在推理+搜索方向，通义实验室搜索团队已提出ZeroSearch、OmniSearch等工作，通过在特定的下游任务进行强化学习训练，让大模型在与 环境交互的过程中学习使用搜索引擎。 该团队认为， 仅在特定任务上训练会导致模型的泛化能力受限 ，难以适应更多场景下的检索推理任务。 受BERT模型在预训练过程中使用的掩码 （Mask） 预测任务启发，MaskSearch引入了 检索增强型掩码预测任务 。 也就是让模型使用搜索工具，预测文本中被遮蔽的部分，在预训练过程中有效学习通用的任务分解、推理等智能体策略，同时熟练掌握搜索引 擎的使用方法，为其后续适配多领域任务奠定基础。 通义实验室 投稿 量子位 | 公众号 QbitAI 为提升大模型" 推理+ 搜 索 "能力，阿里通义实验室出手了。 最新研究开源 全新 通用 预训练框架 —— MaskSearch ，在域内及跨域开放域问答任务上均较基线方法取得显著性能提升。 小模型甚至能媲美大模型表现。 不仅如此，MaskSearch可以兼容 监 督微 调 （ S FT） 和强化学习 （RL） 两种训练方法。 通过加入预训练的二阶段训练，MaskSearch相比只在下游任务进行 ...

Key Points Summary of NVIDIA's Earnings Call Company Overview - **Company**: NVIDIA - **Date of Call**: May 29, 2025 Core Industry Insights - **Industry**: Semiconductor and AI Technology - **Market Impact**: U.S. export controls are expected to significantly affect NVIDIA's revenue, particularly in the Chinese market, with an anticipated loss of $2.5 billion in revenue due to restrictions on the H20 data center GPU [2][4][26]. Financial Performance - **Q1 2026 Revenue**: NVIDIA reported a strong performance with total revenue of $44 billion, a 69% year-over-year increase. Data center revenue reached $39 billion, up 73% year-over-year [4]. - **H20 Revenue**: Confirmed $460 million in H20 revenue, but faced a $4.5 billion expense due to inventory and procurement obligations write-downs [4][26]. - **Gaming Revenue**: Achieved a record $3.8 billion in gaming revenue, a 42% increase year-over-year [2][18]. - **Network Business**: Revenue grew 64% year-over-year to $5 billion, with the Spectrum X product line exceeding $8 billion in annual revenue [2][13][16]. Product and Technology Developments - **Blackwell Product Line**: Contributed nearly 70% of data center computing revenue, with rapid growth and deployment of NVL 70 dual racks [5][6]. - **AI Factory Deployment**: Nearly 100 AI factories are operational, doubling GPU usage across various industries [7]. - **Nemo Microservices**: Widely adopted across industries, enhancing model accuracy and response times significantly [9]. - **Spectrum X and Quantum X**: New products launched to enhance AI factory scalability and efficiency [16]. Market Challenges and Opportunities - **Export Controls**: Anticipated to create an $8 billion negative impact in Q2, with a total estimated impact of $15 billion [3][26]. - **China Market**: Data center revenue from China is expected to decline significantly due to export restrictions, although over 99% of data center computing revenue comes from U.S. customers [2][17]. - **AI Spending Growth**: Projected near $1 trillion in AI spending over the next few years, driven by infrastructure investments [27]. Strategic Partnerships and Collaborations - **Partnerships**: Collaborated with Yum Brands to implement AI in 500 restaurants, with plans to expand to 61,000 [10]. - **Cybersecurity Solutions**: Leading companies like Checkpoint and CrowdStrike are utilizing NVIDIA's AI-driven security solutions [11][12]. Future Outlook - **Growth Confidence**: Despite challenges, NVIDIA maintains confidence in sustained growth for the year, driven by the removal of AI diffusion rules and strong performance in non-China business segments [30][31]. - **Investment in AI Infrastructure**: Significant investments in domestic manufacturing and AI infrastructure are underway, including new facilities in Arizona and Texas [24]. Additional Insights - **Gaming and AI PC Growth**: The gaming sector continues to thrive with a user base of 100 million, and new AI PC products are being introduced [18]. - **Automotive Sector**: Revenue from automotive reached $567 million, a 72% increase, driven by demand for autonomous driving solutions [20]. - **Professional Visualization**: Revenue in this segment was $509 million, with strong demand for AI workstations [19]. This summary encapsulates the key points from NVIDIA's earnings call, highlighting the company's financial performance, product developments, market challenges, and future outlook.

