编辑 | 硬 AI OpenAI最新模型取得突破性进展：具备原创构思能力。 点击 上方 硬AI 关注我们 据介绍，最新模型不仅能总结研究论文或解决数学问题，还能够独立提出新构思，连接不同领域的概念，提出创新性实验 设计，完成需要科学家跨领域合作才能实现的成果，相当于"博士水平AI"。 硬·AI 作者 | 李笑寅 据媒体援引知情人士消息， OpenAI最早将在本周发布代号为o3或o4-mini的新模型， 该模型不仅能总结 研究论文或解决数学问题，还能够独立提出新构思，连接不同领域的概念，提出创新性实验设计。 据介绍，即将推出的新模型能同时利用物理学、工程学和生物学等多个领域的知识，提供跨学科的解决方 案，而科学家通常需要跨领域合作才能实现类似成果，相当"博士水平AI"。 硬·AI OpenAI总裁Greg Brockman在2月的"AI研讨会"活动上曾表示： "我们真正的方向是开发能够花大量时间认真思考重要科学问题的模型，我希望在未来几年内，这将 使所有人的效率提高10倍或100倍。" * 感谢阅读！ * 转载、合作、交流请留言，线索、数据、商业合作请加微信：IngAI2023 * 欢迎大家在留言区分享您的看法 ...

Core Viewpoint - The article discusses the competitive landscape between Zhipu and DeepSeek, highlighting Zhipu's recent product launches and pricing strategies aimed at challenging DeepSeek's dominance in the AI model market [2][10]. Product Launches - On March 31, Zhipu launched the "AutoGLM Thinking Model" and the inference model "GLM-Z1-Air," claiming that Air can match the performance of DeepSeek's R1 model with only 32 billion parameters compared to R1's 671 billion parameters [2]. - The pricing for Zhipu's model is set at 0.5 yuan per million tokens, significantly lower than DeepSeek's pricing, which is 1/30 of DeepSeek's model [2]. Market Dynamics - The article notes a shift in the AI model industry, with some companies, including Baichuan Intelligence and Lingyi Wanyi, experiencing strategic pivots or downsizing, indicating a loss of investor patience with AI startups [3][4]. - Despite the challenges, Zhipu continues to secure funding from state-owned enterprises, positioning itself as a leader among the "six small tigers" in the large model sector [4][6]. Commercialization Challenges - The commercialization of large models remains a significant hurdle for the industry, with Zhipu acknowledging the need to pave the way for an IPO while facing uncertain market conditions [6]. - Zhipu is focusing on penetrating various sectors, including finance, education, healthcare, and government, while also establishing an alliance with ASEAN countries and Belt and Road nations for collaborative model development [6]. Strategic Positioning - Zhipu's CEO emphasizes the company's commitment to pre-training models, despite industry trends moving towards post-training and inference models [3][12]. - The company aims to balance its technological advancements with commercial strategies, ensuring that both aspects support each other dynamically [21]. Future Outlook - The article suggests that Zhipu is optimistic about achieving significant growth in 2025, with expectations of a tenfold increase in market opportunities, while maintaining a stable commercialization strategy [22].

Core Insights - The article discusses the emergence and significance of DeepSeek, a new high-performance reasoning model from China, highlighting its open-source nature and the implications for the AI landscape [3][4][12]. Group 1: DeepSeek Overview - DeepSeek has gained attention for its performance on AI model rankings, raising both interest and concerns [3]. - The model's open-source release of weights and technical details provides valuable insights into reasoning models and their future development [4][12]. Group 2: Training Process - The training of DeepSeek involves three main steps: pre-training on vast datasets, supervised fine-tuning (SFT) with human-generated examples, and reinforcement learning with human feedback (RLHF) [6][9][10]. - The training process is designed to enhance the model's ability to provide accurate and contextually relevant answers, moving beyond simple question-answering to more complex reasoning [11][12]. Group 3: Innovations and Techniques - DeepSeek R1 represents a culmination of various innovations, including self-learning capabilities and multi-stage training processes that improve reasoning abilities [11][13][14]. - The model employs a mixture of experts (MoE) architecture, which allows for efficient training and high performance in reasoning tasks [15][30]. Group 4: Performance and Cost - The cost of training DeepSeek V3 was approximately $5.5 million, with the transition to R1 being less expensive due to the focus on reasoning and smaller-scale SFT [27][29]. - The article notes that the performance of reasoning models has significantly improved, with DeepSeek R1 demonstrating capabilities comparable to leading models in the industry [31][35]. Group 5: Future Implications - The rise of reasoning models like DeepSeek indicates a shift in the AI landscape, necessitating increased computational resources for inference and testing [31][34]. - The open-source nature of these models fosters innovation and collaboration within the AI community, potentially accelerating advancements in the field [36][39].

