⬆️关注 Founder Park，最及时最干货的创业分享 越来越多 AI 出海公司，把达人营销视为最重要的增长杠杆之一。与传统广告投放或内容营销相比，达人营销最大的优势在于它的「活人感」——真实的创 作者在真实的使用场景中展示产品，天然降低了用户的信任门槛。 但问题也恰恰出在这里。 当达人营销走向「规模化」，很多团队会发现一件事： 达人营销很有效，却很难做到真正可控。 不同团队在执行层面拉开的差距，可能直接决定产品的成 败，甚至影响着企业的长远发展。 观察行业成功者的具体实践，可以瞥见典型路径：以 Gamma 为例，通过达人营销，打造可复制的「爆款流水线」，其策略是投入充足的预算，广泛合作 大量达人，最终沉淀 10% 的爆款，带来口碑传播与 90% 的用户增长；Notion 通过长期追踪并量化创作者内容对用户行为的实际影响，用数据反哺策略， 持续筛选并放大高价值创作者关系，将达人合作从短期项目沉淀为稳定的增长资产...... 他们的实践，共同指向一个趋势： 想做好达人营销，需要规模化，需要从单次合作中逐步沉淀方法论经验并转为长期稳定的增长资产。 超 17000 人的「AI 产品市集」社群！不错过每一款有价值 ...

Core Viewpoint - Traditional SaaS applications like CRM and ERP systems will not be replaced but will evolve to serve as the infrastructure for AI Agents, which will enhance the importance of data definition and interpretation within enterprises [2][10][15] Group 1: The Role of AI Agents - AI Agents will not eliminate traditional software systems; instead, they will necessitate a clearer separation between how tasks are performed and the sources of facts [2][10] - The effectiveness of AI Agents is contingent upon their ability to access and understand the correct data from various systems, highlighting the need for accurate and structured input data [2][9] - The emergence of AI Agents creates significant entrepreneurial opportunities for companies that can help businesses manage and structure their unstructured data [3][10] Group 2: Data Management Challenges - A significant portion of enterprise knowledge (80%) exists in unstructured data, which is becoming increasingly difficult to manage [2] - The complexity of data definitions within organizations leads to discrepancies in key metrics like Annual Recurring Revenue (ARR), complicating the role of AI Agents in providing accurate information [7][11] - The traditional approach of consolidating data into warehouses has only partially succeeded, as operational teams still rely on individual systems for real-time transactions [8][10] Group 3: Evolution of Systems - CRM and ERP systems will transition from user-centric interfaces to machine-oriented APIs, allowing AI Agents to interact with these systems programmatically [12][15] - The core value of enterprise systems lies in their ability to encapsulate chaotic data, which will remain essential despite changes in interface and interaction methods [13][15] - The demand for a clear, authoritative source of truth will only increase as AI Agents become more prevalent in business processes [14][15] Group 4: Future of Data Infrastructure - The combination of data warehouses, semantic layers, and governance tools will form the foundation for AI Agent workflows, evolving beyond traditional reporting systems [10][12] - The valuation of AI platforms will increasingly depend on their ability to define and manage facts, rather than just their user interfaces [14][15] - Companies that can create exceptional AI Agent experiences based on reliable data sources will have a competitive advantage in the evolving landscape [15]

Core Insights - Data is a critical risk point for startups, even if it may not serve as a competitive moat [1] - Different types of user data, AI-generated content, and other data categories have varying legal risks and processing requirements [2] - For companies expanding overseas, prioritizing compliance risks is essential due to frequent litigation and infringement disputes [3] Group 1: Workshop Details - The workshop will feature partners Zheng Wei and Sun Qimin from Beijing Xingye Law Firm, focusing on compliance and high-risk issues faced by AIGC startups during international expansion [4] - The event is scheduled for December 18 at 8 PM and will be held online [5] - Participation is limited and requires a screening process for registration [6] Group 2: Data Usage and Compliance - During model training, it is crucial to determine which types of data, such as synthetic data, copyrighted content, and user behavior data, can be utilized [8] - Different data types, including code, images, and audio/video, present unique infringement risks that need to be addressed [8] - Questions regarding ownership of AI-generated content, as well as the delineation of data usage rights and intellectual property for ToB and ToC applications, are critical [10] - Companies must also consider how to manage cross-border data transfer, local storage, and data isolation when expanding internationally [10]

Core Insights - Data is a critical risk point for startups, even if it may not serve as a competitive moat [1] - Different types of user data, AI-generated content, and various media have distinct legal risks and processing requirements [2] - For companies expanding overseas, prioritizing compliance risks is essential due to frequent litigation and infringement disputes [3] Group 1: Workshop Details - The workshop features partners Zheng Wei and Sun Qimin from Beijing Xingye Law Firm, focusing on compliance and high-risk issues faced by AIGC startups during international expansion [4] - The event is scheduled for December 18 at 8 PM and will be held online [5] - Participation is limited and requires a screening process for registration [6] Group 2: Key Discussion Topics - During the model training phase, it is crucial to determine which types of data, such as synthetic data, copyrighted content, and user behavior data, can be utilized [8] - There are specific considerations regarding infringement risks associated with different data types, including code, images, and audio/video [8] - Clarification is needed on the ownership of AI-generated content and the delineation of data usage rights and intellectual property for ToB and ToC applications [10] - Companies must address cross-border data transmission, local storage, and data isolation when expanding their products internationally [10]

(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 ...

