以下文章来源于超级王登科 ，作者DK本人 超级王登科 . 飞机大炮，柴米油盐 人们总喜欢用一句"轻舟已过万重山"，来简化"已过"之前所有的摸索和挣扎。 伟大的产品、公司更是如此。 比如，TikTok。 独响团队成员Piaf（曾在TikTok负责早期用户增长，现独响COO）跟王登科（独响CEO）这样分享： 那段故事里充满了不确定性，却也是团队韧性被推至极限、最让人百转千回的阶段。王登科在记录Piaf视角 下Tiktok那一万个新增用户的文章里，说到： " 很羡慕字节，不是字节产品多厉害，多赚钱，而是在最开始的时候，确实有一个朝气蓬勃，充满干 劲，又愿意跳进水里扑腾的团队，这个团队包括 Piaf，包括卷卷（数美万物CEO任利锋），也包括现在 依然在字节的很多人。 " 勇气，并不体现在你认为是确定的地方，唯吾独尊；而是在自己不确定的地方，不气馁、不放弃、找到那 群"愿意跳进水里扑腾"的人，大家一起坚定地向前冲。 这篇王登科的分享，送给每一位"正在扑腾"的创业者，希望大家都能找到自己翻跃重山的韧性和勇气。 原文： 公众号「超级王登科」 《TikTok 也曾经很在乎那一万个新增用户》 我们团队的 Piaf 在 2017 ...

「锦秋AI实验室」 在这篇测评 《让 AI 分析这波大模型公司宣传战：原来每家都有自己的鲜明人设》 中，我们发现 8 家头部大模型公司的传播中，都有品牌拟人化的策 略。 这是一档专注于探索和评测AI产品在实际场景中应用效果的栏目。 我们正在 用AI 解锁100个效率场景。 下一个场景会是什么？ 这个发现让我们也不禁好奇： 如果锦秋是一个具像化的形象，会是什么样呢？ ➡️ 能否让AI帮我们理解和总结？ 如果要把这个具像化形象设计成为锦秋的IP形象，又会有什么样的结果呢？ ➡️ 能否让AI帮我们生成这个形象？ IP形象生成这件事情，动辄要大几十万的市场预算。 在往期的测评系列中，锦秋AI实验室多次整活，看AI是怎么成为品牌部门的外脑，帮我们做分析、做设计、做报告、做翻译、写文章。 ➡️爱玩AI的我们，为什么不自己做呢？ 依旧抱着"真用""真测"的理念，我们开始了这次有趣的创意实验： 让 AI 帮我们设计品牌IP形象。 我们用三轮测试和 无数次推翻重来 ，记录下了这场人与 AI 的设计博弈： 预想，目前的大模型能力对IP生成可能那么"输入-输出"那么简单。 能否帮市场部门找到通用的方法，解锁AI的又一个使用场景？ ...

Core Insights - The article discusses the evaluation of AI models in the context of the NeurIPS 2025 conference, focusing on how AI can assess research papers through a blind review process [2][10]. Group 1: Evaluation Methodology - The evaluation involved several AI models, including GPT5, Claude 4.5, and others, to conduct blind reviews of selected NeurIPS award-winning papers [7][8]. - Three complementary assessment scenarios were designed: full paper review, abstract-only review, and adversarial review to test the models' sensitivity to different framing [9][10]. Group 2: AI Review Outcomes - In the full paper review, the paper "Gated Attention for Large Language Models" received high scores, with GPT5 rating it as a Best Paper [13][16]. - The paper "1000 Layer Networks for Self-Supervised RL" also received favorable evaluations, with GPT5 giving it a score of 8.3 and recommending it for a poster presentation [21][43]. - The paper "Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?" was rated highly by multiple models, with Minimax even suggesting it as a Best Paper [28][46]. Group 3: Summary of Findings - The AI models generally agreed on the quality of the papers, with most scoring above 8 for technical correctness and significance [30][32]. - However, in adversarial reviews, the same papers faced significant criticism, leading to lower scores and recommendations for rejection, highlighting the models' varying perspectives based on the review context [55][57]. - The evaluations revealed a divergence between human and AI assessments, particularly in the adversarial setting, where AI reviewers were more critical [55][60].

