组委会 发自 凹非寺 量子位｜公众号 QbitAI 「2025人工智能年度榜单」申报 已进入倒计时阶段。 今年是量子位 「2025人工智能年度榜单」评选报名 的 第8年。 八年来，我们见证了技术的突破与落地，产业的融合与重塑，也见证了一批 又一批推动时代前行的企业、人物与产品。 本次评选已经从 企业 、 产品 、 人物 三大维度，设立五类奖项。欢迎企业抓住最后时间，尽快报名！ 让我们共同见证年度之星，点亮未来的方向。 企业榜 产品榜 人物榜 2025 人工智能年度 焦点人物 报名方式 本次评选将于 2025年11月17日 截止。评选结果将于量子位主办的 MEET2026智能未来大会 上正式公布。 2025 人工智能年度 领航企业 2025 人工智能年度 潜力创业公司 扫描二维码即可报名评选： 网页端链接：https://wj.qq.com/s2/23740133/iso8/ 2025 人工智能年度 杰出产品 2025 人工智能年度 杰出解决方案 1、注册地在中国，或主营业务主要面向中国市场； 2、主营业务属于人工智能及相关产业，或已将人工智能广泛应用于主营业务，并在细分领域居于行业领先地位； 3、具备成熟的产 ...

Core Viewpoint - The article discusses the recent actions of SoftBank's Masayoshi Son regarding NVIDIA, highlighting concerns about potential market bubbles and the implications of heavy investments in AI, particularly in OpenAI [3][10][12]. Group 1: SoftBank's Actions - SoftBank has completely liquidated its NVIDIA holdings, cashing out approximately $5.83 billion (around 41.5 billion RMB), which led to a significant drop in NVIDIA's market value by $100 billion overnight [3][10]. - This marks the second time Son has sold off NVIDIA shares, with a previous liquidation in 2019 that has since been viewed as a cautionary tale in the investment community [10][11]. - The decision to sell NVIDIA is seen as a strategic move to free up capital for investments in OpenAI, with SoftBank's CFO stating the need for liquidity to fulfill commitments to OpenAI [12][14]. Group 2: Investment Strategy and Risks - Son's investment strategy appears to focus on concentrated bets on leading companies rather than diversified hedging, which has led to significant losses in past investments, such as WeWork and Coupang [7][11]. - The article raises concerns about the sustainability of OpenAI's business model, noting that it may face substantial losses and cash burn in the coming years, with estimates suggesting losses exceeding $5 billion by 2025 [18][21]. - There are indications that the current market environment may be experiencing a bubble, with warnings from prominent figures about the risks associated with inflated asset prices in the tech sector [18][19]. Group 3: Market Dynamics and Future Outlook - The article suggests that the heavy reliance on circular financing among tech companies, including OpenAI, creates an illusion of unlimited demand while facing real supply constraints, particularly in energy and resources [21][24]. - Historical patterns indicate that SoftBank's stock splits have often coincided with significant market downturns, raising questions about the potential implications of its upcoming stock split [15][18]. - The future of OpenAI is uncertain, with challenges related to energy supply and operational sustainability posing significant risks to its growth and profitability [22][24].

