Workflow
通用人工智能(AGI)
icon
Search documents
微软AI芯片Maia时隔两年上新,号称性能超亚马逊Trainium
第一财经· 2026-01-27 02:43
当地时间1月26日,微软宣布推出第二代人工智能芯片Maia 200。 微软云与人工智能执行副总裁斯科特·格思里表示,Maia 200采用台积电3纳米工艺制造,每颗芯片包含超过 1400 亿个晶体管,为大规模 AI 工作负载量身打造,同时兼顾高性价比。他称Maia 200是微软迄今为止部 署的最高效推理系统,其每美元性能比微软目前部署的最新一代硬件提升了30%。他同时表示,Maia 200的 FP4性能是第三代 Amazon Trainium的3倍。 目前,Maia 200已部署在微软位于 爱荷华州 得梅因附近的美国中部数据中心区域,接下来将部署位于亚利 桑那州凤凰城附近的美国西部3数据中心区域,未来还将部署更多区域。 2026.01. 27 本文字数:1180,阅读时长大约2分钟 作者 | 第一财经 陆涵之 萨蒂亚·纳德拉则在财报中表示,将继续加大对人工智能领域的投资,包括资金和人才,以把握未来巨大的 机遇。他同时表示,微软旗下所有产品中,人工智能功能的月活跃用户已达9亿。 在应用方面,微软超级智能团队将利用 Maia 200 进行合成数据生成和强化学习,以改进下一代内部模型。 在去年11月,微软宣布成立超级 ...
微软AI芯片Maia时隔两年上新,号称性能超亚马逊Trainium
Di Yi Cai Jing Zi Xun· 2026-01-27 02:27
微软云与人工智能执行副总裁斯科特·格思里表示,Maia 200采用台积电3纳米工艺制造,每颗芯片包含 超过 1400 亿个晶体管,为大规模 AI 工作负载量身打造,同时兼顾高性价比。他称Maia 200是微软迄今 为止部署的最高效推理系统,其每美元性能比微软目前部署的最新一代硬件提升了30%。他同时表示, Maia 200的FP4性能是第三代 Amazon Trainium的3倍。 当地时间1月26日,微软宣布推出第二代人工智能芯片Maia 200。 目前,Maia 200已部署在微软位于爱荷华州得梅因附近的美国中部数据中心区域,接下来将部署位于亚 利桑那州凤凰城附近的美国西部3数据中心区域,未来还将部署更多区域。 在应用方面,微软超级智能团队将利用 Maia 200 进行合成数据生成和强化学习,以改进下一代内部模 型。在去年11月,微软宣布成立超级智能团队,由微软消费者人工智能业务负责人穆斯塔法·苏莱曼 (Mustafa Suleyman)领导,旨在实现人文主义超级智能。穆斯塔法·苏莱曼表示,这么做为了摒弃关于 通用人工智能(AGI)竞赛的叙事,而是将其视为一项更广泛、更深刻的人类事业,用于改善人类的生 活 ...
我国在通用人工智能逻辑推理领域实现重大跨越
Huan Qiu Wang Zi Xun· 2026-01-27 01:41
在性能表现上,TongGeometry展现了极高的国产原创技术优越性。相比AlphaGeometry需要庞大的算力 集群,TongGeometry仅需单张消费级显卡(如RTX 4090)即可在最多38分钟内,解决近25年所有的奥 数几何难题,其推理效率和准确率均达到世界顶尖水平。此外,该系统通过创新的"规范化表示"技术, 将搜索空间压缩了几个数量级,有效解决了传统方法中的路径爆炸问题。 论文共同通讯作者、北京大学心理与认知科学学院助理教授朱毅鑫表示,TongGeometry的意义不仅在 于解题速度的提升,更在于它通过模拟人类数学家的直觉和审美,实现了"小数据、大任务"的范式转 化。这种不依赖海量标注数据、通过内部逻辑自我演化的路径,正是通用人工智能(AGI)发展的关 键。"我们的系统不仅能与国际最先进的AI系统对标,更在理解逻辑底层美学和自主发现科学规律方面 走在了前列。"他说。 来源:科技日报 科技日报记者 杨雪 我国科研团队开发出全球首个同时具备自主出题(Proposing)和自动解题(Solving)双重能力的通用 人工智能系统——"通矩模型"(TongGeometry)。相关成果"基于引导树搜索的奥数 ...
