图片来源： DeepMind "如今再想公开发表像Transformer那样的论文，几乎是不可能的。"一位现任研究员感慨道。 据两位知情人士透露，DeepMind对涉及生成式AI的"战略性"论文设置了长达六个月的"冷却期"，研 究人员还需说服多个团队，证明论文的发表价值。 一位接近公司的人士解释称，此举旨在减少研究 人员在"战略或商业上难以获批"的项目上浪费时间。他补充说，DeepMind每年仍发表数百篇论文， 并持续在顶级AI会议上保持影响力。 据七位现任及前任DeepMind研究人员透露，公司已引入更严格的审查机制和行政流程，显著提高了 论文发表的门槛。三位前员工表示， DeepMind尤其避免公开可能被竞争对手利用的技术创新，或可 能使谷歌Gemini大模型在对比中处于劣势的研究。 这一转变标志着DeepMind的战略重心从学术影响力向商业竞争力的倾斜。过去，该机构以发表突破 性论文和吸引全球顶尖AI人才著称——2017年谷歌研究人员发布的Transformer论文奠定了大语言模 型的基础，并直接推动了生成式AI的爆发。然而，随着OpenAI等竞争对手的崛起，DeepMind已成为 谷歌重振AI领导地位 ...

如果您希望可以时常见面，欢迎标星收藏哦~ 来源：内容来自太报，谢谢。 AI运算需求激增，科技巨头加速布局自研加速器。 DIGITIMES调查显示，自研云端ASIC加速器 市场呈新格局，从2023年与2024年的高速成长期，转进稳定成长阶段，预估今年出货量升至496.2 万颗，年增率大幅降至23%。这波成长主要来自生成式AI模型训练成本持续上升、小型语言模型 盛行，英伟达（NVIDIA） GPU价格高涨，以及美国禁令限制中国企业取得高阶AI芯片等因素驱 动。 在生成式AI时代，DIGITIMES指出，自研云端ASIC加速器成云端服务供应商(CSP)竞争关键之 一。自研加速器不仅能够降低能耗，还能帮助CSP掌控AI基础设施的成本与供应链，更能避免受制 于NVIDIA或国际政治因素。此外，透过技术创新，业者能够建立市场壁垒，确保AI产品与服务的 竞争力。因此Google、亚马逊、华为、微软(Microsoft)与Tesla等企业，近年来皆积极投入自研云 端ASIC加速器市场。 DIGITIMES观察，2025年自研云端ASIC加速器市场竞争格局将呈现显著变化，从过去Google独 霸市场，转为Google、亚马逊与 ...

Core Viewpoint - The latest financial report from SenseTime (0020.HK) indicates that the AI industry is on the verge of explosive revenue growth, highlighting the potential for significant commercial opportunities in the near future [1][2]. Group 1: Performance Highlights - SenseTime's generative AI revenue surged to over 2.4 billion RMB, marking a year-on-year increase of 103.1%, and accounting for 63.7% of total group revenue [3][4]. - The company's gross profit reached 1.62 billion RMB with a gross margin of 42.9%, and losses narrowed by 33.7% year-on-year [3]. - Cash reserves stood at 12.75 billion RMB, indicating a strong financial position to support future growth [3]. Group 2: Strategic Framework - SenseTime's "three-in-one" strategy integrates AI infrastructure (large devices), large models, and applications, creating a virtuous cycle that enhances competitive positioning in the generative AI landscape [7][11]. - The large device, located in Shanghai, is recognized as one of Asia's largest AI data centers, with a total operational computing power of 23,000 P, reflecting a 92% year-on-year increase [9][10]. Group 3: Market Potential - The generative AI market is projected to reach a scale of 10 trillion RMB, with significant growth anticipated from 2025 onwards as the industry matures [13]. - Morgan Stanley forecasts that by 2028, generative AI could generate nearly 1.1 trillion USD in total revenue, with enterprise software contributing 401 billion USD and consumer internet spending reaching 683 billion USD [13].

3月31日至4月06日当周重磅财经事件一览，以下均为北京时间： 重磅事件方面，特朗普称计划于4月2日实施对等关税，美联储主席鲍威尔和副主席杰斐逊讲话，欧洲央行行长拉加德讲话，欧洲央行公布3月货币政策会议纪 要，国内成品油开启新一轮调价窗口，微软50周年和小米15周年活动。 数据方面，关注美国3月非农就业和PMI、中国3月PMI、欧元区3月调和CPI报告。财报方面，关注赛力斯、贵州茅台、海天味业、石头科技、海伦司、复星国 际。 美国3月非农就业大概率降温 美国3月非农就业报告将于周五揭晓，根据彭博汇编的数据，市场预期： 3月非农新增就业13.5万人，较前值15.1万人显著下降；预计平均时薪同比增长3.9%，略低于前值的4%，环比涨幅维持在0.3%的水平不变；失业率预计将保 持在4.1%不变。 在非农之前公布的前两天，有小非农之称的APD就业数据公布，市场预期就业人数新增11.9万，较前值7.7万大幅增加。关注ADP就业和非农报告的背离情况。 整体来看，预计美国就业市场继续降温，支撑美联储进一步的降息行动。尽管上周美联储官员大幅下调了今年美国的GDP增长预期，鲍威尔仍很乐观看待就业 前景，他在上周的货币政策新闻发 ...

