Core Viewpoint - DeepSeek has released an upgraded version of its R1 model, named DeepSeek-R1-0528, which shows significant improvements in reasoning, programming, and reducing hallucinations compared to its predecessor [1][3][22]. Model Improvements - The new version retains the base model from December 2024 but has enhanced computational power, allowing for deeper reasoning and more detailed problem-solving [4][6]. - The average token usage for the AIME 2025 test increased from 12K to 23K tokens, resulting in an accuracy improvement from 70% to 87.5% [4][5]. Benchmark Performance - In various benchmarks, DeepSeek-R1-0528 achieved notable scores, such as 87.5% in the AIME 2025 math competition, outperforming its predecessor and showing competitive results against models like OpenAI's and Gemini 2.5 [5][15]. - The model's performance in coding tasks has reached levels comparable to OpenAI's models, with successful outputs in complex coding challenges [10][14]. Reduction of Hallucinations - The hallucination rate in the new model has decreased by 45% to 50%, leading to more reliable outputs in tasks such as summarization and reading comprehension [18]. Creative Writing Capabilities - DeepSeek-R1-0528 has shown improvements in creative writing, producing coherent and logical narratives without the previous issues of "getting stuck" [19][21]. User Reception - While some users express skepticism about the update's impact, many remain optimistic about DeepSeek's potential as a representative of domestic AI technology [22][23].

Group 1 - The Hong Kong stock market showed positive trends on June 4, with the innovative drug sector experiencing significant gains, including over 15% increase for Innovent Biologics and over 7% for Zai Lab and Tigermed [1] - The Hong Kong innovative drug ETF (159567) recorded a trading volume exceeding 1.1 billion yuan for three consecutive trading days, indicating high market enthusiasm [1] - The cost and speed advantages of Chinese teams in drug development are highlighted, with a 2-3 times cost advantage and approximately 2 times speed advantage compared to traditional methods, achieving comparable performance to U.S. teams using AI [1] Group 2 - The innovative drug ETF (159567) tracks the National Index of Hong Kong Innovative Drugs, with 90% of its weight in innovative drug companies, positioning it to benefit from trends such as AI-enabled drug development and the expansion of domestic innovative drugs [2] - The innovative drug ETF (159992) encompasses leading companies in the innovative drug industry chain, benefiting from both AI advancements and the introduction of new healthcare policies [2] - Head innovative drug companies have entered a profitability cycle, with multiple products commercializing and driving rapid revenue growth, supported by the upcoming introduction of a new healthcare payment policy in 2025 [2]

Core Insights - The article highlights the increasing recognition among U.S. pharmaceutical insiders of China's innovative drug sector reaching a "DeepSeek moment," where U.S. companies are investing heavily in Chinese new drugs as a strategic bet on their potential [1][5] - The competitive advantage of Chinese innovation still hinges on breakthroughs in core innovation, despite the current advantages in cost and efficiency [1][4] Group 1: Investment Trends - Pfizer acquired overseas rights for a PD-1/VEGF dual antibody from 3SBio for over $6 billion, including an upfront payment of $1.25 billion [1] - Bristol-Myers Squibb secured co-development rights for a similar drug from BioNTech for $11.1 billion, with an upfront payment of $1.5 billion [1] - Chinese companies have sold overseas rights for innovative drugs to U.S. firms, with total commitments nearing $30 billion [3][12] Group 2: Competitive Landscape - There are currently 35 PD-1/VEGF dual antibodies in development globally, with 20 originating from China, indicating a strong clinical advancement from Chinese firms [2] - Chinese companies have evolved from being five years behind in PD-1 monoclonal antibodies to being approximately three years ahead in the dual antibody space [3] Group 3: Research and Development Efficiency - Chinese firms are leveraging engineering optimization as a competitive advantage, combining different monoclonal antibodies and small molecules to create innovative therapies [4] - The cost and speed advantages of Chinese teams in drug development are significant, with estimates suggesting 2-3 times lower costs and about double the speed compared to U.S. teams [6] Group 4: Market Dynamics - In Q1 of this year, 37% of innovative drug transactions with upfront payments over $50 million originated from Chinese companies, nearly doubling from two years ago [7] - Chinese companies now account for 67% of global transactions in oncology, 60% in cardiovascular diseases, and 50% in endocrine and autoimmune diseases [7] Group 5: Future Outlook - The trend of Chinese innovation in pharmaceuticals is expected to continue expanding into other therapeutic areas, with increasing competition anticipated between Chinese and U.S. firms [10] - The "DeepSeek moment" in Chinese biopharmaceuticals is raising awareness among U.S. markets, prompting significant acquisitions and strategic partnerships [10][12]

