本报讯 记者吕丽报道 "原来完成一个工单平均需要6分钟，现在只要3.5分钟，效率提高了，能集中精力 更好提升服务质量。"记者3月31日获悉，大连市数据局与中国移动大连分公司合作，在全省率先将 DeepSeek大模型引入12345热线平台应用场景，实现服务质量、效率双提升。 责编：杨金凤 审核：徐晓敬 此次12345热线平台升级了智能助手、智能知识库、智能质检三大智能化应用场景，政务服务体验更加 精准、高效、便捷。大连移动相关负责人介绍，12345热线平台借助DeepSeek专有模型能力，在话务端 上线"一键填单"功能，通话过程中，系统自动识别语音信息，直接转化成文字，自动填写工单内容、推 荐承办单位，通话结束后仅点击一个按键就能完成工单；通过DeepSeek实现智能知识库升级，在通话 服务时可根据通话内容实时推送解答话术，大幅提升咨询类问题处理效率和准确性，显著提高诉求直接 答复率。在质检端借助DeepSeek大模型能力实现话务工单由人工抽检到自动质检的模式突破，已实现 工单100%质检。同时增加对话务员响应速度、语言规范性、业务能力多维度评估功能，为整体服务质 量评估、分析提供数据支撑，后续将结合专项培训， ...

对此，中国科学院院士、零一学院创始院长郑泉水表示，"今天睿思芯科发布中国首个全自研RISC-V服 务器CPU，具有特殊的意义。这不仅是一项技术突破，更是一个从源头创新到产业落地的典范。" （图片来源：unsplash） 4月1日消息，钛媒体AGI获悉，RISC-V领先厂商睿思芯科近日在深圳推出灵羽CPU处理器产品。 公司称，这是中国首款全自研高性能RISC-V架构的服务器芯片，从而满足高性能计算、全闪存储与 DeepSeek等开源大语言模型的应用场景。 睿思芯科创始人、董事长兼CEO谭章熹博士表示，灵羽处理器基于睿思芯科全自研的CPU核心IP与片上 网络IP，采用"一芯双核"的设计架构，由32个高性能通用计算核心（CPU）与8个智算专用核心 （LPU）组成，专为DeepSeek等开源大语言模型推理等任务优化，同时通过软硬件结合等技术实现创 新，显著提升运算中能效比以及优化总体拥有成本（TCO）。 睿思芯科称，灵羽处理器计算性能已比肩Intel、AMD等国际主流型号的服务器芯片。 生态方面，目前，灵羽处理器生态体系已拥有超过50家合作伙伴，涵盖关键器件厂商、整机OEM、系 统软件与行业解决方案，包括联想、商汤、 ...

Core Insights - OpenAI CEO Sam Altman announced the upcoming release of a powerful open-weight language model with reasoning capabilities, marking the first such release since GPT-2 [1] - The decision to release this model comes after a long period of consideration, with previous priorities delaying its launch [10] Group 1: Open Weight Model - Open weight refers to publicly available training parameters of AI models, allowing users to modify and redistribute them [2] - This approach is seen as a middle ground between fully closed and fully open-source models, extending the open-source concept into the AI field [2] Group 2: Safety and Evaluation - OpenAI will evaluate the model using a "preparation framework" to ensure safety and reliability before its release [3] - Additional testing and optimization will be conducted due to the potential for modifications post-release [3][4] Group 3: Developer Engagement - OpenAI plans to host a series of developer events to gather feedback and showcase early prototypes, starting in San Francisco and expanding to Europe and Asia-Pacific [3] - Developers interested in participating can sign up through a provided link [3] Group 4: Community Reactions - There is significant interest in the model, with discussions in the comments highlighting the popularity of the OpenAI o1 mini and references to other models like DeepSeek [5][6] - Some community members express skepticism regarding the timeline for the release of GPT-5, which was also mentioned to be coming in "the next few months" [7] Group 5: User Growth and Competition - OpenAI experienced a surge of 1 million new users in five days due to the multimodal image generation capabilities of GPT-4o, leading to increased demand on their GPU resources [8] - Altman acknowledged the competitive landscape, particularly with DeepSeek, and reflected on the lessons learned regarding the importance of open features for viral growth [9]