报告显示此前 R1模型幻觉率在21%左右。 在开源平台HuggingFace上发布R1模型的更新后，5月29日晚，DeepSeek终于发布了官方公告介绍这次版本的具体能力迭代细节，其中包括深度思考能力强 化、幻觉改善和创意写作更好等。 值得一提的是，DeepSeek提到，新版DeepSeek R1 针对"幻觉"问题进行了优化。与旧版相比，更新后的模型在改写润色、总结摘要、阅读理解等场景中，幻 觉率降低了 45%-50% 左右，能够提供更为准确可靠的结果。 所谓幻觉即大模型的"胡说八道"，此前DeepSeek的幻觉率并不低，也被不少用户和开发者所讨论。5月15日，SuperCLUE曾发布最新一轮中文大模型忠实性 幻觉测评结果，显示此前的DeepSeek-R1模型幻觉率在21%左右，在测评的国内模型中排名第五。 doubao-1.5-pr 在测评数据上，官方表示，更新后的R1模型在数学、编程与通用逻辑等多个基准测评中取得了当前国内所有模型中首屈一指的成绩，并且在整体表现上已 接近其他国际顶尖模型，如 o3 与 Gemini-2.5-Pro。 gemini-2.5-pro-p1 gemini-2.5-flash-t ...

图 点击 上方 硬AI 关注我们 英伟达CEO黄仁勋罕见称赞中国AI模型 "DeepSeek R1如ChatGPT般越思考越聪明" ，同时对H20相关影响表示痛 心"500亿中国市场对美关闭" 。尽管英伟达Q2预计将损失H20相关80亿收入，但Blackwell芯片正以周产72,000颗GPU创 纪录速度交付，支撑微软单季处理100万亿Token的"海啸般"推理需求。他信心满满地说："Blackwell 已全面投产，我们 已经开足马力。我们现在拥有多个重要的增长引擎。" 硬·AI 作者 | 龙 玥 他直言，尽管H20出口限制将对英伟达未来的业务产生重大的不利影响，但Blackwell产品线爬坡速度是公 司历史上最快的，推动了数据中心收入同比增长73%。他表示，英伟达在制造良率方面看到显著改善，机 架出货量正以强劲速度交付。GB300也预计本季度末量产发货。 以下为财报电话会要点内容总结： 01 编辑 | 硬 AI 隔夜，英伟达Q1财报重磅出炉。这家AI芯片巨头扛住了特朗普政府出口限制的压力，业绩总体稳健：Q1 营收超预期增69%，Q2料H20收入减少80亿，但总收入指引顶住出口限制压力。 财报发布后，英伟达C ...

美东时间5月28日，英伟达公布了2026财年第一财季业绩。尽管面临美国对人工智能芯片出口管制的压力，这家AI芯片巨头依然 交出了超出市场预期的答卷。 财报显示，英伟达当季营收为440.62亿美元，同比增长69%，环比增长12%；净利润为187.75亿美元，同比增长26%，环比下降 15%。 此前，英伟达发布公告称，第一财季将计提减值55亿美元，但是实际上只减值45亿美元。由于中国市场"抢购"需求，本季度H20 影响小于预期。 财报发布后，英伟达股价在盘后交易中上涨近5%，创下四个月以来新高。截至当日收盘，英伟达股价报134.81美元/股，市值约 为3.288万亿美元。在过去一段时间的回调和喧嚣后，资本市场似乎不再疯狂追逐增速，对于各类冲击的看待更加理性，各大机 构也纷纷上调英伟达目标股价。 Blackwell推动增长 具体看核心业务，作为英伟达的核心增长引擎，数据中心业务本季度实现391亿美元营收，同比增长73%，环比增长10%，占总 营收比例高达89%。 这一增长主要得益于全球对AI基础设施的持续需求，英伟达最新的Blackwell架构产品颇为畅销。英伟达首席财务官Colette Kress 表示，在本季度 ...

