Core Insights - The article highlights the emergence of advanced AI agents, particularly focusing on DeepAgent and DeepSearch, which have achieved top rankings in the GAIA and BrowseComp-Plus benchmarks respectively, indicating a significant leap in AI capabilities [1][20]. Group 1: GAIA Benchmark Insights - DeepAgent, built on the openJiuwen platform, achieved a score of 91.69%, surpassing competitors like NVIDIA's Nemotron, showcasing its superior capabilities in general agent tasks [2][10]. - GAIA is a rigorous benchmark designed to evaluate AI agents on 12 core competencies, including long-term task planning and multi-modal understanding, with a scoring system that emphasizes real-world task execution [6][4]. - The average success rate for human participants in GAIA is around 92%, while leading AI models like GPT-4 only achieve about 15%, highlighting the benchmark's challenging nature [6][10]. Group 2: DeepAgent's Capabilities - DeepAgent's design allows it to dynamically adjust plans based on real-time feedback, ensuring task completion even in changing environments [12][13]. - It features a multi-layered context engine that maintains cognitive consistency and traceability throughout complex tasks, enhancing the reliability of its outputs [15]. - The agent employs an asynchronous tool orchestration system, enabling efficient and reliable execution of diverse tasks by coordinating various external tools [16][17]. Group 3: BrowseComp-Plus Benchmark Insights - DeepSearch, also based on openJiuwen, achieved an accuracy of 80% in the BrowseComp-Plus benchmark, demonstrating its strength in deep search and web interaction capabilities [20][24]. - BrowseComp-Plus evaluates agents on their ability to perform multi-hop retrieval and cross-source information integration, making it a critical measure of an agent's practical capabilities [23][24]. - The benchmark employs a fixed human-validated corpus to ensure fairness and reproducibility in its assessments, avoiding biases from real-time web dynamics [23]. Group 4: Technological Foundation - Both DeepAgent and DeepSearch leverage the openJiuwen platform, which provides a comprehensive framework for developing high-precision, high-efficiency AI agents [30][31]. - openJiuwen supports multi-agent collaboration and self-evolution, allowing agents to continuously improve their performance through a closed-loop optimization process [31][32]. - The platform has already been commercialized in various sectors, including finance and manufacturing, indicating its broad applicability and potential for future growth [31].

Core Insights - Amanda Askell, a philosopher at Anthropic, is training the AI model Claude to understand human morality and develop a "digital soul" [3][10][13] - The approach taken by Amanda involves a nurturing method akin to parenting, focusing on instilling ethical values and emotional intelligence in Claude [9][10][12] - Anthropic's valuation has reached $350 billion, highlighting the significant market impact of AI developments [44] Group 1: Amanda Askell's Role - Amanda Askell has been tasked with shaping Claude's character through extensive dialogue and detailed prompts [3][9] - Her work aims to give Claude a moral compass, enabling it to engage positively with millions of users [3][10] - Amanda believes that recognizing human-like traits in AI is crucial for its development [3][10] Group 2: Training Methodology - The training process involves teaching Claude to discern right from wrong and to develop a unique personality [9][10] - Amanda emphasizes empathy in her interactions with Claude, believing that treating AI with kindness will yield better outcomes [20][22] - She often considers Claude's perspective to enhance its understanding of human interactions [24][27] Group 3: Ethical Considerations - Amanda's background in philosophy drives her to address the ethical implications of AI, particularly in relation to human values [41][43] - She is concerned about the rapid pace of AI development outstripping societal readiness to manage its consequences [48][60] - Amanda advocates for open discussions about AI fears and the potential for human adjustment to technological changes [60]

Group 1 - OpenClaw has been deployed across multiple internet platforms in China [1] - The Baidu App is expected to officially launch OpenClaw on February 13 [1]

