他是中国AI 1.0时代的拓荒者，在深度学习黎明前的黑夜，成为全中国第一个复现AlexNet （深度卷积神经网络模型） 、拿到AI"黄金榔头"的人。 他是 地平线创始团队成员及前智能驾驶总裁 ，却在公司2024年成功上市的高光时刻，转身闯进具身智能的无人区，去验证物理世界的Scaling Law。 他是 Vbot 维他动力 的创始人， 用60人团队 一年量产机器狗，首款 智能 四足机器人 产品发布即引爆，截止1月10日，预订订单量已经达到了6540台，并实现了 以C 端 带 动B端 的销量破局。 他是余轶南博士，Vbot 维他动力创始人兼CEO，他要打造全球领先的消费级机器人产品。 本期「善友探索流」播客，我们邀请到 Vbot 维他动力创始人兼CEO，同时也是混沌学园六期同学——余轶南博士。 在本场深度对谈中，余轶南博士贡献了极具 颠覆性的 三大商业洞见 ： 关于生存哲学 ：硬科技创业不是赌博，而是基于认知的精准判断。只有"锁死下限"，才敢去赌那个"无限的上限"。 关于公司本质： AI公司V.S.科技公司，区别不在技术，而在是否追求Scaling Law（尺度定律）带来的无限增长。 关于组织终局 ：他用了一个 ...

Core Insights - Yang Zhilin, founder and CEO of Moon's Dark Side, has been invited as a keynote speaker at NVIDIA's GTC 2026, marking a significant recognition for him and his Kimi models [1] - The invitation reflects NVIDIA's strategy of identifying emerging trends and endorsing key players in the AI industry [6][7] - Yang's presence at GTC indicates a shift in focus towards the need for advanced reasoning capabilities in AI, moving beyond mere computational power [14] Group 1: Yang Zhilin and Kimi's Journey - Yang Zhilin represents a new pain point in AI development, as the industry seeks to move past the limitations of previous models and explore deeper cognitive capabilities [14] - Kimi faced significant challenges in 2025 due to competition from DeepSeek, which disrupted its business model and led to a decline in user engagement [15][17] - After a period of silence and strategic retreat, Kimi re-emerged with the launch of Kimi K2, showcasing advanced capabilities and reaffirming its position in the market [18][21] Group 2: Market Dynamics and Financial Position - Kimi's recent funding round raised $500 million, bringing its post-money valuation to $4.3 billion, indicating strong investor confidence in its technology and long-term vision [21] - The successful IPOs of competitors in the same sector signal a pivotal moment for AI startups, with Kimi positioned to leverage its cash reserves for continued research and development [21] - Yang Zhilin's focus on technology over marketing has allowed Kimi to regain traction and recognition in the AI community, particularly through its open-source initiatives [21]

量化看市场系列之四：使用"OpenClaw"搭建属 于自己的私域 AI 助理 摘要 OpenClaw（曾用名 ClawdBot、Moltbot）是一款由程序员 Peter Steinberger 开 发的开源 AI 智能体项目，其图标为龙虾形象，该项目曾用名 Clawdbot，因商 标争议于 2026 年 1 月 27 日更名为 Moltbot，其定位为具备长期记忆与软件操 作能力的个人 AI 代理。它如同数字管家，可帮助用户完成处理邮件、编写代 码、发布社交媒体、自动购物等一系列任务。截至本研报发布前，其最新更名 为 OpenClaw。 金融工程 证 券 研 究 报 告 【点评报告】 该项目于 2026 年 1 月发布，发布后迅速在 GitHub 上获得关注，截至同年 1 月 26 日，星标数量已增至 20.7k。OpenClaw 对本地硬件环境友好，支持部署在 Mac、Windows、树莓派等设备上，大部分用户会选择使用 Mac mini 等设备运 行。 与常见的云端大模型不同，OpenClaw 完全运行于本地，拥有与用户等同的系 统权限。用户可通过聊天软件发送自然语言指令，OpenClaw 将在后台调用 ...