Core Insights - Mary Meeker's AI trends report spans 339 pages, covering various aspects of AI technology, applications, and innovations [1] - The report identifies the launch of DeepSeek's reasoning model R1 in January 2025 as a significant event marking the global competition in AI [2] Company Developments - DeepSeek's R1 model achieved performance comparable to OpenAI's models at a lower cost, disrupting the traditional paradigm of expensive closed-source models [3] - Despite initial competition from DeepSeek, OpenAI has seen significant growth, with its valuation reaching $300 billion and active users increasing from 400 million to 800 million [10] - OpenAI's annual revenue surged from $3.7 billion to $12.7 billion, indicating strong market demand for its offerings [10] Market Dynamics - DeepSeek's R1 initially outperformed OpenAI in website traffic and app downloads, but its metrics have since declined [11] - The AI market is experiencing a phase of homogenization and commoditization, with performance differences among leading models becoming minimal [12] - Users prioritize performance and differentiation, with OpenAI maintaining a strong brand presence and Anthropic emerging as a preferred choice for programming models [12] Future Trends - The next phase of AI commercialization may not be a "winner-takes-all" scenario but rather a fusion and reconstruction of platforms and applications [14] - DeepSeek aims to achieve AGI by integrating feedback from its infrastructure, products, and user interactions, positioning itself as a potential horizontal platform for developers [15]

面对挑战，薛先生以BIM技术实施为核心，组建了专业的BIM土建、机电、现场团队，实现了BIM技术 与工程场景的深度融合。在项目实施过程中，技术团队制定各专业BIM建模标准，规范了模型信息，统 一标准化，比如编制了淮南金融广场项目BIM技术实施策划、BIM土建建模标准、BIM机电建模标准、 BIM应用施工方案，指导项目BIM应用实施，确保各方人员在统一制度下高效协作，为后续施工、可视 化管理奠定了基础。 基于丰富的施工管理经验，薛先生领导技术团队通过项目BIM模型基于Revit平台创建工作项目文件， 为保证建模过程的效率、便于应用及检查修改模型，建筑和结构模型采取分层建模的方式搭建，机电模 型采取分专业分系统的方式搭建，在项目施工前，就利用模型对施工方案进行模拟论证，提升了工作效 率。薛先生通过BIM目标应用的制动，形成了包括BIM技术中以三维场地布置、图纸会审、可视化技术 交底等常规应用十项，机房安装指导和信息化平台开发及应用的两项创新型重点应用实践。通过创新的 信息化平台，实施实施过程中发出预警信息200多条，上传质量、安全类问题89条，管理层通过数据进 行宏观调控和精细化管理，为项目的质量、安全、成本、进 ...

由 文心大模型 生成的文章摘要 百度AI搜索宣布全面接入深度思考模型DeepSeekR1-0528，用户可通过PC端、App端免费使用这一升级 后的搜索服务。此前，百度智能云千帆大模型平台已于5月30日正式上架DeepSeekR1-0528大模型，为 企业用户和开发者提供了新的选择。 5月28日晚，DeepSeekR1模型完成小版本试升级，在模型架构、推理能力和开放性等方面取得显著进 展。这一升级引起了业内的广泛关注，腾讯旗下元宝、ima等AI应用迅速响应，率先完成升级并接入 DeepSeekR1-0528最新版。百度紧跟其后，实现了AI搜索对该版本的全面接入，进一步提升了其搜索服 务的智能化水平。 百度AI搜索全面接入DeepSeekR1-0528，为 据官方介绍，DeepSeekR1-0528模型在推理能力上有了显著提升，能够更精准地理解用户意图。这意味 着用户在使用百度AI搜索时，将获得更具个性化和准确性的搜索结果。无论是日常问题咨询，还是复 杂的专业知识查询，该模型都能深入分析用户需求，给出更贴合实际的答案。 在写作表达方面，DeepSeekR1-0528更具人性化风格。其生成的内容信息丰富，不仅涵盖 ...