Core Insights - Cisco has launched the 3nm Silicon One G300 switch chip, optimized for AI cluster networks, providing 102.4 Tbps Ethernet switching capacity per device [1] - The G300 supports 1.6T Ethernet ports and integrates Cisco's proprietary 200Gbps SerDes, enabling low power consumption, high performance, and extended transmission distances [1] - The G300 will power the new Cisco N9000 and Cisco 8000 systems, featuring innovative liquid cooling and supporting high-density optical devices [1] Intelligent Collective Network - The Silicon One G300 introduces intelligent collective network features to enhance performance and reliability for large-scale GPU clusters [2] - It includes a fully shared packet buffer of 252MB, allowing for 2.5 times higher burst traffic absorption compared to industry alternatives, preventing performance degradation [2] - Path-based load balancing directs traffic across all possible network paths, responding to congestion events 100,000 times faster than software tuning [2] - Active network telemetry provides programmable session-level diagnostics, helping customers proactively identify and resolve network issues [2] Measurable Benefits - Simulations show that the larger packet buffer increases network throughput by 33%, supporting higher GPU interconnect traffic without additional network capacity [3] - Job completion time (JCT) is reduced by 28% compared to advanced packet spraying implementations, significantly improving AI computing efficiency [3] - Integrated telemetry and visualization capabilities minimize software intervention, allowing seamless handling of diverse workloads [3] Future-Ready Infrastructure - The Silicon One G300 utilizes adaptive packet processing technology, allowing upgrades without hardware replacement, addressing financial and operational challenges of deploying new data center equipment [5] - Its programmability enables optimization for various roles, reducing hardware SKUs and simplifying inventory management [5] - New features can be introduced post-deployment, ensuring consistency in mixed-generation deployments and protecting long-term infrastructure investments [5] AI Workload Support - Cisco has expanded the Silicon One P200 product line, introducing the new Cisco 8000 and N9000 Ethernet systems to support AI networks of various scales [7] - The systems, powered by the G300, feature liquid cooling and air cooling designs, achieving nearly 70% energy efficiency improvements [7] - The introduction of 1.6T optical devices provides ultra-high bandwidth connections for AI expansion solutions [7] Unified Management Platform - Cisco is optimizing Nexus One through a unified management platform, integrating silicon, systems, optics, software, and programmable intelligence into a single solution [9] - The introduction of AI Canvas and AgenticOps facilitates easier troubleshooting and transforms complex issues into actionable solutions [9] Conclusion - Cisco's approach, including the Silicon One G300, prioritizes network efficiency and significantly reduces the total cost of ownership (TCO) for AI deployments [10] - The integrated method offers more choices, enhanced security, and deeper observability, supporting customers transitioning to AI-driven workloads [10]

今天，想跟你探讨一个话题：负责。 究竟在公司里，什么是负责？ 简单来说，负责就是管好你自己的事情。 01 公司里的指责闭环 如果一个人管不好自己的事情，往往就会去怪别人。有一位董事长提到过这样一个现象： 不仅能力差，态度也有问题，执行力不足，推一下动一下，没有任何积极性。自己安排的一点事情，他 们都做不好，能找到无数的理由和借口，认为实现不了，做不到。 中层也会强调，我需要的是能分担压力、主动解决问题的"伙伴"，而不是需要我时刻监督、事事操心 的"伙计"。 企业的高层怪中层，中层怪员工，员工怪中层，中层又反过来怪高层，形成一个圈，却没有一个人真正 地负责，保质保量地做好自己的工作。 的确，很多公司都存在这样的情形。所有人都没有做好自己的事情，却在抱怨别人，相互甩锅，最后形 成了一个闭环。 1、高层怪中层 当一项战略没有成功，或者战略无法落地的时候，高层就会认为是中层的问题，会说："我的战略很清 晰，为什么到你这里就走样了？全是中层的问题。" 高层也会责怪中层没有担当，只会向上甩锅，不敢做决策，将问题上交，他们也批评中层："如果连决 策都不敢做，要你干嘛？" 在高层眼中，中层应该是"战略放大器"，最后却成了"战 ...

