Jenny 是一个同时理解中美文化、创业、研究与投资的人。她从小在美国长大，2021 年加入 OpenAI，并在 ChatGPT 爆火一周后选择离开，合伙创立了自己的基 金。前一阵她回国，我们借机聊了聊中美 AI 创投之间的差异。 P.S. 本期节目录制于 2025.12.22。几天后，Manus 被微软收购的消息披露。回头再看，Jenny 分享的许多投资思路和对美国市场的判断，其实都有所映照。 本期播客原文约 23000 字，本文经过删减整理后约 7700 字。 曲凯 ：你这两年观察到的几个最主要的 milestones 是什么？ Jenny ：在 23 年，中美都有一个非常明确的共识：投大模型。在美国，就是持续给 OpenAI、Anthropic 这样的公司投钱。这些公司这两年发展得很快，也确实拿 走了行业里大部分的利润。 23 年的另一个共识是，很多人觉得应用只是「套壳」，很轻、很薄，没什么价值。 但到了 24、25 年，这个判断开始发生变化，因为很多应用层公司逐渐做出了自己的特色和护城河，比如 Cursor、Perplexity。 最近两年，Agent 很火。但在真实场景中，AI Agent 的落地依 ...

Core Insights - The article discusses the concept of reward functions in life and how they influence human behavior and satisfaction [4][6] - It emphasizes the importance of understanding one's own desires and motivations to achieve a fulfilling life [9][10] Group 1: Reward Functions - The reward functions developed by society often self-reinforce, leading to a continuous desire for higher achievements, making it difficult to find lasting satisfaction [4] - The best life pursuits should have characteristics such as non-decreasing marginal utility, positive externalities, and the ability to accumulate benefits over time [6] Group 2: Self-Understanding - Understanding oneself is crucial, as it helps in recognizing personal strengths and weaknesses, which can influence one's reward function [9][10] - The article suggests that many individuals struggle with low-quality rewards in childhood, making it challenging to develop sensitivity to positive rewards later in life [9] Group 3: Relationships and Environment - Surrounding oneself with the right people is essential, as they significantly impact an individual's reward function and overall happiness [9] - The article highlights that high salaries may not always reflect contributions but can serve as compensation for sacrifices made [9] Group 4: Happiness and Pain - Happiness is closely linked to expectations for the near future, and emotional issues often stem from cognitive misunderstandings [10][11] - Pain is perceived not just as a negative experience but as an opportunity for deep reflection and growth [11] Group 5: Creativity and Content Creation - Creative thinking can occur in both focused and unfocused states, with the latter often leading to more innovative ideas [14][15] - The article notes that successful content creators often undergo a long accumulation period, emphasizing the importance of continuous creation and improvement [15][16]

AI 来了之后，每天都感慨能做的事情实在是太多了，年底打算再招 1-2 个 FA 同事，和我们一起服务 AI founders。 以下是关于这份工作的一些回答，我们帮你提前拷问曲老师一些问题： Celia： 先一句话介绍一下42章经在做的事情吧。 曲凯： 我们从媒体、叙事、FA、投资等角度全方位服务最好的 AI 时代的创业者，我们想方设法地和最好的 founder 站在一起。 Celia： 你这次想招什么样的人？具体做哪些事情？ 曲凯： 首先希望能参与到 FA 的各项工作之中，比如行业研究、项目 sourcing、BP 制作、对接投资人等。 其次，AI 时代，职位和职位之间的边界会变得越来越模糊，所以如果你能有强烈的好奇心和自驱力，也欢迎 你参与到我们所有的业务之中，你甚至可以把这个岗位当做一个管培生的项目，你会被鼓励做自媒体、你会被 鼓励学习 vibe coding、你会被鼓励组织活动、你也能参与到和一线创始人的各种接触中。 曲凯： 有强烈的好奇心、有很强的自驱力、有自己热爱的事情。 在过去的经历中有拿到过结果，不管是多么冷门、偏门的结果。 有良好的实习或全职经验，有过关的职业素养和沟通能力。 另外我们年初 ...

Tech Ideas 线 | 讨论会 12 曲凯 42章经创始人 莫傑麟 家族办公室资深从业者 核心议题 25 年－二级市场 年终复盘和 26 年展望 关键词 多月不 AT 硬件 Agent 他熟悉硅谷二级市场，我深耕国内一级市场。这种跨视角的对照，使得我们每次凑在一起，往往能碰撞出一些相对超前、也还蛮准确的结论（不少听众反馈，我 们在节目里的很多判断，后来都应验了）。 过去这一年，我们录了三期播客： 《世界怎么就「东升西落」了？聊聊二级市场与 DeepSeek+Manus 的热潮》 《硅谷 AI 大转弯与二级市场的牛市》 《「你觉得 AI 有泡沫吗？」——有》 从今年秋天开始，我们又把这种季度复盘，延展成了一个更小范围、更高密度的形式：Tech Ideas 线上讨论会。 每一期，都会由我、莫傑麟，以及几位长期研究产业 / 投资的朋友共同主持。我们会围绕近期的重点主题，邀请业内朋友们，一起进行小范围的交流讨论。在上一 期活动中，嘉宾的分享中也有不少有启发性、扎实的 insights。 今年的最后一场活动，我们想做一次更完整的收官与展望：复盘 2025 年的一、二级市场，并展望 2026，共同探讨今明两年的 A ...