Core Viewpoint - ByteDance has developed a new gaming agent named Lumine, capable of autonomously playing games like Genshin Impact, showcasing advanced skills in exploration, combat, and puzzle-solving [1][4][16]. Group 1: Agent Capabilities - Lumine can perform complex tasks such as dynamic enemy tracking, precise long-range shooting, and smooth character switching, effectively handling various game scenarios [4][6][10]. - The agent demonstrates strong understanding in boss battles and can solve intricate puzzles, indicating high spatial awareness [6][8][10]. - Lumine is capable of executing GUI operations and can follow complex instructions with clear prior information, enhancing its usability in gaming [12][14]. Group 2: Technical Framework - Lumine is built on the Qwen2-VL-7B-Base model, leveraging multimodal understanding and generation capabilities acquired from extensive training on web data [16]. - The agent employs a unified language space for modeling operations and reasoning, facilitating seamless integration of perception, reasoning, and action [16][19]. - Three core mechanisms are designed for Lumine: Observation Space for visual input processing, Hybrid Thinking for decision-making efficiency, and Keyboard and Mouse Modelling for operational commands [19][22][23]. Group 3: Training Process - The training process consists of three phases: pre-training for basic actions, instruction-following training for task comprehension, and decision reasoning training for long-term task execution [25][27][29]. - Lumine-Base model emerges with core capabilities like object interaction and basic combat, while Lumine-Instruct model achieves over 80% success in short tasks [26][28]. - The Lumine-Thinking model can autonomously complete long-term tasks without human intervention, showcasing its advanced planning and reasoning abilities [30]. Group 4: Performance Evaluation - In comparative tests, Lumine-Base shows over 90% success in basic interactions but lacks goal-oriented behavior in untrained areas [39]. - Lumine-Instruct outperforms mainstream VLMs in task completion rates, achieving 92.5% in simple tasks and 76.8% in difficult tasks, demonstrating superior tactical planning [41]. - Lumine-Thinking completes main story tasks in Genshin Impact with a 100% completion rate in 56 minutes, significantly outperforming competitors like GPT-5 [44][45]. Group 5: Industry Implications - The development of gaming agents like Lumine represents a significant step towards creating general-purpose AI capable of operating in complex 3D environments [50][55]. - Companies like Google are also exploring similar paths with their SIMA 2 agent, indicating a broader industry trend towards utilizing gaming scenarios for training AI [52][56]. - The belief in the eventual transition of gaming agents into real-world applications highlights the potential for embodied intelligence in various sectors [56].

来源：环球网 【环球网科技综合报道】11月16日消息，据科技媒体 golem 报道，谷歌 AI Studio 平台测试的一款未命 名 AI 模型，在破译难辨认历史手稿领域取得重大突破，核心字符错误率仅 0.56%，性能已接近人类专 业研究者水平。 历史学家 Mark Humphries 通过专门开发的基准数据集，对该模型展开系统性测试。在处理五份 18 至 19 世纪的高难度历史手稿时，模型整体字符错误率约为 1.7%，且多数错误集中在标点符号和大小写规 范上，未涉及单词本身的识别偏差。若剔除这类非核心错误，模型字符错误率将降至 0.56%，相当于每 转写 200 个字符仅出现 1 处失误，其准确度已能与专注历史文献转写的专业工作者相媲美。 此次测试的手稿涵盖多种手写风格，不乏字迹潦草、拼写不规范、语法不一致等复杂样本，充分验证了 模型的适应能力。更值得关注的是，该模型超越了单纯的文字转写功能，展现出较强的推理能力。在处 理一份 18 世纪商人日记时，面对未标注计量单位的购糖记录 "145"，模型通过反向核算账本总价，结 合当时英国货币与重量单位的对应关系，成功推断出 "14 磅 5 盎司" 的准确含义。 不 ...

上海人 工智能实 验室 推出了一款革新的多模态生成理解一体化的扩散语言模型 —— Lumina-DiMOO 。基于离散扩散建模（Discrete Diffusion Modeling）， Lumina-DiMOO 打破了多模态任务之间的壁垒，在同一离散扩散框架下，完成从 文本→图像、图像→图像、图像→文本的全栈能力闭环。 从 Chameleon 到 Lumina-mGPT，再到 Janus-Pro—— 主流 "多模态统一模型"，几乎都基于 自回归（AR）架构。这些模型的架构存在显著缺陷： 1. 生成太慢：逐 token 生成，导致图像生成通常需要几分钟； 2. 生成质量受限：图像细节的表现力较弱，尤其是在高分辨率生成时，精细度无法保证； 3. 任务间无法无缝衔接：多模态的生成和理解任务往往分开处理，导致模型的通用性和效率受到制约。 过 去：自回归生成 的瓶颈 然而，Lumina-DiMOO 采用了纯离散扩散框架，彻底解决了上述问题。在这个全新的架构中，我们通过并行化的双向注意力机制和灵活的采样策略，实现了跨多 任务的高效融合，不仅加速了生成过程，还提升了生成质量。 现在：扩散语言模型的崛起 Lumina-Di ...

