进入2025年中，强化学习正在将AI的能力边界向多个维度同时拓展。从OpenAI o3开创的原生工具调用能力，到Claude Opus 4展现的连续编码任务处理能力，AI模 型正在向更实用的维度进化。 更令人瞩目的是，递归自我改进已经从科幻概念变为现实——模型开始参与训练下一代模型，通过优化编译器、改进内核工程甚至调整超参数来加速自身进化。 RL训练范式的兴起正在深刻重塑整个AI产业格局： 硬件需求从预训练的集中式架构转向分布式推理密集型架构，Nvidia NVL72系统专门为此增强内存；基础设施实现去中心化，RL各阶段可在不同地点执行； 组织结构随之重组，OpenAI、Anthropic等巨头纷纷合并推理与训练团队；数据正在成为新的护城河，拥有用户行为数据的企业可通过强化微调构建独特优势，无需 投入巨额预算合成数据。 SemiAnalysis的分析师Dylan Patel和AJ Kourabi写了一篇深度报告，系统剖析了强化学习是如何成为推动AI从当前水平向AGI迈进的关键技术的，以及在这个过程中 需要克服哪些技术、工程和资源挑战。 他特别预测了OpenAI的发展路线图： 即将到来的o4将基于推理成本更低但 ...

近日，由北京大学、北京通用人工智能研究院等单位组成的联合科研团队，开发出全球首个同时具备全手高分辨率触觉感知和完整运动能力的机器人手系统 ——"基于全手触觉的机器人仿生手"(简称F—TAC Hand)，展示了我国在机器人前沿技术领域的创新能力，对推动机器人技术的自主创新和国产化具有重要 意义。相关成果北京时间6月9日在国际顶级学术期刊《自然.机器智能》上发表。 手部既是人类改造自然与外界交互的重要器官，也是实现智能的关键载体。人的手部由27块骨骼和34块肌肉组成，提供了24个自由度的灵活性，具有结构高 度复杂、功能极为精密等特点。因此，对人类手部功能的研究是具身智能与机器人研究的前沿。 实验结果表明，相比没有触觉反馈的系统，F—TAC Hand在面临执行误差和物体碰撞风险时表现出显著的适应性优势，平均成功率从53.5%提升至了100%。 这项研究成果有望推动机器人技术在医疗、工业制造、特殊环境作业等领域的落地应用。论文通讯作者、北京大学人工智能研究院助理教授朱毅鑫表 示："未来，我们将继续深化触觉感知与机器人控制的结合，探索更加智能的体感交互范式，为实现真正意义上的通用人工智能奠定基础。" 人的手部在拿取物体 ...

HUAWEI X HUXIU 在通往通用人工智能（AGI）的路上，如何像其他领域一样实现弯道超车，是业界绕不开的 话题。 在过去的十余年时间里，各项单点技术飞速演进，但随着单点技术演进的边际效应递减和系 统复杂度的提升，系统性能的天花板逐步从单点技术的上限演变成系统工程上限：单点优势 越来越像是精致的零件，提升空间有限；但采用系统工程创新，各个部分完美配合、高效协 同，实现整个系统的效能最优，才有更积极的现实意义。 如何在发挥单点技术优势的同时，以整体视角重新构建路径，通过对复杂系统的极致把控与 再组织、找到新的突破可能？解决这个看似不可能的问题，就有望为我们独立引领最前沿技 术发展创造条件。 近期，虎嗅将推出《华为技术披露集》系列内容，通过一系列技术报告，首次全面详述相关 技术细节，为业界提供参考价值。 我们期待通过本系列内容，携手更多伙伴共同构建开放协作的生态系统，助力昇腾生态在中 国的蓬勃发展。 《华为技术披露集》系列 VOL.13 ：万卡集群 你是否注意到，现在的 AI 越来越 "聪明" 了？能写小说、做翻译、甚至帮医生看 CT 片，这 些能力背后离不开一个默默工作的 "超级大脑工厂"——AI 算力集 ...

当DeepSeek-R1以更低成本实现类似性能突破时，Claude能够连贯工作数小时完成复杂任务时，意味着AI发展已经迈入推理时代，强化学习技术的 重要性不言而喻，将重塑AI产业的技术栈乃至商业模式。 6月8日，AI研究公司SemiAnalysis发布长篇报告《强化学习：环境、奖励破解、智能体、扩展数据》，深度剖析了强化学习的工作原理以及影响 因素，并预测了后续AI发展趋势。 报告表示，强化学习（RL）或成为AGI前最后关键范式，其理密集型特性带来了算力挑战。此外，高质量数据是强化学习护城河，AI设计AI的循 环加速技术迭代。 1. 强化学习（RL）或成为AGI前最后关键范式：强化学习是推动大模型推理能力跃升的核心技术，尤其在思维链（CoT）生成和长 程任务连贯性上表现突出，被视作实现AGI前的终极技术路径。 2. 可验证奖励场景率先商业化：编码、数学等奖励函数明确的任务（如SWE-Bench性能提升30%+）已实现落地，OpenAI的o1、 DeepSeek-R1等模型验证其价值。医疗、写作等非验证领域通过"LLM评判者+人工评分标准"构建奖励函数（如HealthBench医疗 评估），OpenAI、阿里Q ...