Core Insights - Yang Zhiling, founder and CEO of Moon's Dark Side, has been invited as a keynote speaker at NVIDIA's GTC 2026, marking a significant recognition for him and his Kimi models [1][2][27] - The invitation reflects NVIDIA's strategic foresight in identifying emerging trends within the AI industry, as Huang Renxun (NVIDIA's CEO) typically selects speakers who align with future market directions [7][11] Group 1: Yang Zhiling and Kimi's Journey - Yang Zhiling represents a new pain point in AI development, as the industry faces challenges with existing models and the need for innovative solutions [28][30] - Kimi faced significant challenges in 2025 due to the impact of DeepSeek, which threatened its business model and user engagement [33][35] - After a period of silence and strategic retreat, Kimi re-emerged with the launch of Kimi K2, showcasing advanced capabilities and reaffirming its position in the market [38][39] Group 2: Market Position and Financial Developments - Kimi K2.5 was launched in January 2026, demonstrating superior performance in various benchmarks compared to competitors like GPT-5.2 and Claude 4.5 Opus [41][42] - Kimi's successful C-round financing raised $500 million, leading to a post-money valuation of $4.3 billion, indicating strong investor confidence and financial stability [46] - The cash reserves exceeding 10 billion yuan position Kimi well for continued research and development, aiming for leadership in the global SOTA (state-of-the-art) landscape [46]

Core Viewpoint - The collaboration between the company and Volcano Engine has been evolving since 2024, focusing on the development of AI solutions for the smart automotive sector [2] Group 1: Partnership Development - The partnership began with the company's inclusion in the Volcano Engine Automotive Large Model Ecological Alliance [2] - A joint laboratory was established, and the company received HiAgent delivery authorization, indicating an upgrade in their collaboration [2] Group 2: Technological Advancements - The company leverages Volcano Engine's HiAgent and the Kouzi platform, combined with its own operating system technology, to create a comprehensive solution covering the entire chain [2] - In the smart automotive field, the company has launched an AI cockpit solution that enables real-time interaction between the vehicle and cloud brain, achieving a 500ms voice feedback response time and multimodal recognition and recommendation features [2] Group 3: User Experience Enhancement - The GUI Agent developed based on the Volcano Ark Mass platform can autonomously plan, reason, and execute UI interaction operations, significantly enhancing the cockpit interaction experience [2]

Core Insights - The article discusses the competitive landscape among major Chinese tech companies, particularly focusing on their advancements in AI and large language models, highlighting Tencent, Alibaba, and ByteDance's strategies and developments in this space [5][7][10]. Group 1: Tencent's Strategy - Tencent has been cautious in its AI investments over the past two years but began to change its approach in 2025 by hiring top talent and restructuring its AI teams, with the recruitment of Yao Shunyu marking a significant turning point [5][6][10]. - Yao Shunyu identified issues with Tencent's existing large language model, "Hunyuan," noting that its performance was compromised by data contamination from leaderboard-focused training [4][5]. - Tencent's "Yuanbao" AI product, initially lagging behind competitors, saw a significant increase in daily active users after integrating the DeepSeek model, achieving a tenfold growth in just one week [27][44]. Group 2: Alibaba's Approach - Alibaba introduced the concept of "Tongyun," integrating AI, cloud computing, and chip development, positioning itself as a full-stack AI company [7][10]. - Despite lower profitability compared to its competitors, Alibaba plans to increase its investment in AI infrastructure from 380 billion to 480 billion yuan over the next three years [7][10]. - The "Qianwen" app, previously known as "Tongyi," has become a central player in Alibaba's AI strategy, rapidly iterating and developing capabilities to compete with ByteDance's "Doubao" [54][62]. Group 3: ByteDance's Competitive Edge - ByteDance's "Doubao" became the first AI product in China to surpass 100 million daily active users by the end of 2025, showcasing its rapid growth and user engagement [9][33]. - The company has focused on aggressive investment in AI talent and infrastructure, aiming to become a global leader in large models, with a target to rank at least third worldwide [10][66]. - ByteDance's strategy includes leveraging its existing platforms like TikTok to enhance user acquisition for its AI products, demonstrating a strong synergy between its core business and new AI initiatives [9][59]. Group 4: Industry Dynamics - The competition among these tech giants is likened to a "Midway battle," where losing could mean losing the future, emphasizing the high stakes involved in AI development [16][17]. - The article highlights the importance of internal collaboration and resource sharing among different business units within these companies to enhance their AI capabilities [56][58]. - The talent war in AI is intensifying, with companies like Tencent actively recruiting from competitors to bolster their AI teams, reflecting the critical role of human resources in this technological race [66][70].