Core Viewpoint - Data is not necessarily a moat for products, but it is a risk point that startups must take seriously [1] Group 1: Legal Risks and Compliance - Different types of user data, AI-generated content, and various media have distinct legal risks and processing requirements [2] - For companies expanding overseas, it is crucial to prioritize compliance risks, especially given the frequency of lawsuits and infringement disputes [3] - The workshop features partners from Beijing Xingye Law Firm discussing how AIGC startups can navigate compliance and high-risk issues during international expansion [4] Group 2: Data Usage and Rights - During the model training phase, it is essential to determine which types of data, such as synthetic data, copyrighted content, and user behavior data, can be used [8] - There are specific considerations regarding infringement risks for different types of data, including code, images, and audio/video [8] - Questions arise about the ownership of AI-generated content and how to define data usage rights and intellectual property for ToB and ToC applications [10] Group 3: Cross-Border Data Management - Companies must understand how to manage cross-border data transmission, local storage, and data isolation when expanding their products internationally [10]

Group 1 - The Global Developer Pioneer Conference, focusing on embodied intelligence and robotics, will be held in Shanghai from December 12 to 14, 2025, with registration now open for developers in these fields [1][7][12] - The Geek Park Innovation Conference 2026 will take place in Beijing on December 6-7, featuring industry leaders such as He Xiaopeng and Wang Xiaochuan, aimed at fostering connections and opportunities in the AI era [5][6] - The Global Immersion @CES2026 event organized by Geek Park will occur from January 5 to January 11, 2026, in Las Vegas and Los Angeles, targeting tech industry professionals and entrepreneurs [9][11] Group 2 - The BlueChirping Entrepreneurship Camp (Fifth Session - AI) in Beijing will provide high-quality brainstorming sessions and opportunities to engage with top AI entrepreneurs, focusing on practical insights and collaborative creation [18][19][20] - NVIDIA's Startup Acceleration Program is currently recruiting, offering members access to free deep learning training, SDKs, discounts on hardware and software, and opportunities for funding and business connections [26][27]

Core Insights - The recent surge in large models has unexpectedly revitalized the input method sector, previously considered a basic infrastructure, making it attractive by the second half of 2025 [1]. Group 1: Market Developments - In the past two months, there has been a significant increase in news density regarding voice input technologies, with major developments from both domestic and international players [2]. - Domestic advancements include ByteDance's Doubao input method officially launching after internal testing, and WeChat input method continuously iterating on AI-assisted features [2]. - Internationally, Wispr announced a $25 million Series A funding round, bringing its total funding to $81 million, while Typeless gained attention on Product Hunt [2]. Group 2: Competitive Landscape - The voice input market can be categorized into three main camps: 1. Desktop SaaS players like Wispr and Typeless, focusing on productivity for core office users. 2. Mobile giants like Doubao and WeChat, leveraging vast ecosystem traffic for social interactions. 3. Low-cost indie developers represented by Whisper Keyboard and Lightning Say, focusing on localized or independent development [4]. Group 3: Product Performance - A subjective testing scenario revealed Typeless as the best desktop input method and Doubao as the best mobile input method, with specific strengths in handling complex language and context [6]. - Typeless achieved a processing time of 3.05 seconds, effectively removing filler words and correcting formats, while Doubao excelled with a 2.05-second response time, accurately interpreting context [6][13]. - WeChat input method, with a rapid 1.08 seconds response time, remains dominant in casual communication despite some limitations in professional formatting [13]. Group 4: User Experience Insights - The user experience of third-party voice input methods on iOS is often hindered by permission issues, requiring app switches for voice input [8]. - Doubao's voice model demonstrates superior performance in speed and accuracy, particularly in Chinese, although it faces challenges on iOS due to Apple's privacy restrictions [8][42]. - Typeless offers the best output quality for desktop users, providing high accuracy and innovative interaction features, while Lightning Say, despite its speed, struggles with professional terminology [8][60]. Group 5: Technological Evolution - The voice input sector is experiencing a paradigm shift from traditional automatic speech recognition (ASR) to models that understand and reconstruct language, enhancing user interaction [63]. - This evolution allows for greater tolerance of user errors, enabling a more natural and intuitive communication interface, transforming input methods into tools for thought rather than mere transcription [64][65].