Core Viewpoint - High-tech companies consistently develop roadmaps to mitigate risks associated with technology planning and adoption, especially in the semiconductor industry, where performance and capacity limitations can hinder business operations [1][2]. Group 1: Nvidia's Roadmap - Nvidia has established an extensive roadmap that includes GPU, CPU, and networking technologies, aimed at addressing the growing demands of AI training and inference [3][5]. - The roadmap indicates that the "Blackwell" B300 GPU will enhance memory capacity by 50% and increase FP4 performance to 150 petaflops, compared to previous models [7][11]. - The upcoming "Vera" CV100 Arm processor is expected to feature 88 custom Arm cores, doubling the NVLink C2C connection speed to 1.8 TB/s, enhancing overall system performance [8][12]. Group 2: Future Developments - The "Rubin" R100 GPU will offer 288 GB of HBM4 memory and a bandwidth increase of 62.5% to 13 TB/s, significantly improving performance for AI workloads [9][10]. - By 2027, the "Rubin Ultra" GPU is projected to achieve 100 petaflops of FP4 performance, with a memory capacity of 1 TB, indicating substantial advancements in processing power [14][15]. - The VR300 NVL576 system, set for release in 2027, is anticipated to deliver 21 times the performance of current systems, with a total bandwidth of 4.6 PB/s [17][18]. Group 3: Networking and Connectivity - The ConnectX-8 SmartNIC will operate at 800 Gb/s, doubling the speed of its predecessor, enhancing network capabilities for data-intensive applications [8]. - The NVSwitch 7 ports are expected to double bandwidth to 7.2 TB/s, facilitating faster data transfer between GPUs and CPUs [18]. Group 4: Market Implications - Nvidia's roadmap serves as a strategic tool to reassure customers and investors of its commitment to innovation and performance, especially as competitors develop their own AI accelerators [2][4]. - The increasing complexity of semiconductor manufacturing and the need for advanced networking solutions highlight the competitive landscape in the AI and high-performance computing sectors [1][4].

根据国泰君安的研报，自从 DeepSeek 爆火之后，接入他们大模型的需求在短时间内迅速增加。从 2 月初至 今，腾讯、百度、阿里等互联网大厂，不仅在各自的云计算平台上线了 DeepSeek 模型。在直接面向用户的 业务上，即使这些巨头都拥有自己的大模型，但依然让旗下的部分应用接入了 DeepSeek。其中，包括月活 跃用户量达 13.8 亿的微信，以及曾因广告收入受影响、对 AI 搜索存在顾虑的百度。 除了互联网大厂，吉利、一汽大众等几十家车企、华为等主流手机厂商、三大电信运营商，也都在短时间 内完成了接入。甚至有些银行、券商、公募基金，以及国内部分地区的各类政府部门，也加入了这个行 列。比如，有些银行把 DeepSeek 应用到了面向用户的智能客服上。深圳、广州、呼和浩特、无锡等地的政 府，也宣布在政务系统中接入了 DeepSeek 模型，希望提升政务办公效率和群众办事体验。 那么，从汽车品牌到券商甚至政府，为什么大家纷纷都想要接入 DeepSeek？ ▲ 近日，吉利汽车正式官宣其自研大模型与 DeepSeek 已完成深度融合。| 图源：吉利汽车集团微信公众号 财新的报道指出，腾讯等大厂积极接入 Deep ...

点击 上方 硬AI 关注我们 如果Scaling Law继续有效， 继续看好AI系统组件供应商（如芯片、网络设备等），谨慎对待那些不得不持续投入巨额资 本支出的科技巨头。如果预训练缩放停滞： 看好科技巨头（因为自由现金流将回升），并关注那些拥有大量用户、能够 从推理成本下降中获益的应用类股票。 硬·AI 作者 |硬 AI 编辑 | 硬 AI 还抱着"越大越好"的AI模型不放？华尔街投行巴克莱最新研报给出了一个颠覆性的预测： AI行业正经历一 场"巨变"（Big Shift），"推理模型"和"Agent"将成为新时代的弄潮儿，而"大力出奇迹"的传统大模型， 可能很快就要过气了！ 这场变革的核心，是AI模型从"死记硬背"到"举一反三"的进化。过去，我们追求更大的模型、更多的参 数、更海量的训练数据，坚信"量变产生质变"。但现在，巴克莱指出，这条路可能已经走到了尽头。 算力无底洞、成本高企、收益却难以匹配……传统大模型的"军备竞赛"让众多科技巨头苦不堪言。更要命 的是，用户真的需要那么"大"的模型吗？在许多场景下，一个更"聪明"、更会推理的小模型，反而能提供 更精准、更高效的服务。 这究竟是怎么回事？对于投资者来说 ...

