Core Insights - Alibaba's personal AI assistant "Qianwen" is expanding into the AI hardware sector, planning to launch multiple AI hardware products globally this year [1][2] - The first product, an AI glasses named after Qianwen, will be unveiled at the 2026 Mobile World Congress in Barcelona, with pre-orders starting on March 2 [1][2] - Qianwen aims to integrate software and hardware across various devices, enhancing its ability to capture information from the physical world and understand user intent in complex scenarios [2] Product Launches - Qianwen will release not only AI glasses but also AI rings and AI headphones later this year, targeting the global market [2] - The AI assistant's capabilities, such as ordering food and hailing rides, will seamlessly connect to the new hardware devices [2] User Engagement - During the recent Spring Festival, Qianwen's app became a national-level AI assistant, with users placing nearly 200 million orders through the app, averaging one order for every ten people in the country [2] - The app's daily active users have surpassed 73 million, approaching the user base of Doubao [2]

Core Insights - A new reasoning framework for agents called DualPath has been introduced, which addresses I/O bottlenecks in long-text reasoning scenarios by optimizing the speed of loading KV-Cache from external storage [1][3]. Group 1: DualPath Framework - DualPath changes the traditional Storage-to-Prefill loading mode by introducing a second path, Storage-to-Decode, allowing for more efficient data handling [3][6]. - The framework utilizes idle storage network interface card (SNIC) bandwidth from the decoding engine (DE) to read caches and employs high-speed computing networks (RDMA) to transfer data to the prefill engine (PE), achieving global pooling of storage bandwidth and dynamic load balancing [3][13]. Group 2: Performance Improvements - In tests with a production-level model of 660 billion parameters, DualPath demonstrated a remarkable increase in offline inference throughput by 1.87 times and an average increase in online service throughput by 1.96 times [3][14]. - The framework significantly optimizes first token latency (TTFT) under high load while maintaining stable token generation speed (TPOT) [5][14]. Group 3: Technical Innovations - DualPath allows KV-Cache to be loaded into the decoding engine first, which is then transmitted to the prefill engine, alleviating bandwidth pressure on the prefill side [7][9]. - The architecture includes a central scheduler that dynamically allocates tasks based on I/O pressure and computational load, preventing congestion on any single network interface or computational resource [14][18]. Group 4: Research and Development - The first author of the paper, Wu Yongtong, is a PhD student at Peking University, focusing on system software and large model infrastructure, particularly in optimizing inference systems for large-scale deployment [15][16].

【算力闭环：即将发布的DeepSeek V4"海狮轻型版"引爆国产AI产业链的"价值重估"】 2026年新春，AI领域迎来双重变奏。 一边是DeepSeek V4"海狮轻型版"即将发布，以1M超长上下文窗口惊艳业界；另一边则是路透社爆出重磅消息： DeepSeek在V4的预发布阶段，未按惯例向英伟达、AMD开放，而是将早期访问权限独家授予了华为等国内供应 商。 这一"反常"举动，将国产大 【算力闭环：即将发布的DeepSeek V4"海狮轻型版"引爆国产AI产业链的"价值重估"】 2026年新春，AI领域迎来双重变奏。 一边是DeepSeek V4"海狮轻型版"即将发布，以1M超长上下文窗口惊艳业界；另一边则是路透社爆出重磅消息： DeepSeek在V4的预发布阶段，未按惯例向英伟达、AMD开放，而是将早期访问权限独家授予了华为等国内供应 商。 这一"反常"举动，将国产大模型与国产算力的协同推到了聚光灯下。 "海狮轻型版"的命名，似乎暗示着某种轻盈与敏捷。 DeepSeek选择华为，不仅是技术适配的选择，更是战略安全的考量。 此前，DeepSeek已在昇腾平台成功完成模型迁移，通过KernelCAT工具将推理 ...