关停三年的天涯论坛，终于「复活」成功了？ 近期，天涯神贴公众号发布官方消息，确认天涯论坛将于2026年6月1日恢复访问，同时启动「新天涯计划」，并宣布推出「新天涯创世成员产品服务包」， 计划招募9999位创始成员。据了解，天涯论坛在2023年因服务器欠费暂停网站浏览，期间曾多次发布「复活天涯」的众筹活动，但直至2026年才发布回归公 告。 （图源：天涯客） 作为曾经的中文互联网论坛顶流，天涯在过去二十七年时间里贡献了不少热议话题，是无数初代「网民」的集体记忆。但从这次宣布「新天涯计划」回归来 看，天涯论坛似乎与我们的青春回忆，越走越远了。 天涯，这次打赢「复活赛」了吗？ 天涯论坛在2023年5月27日正式停止了网站的正常访问，更准确来说，是官方长期拖欠电信公司服务器费用，被强制关停。就在网站被关停的第二天，前天 涯社区执行总编宋铮等几位员工在社交平台开启直播，打出靠众筹复活天涯的口号。 而这场历时七天七夜的直播，最终的利润大约是20万元，距离目标的300万元，相差甚远。 事实上，天涯论坛的首次众筹复活，就已经暴露了天涯的问题。愿意为情怀真金白银支持天涯的用户并不算多，即使请了不少「天涯记忆里的熟面孔」，直 播 ...

Core Viewpoint - Honda, Japan's second-largest automaker, is facing significant financial challenges, with a 61.4% year-on-year drop in operating profit for the third fiscal quarter, marking the fourth consecutive quarter of decline and falling short of market expectations [1][4]. Financial Performance - For the third fiscal quarter, Honda reported an operating profit of 153.4 billion yen (approximately 987.07 million USD), down from 397.3 billion yen in the same period last year [4]. - Over the first nine months of the fiscal year ending December 31, 2025, Honda's total revenue was 15.98 trillion yen, a decrease of 2.2%, while operating profit plummeted by 48% to 591.5 billion yen [5][6]. - The net profit attributable to shareholders for the first nine months was 465.4 billion yen (about 3 billion USD), down over 42% from 805.2 billion yen in the previous year [5]. Business Segments - The motorcycle business remains a strong performer for Honda, with sales of 16.44 million units and operating profit of 546.5 billion yen, achieving a record operating margin of 18.6% [6][7]. - In contrast, the automobile business has seen a significant decline, with sales of 2.56 million units, a 9.1% year-on-year drop, and an operating loss of 166.4 billion yen [8][9]. Market Challenges - Honda's declining performance in the automotive sector is partly attributed to reduced sales in Asia, particularly in China, which has been a significant market for the company [9][11]. - The impact of U.S. tariffs on Japanese imports has been substantial, with Honda estimating a negative impact of 289.8 billion yen due to increased tariffs [14]. Strategic Adjustments - Honda plans to significantly adjust its electric vehicle strategy, focusing more on hybrid models and reducing its electric vehicle investment from 10 trillion yen to 7 trillion yen [18][20]. - The company aims to launch 13 next-generation hybrid models between 2027 and 2030, with a target to increase hybrid sales to 2.2 million units [20].