机器之心发布 机器之心编辑部 EverMind 团队近日宣布正式发布其旗舰产品 EverMemOS，这是一款面向人工智能智能体的世界级长期记忆操作系统。它旨在成为未来智能体的数据基础 设施，为 AI 赋予 持久、连贯、可进化的 "灵魂"。 近期， EverMemOS 在 LoCoMo 和 LongMemEval-S 等最主流的长期记忆评测集上，其表现已显著超越此前工作，成为新的 SOTA。 记忆能力：决胜下一代 AI 的分水岭 灵感源自人类大脑的记忆机制 EverMind 团队来源于 盛大集团（Shanda Group） ，这一曾引领中国数字创新浪潮的科技和投资集团。他们的灵感来自人类大脑的记忆机制：从感官信 号编码、海马体索引到皮层长期存储，前额叶与海马体协同完成记忆的形成与提取。这种「类脑」理念，成为 EverMemOS 设计的核心，让 AI 能够像人类 一样 思考、记忆与成长。 受限于 LLMs 固定的上下文窗口，AI 在长时程任务中会频繁'失忆'。这不仅导致记忆断裂、事实矛盾，更让深度个性化和知识一致性成为空谈。AI 无法利 用历史交互数据来理解用户，也无法保留上下文的中间数据，这使其应用价值大打折扣 ...

和任何人，去任何地方！ 复旦 大 学携 手阶 跃星辰 打破 "复制粘贴" 魔咒，重磅推出 全新 AI 合照生成模型 WithAnyone —— 只需上传照片，就能一键生成自然、 真实、毫无违和感的 AI 合照！ W ithAnyone 是什么？ WithAnyone 是一种个性化 AI 合照生成方法：给定一个人的一张照片，WithAnyone 可以生成 ta 的各种角度、各种表情；给定几个人的各一张照片，WithAnyone 可 以让他们同框，生成一张毫无违和感的 AI 合照。 以往的方法，无论是现象级的 InstantID，还是火出圈的 PuLID，都面临着一个困境：给它什么样的表情、角度，它就生成什么表情、角度，做不到一点变化。 而 WithAnyone 打破了这一魔咒！真的可以拿 ID 一致性与 ID 可控性双料冠军。 WithAnyone 效果如何？ 空口无凭，且看效果！ 还有多人合照： 真可谓是相当惊艳了！ 我们也试了一下更多人的场景： 三姐妹戴着同款墨镜其乐融融，三兄弟西装革履装酷耍帅。 过去的模型想做到生成得像，全靠 "复制粘贴 "，参考图的表情神态、头部姿态，照搬到生成图上，没有一点变化。而 ...

Group 1 - Zhengzhou High-tech Investment Fund, in collaboration with multiple entities, has established the Henan Huirong Artificial Intelligence Industry Investment Fund with a total scale of 1 billion yuan, focusing on AI and advanced technology sectors [2] - The fund aims to support the strategic layout of Zhengzhou as a national important future industry pilot zone, emphasizing core technology breakthroughs and deep application scenarios [2] - The High-tech Investment Fund has created a capital matrix consisting of 26 funds with a total scale exceeding 17 billion yuan, leveraging over 50 billion yuan in industrial implementation [2] Group 2 - Shantou Tianze High-tech Equity Investment Fund, the first industrial investment fund in Shantou High-tech Zone, has been established, focusing on new energy, new materials, and health sectors [3] - The fund employs an innovative investment model to address industry chain shortcomings and accelerate technology transformation [3] - Chip Link Capital announced the completion of fundraising for its first main fund, totaling 1.25 billion yuan, targeting hard technology sectors such as semiconductors and AI [4] Group 3 - Shenzhen Deep Investment Control and Bank of Communications Shenzhen Branch have signed a cooperation agreement for a 2 billion yuan fund focusing on Shenzhen's "20+8" industrial clusters [5] - The fund aims to provide long-term capital support for hard technology enterprises, integrating various resources to foster high-quality development [5] - Jiangsu Province's strategic emerging industry mother fund is set to invest in a 2 billion yuan AI-focused fund, targeting AI infrastructure and applications [6] Group 4 - Anker Innovations plans to issue H-shares and apply for listing on the Hong Kong Stock Exchange to enhance its global strategy and competitiveness [7] - Yushu Technology has completed its IPO counseling report and plans to apply for an IPO in mainland China, with a focus on consumer and industrial robotics [8][9] - Moore Threads has initiated its IPO process on the Sci-Tech Innovation Board, focusing on GPU technology for AI and digital twin applications [9] Group 5 - Jiliu Technology has completed nearly 100 million yuan in Series B financing, with funds aimed at product development and talent acquisition in the AI computing cluster sector [10][11] - The company has developed a high-performance AI infrastructure system, Galaxy HPAC, to support large model training and inference [10][11] - Yangli Pharmaceutical has completed a significant Series B financing round, focusing on innovative drug development for oncology and chronic diseases [12]