Core Viewpoint - The article discusses the evolution of the content industry from efficiency to emotional engagement, emphasizing that AI will not replace creators but will enhance their value through interactive and imaginative content creation [4][5][51]. Group 1: Company Overview - Jinqiu Fund is the lead investor in the Series A financing of Xiangfa Liu, focusing on long-term investments in groundbreaking AI startups [5][8]. - Xiangfa Liu, founded in 2023, aims to build an AI co-creation content universe where users actively participate in creating characters, worlds, and stories [6][8]. - The core product "Zao Meng Ci Yuan" (Dream Dimension) has over 10 million users and an average daily interaction time exceeding 100 minutes, making it one of the most engaged AI content platforms [7][9]. Group 2: Interactive Content Ecosystem - The interactive content ecosystem primarily targets younger audiences, including teenagers and young adults, with a focus on storytelling rather than gameplay mechanics [16][18]. - The main challenge in this ecosystem is the insufficient content supply, which limits the engagement of younger users [19]. - AI advancements, particularly in large language models (LLMs) and multi-modal content, are transforming user interactions and enabling imaginative content creation [21][22]. Group 3: Monetization and IP Development - Creators on the platform can earn revenue through various means, including direct tips, subscriptions, and merchandise sales, with a focus on developing characters into influential IPs [23][25]. - The platform emphasizes the importance of continuous content creation around a single IP to build a rich narrative universe, enhancing user engagement and emotional connection [27][26]. - The shared ownership model of IP between the platform and creators fosters a collaborative environment for content development [28]. Group 4: Future Directions and Innovations - The company is focused on developing tools that lower the barrier for creators, allowing them to produce high-quality content with minimal technical skills [36][37]. - Upcoming features include innovative gameplay mechanics and the ability to convert user interactions into music and short films, expanding the creative possibilities [44][45]. - The strategic integration of tools, content platforms, and IP development is seen as a holistic approach to building a sustainable content ecosystem [47][49].

Core Insights - The article discusses the launch of Doubao Mobile Assistant, an AI Agent developed in collaboration with ZTE, designed to enhance mobile efficiency by automating complex tasks across applications [1][4][6] - The testing revealed both impressive capabilities and limitations, highlighting the need for further optimization and user experience improvements [5][30] Group 1: Product Features and Performance - Doubao Mobile Assistant operates as a system-level agent, capable of executing tasks without user intervention, utilizing visual recognition and contextual understanding [1][8] - During testing, the assistant demonstrated strong memory retention and task execution over extended periods, successfully navigating multiple applications [14][30] - The assistant's ability to adapt to user commands and switch between tasks was noted, particularly in scenarios involving complex navigation and information retrieval [15][23][30] Group 2: User Experience and Limitations - Users experienced delays in operation, particularly in tasks requiring rapid sequential actions, which affected overall efficiency [5][34] - Certain applications were not compatible, limiting the assistant's functionality and user engagement [34][36] - The assistant's performance varied based on task complexity, with some tasks requiring manual input due to recognition inaccuracies [18][34] Group 3: Future Implications and Industry Impact - The article suggests that Doubao Mobile Assistant represents a significant shift towards proactive AI agents that can manage user tasks autonomously [37][41] - The potential for integrating user context and enhancing AI capabilities is highlighted, indicating a future where AI can operate seamlessly across various applications [39][41] - The competition for user interaction points is expected to intensify, as the assistant aims to unify task management across disparate applications [47][49]