S-GRPO的全称为序列分组衰减奖励策略优化（ S erial- G roup Decaying- R eward P olicy O ptimization），旨在提升大语言模型 （LLM）的推理效率和准确性，解决 冗余思考 问题。 S-GRPO团队 投稿 量子位 | 公众号 QbitAI AI回答问题太慢太长且无用，有没有能让大模型提前停止思考的方法？ 华为提出了首个在Qwen3上还有效的高效推理方法—— S-GRPO ，突破了思维链 「冗余思考」 瓶颈。 通过 "串行分组 + 衰减奖励" 的设计，在保证推理准确性的前提下，让模型学会提前终止思考， 推理提速60% ，生成更精确有用的答案。 S-GRPO适合作为当前Post Training（训练后优化）范式中的最后一步，在确保模型预先存在的推理能力不受损害的情况下， 使能模型在思 维链的早期阶段即可生成质量更高的推理路径，并在思考充分后隐式地提前退出 。 S-GRPO对单条完整推理路径进行分段截断 OpenAI o1, Deepseek-R1等推理模型依赖Test-Time Scaling law解决复杂的任务。 然而，过长的思维链序列的生成也显著增加了 ...

点击招商研究小程序查看PDF报告原文 事件： 英伟达发布FY26Q1季报，本季营收440.62亿美元，同比+69.18%/环比+12.03%，剔除H20相关 费用影响后毛利率为71.3%，按剔除影响后同比-2.2pcts/环比-7.6pcts。综合财报及交流会议信 息，总结要点如下： 评论： 1、FY26Q1营收符合指引预期，H20禁售费用影响低于公司预期。 FY26Q1营收440.62亿美元，同比+69.18%/环比+12.03%，超指引预期（430亿美元）； FY26Q1 non-GAAP毛利率为61%，剔除H20相关费用影响后为71.3%，按剔除影响后同 比-2.2pcts/环比-7.6pcts，毛利率下降主要系H20带来的45亿美元费用影响；非GAAP运营费用 增长6%，系薪酬及员工增加。4月5日英伟达接到美国政府通知，表示H20出口到中国市场需要许 可证，公司在FY26Q1因H20产品库存过剩和采购义务而产生45亿美元的费用，由于部分原材料可 以重复使用所以45亿美元费用低于公司预期，FY26Q1 H20产品营收为46亿美元，在FY26Q1无法 额外交付约25亿美元的H20产品。 2、数据中心部门网 ...

Core Insights - Nvidia reported a 69% year-over-year revenue growth for FY26Q1, reaching $44 billion, driven by a significant increase in data center revenue, which grew 73% to $39 billion [1] - The company confirmed $4.6 billion in H20 revenue for the first quarter, but faced $2.5 billion in unfulfilled shipments, leading to a $4.5 billion impairment charge [1][3] - For Q2, Nvidia expects total revenue of $45 billion, factoring in an $8 billion reduction in H20 revenue due to export restrictions [1][8] Group 1: Financial Performance - Nvidia's overall revenue for FY26Q1 was $44 billion, a 69% increase year-over-year [1] - Data center revenue reached $39 billion, marking a 73% increase compared to the previous year [1] - The company anticipates Q2 revenue of $45 billion, with a potential variance of ±2% [1] Group 2: H20 Revenue and Impairment - H20 revenue for Q1 was confirmed at $4.6 billion, with $2.5 billion in shipments unfulfilled [3] - An impairment charge of $4.5 billion was recorded, primarily related to inventory and procurement commitments [3] - Future H20 revenue is expected to decrease by $8 billion in Q2 due to export restrictions [1][3] Group 3: Market Insights - Nvidia highlighted the importance of the Chinese market, noting it as a key player in the global AI landscape [1] - The company expressed concerns that isolating Chinese chip manufacturers from U.S. competition could enhance their international competitiveness [1] - Nvidia estimates a potential market size of $50 billion that may remain uncovered due to current export restrictions [3] Group 4: AI Infrastructure and Growth - AI is viewed as a transformative technology across various industries, requiring substantial infrastructure for deployment [4][5] - The company is entering a new phase of AI adoption, with inference capabilities becoming a critical component of computational workloads [5] - Nvidia is focusing on enterprise AI solutions, with products designed for local deployment and integration with existing IT systems [15] Group 5: Future Outlook - The demand for inference AI is experiencing exponential growth, indicating a significant shift in the AI landscape [9] - Nvidia is expanding its supply chain and production capacity to meet increasing customer demand for AI infrastructure [7] - The company is optimistic about future growth, driven by advancements in AI technology and infrastructure development [9][14]