Core Viewpoint - The National Supercomputing Internet Platform announced the launch of Alibaba's open-source inference model QwQ-32B API interface service, offering users 1 million free tokens [1] Group 1: Product Launch - The QwQ-32B is the latest inference model released by Alibaba's Qwen team, built on Qwen2.5-32B with reinforcement learning [1] - The API service for QwQ-32B will be available starting this week [1] Group 2: Performance Metrics - According to official benchmark results, QwQ-32B performs comparably to DeepSeek-R1 on the AIME24 assessment set for mathematical capabilities and significantly outperforms o1-mini and similarly sized R1 distilled models in code evaluation on LiveCodeBench [1]

Core Insights - Alibaba has released and open-sourced a new inference model named Tongyi Qianwen QwQ-32B, which features 32 billion parameters [1] - The performance of this model is comparable to that of DeepSeek-R1, which has 671 billion parameters, with 370 billion of those being activated [1]

Market Overview - The main board led the gains with notable increases in stocks such as Demingli (+6.12%), Heertai (+4.03%), and Yingfangwei (+2.86%) [1] - The Sci-Tech Innovation Board saw significant growth, particularly with Yuncong Technology-UW (+19.98%) and Tiande Yu (+13.56%) [1] - Active sub-industries included SW Digital Chip Design (+0.55%) and SW Passive Components (+0.33%) [1] Domestic News - CINNO Research reported that the total investment in China's semiconductor industry for 2024 is projected to be 683.1 billion RMB, a decrease of 41.6% year-on-year, although semiconductor equipment investment grew by 1.0% to 40.23 billion RMB [1] - A strategic cooperation agreement was signed between Jinghe Integration and Sitwei, marking a significant upgrade in their partnership, with plans for monthly delivery capabilities of 15,000 and 45,000 Stacked wafers in different phases [1] - DeepSeek announced the release of three optimized parallel strategies, enhancing GPU utilization through detailed computational and communication optimizations [1] - Chip Origin announced the launch of its latest AI image processing IP series, including AINR1000, AINR2000, AISR1000, and AISR2000, aimed at various sectors such as automotive and consumer electronics [1] Company Announcements - Huahai Chengke reported a revenue of 332 million RMB for 2024, a year-on-year increase of 17.21%, with a net profit of 40.8 million RMB, up 28.97% [2] - Tiancheng Technology announced a revenue of 381 million RMB for 2024, reflecting a 12.32% year-on-year growth, and a net profit of 76.84 million RMB, up 31.19% [2] - Weidao Nano reported a revenue of 2.7 billion RMB for 2024, a significant increase of 60.74%, attributed to growth in the photovoltaic and semiconductor sectors [2] - Chip Source Micro reported a revenue of 1.77 billion RMB for 2024, a 3.09% increase, with a net profit of 211 million RMB, supported by successful validation of its high-temperature sulfuric acid cleaning machine [2] International News - CounterPoint Research projected that global TV shipments will reach 230 million units in 2024, a 2% year-on-year increase, with China surpassing South Korea in shipments for the first time [2] - CTV Finance reported that global smart glasses sales are expected to reach 2.983 million units in 2024, with a projected fourfold increase in 2025, as over 40 companies, including Apple and Google, enter the market [2] - Nvidia announced that demand for inference is accelerating, driven by new models like DeepSeek R1 and OpenAI o3, with the high-end Blackwell Ultra chip expected to launch in the second half of the year [2] - TrendForce reported that global DRAM industry revenue is expected to exceed 28 billion USD in Q4 2024, a 9.9% increase from the previous quarter, driven by rising contract prices for Server DDR5 [2]

从大语言模型到推理模型。 文丨 贺乾明 但 OpenAI CEO 山姆·阿尔特曼（Sam Altman）的好心情很快就被打断。在他宣布 o1 全量上线的推文下， 排在第一的评论是："到底什么时候能用上新的语音功能？？" 他立刻反击："能不能先花几个星期感谢感 谢这魔法般的智能，然后再要新玩具？" 这位用户追着阿尔特曼要的不是什么新玩具，是 OpenAI 在今年 5 月就允诺即将到来的 GPT-4o 端到端语 音功能。在当时的现场演示中，这个新的 AI 声音自然、反应极快，还知道什么时候插话，让旁人难辨真 假。按官方时间表，上千万 ChatGPT 付费用户本将在几周内用上这功能，但一直被跳票到现在。 过去一年里，OpenAI 的产品都是类似的 "期货"：GPT-4 已上线一年多，OpenAI 的下一代模型 GPT-5 依 然没有发布迹象。OpenAI 今年初发布的视频模型 Sora 也没有大规模开放，到现在都只有少数被他们挑选 的行业人士实际用过。 行业第一的跳票一次次磨损着资本市场对 AI 大模型的耐心。一些中国科技巨头和大模型公司今年年中暂 缓训练基础模型，把更多资源投到应用开发，或把 GPU 算力租给外部 ...