央视网消息：目前，第二十七届中国国际高新技术成果交易会正在举办，本次高交会上，具身智能机器人成为关注的热点。在现场，当灵巧手 遇上手语大模型，机器人手语交流员就"上岗"了。 高校科研成果加速创新 在中山大学展台，记者看到了这套"心脏性猝死综合防治人工智能可穿戴预警设备"，它不仅解决了传统设备穿戴不舒适、信号噪声大且弱等短 板，还可以通过深度学习算法，精准识别异常心电特征，预测猝死风险，为心血管疾病早期干预提供依据。 和记者互动的就是搭载了我国首个手语具身智能的手语交互系统的具身智能机器人，通过它搭载的具身智能模型，可以让普通的具备灵巧手功 能的机器人，拥有手语交互能力。 哈尔滨工业大学（深圳校区）计算机应用研究中心吕俊增介绍，手语库在国际上没有公开的数据集，他们通过手语合成系统，手工录入5000多 条手语到动作之间的交互。 作为科研项目，团队从2008年就开始设计这套手语库。随着具身智能的发展，他们将这套手语库同机器人灵巧手相结合，已经可以适配多款不 同自由度的机器人灵巧手。 此外，在本届高交会期间，超百所高校、科研院所也携前沿科研成果集中亮相。 这款彩色材料叫零功耗智能控温薄膜，是一种可以实现"冬暖夏凉"的 ...

Market Overview - The A-share market experienced both positive and negative movements during the trading week from November 10 to 14, with the Shanghai Composite Index reaching new highs in the first four days before a significant drop on Friday erased gains across many indices [2][4]. Index Performance - The Shanghai Composite Index saw a weekly decline of 0.18%, with a year-to-date increase of 19.06% [5]. - The ChiNext Index dropped by 3.01% for the week, but it has a year-to-date increase of 45.29% [5]. - The total trading volume in the market fell below the psychological threshold of 2 trillion yuan, indicating a potential limitation on market rebound height if it continues to shrink [10]. Market Sentiment and Trends - The market is currently characterized by a "slow bull" trend, although upward momentum is diminishing, requiring consolidation before further advances [8]. - Investors are becoming more cautious, leading to a consensus on market volatility, which may result in further fluctuations or downward corrections [8]. - The recent trading patterns suggest a tendency for "Friday declines" to be followed by "Monday recoveries," indicating a potential for short-term market corrections [8]. Economic and Earnings Outlook - Short-term economic recovery is expected to remain weak, with corporate earnings likely to continue their upward trajectory, supported by narrowing declines in PPI and improved earnings growth in Q3 compared to Q2 [9]. - The liquidity environment is anticipated to remain loose, despite a slowdown in capital inflows into the stock market [9]. Upcoming Events and Policy Implications - Significant events that could influence market sentiment include the upcoming China International Photovoltaic and Energy Storage Conference and the Nvidia Q3 earnings report, which is expected to be a key indicator for AI-related stocks [12][13]. - There are expectations for potential adjustments in LPR rates by the end of the year, driven by policies aimed at boosting domestic demand and stabilizing the real estate market [15].