2025 年，是我们 「All in AI」的第三年。 2023 年，我们发布了 20 期内容，陪大家一起从 0 开始，搞清楚 AI 到底是什么： 2023 文章、播客合集 2024 年，市场一度遇冷。但我们仍然保持乐观，发布了 34 期内容： 2024 文章、播客合集 到了今年，随着年初 DeepSeek 和 Manus 的发布，AI 真的变成了街头巷尾都会聊起的大众话题。 我们也保持节奏，更新了 22 期播客、18 篇文章，3 次被小宇宙首页推荐，播客订阅数也增长到了近 11 万。 以下是我们全年的播客合集（按分享量排序）： 1. 组织能力才是 AI 公司真正的壁垒 | 对谈 Palona AI 联创任川 这是我们的第 50 期节目，也是我今年最有成就感的一期。 在聊过这么多创业者、看过这么多公司后，我们越来越清晰的一个判断是：在 AI 时代，组织能力的重要性被大大低估了。在这期节目里，我们就把硅谷最 AI Native 的组织方式带给了大家。如果能帮助国内创业者和公司往前一步，那就善莫大焉了。 （ 推送文字稿传送门 ） 2. 世界加速分化下，我们的机会在哪里？ | 对谈绿洲资本合伙人张津剑 津剑是我们的 ...

Core Insights - Dify has successfully established itself as a leading open-source project in the AI field, surpassing many expectations in its growth over the past two years [2][3][4] - The company adopted three core strategies from the beginning: open-source, B2B focus, and globalization, which have proven to be effective [3][4] Market and Technological Changes - The AI landscape has undergone three significant shifts over the past two years, with Dify evolving its offerings accordingly [5][6] - In 2023, Dify launched its first version, which was user-friendly and gained traction quickly due to the rising interest in AI [6] - By 2024, Dify introduced its core capability, workflow, and began building a plugin ecosystem, attracting paying enterprise customers [6] - By 2025, advancements in models, particularly in open-source capabilities and multi-modal functionalities, validated Dify's initial assumptions about the need for an intermediary layer [6][10] Competitive Landscape - Dify differentiates itself from competitors like LangChain by targeting a broader user base, including those with minimal technical skills [9][10] - The company has faced competition from various players, including large tech firms and startups, but has maintained its unique positioning by focusing on process integration within enterprises [12][17] - Dify's approach to product development emphasizes solving workflow issues and connecting LLMs with enterprise tools and data [17][18] Product Development and Engineering - Dify's engineering focus is seen as a key asset, with a strong emphasis on layered design and understanding user business scenarios [31][32] - The company believes that the most valuable aspect of its product is the engineering behind it, which requires significant cognitive effort and user collaboration [32][35] - Dify's workflow product is designed to ensure stability and reliability, allowing for gradual advancements in AI capabilities over time [38][39] Future Outlook - Dify envisions a future where its platform serves as an intelligent operating system for enterprises, integrating various capabilities and facilitating human-agent collaboration [56][57] - The company recognizes the importance of addressing the "last mile" issues in AI applications, focusing on building infrastructure that bridges the gap between model capabilities and human usability [72][73] - Dify's success in markets like Japan is attributed to its adaptability to local business structures and the scarcity of technical personnel [64][66] User Engagement and Market Penetration - Approximately 20% of Fortune 500 companies are currently using Dify, highlighting its significant market penetration [60] - The open-source model has been crucial for Dify's growth, enabling rapid dissemination and adoption of its technology [62][63]

(一) 为什么独立的 Vibe Coding 必死，但 Personal Software 会火？ Personalized software (个性化软件) 这个方向最近很火。 蚂蚁的灵光上线后，听说一天内的数据就达到了团队定的年度目标。 Replika 的创始人 Kuyda 最近也再次创业，做了一个叫 Wabi 的产品，定位是 Youtube for Apps，一个 mini app 的集合平台。(类似方向的产品国内还有马卡龙、 Youware 等等) 姚顺雨（前 OpenAI 研究员）曾反反复复表达过一个观点，我印象很深： 「创业公司最大的机会，在于设计不同的交互方式。」 于是这期就索性围绕「交互」这个主题展开，分享一下我们最近观察到的一些机会。 目录 我最近听了她的两期播客，很有意思，摘录一些 insights： 1. 软件的未来将是「应用的 YouTube 化」。 在 Kuyda 看来，软件行业正在经历内容行业当年的变迁。 就像视频从专业制作走向人人可拍，软件开发也将从全球「2000 万开发者」的特权，泛化为「80 亿创作者」的日常媒介。 在这个语境下，未来的软件，更多会像快消品，它不再需要追求 S ...