Core Viewpoint - Morgan Stanley's report indicates that despite policy tightening pressures in the first half of 2025, the U.S. stock market is expected to enter a more optimistic scenario in the second half of 2025 and into 2026 [1] Economic Outlook - The firm does not foresee a recession but anticipates seven interest rate cuts in 2026, which will support above-average valuations [1] - The S&P 500 index target price is set at 6,500 points for the next 12 months, corresponding to an earnings per share (EPS) of $302 and a forward price-to-earnings (P/E) ratio of 21.5 times [1] - Projected EPS for 2025 is $259 (7% growth), for 2026 is $283 (9% growth), and for 2027 is $321 (13% growth) [1] Industry Allocation - **Overweight Sectors**: Financials, Energy, and Utilities are rated as overweight. The financial sector is expected to see a recovery in M&A and capital market transactions by 2028, with potential for accelerated stock buybacks due to regulatory easing [3] - The energy sector is linked to oil price movements, with geopolitical tensions potentially disrupting supply and raising prices. The sector's free cash flow (FCF) margins are significantly above historical averages [3] - Utilities historically perform well in late-cycle phases due to their defensive characteristics and are expected to benefit from rising interest rates and energy capacity concerns [3] Neutral Sectors - Technology, Healthcare, Communication Services, Materials, Real Estate, and Industrials are maintained at neutral. The technology sector shows significant internal differentiation, with AI-related stocks performing well, while hardware faces challenges from weak consumer demand [4] Underweight Sectors - Consumer Discretionary and Consumer Staples are rated for reduction due to poor pricing power and tariff risks. The consumer products sector faces significant cost pressures, with tariffs impacting EBITDA margins by 10% to 70% [4]

Group 1 - The core idea of embodied intelligence emphasizes that cognition is influenced by the agent's perception and actions, suggesting that intelligence arises from the interaction between the agent's body and the surrounding environment, rather than solely from brain function [1][2] - Embodied intelligence theory has profound implications across various fields such as cognitive science, psychology, anthropology, and art, leading to the emergence of sub-disciplines like embodied cognition and embodied psychology [1][2] - The transition from traditional disembodied intelligence to modern embodied intelligence marks a significant shift in artificial intelligence research, where the latter integrates physical interaction with the environment for learning and decision-making [2][3] Group 2 - The history of artificial intelligence has evolved through three stages: the first generation focused on knowledge-based reasoning models, the second generation introduced data-driven models, and the third generation, marked by the emergence of large language models, represents a new phase of development [3][4] - The introduction of large language models in 2020 has enabled machines to achieve free interaction with humans in open domains, indicating a significant step towards general artificial intelligence [4][5] - Despite advancements in language generation, there are still limitations in achieving domain generality across various tasks, particularly in complex areas like medical diagnosis, highlighting the need for embodied intelligence to bridge these gaps [5][6] Group 3 - The concept of embodied intelligence was first proposed in the field of robotics, emphasizing the importance of the interaction between the body and the environment in intelligent behavior [6][7] - Embodied intelligence has driven advancements in robotics technology, shifting from single-modal perception to multi-modal perception, which is crucial for applications like autonomous vehicles [8][9] - The integration of the agent concept in embodied intelligence allows robots to combine thinking, perception, and action, facilitating tasks in both digital and physical worlds, and enhancing the efficiency of robotic development through simulation [9]

近年来以千亿参数级的大模型为代表的人工智能技术在文本生成、图像理解和多模态推理等领域取得了 突破性进展，然而大模型的发展面临着"虚实鸿沟"的挑战：尽管其在数字世界中表现卓越，但如何将这 种能力转化为物理世界的实际价值，仍是横亘在行业面前的难题。 上海码极客／成都考拉悠然创始人、董事长申恒涛近日在接受记者采访时表示，大模型的终极价值不在 于参数规模的比拼，而在于能否实现从虚拟到现实的"能力落地"。空间智能技术的突破为实现这一目标 提供了关键路径。作为人工智能领域的下一个前沿技术方向，空间智能被认为是实现通用人工智能的关 键一环，其技术优势正在重塑人工智能与物理世界的互动模式，推动行业从"数字想象"迈向"物理实 效"的新阶段。 上海码极客／考拉悠然近日联合同济大学发布了悠然无界大模型以及MAGX空间智能体产品家族。"AI 的真正落地，要让数字世界和物理世界融合，希望通过悠然无界大模型，让每一个物理世界中的智能体 都能感知、理解、执行。"申恒涛表示，上海码极客依托多模态大模型技术优势资源，正在联合人工智 能产业链上下游企业，打造集"多模态世界模型+智能体硬件+行业应用"于一体的全栈空间智能技术体 系，建设开放、协作 ...

