AI 新闻日报 2025-08-23：全球与中国 AI 头条 Top 10

**MIT Study Reveals Low ROI on AI Investments:** A recent MIT study analyzing 300 public AI initiatives found that 95% of organizations saw no return on their AI investments, despite significant spending.

**AI Thought Leader Warns of Catastrophe:** Carnegie Mellon's Anand Rao has cautioned that unchecked global AI investment could lead to societal catastrophe, citing risks like worker displacement, increased emissions, and eroded trust.

**Meta Restructures AI Unit Again:** Meta is undergoing its fourth AI restructuring in six months, splitting its Superintelligence Labs into four distinct groups as CEO Mark Zuckerberg intensifies efforts toward artificial general intelligence.

**Google Debuts Pixel 10 with AI Coaches:** Google has unveiled its Pixel 10 smartphone lineup, featuring new AI capabilities including a photography coach and a health coach integrated into Fitbit Premium.

**Cohere Secures $500 Million in Funding:** Enterprise generative AI unicorn Cohere has successfully raised $500 million in a new funding round to expand its AI platform for large organizations.

**Broadcom Ships New AI Data Center Chip:** Broadcom has begun shipping a next-generation AI chip designed to enable ultra-fast connectivity between GPUs in data centers, a critical upgrade for scaling AI infrastructure.

**Xiaomi Unveils Advanced AI Voice Model:** Xiaomi has announced a new AI voice model optimized for in-car and smart home experiences, featuring faster response times, offline capabilities, and context-aware control.

**OpenAI Pursues Record Valuation:** OpenAI is reportedly in talks for a stock sale that could give it the world's largest valuation for a private company.

**Education Systems Need AI Overhaul:** Experts suggest that A-levels and GCSEs require a significant overhaul to adapt to the rapid advancements in generative AI.

**Meta and Google Cloud Ink $10 Billion AI Deal:** Meta has signed a six-year, US$10 billion cloud computing partnership with Google Cloud to bolster its AI infrastructure through enhanced servers, storage, and networking capacity.

**麻省理工学院研究揭示AI投资回报率低：** 麻省理工学院最近一项针对300个公共AI项目的研究发现，尽管投入巨大，但95%的机构在AI投资上未能获得回报。

**AI思想领袖警告全球竞赛恐酿灾难：** 卡内基梅隆大学的Anand Rao警告称，不受限制的全球AI投资可能导致社会灾难，理由是工人流失、排放增加和信任受损等风险。

**Meta再次重组AI部门：** Meta在六个月内第四次重组其AI部门，将其超级智能实验室拆分为四个不同的团队，首席执行官马克·扎克伯格正在加紧推进通用人工智能的开发。

**谷歌发布Pixel 10手机，内置AI教练：** 谷歌推出了Pixel 10智能手机系列，新增了AI功能，包括摄影教练和集成到Fitbit Premium中的健康教练。

**Cohere获5亿美元融资：** 企业级生成式AI独角兽Cohere成功在新一轮融资中筹集了5亿美元，以扩大其面向大型企业的AI平台。

**博通出货新型AI数据中心芯片：** 博通已开始出货下一代AI芯片，旨在实现数据中心GPU之间超高速连接，这是扩展AI基础设施的关键升级。

**小米发布先进AI语音模型：** 小米宣布了一款针对车载和智能家居体验优化设计的新型AI语音模型，该模型具有更快的响应时间、离线功能和情境感知语音控制。

**OpenAI寻求创纪录估值：** 据报道，OpenAI正在就股票出售进行谈判，这可能使其成为估值最高的私营公司。

**教育系统需进行AI改革：** 专家建议，A-level和GCSE考试需要进行重大改革，以适应生成式AI的快速发展。

**Meta与Google Cloud达成100亿美元AI协议：** Meta与Google Cloud签署了一项为期六年、价值100亿美元的云计算合作协议，通过增强服务器、存储和网络容量来加强其AI基础设施。

**DeepSeek Hints at China's Next-Gen AI Chips:** Chinese AI startup DeepSeek has indicated that China is on the verge of unveiling its own "next generation" chips for AI stacking, fueling speculation about domestic breakthroughs.