量子位智库《AI APP TOP 50 用户新增》 作为首个中小学生专属教育AI，其成功密码在于对教育本质理解的深度。 小猿AI底层是猿辅导集团十余年来积累的高质量数据集，其中包括以500万小时+教学视频、20亿+题库、1万+知识点、100万+试卷构成的内容数据，全球5 亿用户积累的300亿学情数据。这些数据成为高精度学情分析、个性化学习路径规划的坚实基础。 央广网北京5月30日消息 QuestMobile AI产业研究院最新报告揭示，2025年3月移动端AI应用月活用户突破6.47亿，意味着超半数中国网民已迈入"AI原生应 用"时代。在这场技术浪潮中，教育AI强势突围。 猿辅导集团副总裁王向东曾强调："大模型时代，教育AI绝不仅是参数与速度的比拼，更是对教育本质理解的较量。"这一判断在小猿AI的市场表现中得到验 证：据量子位智库5月19日数据，这款专为中小学生打造的教育AI，上线首月即登顶教育领域AI用户新增榜首，并跻身全网AI应用新增下载TOP 10。 小猿 AI APP 凭借全场景学习解决方案，为中小学生构建起覆盖作业检查、学情诊断、错因分析、1v1讲解等 100 多个关键学习场景的智能学习生态。其 中 ...

5月29日，腾讯旗下腾讯元宝、ima、搜狗输入法、QQ浏览器、腾讯文档、腾讯地图、腾讯乐享等多款 AI应用宣布接入DeepSeek R1-0528，用户在不同产品中，选择DeepSeek模型R1深度思考，都可以体验 到DeepSeek R1-0528最新的深度思考、编程和长文本处理等能力，免费且不限量使用。同时，腾讯云也 上线了DeepSeek-R1-0528，企业和开发者可以通过腾讯云直接调用API接口，获得稳定优质的服务；也 可以通过腾讯云智能体开发平台内置的RAG、工作流和智能体开发的能力，快速搭建专属智能体应 用；此外，还能通过腾讯云TI平台对模型进行精调。（全天候科技） ...

DeepSeek的R1模型自1月发布以来，曾因多项指标优于西方竞争对手而引发全球关注，并导致科技股暴 跌。然而，自V3模型发布后，DeepSeek的热度有所下降，市场更关注其R2模型的发布。此前有消息 称，R2将采用混合专家模型（MoE），参数量预计达1.2万亿，单位推理成本较GPT-4大幅降低。但截 至目前，DeepSeek仍未正式确认R2的发布时间。 R1推理模型的最新版本R1-0528再次引发了全球媒体的广泛关注。 路透社报道称，此次发布加剧了与美国竞争对手如OpenAI的竞争。在由加州大学伯克利分校、麻省理 工学院和康奈尔大学研究人员开发的LiveCodeBench排行榜中，DeepSeek的R1-0528在代码生成方面仅 略低于OpenAI的o4 mini和o3推理模型，但超过了xAI的Grok 3 mini和阿里巴巴的Qwen 3。 CNBC指出，DeepSeek此次升级模型发布低调，但其推理能力使其能够执行更复杂的任务。《华尔街日 报》中文版则提到，DeepSeek的低成本、高性能R1模型自年初以来吸引了全球目光，推动了中国科技 股的上涨，也显示出市场对中国AI能力的乐观预期。 DeepSeek ...

特朗普政府关税的命运一波三折，先是被美国国际贸易法院的裁决"叫停"，特朗普政府很快提起上诉， 之后上诉法院又"叫停"了贸易法院的裁决。美东时间周四，美国联邦上诉法院批准了特朗普政府的请 求，暂时中止前一日一家下级法院禁止执行美政府多个关税行政令的裁决。 美国总统特朗普在其第二任期内首次会见了美联储主席鲍威尔。特朗普批评鲍威尔未能降低借贷成本。 鲍威尔则坚称，任何决定都不会基于总统的要求。根据美联储的一份声明，特朗普邀请鲍威尔与他会 面，"讨论经济发展，包括增长、就业和通胀。"鲍威尔没有讨论他对货币政策的预期，只是强调政策路 径将完全取决于即将发布的经济信息及其对前景的意义。鲍威尔表示，他和FOMC的同事们将完全基于 谨慎、客观和非政治的分析做出政策决定。 美股方面，道指涨0.28%，纳指涨0.39%，标普500指数涨0.4%。 大型科技股多数上涨，亚马逊、Meta、特斯拉涨幅不足1%，英伟达涨超3%，博通涨超1%。 热门中概股多数收涨，纳斯达克中国金龙指数涨1.44%，京东涨4.31%，小鹏涨3.98%，新东方涨 3.61%，再鼎医药涨3.4%，百胜中国涨2.97%，携程涨2.2%，理想涨1.1%，阿特斯太阳 ...