Industry Development - Yibin is constructing a "4+4+4" modern industrial system, focusing on characteristic advantageous industries and strategic emerging industries, transitioning from a "black and white" to a "blue and green" industrial structure [1] - The city has nine "billion-level" enterprises and over 1,000 industrial enterprises above designated size, with major industries including liquor, power batteries, and digital economy [1][10] - Yibin has been recognized as a national advanced manufacturing city and has established itself as a hub for the power battery industry, with a complete industrial chain and strong supporting capabilities [1][10][12] - The city has seen significant growth in employment, with over 100,000 new jobs created in the past three years, largely benefiting local university graduates [11][15] Employment - Yibin is enhancing employment guidance services, focusing on improving students' competitiveness from the source, aligning university programs with leading industries [3][4] - The city has established a "one person, one file, one strategy" support mechanism for key groups of graduates, providing tailored assistance for those who are unemployed [4][11] - Yibin has created 60 "Yijiu Employment" stations and 15 gig economy markets, successfully incubating 1,384 entrepreneurial projects that have generated employment for over 29,000 people [4][11] Ecological Development - Yibin is committed to building an ecological barrier in the upper reaches of the Yangtze River, focusing on green and low-carbon development [6][7] - The city has achieved a 100% excellent water quality rate in 22 national and provincial monitoring sections, with significant biodiversity protection efforts [7] - Yibin has established 21 national-level green factories and 17 "zero-carbon" factories, with a growing number of electric vehicles and renewable energy initiatives [7][12] Urban Development - Yibin is undergoing significant urban renewal, with a focus on improving infrastructure and enhancing the quality of life for residents [14][17] - The city has implemented a comprehensive planning strategy for its urban area, optimizing transportation and public services to support economic growth [14][17] - Yibin has been recognized as one of China's most livable cities, reflecting its commitment to high-quality urban development [17][18]

（文章来源：中国新闻网） 据悉，长三角各地商会、企业界人士，专家学者和政府相关部门负责人近500位与会。上海市政协原副 主席周汉民从DeepSeek崛起与《哪吒2》成功探索看人工智能时代的品牌建设，认为人工智能品牌成功 的核心要素是技术创新，是文化符号与科技产品的深度融合，未来人工智能品牌应重点关注技术伦理合 规，注重"五精"，即精妙构思、精确设计、精巧营销、精准定价、精良服务。 德国品牌应用科学大学校长、教授单凡从欧洲视角看中国品牌建设，认为近年来中国品牌的全球化发展 非常显著，特别是在欧洲和德国的发展，可以说是又快又深，在品牌文化和情感链接上做得越来越出 色。 中国科学技术大学客座教授方明认为，认知大模型掀起了人工智能第四次浪潮，认知大模型初步实现智 能涌现，标志着通用人工智能(AGI)时代的到来，将带来解决人类刚需的全新机遇。上海市在沪企业(商 会)联合会会长王均豪认为，新质生产力不仅体现在技术层面，文化也是新质生产力的一部分，真正把 新质生产力做到国际领先，这才是中国企业未来要拓展的新空间。 多位专家学者以及企业家从不同视角对长三角地区人工智能品牌发展的理论与实践进行梳理与总结， 对"十五五"长三角人 ...

悟道 1.0 发布时，学术界对" 大模型是通往 AGI 的技术路线 "尚未得出统一结论。 现在的具身智能，也处于这个阶段。 作者 | 王启隆 出品丨AI 科技大本营（ID：rgznai100） 大模型的热潮之下，一种微妙的瓶颈感，正成为行业共识。 "过往所说的 '百模大战'，更多是大语言模型的竞争，" 智源大会前夕， 智源研究院院长王仲远 在 与 CSDN 的对话中，开门见山地指出了问题的核 心，"而大语言模型受限于互联网数据的使用，性能虽然还在提升，但速度已大不如前。" 出路何在？在王仲远看来，AI 要突破天花板，就必须在"读万卷书"（互联网数据）后，去"行万里路"（物理世界）。 这并非孤立的判断。今年三月， 英伟达 CEO 黄仁勋就在 GTC 大会上为 AI 的下半场指明了方向 ：打造"AI 工厂"，迎接"物理 AI"时代，让 AI 走出屏 幕，与现实世 界交互。 思考趋于一致，行动便接踵而至。6 月 6 日，CSDN 在北京智源大会现场，见证了王仲远在他的主题演讲中给出的答案。如果说 2021 年的"悟道"系列 代表着对技术路径的探索（" 道 "），那么他所揭晓的全新"悟界"系列，则亮明了新的野心——用 ...