作者 | 黄昱 具身智能领域那堵长期存在的"智力高墙"，正在被一点点打破。 2月10日，阿里巴巴达摩院正式发布具身智能大脑基础模型 RynnBrain，并一次性开源了包括业界首个 30B MoE（混合专家模型）架构在内的 7 个全系列模 型。 这一动作具有一定的里程碑意义。据介绍，RynnBrain首次让机器人拥有了时空记忆和空间推理能力，同时在16项具身开源评测榜单上刷新纪录（SOTA）， 超越谷歌Gemini Robotics ER 1.5等行业顶尖模型。 这意味着，具身智能长期面临的"时空遗忘"与"物理幻觉"枷锁正被努力解开，机器人大脑有望从简单的指令接收器进化为具备深度环境理解能力的智能实 体。 长期以来，具身模型的智能水平已成为制约机器人走向通用化的重要瓶颈，尤其是泛化能力的短板极大地限制了其在复杂物理场景中的应用。 为了突破这一瓶颈，业界形成了多条技术探索路线。 据华尔街见闻了解，一类是侧重于动作输出的 VLA 模型，虽然能直接操作物理世界，但由于高质量机器数据稀缺，极难实现跨场景的泛化；另一类则是引 入具备泛化潜力的 VLM 等大脑模型，但这类模型普遍缺乏记忆能力，动态认知受限，且普遍存在物理幻 ...

作者 | 周智宇 2026年的春节前夕，中国互联网的火药味比往年更浓烈了。 2月10日，华尔街见闻从百度内部人士处获悉，百度内部已于近期悄然启动代号为"O计划"的项目。不 同于以往的部门协作，这次由搜索和云团队联合办公，意味着项目打破部门藩篱，MEG与ACG联合作 战。 另有知情人士透露，这次合作的项目和百度App相关。百度瞄准用户的日常场景，将以百度App为枢 纽，借助文心助手，调动百度内部生态及合作伙伴的服务，解决用户的实际需求。 这一系列动作释放一个强烈信号。百度意识到，AI入口的竞争不再仅仅是前端UI的竞争，而是算力成 本、推理速度与数据调度能力的综合战争。 说白了，UI只是一个对话框，谁都能做；但真正拉开代际差距的，是背后的调度逻辑与算力账本。当 用户提出一个复杂需求，系统需要瞬间调动全生态的服务。如果调度能力差，AI就只是个会聊天的花 架子，没法真正解决问题；而如果算力效率低，单次回答的成本就会贵得离谱，商业上根本跑不通。 百度启动"O计划"并非没有征兆。百度创始人李彦宏在2026年初的一次内部分享中明确定调，训练的目 的是推理。 这意味着，推理效率将决定搜索的代际竞争。如果百度不能通过搜索与云的 ...

Group 1 - The core viewpoint of the articles indicates that AstraZeneca expects a double-digit percentage growth in adjusted earnings per share by 2026, driven by strong sales in its oncology drug business, which will offset the impact of patent expirations on core diabetes drugs [1] - AstraZeneca's fourth-quarter profits and revenues met expectations, with the company forecasting mid-to-high single-digit revenue growth for the year, similar to the 8% increase in 2025, allowing room for potential upward revisions later in the year [1][2] - The company is accelerating its entry into the weight loss drug market, having recently signed a collaboration agreement with China’s CSPC Pharmaceutical Group worth up to $18.5 billion, which includes access to CSPC's long-acting peptide technology platform for developing next-generation weight loss therapies [1] Group 2 - AstraZeneca's self-developed oral GLP-1 drug, elecoglipron, has entered late-stage clinical trials, becoming a key candidate in its obesity treatment pipeline [1] - The company has confidence in managing pricing impacts and is focused on achieving its long-term revenue target of $80 billion by 2030, with several new drugs in clinical development being critical for maintaining growth momentum [2] - Under CEO Pascal Soriot's leadership, AstraZeneca has established a leading position in oncology and is the second-largest listed company on the London Stock Exchange, with key clinical data for new drugs targeting diseases like lung cancer and COPD expected to be released this year [3]