Group 1 - The emergence of the Chinese startup DeepSeek, which offers high-performance and low-cost generative AI models, has surprised the world and raised expectations for similar companies in Japan [1][2] - The success of DeepSeek is attributed to China's reform and opening-up policies, which have fostered a competitive environment where young people are driven to develop technology and start businesses [1] - In contrast, Japan's comfortable lifestyle and reduced competition have led to fewer ambitious young entrepreneurs, contributing to the lack of similar startups like DeepSeek [1] Group 2 - Japan lags behind China in terms of the number of entrepreneurs and funding in the startup sector, with a lack of growth observed over the past 30 years due to various reasons [2] - The absence of acceptance for new innovations in Japanese society is highlighted, suggesting that there are innovative seeds that could thrive if the environment were more conducive [2] - To create companies like DeepSeek, it is essential for both private and government sectors in Japan to promote the latest technologies and information to raise public awareness [2]

Core Insights - DeepSeek has become the fastest-growing AI tool globally, surpassing OpenAI's ChatGPT in monthly website visits, reaching 525 million visits in February 2025 compared to ChatGPT's 500 million [1] - DeepSeek's market share has increased from 2.34% to 6.58%, making it the third-largest player in the AI tools market, following ChatGPT and Canva [2] - The launch of DeepSeek's open-source reasoning model, DeepSeek-R1, has significantly impacted the global AI industry landscape, leading to a shift in capital markets [1][2] Market Performance - DeepSeek's chatbot has a total visit count of 792.6 million and 136.5 million unique users, capturing 12.12% of the chatbot market, ranking second [2] - The total visits to AI tools globally reached 12.05 billion in February, with 3.06 billion unique visitors interacting with various AI tools [2] - Following the release of the DeepSeek R1 model, user growth surged by 125 million in January, with over 80% of new users joining in the last week of that month [2] Technological Advancements - The recent release of DeepSeek-V3, with 685 billion parameters, has significantly improved its capabilities in coding, mathematics, and reasoning, rivaling top models like Anthropic's Claude 3.7 [4] - DeepSeek's open-source model allows for widespread accessibility and usage, contrasting with commercial models that charge fees [5] - The performance leap in DeepSeek-V3 suggests potential for future major updates, indicating ongoing innovation within the company [4][5] Industry Impact - DeepSeek's rapid rise is reshaping the AI industry, indicating a more competitive landscape ahead [3] - The introduction of open-source models by DeepSeek is expected to trigger a wave of similar developments in China, raising concerns about the technological gap between the US and China [5] - The efficiency and cost-effectiveness of DeepSeek's models position it as a strong competitor in the AI market, potentially marking 2025 as a pivotal year in US-China AI competition [5][6]

Core Viewpoint - The restaurant industry is experiencing a significant transformation driven by artificial intelligence (AI), enhancing customer experience and operational efficiency, marking a shift from labor-intensive to high-efficiency models [1][6][8]. Group 1: AI Integration in Restaurants - AI technologies, such as robotic waiters and smart inspection systems, are being integrated into various aspects of restaurant operations, improving service delivery and management efficiency [2][4][6]. - The introduction of AI tools has led to a 30% increase in inspection efficiency for restaurant management, showcasing the potential of AI in enhancing operational workflows [4][5]. Group 2: Customer Experience Enhancement - The deployment of AI, such as the robot dog at Hu Da Restaurant, not only addresses labor shortages but also provides an engaging experience for customers, transforming waiting times into enjoyable interactions [2][3]. - The focus on emotional value through innovative AI applications is reshaping brand image and customer engagement strategies in the restaurant sector [3][5]. Group 3: Future Trends in the Industry - The restaurant industry is transitioning towards a more digital and intelligent future, with AI expected to play a crucial role in supply chain management, food safety, and personalized customer experiences [8]. - The ongoing evolution of AI applications indicates that the restaurant sector is moving from the development phase to a more mature stage, with significant potential for further advancements [7][8].