以下文章来源于四木相对论 ，作者关注AI的 四木相对论 . 写技术、写产品、写未来。写正在改变世界的人和事。 「Jinqiu Spotlight」 追踪锦秋基金与被投企业的每一个光点与动态， 为创业者传递一线行业风向。 锦秋基金已完成对 ArtArch 的投资。 锦秋基金，作为12年期的 AI Fund，始终以长期主义为核心投资理念，积极寻找那些具有突破性技术和创新商业 模式的通用人工智能初创企业。 创作正在从"拍-编-剪"时代，进入"用想象力构建作品"的时代。 一年完成两轮融资，前字节智能创作部门工程研发负责人黄严创立的 ArtArch， 正在用一套自研的"想象力引 擎"破解 AI 创作的最后一公里难题。 举个例子，这套引擎能将"一个小孩在雨里奔跑，突然抬头看到一条巨龙掠过天空"这样一句话，自动拆解成包含 运镜、光影、节奏的完整五镜分镜，编剧、导演、分镜师、特效师的反复沟通，被压缩成了几秒钟的系统自动编 排。 "真正的创作不是 prompt，而是大模型、多模态能力、工程化结构和创意逻辑的一体化协同。" 黄严所说的这套体系，本质上是在重新定义 AI 时代的创作范式： 让技术门槛消失，让想象力直接变成可执行的 ...

Group 1: Recent Financing Activities - Astribot completed a multi-hundred million yuan A++ round financing led by Guoke Investment and Ant Group, with participation from various notable financial institutions and industry capital, including continued support from Jinqiu Fund, which was the lead investor in the A round [1] - Lingqi Wanwu secured nearly 100 million yuan in three rounds of financing over four months, with the latest round led by Jinqiu Fund and participation from several other investors, focusing on a dual architecture model for human motion capture data [2] - Micronucleus completed over 100 million yuan in B round strategic financing led by BlueRun Ventures, showcasing strong market consensus on its 3D-CIM™ technology for AI computing applications [3] - VideoTutor announced the completion of a seed round financing of 11 million USD, led by YZi Labs, targeting K12 education with personalized video generation [4] - NemoVideo raised nearly 10 million USD in Pre-A and angel rounds, focusing on video creator tools and building a video production agent platform [5] Group 2: Technological Innovations - Yushu Technology launched a full-body remote operation platform that utilizes motion capture and real-time transmission systems to replicate human movements with a humanoid robot, demonstrating its application in various scenarios [8] - Diguo Robot introduced the S600, a high-performance development platform for embodied intelligent robots, and announced plans for a comprehensive development platform that integrates hardware and software [9] - Lingqi Wanwu released a demo video showcasing its algorithm in collaboration with Yushu's robot, achieving near-human fluidity in executing household tasks [10] Group 3: Industry Insights and Trends - Leonis Capital published a benchmark report analyzing the fastest-growing AI startups, highlighting a shift in capital investment towards computing power and data rather than human resources [14] - The first "Jinqiu Conference" featured discussions on entrepreneurial opportunities and trends in AI investment for 2025, with insights from various industry leaders [17]

以下文章来源于极客公园 ，作者金光浩 极客公园 . 用极客视角，追踪你最不可错过的科技圈。欢迎同步关注极客公园视频号 「Jinqiu Spotlight」 追踪锦秋基金与被投企业的每一个光点与动态， 为创业者传递一线行业风向。 锦秋基金已完成对Hogi 的投资。 锦秋基金，作为12年期的 AI Fund，始终以长期主义为核心投资理念，积极寻找那些具有突破性技术和创新商业 模式的通用人工智能初创企业。 最近 AI 圈出了一款有趣的产品：来自Hogi的 「OiiOii」，一款专注 AI 生成动画的 Agent。 它异常火爆，7210 个内测名额很快被抢光，闲鱼上免费邀请码被炒到 30 块，内测群有2万人，甚至据说内测用 户里还出现了全网 2000w 的顶级创作者。 出现这种现象级产品传播的背后原因，本篇文章（原创：极客公园）总结道： 技术上，Sora2 与 nanobanana2 让"人物一致性"这一 AI 视频动画的最大痛点被攻破，技术窗口正式打开， OiiOii 成为最快把前沿能力产品化、吃到红利的玩家。 需求上，在短视频时代，人人都有视觉表达需求，而 OiiOii 用简单工具补上专业产能的缺口，让动画创作从 ...