Group 1 - Avalon GloboCare Corp. has entered into definitive agreements for the issuance and sale of 6,372,550 shares of common stock and associated warrants at a purchase price of $0.51 per share [1][3] - The offering is expected to generate approximately $3.25 million in gross proceeds before fees and expenses, with potential additional proceeds of about $6.5 million from the warrants if fully exercised [3] - The Series A-1 warrants will expire five years from stockholder approval, while the Series A-2 warrants will expire eighteen months from stockholder approval [1] Group 2 - H.C. Wainwright & Co. is acting as the exclusive placement agent for the offering [2] - The securities are offered in a private placement under Section 4(a)(2) of the Securities Act and are not registered under the Act or applicable state laws, being available only to accredited investors [4] - The company plans to use the net proceeds for repayment of outstanding debt and for working capital and general corporate purposes [3][4] Group 3 - Avalon GloboCare Corp. is focused on developing precision diagnostic consumer products and generative AI publishing and software [1][6] - The company is advancing AI systems through its subsidiary, Avalon Quantum AI LLC, and is also marketing the KetoAir™ breathalyzer device registered as a Class I medical device [6]

王阿姨在随园养老社区照顾老人，爱说爱笑。 直到马年春节大消息传来：阿里巴巴全新一代自研大模型——千问Qwen3.5-Plus除夕正式开源。凭 借全面革新的底层技术与卓越的性能表现，千问3.5一经发布便登顶全球最强开源大模型的宝座，多项 核心能力媲美或超越了GPT-5.2、Gemini-3-pro等全球顶尖的闭源模型，实现了从"能聊天"到"看懂并操 作现实世界"的性能跨越。 杭州在通用人工智能（AGI）关键赛道上的前瞻布局与深厚潜力，瞬间改变了我们的生活。随之而 来的春节假期期间，全国用户说了50亿次"千问帮我"，让千问一跃成为国民级AI助手；在千问上"一句 话下单"使用近2亿次，让"千问帮我"买奶茶、订机票酒店、买电影票和景点门票，甚至被称为"新年 俗"。相比传统点选式下单，"一句话下单"操作更简单，极大提升了下单便捷性，即使是不会操作App 的老年人群，也能享受AI生活带来的便利。春节假期期间，有超过400万的60岁以上用户让千问帮忙下 单。 春节里，我像去年一门心思琢磨DeepSeek一样，用相同的指令操作不同的AI App，比较它们的响 应方式，多大程度能满足我对文字资料、图片、视频的功能需求，争取找到 ...

Core Viewpoint - Yunzhisheng (09678) has launched its intelligent document foundation model "Unisound U1-OCR," marking a significant advancement in the OCR technology, transitioning from basic character recognition to understanding business logic [1] Group 1: Product Launch - On February 26, Yunzhisheng officially introduced the "Unisound U1-OCR," which is the first industrial-grade intelligent document foundation model [1] - The model signifies the beginning of the OCR 3.0 era, enhancing document understanding by not only recognizing text but also extracting deep semantic insights and automatically classifying documents [1] Group 2: Technological Advancement - Unisound U1-OCR is recognized as a state-of-the-art (SOTA) document understanding model, achieving top performance in multiple authoritative tests [1] - The model overcomes the limitations of traditional models that only read text without understanding layout, enabling it to comprehend complex documents like a human expert [1] Group 3: Strategic Importance - The launch of Unisound U1-OCR represents not only an innovation in document intelligence but also a crucial step for Yunzhisheng towards achieving Artificial General Intelligence (AGI) [1]

智通财经APP获悉，Anthropic与美国国防部之间的矛盾将于本周五迎来最终时限。由于双方在AI军事应 用伦理条款上存在根本性分歧，五角大楼向Anthropic下达最后通牒：必须在当地时间周五下午5:01前无 条件接受政府提出的"所有合法用途"条款，否则将被列入"供应链风险"黑名单，并可能面临冷战时期 《国防生产法》的强制约束。 这场对峙源于Anthropic对其AI模型Claude在军事应用中设置的两条核心安全红线：禁止用于对美国民众 的大规模监控，以及禁止集成于完全自主的武器系统。尽管Anthropic已是五角大楼的供应商，并于去 年7月签署了价值2亿美元的国防合同，其Claude模型更是首个获准部署于政府机密网络的AI工具，但公 司坚称，无法在违背原则的前提下接受国防部提出的最新妥协方案。 "我们不能违背良心同意他们的要求。" Anthropic首席执行官Dario Amodei在周四的一份声明中强硬回 应，称五角大楼的威胁不会改变公司的立场。 "我们不会让任何公司左右我们的作战决策方式，"Parnell在社交平台X上发出最后警告。 这一僵局迅速改变了美国国防AI的竞争格局。据报道，在Anthropi ...