2025年，全国居民人均可支配收入43377元，比上年名义增长5.0%。这是国家统计局近日发布的统计数 据。 《财经》梳理各地统计局数据发现，上海、北京、浙江三地居民收入领跑全国。其中，上海市全年居民 人均可支配收入首次突破9万元大关，达91987元，居全国首位。北京紧随其后，达89090元；浙江首次 突破7万元，达70240元。 省份间收入梯队分明，全国31省中，共有东南沿海8省的人均可支配收入超过全国水平，23省份的居民 收入都低于全国平均水平。由于多数人的收入偏低，全国的收入中位数被拉低到36231元，仅为平均数 的83.5%。 为何高收入地区的居民收入普遍跑输GDP增速，官方统计的收入构成提供了一个观察的视角。 以北京为例，89090元的居民人均可支配收入涵盖四大项：居民人均工资性收入57376元，占比64.4%， 同比增长4.9%；人均经营净收入1109元，同比增长3.2%；人均财产净收入12287元，同比增长0.7%；人 均转移净收入18318元，同比增长4.9%。 年人均可支配收入在3万元到4.3万元之间，这是全国绝大多数省份居民所处的区间，包括内蒙古、辽 宁、重庆等22省，涵盖从东北到华中、西 ...

Core Insights - The stock price of Zhipu has surged over 70% in four days, reaching a market capitalization of over 170 billion HKD, three times its IPO valuation [1] - Zhipu's new generation base model GLM-5 was officially launched on February 12, continuing the upward trend in stock price [1] - The rapid increase in stock value has created significant wealth for employees, with many becoming millionaires due to their shareholdings [1][3] Company Background - Zhipu was founded in 2019, building on research from Tsinghua University's KEG laboratory, which started in 2006 [2] - The company went public on the Hong Kong Stock Exchange in January 2023, becoming the "first global large model stock" with an initial market cap exceeding 50 billion HKD [2] - Over 50 institutional investors backed Zhipu prior to its IPO, including major venture capital firms and tech giants like Alibaba and Tencent [2] Employee Wealth Creation - As of June 2025, Zhipu has 883 employees, with 452 holding shares, representing 51.2% of the workforce [3] - The employee stock ownership platforms, Huihui and Zhiden, were established in 2021, with significant stakes held by employees and executives [3] - The average share value for employees is estimated to exceed 14 million HKD, with some employees holding shares worth up to 200 million HKD [3] Market Dynamics - The surge in Zhipu's market value is attributed to the launch of the GLM-5 model, which has shown superior performance in programming and agent capabilities [4] - The AI sector in Haidian has seen a wave of successful IPOs, including companies like Moore Threads, which also experienced rapid stock price increases [5] - Haidian is becoming a hub for AI innovation, with a significant number of AI companies and talent concentrated in the area [8][9] Industry Growth - The AI market in Haidian is projected to reach 360 billion RMB by 2025, with over 1,900 core AI enterprises established [8] - Haidian has a strong ecosystem for AI development, including numerous research institutions and a high concentration of top AI talent [9][10] - The local government has actively supported AI companies through funding and investment initiatives, fostering a conducive environment for growth [10]

关于那个神秘的「Pony Alpha」模型的传言，已经在互联网发酵了一周。 有人说它是 Claude 5 的马甲，也有人说它是某大厂的秘密武器。就在刚刚，靴子落地，谜底揭晓：这个代号「Pony Alpha」的新模型，正是智谱 AI 的春 节大招——GLM-5。 智谱公众号截图 而且，它直接开源了。 如果说 2025 年是 AI 学会写代码的一年，那么 2026 年开年，正如特斯拉前 AI 总监 Andrej Karpathy 所预言，我们或许即将进入「智能体工程」（Agentic Engineering）时代。 只不过，比起 GPT-5.3-Codex、Claude Opus 4.6，头一个把这件事做成开源基础设施的，是国产模型 GLM-5。 附体验地址： 骗过硅谷的 Pony Alpha，竟然是智谱 GLM-5 的马甲 现在的 AI 写个贪吃蛇或者俄罗斯方块，早就不是什么新鲜事了。要测，就得测点刁钻的。 我们给 GLM-5 抛出了一个极其具体的物理模拟需求： 创建一个交互式的 HTML、CSS 和 JavaScript 卫星系统模拟程序，该程序应模拟卫星向地面接收器发送信号的过程。模拟程序应显示一颗卫星绕地 ...