阿里正式发布!千问最强模型来了
券商中国· 2026-01-27 01:06
Core Viewpoint - Alibaba has officially launched its flagship reasoning model Qwen3-Max-Thinking, which sets new global records in various key performance benchmarks, including scientific knowledge, mathematical reasoning, and code programming [1][3]. Group 1: Model Specifications - The Qwen3-Max-Thinking model has over 1 trillion parameters and a pre-training data volume of 36 trillion tokens, making it Alibaba's largest and most capable reasoning model to date [3]. - The model has achieved state-of-the-art (SOTA) performance in 19 recognized large model benchmark tests, comparable to models like GPT-5.2-Thinking-xhigh, Claude Opus 4.5, and Gemini 3 Pro [3]. Group 2: Enhanced Capabilities - Qwen3-Max-Thinking has improved its native agent capabilities for autonomous tool invocation, allowing it to intelligently combine tools for various tasks [3]. - The model's adaptive tool invocation ability can be experienced on QwenChat, where it autonomously selects core agent tools such as search, personalized memory, and code interpreter, while also reducing model hallucinations [3]. Group 3: User Accessibility - Developers can experience the Qwen3-Max-Thinking model for free on QwenChat, while enterprises can access the new model's API services through Alibaba Cloud [4]. - The Qwen APP has integrated the new model and is set to provide users with access to over 400 AI service functions, enhancing user interaction with various Alibaba ecosystem services [6][7]. Group 4: Future Developments - Alibaba plans to gradually introduce agentic-AI features in the Qwen application to support shopping functionalities across its platforms, with a focus on enhancing the model's understanding and discernment capabilities [7]. - The company is investing significantly in AI infrastructure, with a projected investment of 380 billion yuan, aiming to develop both services and the underlying technology [9].
南方科技大学孟庆虎:马斯克关于“Optimus三年内做手术”的大饼烙不熟
Group 1 - Tesla's CEO Elon Musk predicts that Optimus, Tesla's humanoid robot, will surpass the best human surgeons within three years on a large scale [1] - Meng Qinghu, a professor at Southern University of Science and Technology, disagrees with Musk's assertion, stating that it is impossible to achieve such advancements in three to five years due to limitations in data and hardware capabilities [1] - Meng emphasizes that current AI models, particularly in robotics, are still limited in their ability to perform complex tasks autonomously and require significant improvements in precision and dexterity compared to humans [1] Group 2 - There is a common misconception that large models, such as ChatGPT and others, are capable of performing complex physical tasks, while in reality, they are primarily effective in text generation [2] - Meng clarifies that the current AI models are primarily two-dimensional and lack the three-dimensional understanding necessary for advanced image processing, which contributes to their limitations in physical interactions [2] - The development of "scene intelligence" is proposed as a more practical approach for AI applications, focusing on solving specific problems efficiently with minimal computational resources [3] Group 3 - Meng argues that achieving general artificial intelligence (AGI) requires transforming all scenarios into intelligent systems, which is currently hindered by a lack of high-quality data [4] - He suggests that the term "scene intelligence" should replace "general models," as it can address specific issues immediately, while general models remain constrained by data and hardware limitations [4] - Meng predicts that the capability for robots to perform surgeries successfully may take five to ten years, likely involving a combination of humanoid robots and AI agents working alongside human experts [4]
马斯克喊出25万亿估值!Optimus Gen 3必须救活特斯拉!
Sou Hu Cai Jing· 2026-01-26 11:50
当地时间2026 年 1 月 18 日,美国科技创投圈知名人士 Jason Calacanis 在 X 平台抛出重磅观点:"未来 没人会记得特斯拉造过汽车,人们只会记得它量产了 10 亿台 Optimus 人形机器人"。 这一论断迅速得到特斯拉CEO 埃隆・马斯克的正面回应,后者直言 "这一愿景很可能成真"。马斯克同 时明确表态,特斯拉将全面聚焦 Optimus 人形机器人研发与量产,这项业务有望推动公司估值突破 25 万亿美元,成为全球顶尖机器人巨头,且机器人业务的价值占比将远超当前的汽车板块。 数据援引新华网1月3日披露 值得注意的是,特斯拉此番战略转向的背后,是其汽车主业面临的严峻增长压力。2025 年,特斯拉交 出了一份近年最差经营成绩单:全球汽车销量首次被比亚迪超越,且销量连续两年下滑,净利润大幅缩 水,新能源汽车核心业务增长陷入瓶颈。 市场分析认为,押注Optimus 是特斯拉在增长困境下的一次战略豪赌,但该人形机器人能否顺利量产、 能否实现商业化落地、能否真正扛起业务救赎的大旗,目前仍存在多重不确定性。 ▍特斯拉的救命稻草为何是Optimus 人形机器人? 公开数据显示,2025年特斯拉全球交付 ...
商道创投网·会员动态|曦望芯科·完成近30亿元战略融资
Sou Hu Cai Jing· 2026-01-26 11:05
《商道创投网》2026年1月26日从官方获悉:杭州曦望芯科智能科技有限公司(以下简称"曦望")近日 完成了由三一集团旗下华胥基金、范式智能、杭州数据集团、正大机器人、协鑫科技、游族网络、北京 利尔等产业投资方,以及无极资本、IDG资本、心资本、高榕创投、中金资本、普华资本、松禾资本、 易方达资本、工银投资、海通开元、越秀产业基金、银泰投资、国元基金、粤民投、华民投等国内知名 VC/PE机构联合投资的近30亿元战略融资,同时获得诚通混改基金、杭州金投、杭州高新金投等国资背 景资本的鼎力加持。 《商道创投网》创业家会员·单位简介 《商道创投网》创投生态圈·本轮投融观点是什么? 曦望(Sunrise)是一家专注于高性能GPU及多模态场景推理芯片研发与商业化的国产全栈自研人工智 能算力芯片企业。其前身是商汤大芯片部门,2024年底分拆独立运营。凭借八年技术沉淀、二十亿研发 投入及两代量产芯片的工程化验证,曦望已成为国产GPU替代的核心力量。公司通过技术创新和深度产 业协同,致力于为千行百业提供成本降低十倍、能效比突破的智能算力基石,推动各行业的智能化发 展,并助力实现通用人工智能(AGI)的普惠化目标。 商道创投网创 ...