2月25日，自治区党委书记、自治区人大常委会主任陈刚考察香港科技大学，发表人工智能主旨演 讲，并出席"桂港链通 智汇东盟"人工智能恳谈会。 当天，陈刚首先参观香港生成式人工智能研发中心，观摩香港HKGAI大模型展示，了解HKGAI模 型在政务、出行、环保、法律、娱乐等场景应用效果，以及人工智能+金融的实践案例。 陈刚随后来到该校学术大楼与师生交流，以"AI向善，共治共行——从人类的进化引发的思考"为主 题发表主旨演讲，深入浅出地介绍了人工智能面临的安全挑战，阐述了AI向善，进而实现人类与人工 智能共生共荣的观点思路，并以问答形式与香港科技大学师生交流互动，现场气氛热烈。陈刚表示，当 前，人工智能正以前所未有的广度与深度重塑全球经济、社会结构和治理格局。但人工智能在带来科技 进步和产业发展重大跃升的同时，也伴随着复杂深刻的风险挑战。我们必须坚持以人为本、智能向善， 守牢人工智能安全，既"踩油门"又"踩刹车"，提升人工智能数据安全和多样性，有效应对各种风险挑 战。广西期待在人工智能伦理安全、数据跨境安全合规、人工智能赋能千行百业等方面进一步加强桂港 协同，真诚欢迎香港科技大学师生积极参与到桂港人工智能合作中来，以 ...

如果进展顺利，阶跃星辰有望成为第三家IPO的通用大模型独角兽。此前智谱和MiniMax股价大涨，激发一级市场投资人"害怕错过"的情绪，对创 业大模型公司重新有了投资热情 文｜《财经》记者 刘以秦 编辑 | 谢丽容 根据公开信息，阶跃星辰目前共完成2轮融资，今年1月，阶跃星辰完成超50亿元B+轮融资，2024年12月，该公司完成数亿美元B轮融资，上海国 投先导基金、浦东创投、徐汇资本、无锡梁溪基金、厦门国贸等多家国资参投。另外，五源资本、启明创投、腾讯等多家机构也参与投资。 如果进展顺利，阶跃星辰有望成为中国第三家在港交所上市的AI大模型公司。 今年1月7日，大模型公司智谱科技在港交所挂牌上市，成为港股"大模型第一股"，其IPO募资总额超43亿港元。1月9日，大模型公司MiniMax在港 交所挂牌上市，募集资金总额约55.4亿港元，两家公司上市后市值一路上涨。 《财经》独家获悉，AI独角兽公司阶跃星辰正在进行新一轮Pre-IPO融资，本轮融资分两拨交割，第一拨投前估值约40亿美元，计划融资20亿-30 亿元人民币，目前已经确定由一家运营商领投，金额约12亿元人民币，其他老股东跟投，计划在3月6日前完成；第二拨投 ...

据美国"政治"新闻网报道，五角大楼近期对Anthropic展开极限施压，要求Anthropic公司允许军方无限制 使用其旗下AI大模型Claude。美国国防部长赫格塞思威胁称，如果Anthropic未能在美国时间27日下午5 点01分前遵守要求，将被认定为供应链风险企业。 针对军方施压，阿莫迪26日发表声明称：Anthropic公司理解军事决策应由美国国防部而非私营企业作 出。我们从未对具体军事行动提出异议，也从未试图临时限制技术应用。然而某些应用场景超出了当前 技术安全范畴。以下两类应用场景从未纳入我们与国防部的合同范畴，且我们认为现阶段仍不应纳入： 一是在美国国内的大规模监控，二是完全自主武器（即彻底排除人类干预、自动选择并攻击目标的系 统）。 来源：环球网 【环球网报道】"这些威胁不会改变我们的立场。我们无法违背良知接受军方要求。"人工智能初创公司 Anthropic首席执行官达里奥·阿莫迪当地时间2月26日发表声明，拒绝美国军方要求。 另一方面，阿莫迪在声明中又留有回旋余地。他声称，选择最符合其要求的承包商是国防部的权利。但 鉴于Anthropic的技术为美军带来巨大价值，我们希望他们能重新考虑。我们 ...