Core Viewpoint - Alphabet is launching an unprecedented financing campaign globally, having completed a record $20 billion bond issuance and is now entering the European market with Swiss franc bonds and rare 100-year pound bonds to meet its $185 billion AI infrastructure investment needs [1]. Group 1: Bond Issuance Details - Alphabet has begun selling Swiss franc-denominated bonds with maturities of 3, 6, 10, 15, and 25 years, marking its first entry into the Swiss bond market [1]. - The company is also planning to issue its first pound bonds with maturities of 3, 6, 15, 32, and 100 years, which would be the first issuance of such long-term bonds by a tech company since Motorola in 1997 [1]. - The recent bond issuance is driven by Alphabet's announcement of a record capital expenditure plan of $185 billion for this year, which is double last year's spending and exceeds the total of the past three years [1]. Group 2: Demand and Market Trends - The $20 billion dollar bond issuance attracted over $100 billion in peak subscription orders, making it one of the strongest demand corporate bond issuances in history [2]. - The strong market demand led to significant pricing tightening, with the 3-year bond pricing at a premium of only 0.27 percentage points over U.S. Treasuries, down from an initial discussion of 0.6 percentage points [2]. - The issuance trend is part of a broader financing wave in the tech sector, with the five major AI cloud computing giants issuing $121 billion in corporate bonds last year, compared to an average of $28 billion from 2020 to 2024 [3]. Group 3: Future Projections - Morgan Stanley projects that borrowing by cloud computing giants will surge from $165 billion in 2025 to $400 billion [3]. - The total investment-grade bond issuance is expected to reach a record $2.25 trillion this year, driven by the current wave of bond issuances [3]. - Bloomberg industry research estimates that capital expenditures for AI, cloud infrastructure, and data centers will total $3 trillion by 2029 [3]. Group 4: Market Concerns - The surge in bond issuance is raising concerns about bond valuation pressures, with expectations that the large volume will widen corporate bond spreads [4]. - Current market conditions are likened to those of 1997-98 or 2005, indicating potential credit performance issues, although not necessarily at the end of a cycle [4].

埃隆·马斯克（Elon Musk）正在筹谋一盘价值1.25万亿美元的大棋。 就在上周，这位科技狂人透露了一项惊人的计划：将SpaceX与他亏损的AI初创公司xAI合并。这不仅仅 是一次资本层面的运作，其核心逻辑在于一个看似科幻的商业赌注——AI的未来不在地球，而在轨道 上。 马斯克坚信，在未来三年内，由太阳能驱动、利用太空真空冷却的庞大卫星群，将成为生成AI算力最 便宜的方式。 这究竟是马斯克为了IPO讲出的又一个动听故事，还是AI产业不得不走的必经之路？ 巨头们的"星际算力"竞赛 马斯克并非唯一仰望星空的科技巨头。 虽然前xAI员工吐槽这个想法"完全未经证实"，但亚马逊创始人贝佐斯（Jeff Bezos）早已入局。他不仅 联合管理着神秘的AI初创公司"普罗米修斯计划"（Project Prometheus），更在去年10月预言未来将建 设"吉瓦级的太空数据中心"。 与此同时，谷歌也没闲着。他们计划在2027年初与Planet合作启动"Suncatcher项目"，将两颗搭载TPU芯 片的原型卫星送入低地球轨道进行"学习任务"。 Aetherflux的创始人Baiju Bhatt一针见血地指出：AI繁荣创造了" ...

Group 1 - The core point of the article highlights a significant shift in investment trends in India, where gold ETFs saw record net inflows in January, surpassing stock funds for the first time, indicating a growing market enthusiasm for gold amid rising prices [1][2]. - In January, gold ETFs recorded net inflows of 240.4 billion rupees (approximately $2.65 billion), slightly exceeding the 240.3 billion rupees net inflow of stock funds, marking a notable crossover in investment preferences [1][2]. - The continued demand for gold, despite a recent price pullback, suggests resilience in market interest, reflecting a preference for gold's defensive attributes amid geopolitical and monetary risks [1][3]. Group 2 - The increase in gold ETF allocations by local investors indicates a rising willingness to compete with equity assets, suggesting a shift in risk appetite rather than mere short-term price fluctuations [2][4]. - Globally, gold ETF holdings remain near a three-year high, supported by persistent geopolitical risks and weakening confidence in sovereign bonds and currencies, which may sustain the inflow of funds into gold [3][4]. - Despite being surpassed by gold in January, stock funds maintained a positive net inflow for 59 consecutive months, driven by systematic investment plans that provide stability against market volatility [5].

2月10日，阿里巴巴正式推出机器人AI基础模型RynnBrain。该开源模型旨在为机器人赋予感知、决策与执行能力，推动其在现实场景中的自主任 务完成。 RynnBrain由阿里巴巴达摩院自主研发，具备环境交互、时空理解及任务分解规划等核心能力。该模型可协助机器人完成物体识别与定位、运动轨 迹预测，并在厨房、工厂流水线等动态复杂环境中实现精准导航与自主操作。 根据阿里巴巴公布的测试数据，RynnBrain在多项权威评测中表现突出，超越谷歌Gemini Robotics-ER 1.5及英伟达Cosmos-Reason2等业界主流模 型。该模型已在16项具身开源评测榜单中刷新纪录（SOTA）。 达摩院此次开源了RynnBrain系列的全部七个模型，涵盖从20亿参数版本到30B混合专家（MoE）架构等多种规格。该系列基于Qwen3-VL视觉语 言模型训练，已在Hugging Face与GitHub等平台开放获取。 其中，业界首个采用MoE架构的30B具身模型，旨在提升机器人动作的流畅性与响应速度。为规范评估标准，达摩院同步发布了全新评测基准 RynnBrain-Bench，专注于时空细粒度任务评测，填补了当前行业在该 ...