国际清算银行:人工智能与数字金融需在创新与金融稳定间寻求平衡
但张涛在演讲中警示,人工智能与数字金融在带来效率提升和市场整合机遇的同时,也对金融稳定构成 多维度挑战。在市场层面,人工智能驱动的快速交易可能加剧短期价格波动,而代币化资产在市场承压 时的快速流转,可能引发流动性紧张,破坏市场秩序;在操作层面,相关技术对少数专用硬件、云计算 服务提供商及数据供应商的高度依赖,以及数字金融领域共享平台的系统重要性,使其面临的运营中 断、网络攻击等风险可能扩散至整个金融体系;在风险传播层面,同质化的人工智能模型与决策规则会 增强机构行为的关联性,代币化平台构建的跨市场依赖关系,可能加速风险传染,放大冲击效应。此 外,技术变革还提升了风险的复杂性,给监管机构的风险识别与管理工作带来巨大考验。 面对跨司法管辖区运作的技术服务与各国分散的金融稳定责任之间的矛盾,张涛强调国际合作的必要 性。他指出,构建涵盖问责制、风险管理、监督等要素的全球治理框架是关键,需重点解决人工智能领 域的数据治理、第三方依赖,以及数字金融和代币化过程中的准入安排、运营责任界定等问题。作为国 际央行合作的重要平台,BIS正通过搭建交流渠道、支持学术研究、运营创新中心等方式,促进各国央 行协作,并联合金融稳定理事会 ...
再见了, OpenAI!三年老用户忍痛卸载ChatGPT
猿大侠· 2026-01-26 04:11
可以上下滚动的图片 毫无疑问,现在判断谁是最后赢家,还为时过早。 编辑:KingHZ 【导读】 从ChatGPT惊艳问世到如今广告缠身,OpenAI乌托邦梦碎!谷歌和Anthropic强势反 扑,达沃斯论坛上互怼升级,这不是AGI的星辰大海,而是残酷的商业战场。 OpenAI全球首家实现AGI! 只不过,这个AGI可能要贻笑大方了。 奥特曼口中的「口袋里的博士级」AGI,不是星辰大海般的「通用人工智能」(Artificial General Intelligence),而是「广告生成收入」(Ad-Generated Income)。 当前,AI竞赛空前激烈,赌注之高前所未有,而OpenAI在ChatGPT里塞广告,未免操之过急,被普 遍认为是昏招: 但不可否认,OpenAI这次无疑成了硅谷AI巨头中的「眼中钉」。 10多年从业经验的科技记者、The Verge前副主编Alex Heath,在达沃斯与多名AI领袖的交谈之 后,他留下了一个印象: 整个行业似乎已集体决定联合起来对付OpenAI。 不过,OpenAI的投资人、Khosla Ventures合伙人Ethan Choi深度复盘了2026开年 AI行业的 ...
“大模型六小虎”之一阶跃星辰B+融资超50亿,多地国资参投
Guan Cha Zhe Wang· 2026-01-26 02:49
Core Insights - StepFun has completed a B+ round financing exceeding 5 billion yuan, setting a record for the largest single financing in China's large model sector in the past 12 months [1] - The company was founded in April 2023 by former Microsoft global vice president Jiang Daxin and is recognized as a leading startup in the AI large model field, alongside other notable companies [1] - StepFun focuses on the development of General Artificial Intelligence (AGI) and has released 22 self-developed foundational models, including significant models with hundreds of billions of parameters [1] Investment Details - The financing round included participation from several state-owned investment institutions, highlighting strong institutional support [3] - Shanghai Guotou Xiandai Private Equity Fund Management Co., Ltd. is a wholly-owned subsidiary of Shanghai State-owned Capital Investment Co., Ltd., focusing on early-stage investments in key industries [3] - China Life Equity Investment Co., Ltd. aims to leverage resources for long-term value investments, with a focus on the healthcare sector through its flagship fund [4] - Shanghai Pudong Venture Capital Co., Ltd. is the first venture capital institution in Pudong New Area, primarily investing in high-tech industries [4] - Shanghai Xuhui Capital Investment Co., Ltd. focuses on early-stage investments in AI and health sectors to support regional economic development [4] - Wuxi Liangxi Fund is a core platform for financial and industrial integration, with significant investments in AI and technology innovation [4] Company Background - StepFun is part of a group of emerging companies referred to as the "six small tigers" in the domestic large model sector, indicating its competitive positioning [1] - The core team includes top experts from the AI field, enhancing the company's research and development capabilities [1] - The company has successfully implemented its products in various industries, including telecommunications and finance, demonstrating its market applicability [1]