Core Viewpoint - The company has shown a significant recovery in revenue growth and a notable increase in the contribution of generative AI to its overall income, while also managing to reduce losses substantially [2]. Financial Performance - In 2024, the company achieved a revenue of 3.77 billion yuan, representing a year-on-year growth of 10.8% compared to a decline of 10.6% in 2023 [2]. - Generative AI revenue reached 2.4 billion yuan, marking a 103% increase year-on-year, and its share of total revenue rose from 35% in 2023 to 64% in 2024 [2]. - The gross margin was 42.9%, slightly down from 44.1% in 2023 [2]. - Adjusted net loss was 4.31 billion yuan, narrowing by 33.7% year-on-year, while EBITDA loss was 3.14 billion yuan, a reduction of 42.3% [2]. - R&D expenses (excluding SBC) were 4.05 billion yuan, up 19.2% year-on-year, while sales expenses decreased significantly by 20% [2]. Cash Flow and Capital Management - The company experienced a net cash outflow from operating activities of 3.93 billion yuan, which widened by 690 million yuan year-on-year, but improved cash turnover efficiency with the cycle reducing from 334 days to 228 days [3]. - Trade receivables increased by 19% year-on-year, and the net loss from financial asset impairments and contract assets decreased from 1.84 billion yuan to 780 million yuan [3]. - Capital expenditures fell to 1.13 billion yuan, down 29.9% year-on-year, while total cash reserves increased to 12.75 billion yuan, up 11.8% [3]. Business Highlights - The company is focusing on a synergistic development model of "large devices + large models + AI applications," with a 1+X architecture to concentrate resources [4]. - The self-owned 5A-level intelligent computing center in Shanghai has increased its computing power to 23,000 PetaFlops, ranking third in China's GenAI IaaS market with a share of 13.3% [4]. - The company launched the "Riri New" 5.0 model in 2024, maintaining industry leadership in multi-modal capabilities, and plans to release the "Riri New" 6.0 model in 2025, which is expected to match international standards [4]. AI Applications - The reduction in inference costs is accelerating the deployment of AI applications [5]. - In productivity tools, customer willingness to pay increased sixfold in 2024 compared to 2023 [5]. - The company is advancing its autonomous driving solutions, with the first mass-produced models based on the domestic Horizon J6M computing platform expected to be delivered in 2025 [5]. - Monthly user engagement for interactive tools has increased eightfold compared to 2023 [5].

Core Viewpoint - The article discusses the current wave of AI entrepreneurship driven by generative AI technologies like ChatGPT, highlighting both the unprecedented opportunities and significant challenges faced by entrepreneurs in this field [1]. Group 1: AI Entrepreneurship Landscape - The surge in AI startups is characterized as a golden age for entrepreneurs, with a multitude of opportunities arising from technological advancements [1]. - Despite the opportunities, the competitive landscape is fierce, making it difficult for startups to establish themselves [1]. - Key challenges include the difficulty of translating advanced AI models into practical applications, market education for users, and the need for compelling narratives and solid data to attract investors [1]. Group 2: Case Study of an AI Entrepreneur - The story of an entrepreneur named Gao Feng illustrates the struggles and eventual pivot in AI entrepreneurship, transitioning from an e-commerce supply chain platform to developing AI products for children [2][3]. - After facing significant setbacks due to the COVID-19 pandemic, Gao Feng found inspiration in his personal life, leading to the creation of an AI companion for children [2]. - The team focused on integrating advanced AI technologies into their products, emphasizing user needs and product positioning, which ultimately led to successful funding and product development [3]. Group 3: Insights from "AI Entrepreneurship Symphony" - The book "AI Entrepreneurship Symphony" features interviews with over 20 leading AI entrepreneurs, showcasing diverse backgrounds and experiences in the AI sector [4][5]. - It covers various hot AI sectors, including AI image generation, smart programming, and AI in education, highlighting successful case studies and innovative products [5]. - The narratives provide valuable insights into the entrepreneurial journey, including the challenges of funding, product iteration, and market positioning [6][17]. Group 4: Practical Takeaways - The experiences shared in the book serve as a practical guide for aspiring entrepreneurs and investors, offering lessons learned and strategies for navigating the AI landscape [16][17]. - The stories reflect a blend of technical innovation and business acumen, providing a comprehensive view of the current AI entrepreneurial ecosystem [6][17]. - The book aims to inspire readers by showcasing the resilience and creativity of AI entrepreneurs, encouraging them to pursue their own ventures in this dynamic field [18].