这是莫傑麟第六次做客我们的播客。我们几乎每隔几个月，就会一起复盘一次 AI 市场的最新动向。这一次，我们聊的是近期热度最高的话题之一： AI 泡沫。泡沫 只是情绪化的表象，更值得探讨的，是泡沫之下结构性的变化。 本期播客原文约 17000 字，本文经过删减整理后约 6600 字。 曲凯： 我问个最直接的问题，你觉得现在的 AI 有泡沫吗？ 莫傑麟： 有。 曲凯： 这么直接？你不分类讨论一下吗（笑）？ 莫傑麟： 哈哈哈如果把泡沫理解成「预期高于现实」，那肯定是有的。 但有泡沫不一定是坏事，反而能推动行业发展。泡沫也不一定随时会破。 曲凯： 但我觉得需要拆开看，把价值和价格分开来说。 从价值来看，我不觉得现在的 AI 有问题。 最近关于泡沫的讨论，大多发生在小红书和 Twitter，但我们身边真正做 AI 的人，几乎没人谈这事。我还特意问了几个人，结果大家都非常鄙视我，说：「AI 哪 有什么泡沫？发展得挺好，我们都很兴奋。现在的智能水平也已经够高了。」 所以从价值层面看，没什么泡沫。 但从价格层面看，就得进一步细分，拆成中国、美国、一级、二级来看。 价格体现在一级市场是估值，二级市场是市值。 今年国内一级市场整体 ...

Core Insights - The article discusses the insights shared by Ilya regarding the future of AI and the development of superintelligence, emphasizing a shift back to research-focused approaches after a period of scaling [3][5]. Group 1: Era Transition - Ilya outlines a clear timeline for the evolution of AI, indicating that the period from 2012 to 2020 was focused on research, while 2020 to 2025 is characterized by scaling, particularly after the emergence of GPT-3. He predicts that from 2025 onwards, the limitations of pre-training scaling laws will necessitate a return to research-focused methodologies [5]. Group 2: SSI's Strategy - Ilya's company, SSI (Safe Superintelligence), plans to focus on developing superintelligence without intermediate products, aiming to avoid market distractions that lead to compromises in quality [3][4]. Group 3: Learning Mechanisms - Ilya emphasizes the importance of developing a value function in AI, which allows for more intuitive learning processes similar to human decision-making. He believes that breakthroughs in this area could significantly enhance AI efficiency [6][10]. Group 4: Reinforcement Learning (RL) Insights - Ilya presents a contrarian view on RL, suggesting that it may hinder the capabilities of models by forcing them to conform to narrow human-defined metrics, potentially sacrificing broader intelligence [7][8]. He also notes that RL is becoming more resource-intensive than pre-training, indicating a shift in the industry's focus [8]. Group 5: Empathy and AI - Ilya argues that empathy could be a crucial element in developing superintelligence, proposing that AI should be designed to care for sentient life, akin to how human evolution has hardcoded empathy into our brains [13][14][19]. Group 6: Language and Research Direction - The language used in the AI field can shape research directions, with terms like AGI and scaling influencing the focus of development. Ilya warns against the overemphasis on these buzzwords, which may lead to neglecting other important aspects of intelligence [20][22]. Group 7: Market Dynamics - Ilya predicts that the future AI market will not be dominated by a single entity but will instead feature specialization, where companies focus on specific applications, creating a balanced ecosystem similar to biological evolution [22].

Group 1 - The core viewpoint of the article discusses the current state of the AI market, highlighting that the growth from 2023 to 2024 relies on the scaling law and the consensus around AGI, while there is no unified judgment on RL scaling law since 2025 [5] - AI models are developing in a stepwise manner, while applications are experiencing pulsed advancements, indicating a subtle blank period currently [5] - There is uncertainty regarding the continued enhancement of intelligence, but the acceleration of application deployment is assured [5] Group 2 - The narrative logic is changing, suggesting that while prices that rose previously may have bubbles, the intrinsic value of AI remains intact [5] - Several unresolved questions about the future development of AI, including whether to buy or short Nvidia, the opportunities in multimodal applications, and the feasibility of embodied production and deployment, are raised [5] - An online discussion meeting is scheduled for November 29, aiming to engage in these topics with interested participants [5]