Core Insights - The article emphasizes the shift from "scaling up" large language models (LLMs) to "increasing capability density," highlighting the limitations of simply adding more computational power and data to larger models [2][3] - A new concept called "Densing Law" is introduced, which indicates that the capability density of LLMs is exponentially increasing, approximately doubling every 3.5 months [18][19] Group 1: Transition from Scaling Law to Densing Law - The article discusses the evolution from Scaling Law, which led to the development of large models like GPT-3 and Llama-3.1, to the need for improved inference efficiency [10] - Two core questions are raised: the ability to quantitatively assess the quality of different scale LLMs and the existence of a law reflecting LLM efficiency trends [10] - A quantitative evaluation method based on a reference model is proposed to address the non-linear relationship between capability and parameter size [11][12] Group 2: Capability Density and Its Implications - Capability density is defined as the ratio of effective parameter size to actual parameter size, allowing for fair comparisons across different model architectures [13] - The article notes that if the density (ρ) equals 1, the model is as efficient as the reference model; if greater than 1, it indicates higher efficiency [15] - A comprehensive evaluation of 51 mainstream open-source foundational models reveals that capability density has been increasing exponentially over time, leading to the establishment of the Densing Law [17] Group 3: Insights from Densing Law - The article identifies three key insights: 1. Data quality is a core driver of the Densing Law, attributed to the explosive growth in pre-training data and its quality [19] 2. Large models do not necessarily equate to high density, as training costs and resource limitations can hinder optimal performance [19] 3. The Densing Law reflects a pursuit of computational efficiency akin to Moore's Law in integrated circuits [19] Group 4: Predictions and Implications - The article predicts that the actual parameter size required to achieve the same performance level will decrease exponentially over time, with a case study comparing MiniCPM and Mistral models illustrating this trend [21] - It also notes that inference costs will decrease exponentially, with recent technological advancements in infrastructure contributing to this reduction [22][23] - The combination of Densing Law and Moore's Law suggests significant potential for edge-side intelligence, with the effective parameter scale on fixed-price hardware expected to double approximately every 88 days [24] Group 5: Acceleration of Density Growth Post-ChatGPT - Following the release of ChatGPT, the growth rate of model density has accelerated, with a notable increase in the slope of density growth trends [25] - Factors contributing to this acceleration include increased investment in LLM research, a thriving open-source ecosystem, and the proliferation of high-quality small models [28] Group 6: Challenges in Model Compression - The article cautions that compression techniques like pruning, distillation, and quantization do not always enhance density, as many compressed models exhibit lower density than their original versions [30] - It emphasizes the importance of ensuring that compressed models undergo sufficient training to maintain or improve capability density [30] Group 7: Future Directions in Model Training - The discovery of Densing Law suggests a fundamental shift in training paradigms, moving from a focus on size to efficiency per parameter [32] - Key dimensions for enhancing density include efficient architecture, advanced data engineering, and the collaborative evolution of large and small models [33][34][35]

「Jinqiu Select」 跨越语言与时差，传递科技圈最值得被听到的声音。 < Overview > CB Insights 发布的 《2025 Future Tech Hotshots：Scouting Reports》 报告，结合生成式 AI 分析与专有 Mosaic 评分体系，从全球海量初创企业中 遴选出 45 家最具潜力的科技公司。 这些崭露头角的科技新势力，覆盖企业科技、金融服务、医疗健康、工业、法律、零售与供应链六大领域，累计融资超 28 亿美元，平均 Mosaic 评分达 791（远超平均值 370），其中超 70%已进入商业化部署阶段。 < 六大行业的差异化特征 > 企业科技（22 家） ：数量最多，聚焦 AI 基础设施与开发者工具，平均融资规模最大 医疗健康（6 家） ：语音 AI 与临床工作流自动化主导，HIPAA 合规是入场券 工业（6 家） ：机器人与地理空间 AI 崛起，硬科技属性最强、验证周期最长 法律（2 家） ：AI 进入司法推理与合同审查，专有数据是关键护城河 零售与供应链（2 家） ：消费级 AI 应用与物流决策优化，离 C 端最近 AI 正从 "能回答" 演化为 "能 ...