作者丨刘燕秋 来源丨投中网 移动互联网时代，创业公司的人才需求旺盛，与之相比，这一波 AI 创业的一大典型特征是，几十甚至十几人的小团队便能搞 定一切。这些务实的创始人们也并不恋战，当资本方递来橄榄枝，往往该卖就卖，见好就收。 这不眼看着，又有一批身价过亿的富豪诞生。 3 月 27 日，外媒消息称，英伟达计划以数亿美元收购阿里前副总裁贾扬清创 立的 Lepton AI 。对此，贾扬清很有分寸地回应媒体称"无法评论"。 将投中网设为"星标⭐"，第一时间收获最新推送 又一个AI造富的案例。 Lepton AI 的主要业务，是出租搭载英伟达 GPU 的服务器。可以这么理解，它扮演的是英伟达生态的补充者，帮助开发者租 赁使用英伟达 GPU 。其价值在于，可以在无需签订长期云合同的情况下，为中小客户提供 GPU 资源，同时为英伟达拓展车 企、科研机构等非云厂商客户。考虑到如今英伟达也试图进入服务器租赁市场来分一杯羹，甚至与 AWS 和谷歌等云服务商竞 争，直接下场买公司也并不稀奇。 Lepton AI 只是这波 AI 收购潮中的一朵浪花。生成式 AI 的概念火了两年多，一笔笔 AI 收购交易已然浮出水面。这个月就 有两 ...

Overall Overview - The report covers insights from 1,042 valid responses across 120 cities, with a significant representation from first-tier cities like Beijing, Shanghai, Guangzhou, and Shenzhen [26] - The internet industry leads with over 33% of respondents, followed by software (21.11%) and manufacturing (18.23%), indicating a diverse industry landscape [29][30] - Companies with 51-100 employees increased to 36.29%, a rise of 13.3% from the previous year, while those with over 501 employees decreased to 20.39%, down 14.09% [32][33] - Over 60% of respondents work 50 hours or more per week, reflecting high work intensity in the industry [41] Project Teams - Project teams are primarily composed of 3-10 and 11-30 members, with a trend towards smaller, more flexible teams [3] - The average project duration is concentrated in the 1-3 months and 6 months range, with an increase in short-term projects and a decrease in long-term projects [3] - The delivery rate of projects is complex, with a significant portion falling within the 40%-80% range, indicating room for improvement in project management [3] Project Management - Agile methodologies are used by approximately 49.47% of teams, with Kanban being favored, while the waterfall method is declining [4] - Major challenges include demand management, estimation accuracy, and quality control, with Gantt charts and Kanban being popular visualization tools [4] - The adoption of external project management tools is hindered by usage habits and lack of decision-making authority [4] Job Details - Different roles exhibit unique characteristics in coding abilities, work methods, and learning directions, with project managers focusing on progress control and product managers emphasizing requirement organization [4] Summary and Outlook - The report highlights various challenges in IT project management, suggesting that companies should optimize processes, enhance personnel capabilities, and allocate resources effectively to improve project management standards [4]

在去年高增长、高基数的基础上，2024年生成式AI业务依然保持了极高的增速。这不仅彰显了以大模 型技术为基础的业务所具有的强劲生命力，也验证了商汤"AI基础设施(大装置)-大模型(日日新)-应用"三 位一体战略的可行性。该战略构建了完整的商业闭环：AI大装置为应用落地提供算力支撑，应用反馈 又指导基础设施升级；大模型能力与基础设施相互促进，实现算力效率的持续优化；同时，多元化的技 术路径加速了应用场景渗透，而应用实践又推动大模型向精细化方向发展。 特别值得关注的是，2024年生成式AI业务在总收入中的占比从34.8%提升至63.7%，是商汤集团最大的 业务收入。这标志着公司彻底完成从视觉AI向生成式AI战略转型的全面胜利，其收入结构变革已形成 不可逆的良性发展态势。 一、"三位一体"战略持续巩固技术领先地位，训练效率优于DeepSeek公开报告 商汤科技依托"大装置-大模型-应用"的深度协同创新，形成核心技术差异化优势，并持续巩固技术领先 地位。 2024年，商汤生成式AI业务收入突破24亿元，同比增长103.1%，连续两年保持三位数增长。这一强劲 表现推动商汤2024年全年收入同比增长10.8%，达到37. ...