2月3日，港股早盘探底后小幅反弹，午后保持窄幅震荡走势，截止收盘，恒生指数涨0.22%报26834.77 点，恒生科技指数跌1.07%报5467.26点，国企指数跌0.3%报9053.11点，红筹指数涨1.06%报4290.3点。 盘面上，大型科技股普遍收跌，阿里巴巴跌1.41%，腾讯控股跌2.92%，京东集团跌0.72%，小米集团跌 1.31%，网易涨0.3%，美团跌1.74%，快手跌4.61%，哔哩哔哩跌2.54%；大模型概念股走强， MINIMAX-WP涨近11%，智谱涨超8%；黄金股反弹，万国黄金集团涨超13%，紫金黄金国际涨超7%， 赤峰黄金涨超7%，灵宝黄金涨超5%；东鹏饮料港股首日涨超1%。 企业资讯 云工场(02512.HK)：附属中标人工智能产业基地二期项目，合约价值为人民币5.2亿元。 中兴通讯(00763.HK)：拟出资1.17亿元认购建兴湛卢基金份额，可投资扶持更大范围的创新业务及上下 游产业链企业。 小鹏汽车(09868.HK)：1月共交付新车20011辆。 君圣泰医药（核心股）-B(02511.HK)：任命Filip Surmont博士为首席医学官，强化心肾代谢系统疾病 (CKM) ...

阶跃星辰在大模型研发与应用落地领域的快速发展，离不开上海完善的人工智能产业生态与国资力 量的强力支撑。今年以来，上海市人工智能企业轮番登陆资本市场，上海市政府主导的国家资本早已在 该赛道提前布局，且上海国资敢于在企业发展关键阶段扮演领投方角色，以此吸引更多社会资本入局， 形成合力共同支持硬科技企业发展。 目前，上海国资已在沪逐步构建起覆盖底层芯片算力、中间层模型工具、上层行业应用的完整AI 生态版图。 2月2日，阶跃星辰正式发布新一代开源Agent基座模型Step 3.5 Flash。该模型专为实时Agent工作流 场景打造，实现了推理速度、智能水平与使用成本的兼顾，在单请求代码类任务中，其最高推理速度可 达每秒350个token，为市场提供了"更快更强更稳"的Agent底层模型选择。 据悉，Step 3.5 Flash采用稀疏MoE架构设计，模型总计拥有1960亿参数，且每个token仅激活约110 亿个参数，在保障模型核心能力的同时，实现了推理效率的显著提升，也为Agent类应用打造了更高 效、更具性价比的底层模型方案。目前，华为昇腾、沐曦股份、壁仞科技、燧原科技、天数智芯、阿里 平头哥等多家头部芯片厂商 ...

格隆汇2月3日丨天溯计量(301449.SZ)在投资者互动平台表示，公司基于大模型的计量检测智能机器人 研发按计划稳步推进中。 ...

截至2026年2月3日收盘，科创芯片设计ETF天弘(589070)换手10.23%，成交6304.25万元，市场交投活跃。跟踪的上证科创 板芯片设计主题指数(950162)强势上涨2.19%，成分股新相微上涨9.49%，芯原股份上涨9.11%，普冉股份上涨8.24%，国芯 科技，中微半导等个股跟涨。 资金流入方面，科创芯片设计ETF天弘(589070)昨日（2月2日）获资金净流入6671万元。 【产品亮点】 【更多产品】 | 么 玩星 | | | | --- | --- | --- | | 159977 | 创业板ETF天弘 | 创业板 | | 589860 | 科创综指ETF天弘 | 科创板 | | 159360 111 8 | 中证A500ETF天弘 | A股风向标 | | 159128 | 港股科技ETF天弘 | T0+港股科技 | | --- | --- | --- | | 159281 | 港股通央企红利 | T0+高股息 | | | ETF天 54 | | | 159998 | 计算机ETF | AI算力+软件 | | 159770 | 机器人ETF | 机器人 | | 159310 | 芯片ET ...