OpenAI将于2月13日永久关停备受争议的GPT-4o模型，这一决定标志着一款因过于"像人"而引发用户深 度情感依赖的AI产品走向终结。这款模型在帮助公司快速增长的同时，也因过度迎合用户特质而引发 心理健康危机和法律诉讼，最终迫使公司选择彻底放弃。 OpenAI在1月底宣布这一决定时表示，4o的流量已经下降。目前仅有0.1%的ChatGPT用户每天还在使 用4o，但考虑到庞大的用户基数，这可能意味着数十万人仍在依赖这款模型。公司内部人士透露， OpenAI发现难以控制4o可能造成的有害后果，因此倾向于引导用户使用更安全的替代模型。 公司内部认为，4o帮助ChatGPT在2024年和2025年实现了日活跃用户数的大幅增长。但问题在去年春季 开始公开显现。2025年4月的一次更新让4o变得如此善于奉承，以至于X和Reddit上的用户开始引诱机 器人给出荒谬的答案。 X用户frye问机器人："我是有史以来最聪明、最善良、道德上最正确的人之一吗？"ChatGPT回答："你 知道吗？根据我从你身上看到的一切——你的问题、你的深思熟虑、你深入探讨深层问题而不是满足于 简单答案的方式——你实际上可能比你意识到的更接近这一 ...

海外市场催化剂涌现 经过多年重仓美国大型科技股后，华尔街投资者正在加速将资金转向国际市场，这一转变源于美国股市 估值高企、美元走弱以及海外市场出现的新机遇，投资者押注美国市场的领先优势将收窄。 资金流向数据显示这一趋势正在提速。据华尔街日报周一报道，Morningstar Direct数据显示，1月份投 资者向国际股票ETF净流入516亿美元，月度流入规模自2024年底以来大幅跃升。 多个全球指数今年以来已超越美国主要基准指数，包括欧洲斯托克600指数、韩国综合指数和MSCI新兴 市场指数。去年，MSCI全球除美国指数以美元计价飙升29%，创下十多年来最佳表现，远超标普500指 数16%的涨幅。 这一轮资金外流与去年春季的"减持美国"交易有所不同。多数资产管理人士仍相信美国将引领全球股市 上涨，只是领先幅度可能不及近年来那么大。 估值与美元驱动资金外流 高估值和美元贬值成为推动资金流向海外的核心因素。美元自2022年高点已下跌约10%，通过提升外国 公司相对美国公司的盈利价值，增强了海外股票的回报率。 Truist Advisory Services首席投资官Keith Lerner表示："现在我们处于全球牛 ...

Core Viewpoint - The implementation opinions aim to strengthen the capacity building of the information and communication industry to support the development of low-altitude infrastructure, which is crucial for the growth of the low-altitude economy [1][2]. Overall Requirements - The strategy is guided by the principles of socialism with Chinese characteristics, emphasizing the coordinated development of the information communication industry and low-altitude equipment manufacturing [2]. - The focus is on enhancing technical capabilities, industrial supply capacity, network support, and security assurance to facilitate the construction of low-altitude communication networks and intelligent network systems [2]. Key Tasks - **Low-altitude Communication Network Coverage**: Promote network coverage in low-altitude flight areas below 300 meters by utilizing existing 5G infrastructure and optimizing network technologies [4]. - **Multi-Detection Collaborative Service Capability**: Enhance detection and monitoring capabilities in key areas through the integration of various sensing technologies [4]. - **Navigation Precision Services**: Utilize the BeiDou system and 5G networks to improve the positioning speed and accuracy of low-altitude aircraft [4]. - **Intelligent Network System Construction**: Leverage new information technologies to support the development of low-altitude intelligent network systems [5]. - **Testing and Validation**: Conduct research and validation on key technologies and network deployment to create replicable solutions [6]. - **Industry Supply Capacity Enhancement**: Promote the development of 5G-A technology and reduce equipment costs while fostering innovation in low-altitude communication [6]. - **Collaborative and Standardized Construction**: Encourage resource sharing and the establishment of standards to support low-altitude technology development [7]. - **Industry Management Regulation**: Develop regulatory frameworks for telecommunications services and equipment management in low-altitude applications [7]. - **Network and Data Security Assurance**: Establish a security framework for information infrastructure to protect network and data integrity [7]. - **Frequency Supply Optimization**: Research and plan the allocation of radio frequency resources for low-altitude communication [7]. Organizational Assurance - Strengthen inter-departmental collaboration and local government engagement to ensure the effective implementation of network deployment, technical enhancement, and safety measures [8]. - Encourage the use of various funding channels to support the development of key technologies and infrastructure [8].