**ByteDance and DeepSeek Unveil New AI Models:** Chinese tech giants ByteDance and DeepSeek have announced new AI models, with ByteDance releasing three variants of its Seed-OSS-36B and DeepSeek upgrading its V3 model, challenging Western AI dominance.

**DeepSeek V3.1 Optimized for Chinese Chips:** DeepSeek's newly released V3.1 model has been specifically designed "for the home-grown chips to be released soon," suggesting crucial hardware-software coordination within China.

**China Advances Towards AI Self-Sufficiency:** Disclosures hint at China's significant progress in building a self-sufficient AI stack with domestic technologies, potentially mitigating the impact of US chip export restrictions.

**Chinese AI Firms Embrace Open-Source Strategy:** Most Chinese AI models are open source, reflecting a strategy by Chinese AI firms to drive broader adoption and close the gap with their US rivals.

**Alibaba and Baidu Release New AI Iterations:** Tech heavyweights Alibaba and Baidu continue to advance the Chinese AI industry by releasing several iterations of their Qwen and Ernie AI models, respectively.

**Nvidia Reportedly Developing New AI Chip for China:** Nvidia is reportedly preparing a new AI chip, tentatively named B30A, for the Chinese market, which is expected to outperform its current China-specific H20 model.

**Chinese Authorities Discourage Nvidia H20 Chip Use:** Chinese authorities have reportedly discouraged local companies from deploying Nvidia's H20 chip and instructed major firms to halt purchases for national security reasons.

**China Leads AI-Fueled Data Center Boom:** China, alongside the US, is at the forefront of a global surge in AI-fueled data center construction, with AI projected to drive a US$6.7 trillion global spending boom by 2030.

**Hangzhou Startups Aim for Global AI Dominance:** A cluster of elite AI startups in Hangzhou, dubbed the "six dragons," are working intensely to compete with Silicon Valley and achieve global AI dominance.

**DeepSeek暗示中国即将推出下一代AI芯片：** 中国人工智能初创公司DeepSeek表示，中国即将推出自己的“下一代”AI堆栈芯片，引发了对国内技术突破的猜测。

**字节跳动和DeepSeek发布新型AI模型：** 中国科技巨头字节跳动和DeepSeek宣布推出新型AI模型，其中字节跳动发布了Seed-OSS-36B的三个变体，DeepSeek升级了其V3模型，挑战了西方AI的主导地位。

**DeepSeek V3.1针对国产芯片优化：** DeepSeek最新发布的V3.1模型专门设计“用于即将发布的国产芯片”，表明中国国内在软硬件协同方面取得了关键进展。

**中国迈向AI自给自足：** 相关披露表明，中国在构建包含国内技术的自给自足AI堆栈方面取得了重大进展，这可能有助于减轻美国芯片出口限制的影响。

**中国AI企业拥抱开源策略：** 大多数中国AI模型都是开源的，这反映了中国AI企业旨在推动更广泛的应用并缩小与美国竞争对手差距的策略。

**阿里巴巴和百度发布AI新版本：** 科技巨头阿里巴巴和百度通过发布其Qwen和Ernie AI模型的多个迭代，继续推动中国AI产业的发展。

**英伟达据报道为中国市场开发新型AI芯片：** 据报道，英伟达正在为中国市场准备一款代号为B30A的新型AI芯片，预计其性能将超越目前针对中国市场的H20型号。

**中国当局不鼓励使用英伟达H20芯片：** 据报道，中国当局已不鼓励国内公司部署英伟达的H20芯片，并指示主要企业出于国家安全原因暂停采购。

**中国引领AI驱动的数据中心建设热潮：** 中国与美国一道，正处于全球AI驱动数据中心建设热潮的最前沿，预计到2030年，AI将推动全球数据中心支出激增6.7万亿美元。

**杭州初创公司力争全球AI主导地位：** 杭州的一批被称为“六大金刚”的顶尖AI初创公司正在加紧工作，力求超越硅谷，实现全球AI领域的主导地位。