Group 1 - The core concept of AI Agents is that they are intelligent entities capable of perceiving their environment, making decisions, and executing actions, built on large language models (LLMs) with multi-modal interaction and autonomous execution capabilities [1][12][19] - DeepSeek and other reasoning models enhance the reasoning, decision-making, planning, execution, and tool-calling abilities of AI Agents, promoting multi-modal and personalized interactions while supporting localized deployment for data privacy [1][28][31] - The current state of the AI Agent industry shows a strong inclination towards B-end applications, with high adoption rates among medium-sized enterprises and diverse C-end applications across various sectors such as healthcare, human resources, and retail [2][4][5] Group 2 - Manus, a general AI agent, consists of a multi-agent collaborative architecture and a dynamic task scheduling engine, enabling full-chain autonomous execution of tasks across multiple fields, while also presenting ethical and governance challenges [2][34] - AI Agents impact business operations by optimizing processes, assisting in decision-making, and driving innovation, with deployment models including direct use, client API calls, and localized private deployment [2][5][6] - The emergence of high-quality reasoning models like DeepSeek positions China as a potential leader in the global AI Agent market, shifting from being a rule-taker to a co-governor in AI standards [28][32][34]

Group 1 - The core idea presented by Zhang Ping emphasizes the need for imagination driven by enterprise demand in the development of 6G technology, suggesting that 6G should learn from DeepSeek to foster more innovation [1] - Zhang Ping highlights that 6G applications require higher innovation capabilities from suppliers, and there is often a disconnect between the diverse needs of demand-side and the perceived capabilities of supply-side [2] - He stresses the importance of forming an industrial chain and ecosystem for successful innovation, indicating that no single company can complete a project independently and profitably [2] Group 2 - The integration of artificial intelligence (AI) with communication is identified as a key focus for 6G development this year, with AI expected to revolutionize network applications [2] - Zhang Ping notes that the current high costs and power consumption of AI must be addressed as AI moves towards edge applications, with 6G potentially enhancing the efficiency of intelligent agents in network operations [2] - Beijing is set to play a leading role in the 6G process, with plans to establish a high-standard 6G laboratory and promote key technologies in smart cities, intelligent transportation, and emergency communications by 2025 [3]

出品｜虎嗅科技组 作者｜宋思杭 编辑｜苗正卿 头图｜智谱Openday现场 那个在一个月内斩获三轮国资的"国家队"智谱，终于开启与DeepSeek正面交锋。 但也并非所有大模型厂商都在摒弃预训练、拥抱DeepSeek。对此，智谱创始人兼CEO张鹏告诉虎 嗅，"我们仍然会坚持预训练，但更深一层的问题是，目前关于预训练有很多没有被关注到的点，比如 预训练的架构、数据使用效率等等。" 实际上，自2025开年以来，DeepSeek对所有基座模型形成的降维打击，正在持续给整个行业带来余 震。"DeepSeek做得那么好，为什么还要投你？"类似这样的灵魂拷问，却从未在过去两年从投资人的口 中传出。相反，"大模型六小虎"曾是他们摘不到的星星。 而从整个第一季度来看，资本正逐渐对大模型初创公司失去耐心。可以说，除了智谱连续拿到地方产投 融资，目前并无市场化资本进入到其余大模型企业当中。 与此同时，这种降维打击所暴露出的另一面是，即使是用百亿元砸出来的预训练，也没能让基座模型厂 商形成护城河。 商业化难题，是大模型企业这三年来都没能绕过去的槛。继拿到多笔国资后，智谱也即将开启新旅程。 在今日的发布会现场，CEO张鹏首次正面回应 ...

AI基础设施建设再度迎来高峰，由此引发产业链层面的价格抬升。 近日，国盛证券发布报告指出，云计算、IDC、服务器、GPU及算力租赁等领域均在不同程度上出现价 格上升的信号，验证了云上游行业景气度提升。 21世纪经济报道记者通过采访调查发现，本轮云上游价格抬升，是建立在去年算力市场进入半饱和状 态，租赁和芯片价格有所调整的背景条件下。本次回升，是因DeepSeek出现，重新激起了算力市场部 署热情，尤其是针对AI推理侧。 而目前国内AI芯片厂商在AI推理侧已经具备一定能力积累，有望受益于此轮行情。至于其具体进度， 则有待国内计算产业链共同推进生态和标准化工作的完善。 云上游涨价 国盛证券指出，近期云上游涨价现象已在多个领域显现。以AWS为例，3月4日H100云计算调用价格相 比两周前平均上涨约22%；以英伟达参股的算力租赁公司CoreWeave为例，截至2025年3月，8卡H100服 务器租赁价格为49.24美元/小时，相较于2023年4月的38.08美元/小时价格上涨29.31%，反映了企业对实 时推理算力服务的迫切需求。 在DeepSeek热潮带动下，国内大厂对英伟达H20芯片需求旺盛。报告举例称，目前8卡 ...