和代理工作流程。

相关链接：

• https://www.ithome.com/0/872/220.htm

Tongyi Qianwen Launches Qwen3-Coder-Flash, Optimized for Code Platforms and Function Calls

Brief:

• Alibaba's Tongyi Qianwen has launched the new Qwen3-Coder-Flash model. 🚀
• This model is optimized for platforms such as Qwen Code, Cline, Roo Code, and Kilo Code.
• The model features a specially designed function call format to support seamless function calls and agent workflows. 💡✨

Related Links:

• https://www.ithome.com/0/872/220.htm

报告称Anthropic已成企业AI首选，市场份额超越OpenAI

简报：

• 根据Menlo Ventures的最新报告，Anthropic在企业大型语言模型市场的份额已达到32%，首次超过OpenAI的25%，与两年前OpenAI占50%、Anthropic仅占12%的情况形成鲜明对比。
• Anthropic的增长主要得益于其Claude 3.5及后续模型的推出，在编程领域，其企业市场份额更是高达42%，远超OpenAI的21%。
• 尽管企业市场份额下降，OpenAI在消费者市场依然保持强势，其ChatGPT每日处理超过25亿条用户提示。
• 报告指出，企业因数据安全和性能稳定等考量，愈发倾向于使用封闭模型，而近期OpenAI因允许共享对话被搜索引擎索引而叫停试验，也凸显了企业对隐私风险的担忧。
• 市场竞争依然激烈，例如苹果公司被报道正同时考虑采用Anthropic或OpenAI的技术来升级其Siri服务。

相关链接：

• Anthropic 超越 OpenAI 成为企业 AI 模型的新宠
• OpenAI 被曝向搜索引擎公开 ChatGPT 共享对话内容，后因隐私风险叫停
• Apple may power Siri with Anthropic or OpenAI models amid AI struggles
• 輸錯一句話後果慘！專家示警「5類資訊」千萬別對ChatGPT說

Report: Anthropic Becomes Enterprise AI Preferred Choice, Surpassing OpenAI in Market Share

Briefing:

• According to Menlo Ventures' latest report, Anthropic's market share in the enterprise large language model (LLM) sector has reached 32%, surpassing OpenAI's 25% for the first time. This stands in stark contrast to two years ago when OpenAI held 50% and Anthropic only 12%. 📈
• Anthropic's growth is primarily attributed to the launch of its Claude 3.5 and subsequent models. In the programming domain, its enterprise market share is even higher at 42%, significantly exceeding OpenAI's 21%.
• Despite a decline in enterprise market share, OpenAI maintains a strong presence in the consumer market, with its ChatGPT processing over 2.5 billion user prompts daily.
• The report indicates that enterprises are increasingly opting for closed models due to considerations such as data security and stable performance. OpenAI's recent halt of an experimental feature that allowed shared conversations to be indexed by search engines also underscores enterprise concerns about privacy risks. 🔒
• Market competition remains fierce; for instance, Apple is reportedly considering adopting technology from either Anthropic or OpenAI to upgrade its Siri service. 🤔

Related Links:

• Anthropic Overtakes OpenAI as Enterprise AI Model Favorite
• OpenAI Reportedly Exposed ChatGPT Shared Conversations to Search Engines, Later Halted Due to Privacy Risks
• Apple may power Siri with Anthropic or OpenAI models amid AI struggles
• Severe Consequences for One Wrong Phrase! Experts Warn "5 Types of Information" Must Not Be Told to ChatGPT

AI芯片缺陷检测平台SixSense完成850万美元融资

简报：

• 新加坡深科技初创公司 SixSense 完成850万美元A轮融资，公司总融资额达1200万美元。
• 本轮融资由 Peak XV 的 Surge（前身为Sequoia India & SEA）领投，Alpha Intelligence Capital 和 FEBE 等参投。
• SixSense 成立于2018年，其AI驱动的平台旨在帮助芯片制造商实时预测和检测潜在的芯片缺陷，以提高产量和质量。
• 该平台已被 GlobalFoundries 和 JCET 等大型半导体制造商应用，据称可将人工检查工作减少90%。

相关链接：

• 新加坡AI初创公司SixSense获850万美元融资

AI Chip Defect Detection Platform SixSense Secures $8.5 Million Funding

Briefing:

• Singapore-based deep tech startup SixSense has completed an $8.5 million Series A funding round, bringing its total funding to$12 million. 🚀
• This round was led by Peak XV's Surge (formerly Sequoia India & SEA), with participation from Alpha Intelligence Capital, FEBE, and others.
• Founded in 2018, SixSense's AI-powered platform aims to help chip manufacturers predict and detect potential chip defects in real-time, thereby improving yield and quality. ✨
• The platform has been adopted by major semiconductor manufacturers like GlobalFoundries and JCET, reportedly reducing manual inspection efforts by up to 90%. 💡

Related Links:

• Singapore AI Startup SixSense Secures $8.5 Million Funding

网信办就H20芯片安全风险约谈英伟达

简报：

• 中国国家互联网信息办公室（网信办）于7月31日发布公告，称已就H20芯片的“漏洞后门安全风险问题”约谈英伟达（NVIDIA）。
• 公告指出，因美国有议员呼吁出口的先进芯片应配备“追踪定位”和“远程关闭”功能，网信办对此表示安全关切。
• 网信办依据《网络安全法》、《数据安全法》等规定，要求英伟达就H20芯片相关安全风险做出说明，并提交证明材料。
• 此前，美国政府已决定允许英伟达向中国出口H20芯片，英伟达也已向台积电下达大额订单。

相关链接：

• 中國網信辦今約談輝達　稱H20晶片有「漏洞後門安全風險」
• 新聞幕後／AMD MI350可望彎道超車！與輝達雙霸並立　重寫AI版圖

CAC Summons NVIDIA Over H20 Chip Security Risks

Brief:

• The Cyberspace Administration of China (CAC) announced on July 31 that it had summoned NVIDIA regarding "vulnerability backdoor security risks" concerning its H20 chip. 🚨
• The announcement noted that the CAC expressed security concerns following calls from some U.S. lawmakers for advanced chips exported to be equipped with "tracking and positioning" and "remote shutdown" capabilities.
• Pursuant to regulations such as the "Cybersecurity Law" and "Data Security Law," the CAC has demanded NVIDIA provide explanations and submit supporting documentation regarding the H20 chip's security risks.
• Previously, the U.S. government had decided to permit NVIDIA to export H20 chips to China, and NVIDIA had already placed substantial orders with TSMC. 💡

Related Links:

• China's Cyberspace Administration Summons NVIDIA Today, Citing H20 Chip's 'Vulnerability Backdoor Security Risks'
• News Behind the Scenes / AMD MI350 Poised to Overtake NVIDIA! Forming a Dual Dominance to Reshape the AI Landscape 🚀