这一技术突破直接推动国家电网台区线损治理水平跃升——应用该平台的区域，计量设备故障率平均降 低35%，线损治理成本下降28%，为电网企业年节省运维成本超亿元。 市场版图：从区域深耕到全国布局 截至2024年，致成电子计量诊断平台已覆盖浙江、北京、上海、四川等13个省份，累计服务用户超2亿 户，占全国智能电表用户总数的34.33%。其中，浙江作为核心市场，服务规模达数千万户，占全省智 能电表用户的近四成，成为区域数字化标杆案例。值得关注的是，致成电子在西南、西北等能源富集地 区的市场渗透率快速提升。在四川，平台深度融入当地"数字电网"建设，助力国网四川电力实现全省低 压台区拓扑自动绘制与故障快速响应;在甘肃、宁夏等新能源高占比省份，平台通过融合光伏出力数 据，创新性解决分布式能源计量偏差难题，相关技术方案被纳入《国家电网新能源计量技术白皮书》。 在国家"双碳"战略与能源数字化转型的浪潮中，电力计量设备的精准诊断与智能化运维成为电网企业降 本增效的核心命题。作为国内领先的能源数字化解决方案提供商，杭州致成电子科技有限公司凭借自主 研发的基于混合推理模型的电力计量设备故障诊断平台，在细分领域实现跨越式发展。截至20 ...

Core Insights - The article discusses the recent updates from DeepSeek and NVIDIA's performance, highlighting that NVIDIA's results were better than expected, with significant implications for the AI market and the company's future growth prospects [1][2]. Group 1: DeepSeek and NVIDIA Performance - DeepSeek has updated to version R1-0528, indicating improved reasoning capabilities, but it does not consider this version worthy of being labeled as "R2," suggesting that a true R2 will likely be based on V4 rather than V3 and will require more time for development [1]. - NVIDIA's H20 segment had an overestimation of $1 billion, reducing the total from $5.5 billion to $4.5 billion, as some materials can be reused, which was previously underestimated [1]. - The CFO mentioned that prior to April 9, NVIDIA shipped $4.6 billion worth of H20, but the export ban resulted in a $2.5 billion shortfall in Q1 shipments, leading to a significant capital expenditure gap of nearly $20 billion [1]. Group 2: Revenue and Market Demand - The H20 segment is expected to impact $8 billion in revenue for Q2, but if adjusted, Q2 revenue could exceed $54 billion, indicating strong demand for Blackwell GPUs, which accounted for 70% of DC revenue [1]. - Blackwell GPU revenue is projected to increase from $10 billion in Q1 to $30 billion in Q2, showcasing a rapid transition from H to B cards [1]. - The CEO emphasized a dramatic increase in reasoning demand, citing examples from major companies like Microsoft and Google, and noted that the emergence of agents will further drive this demand [1]. Group 3: Importance of the Chinese Market - The CEO reiterated the significance of the Chinese market, stating that the $50 billion market has effectively closed off U.S. industrial opportunities due to export bans, impacting the Hopper data center business [2]. - The article suggests that regardless of U.S. chip availability, AI development in China continues to progress, raising concerns about whether the largest AI market will operate on U.S. platforms [2]. - The export restrictions are seen as a catalyst for innovation among Chinese chip manufacturers, which could enhance their competitiveness internationally while potentially weakening the U.S. position [2].