以创新为主线和以安全为底线、统筹创新与竞争、兼顾效率与公平优化人工智能大模型市场竞争路径。 再次，通用大模型难以达到监管透明度要求。通用大模型训练与推理过程的不可解释性、不可预测性， 使得要求算法公开的救济手段难以实施，制度上也缺乏对算法透明度的强制性审查要求，监管机构难以 追溯相关反竞争行为的证据。同时，我国现行反垄断相关制度仍以市场份额、价格协议等作为垄断行为 认定标准，难以应对人工智能领域特有的数据垄断、算力垄断、算法共谋等新型竞争问题。例如，通用 大模型训练所需的海量多模态数据可能聚集形成"数据孤岛"，但现行反垄断监管框架中缺乏对数据聚集 的审查的触发标准和可量化指标。 关于大模型的开源与闭源两种软件开发模式的争议从未停止。开源指开放源代码，将源代码公开发布并 允许任何人查看、修改和使用；闭源则不公开源代码，只对外发布编译后的软件。选择闭源路径的大模 型主要有OpenAI的GPT系列、Anthropic的Claude以及谷歌的Gemini。闭源大模型的模型代码及数据通常 不对外开放，研发者可以通过提供API访问和企业解决方案来实现盈利，进而保证相关技术能够持续研 发并改进模型。与此路径相反的是以深度求 ...

Core Viewpoint - The article discusses a heated debate among industry leaders at the WAIC 2025 forum regarding the evolution of large model technologies, focusing on training paradigms, model architectures, and data sources, highlighting a significant shift from pre-training to reinforcement learning as a dominant approach in AI development [2][10][68]. Group 1: Training Paradigms - The forum highlighted a paradigm shift in AI from a pre-training dominant model to one that emphasizes reinforcement learning, marking a significant evolution in AI technology [10][19]. - OpenAI's transition from pre-training to reinforcement learning is seen as a critical development, with experts suggesting that the pre-training era is nearing its end [19][20]. - The balance between pre-training and reinforcement learning is a key topic, with experts discussing the importance of pre-training in establishing a strong foundation for reinforcement learning [25][26]. Group 2: Model Architectures - The dominance of the Transformer architecture in AI has been evident since 2017, but its limitations are becoming apparent as model parameters increase and context windows expand [31][32]. - There are two main exploration paths in model architecture: optimizing existing Transformer architectures and developing entirely new paradigms, such as Mamba and RetNet, which aim to improve efficiency and performance [33][34]. - The future of model architecture may involve a return to RNN structures as the industry shifts towards agent-based applications that require models to interact autonomously with their environments [38]. Group 3: Data Sources - The article discusses the looming challenge of high-quality data scarcity, predicting that by 2028, existing data reserves may be fully utilized, potentially stalling the development of large models [41][42]. - Synthetic data is being explored as a solution to data scarcity, with companies like Anthropic and OpenAI utilizing model-generated data to supplement training [43][44]. - Concerns about the reliability of synthetic data are raised, emphasizing the need for validation mechanisms to ensure the quality of training data [45][50]. Group 4: Open Source vs. Closed Source - The ongoing debate between open-source and closed-source models is highlighted, with open-source models like DeepSeek gaining traction and challenging the dominance of closed-source models [60][61]. - Open-source initiatives are seen as a way to promote resource allocation efficiency and drive industry evolution, even if they do not always produce the highest-performing models [63][64]. - The future may see a hybrid model combining open-source and closed-source approaches, addressing challenges such as model fragmentation and misuse [66][67].

Core Insights - The conversation emphasizes the importance of sustainable value in the next generation of AI, highlighting the confusion and uncertainty that often accompany major technological shifts [3][4] - Kevin Scott argues that the current era is the best time for entrepreneurs, advocating for active exploration and product development rather than passive observation [5] - The discussion also touches on the balance of value creation between startups and established companies like Microsoft, suggesting that both can benefit from new AI capabilities [6][7] Group 1: AI Value and Product Development - Kevin Scott believes that while models are valuable, their worth is realized only when connected to user needs through products [6] - The conversation stresses that product quality is paramount, and that successful exploration requires rapid iteration and responsiveness to data and feedback [5][6] - The scaling law in AI is not seen as having a limit currently, with Scott asserting that AI capabilities will continue to expand [8] Group 2: Data and Efficiency - The importance of high-quality data is highlighted, with synthetic data becoming increasingly significant in model training [9][10] - There is a noted gap in the ability to evaluate the impact of specific data on model performance, indicating a need for better assessment tools [9][10] Group 3: Future of AI Agents - The future of AI agents is discussed, with expectations for improved memory and task execution capabilities, allowing them to handle more complex tasks autonomously [21][22] - The interaction model between humans and agents is expected to evolve, moving towards more asynchronous operations [22] Group 4: Industry Dynamics and Trends - The conversation reflects on the dual existence of open-source and closed-source solutions in AI, suggesting that both will coexist and serve different needs [15] - The role of engineers and product managers is expected to change, with a greater emphasis on specialization and collaboration with AI agents [18][19] Group 5: AI's Impact on Technical Debt - Kevin Scott expresses optimism that AI can help mitigate technical debt, transforming it from a zero-sum problem to a non-zero-sum opportunity [31] - The potential for AI to accelerate product development and reduce the burdens of technical debt is seen as a significant advantage [30][31]

Core Viewpoint - Google is shifting its strategy regarding the Android Open Source Project (AOSP) by moving development to an internal codebase, which raises concerns about the potential closure of AOSP while still maintaining its commercial interests [2][3][4][5]. Group 1: AOSP Overview - AOSP was initially created to enhance the Android experience through contributions from manufacturers and developers, aiming to compete with iOS [6]. - The project has evolved into two branches: the public AOSP branch, accessible to anyone, and the internal development branch, limited to companies that have signed agreements with Google [10]. Group 2: Strategic Shift - By concentrating development on the internal codebase, Google aims to reduce operational costs and compel OEM manufacturers to pay for access to the latest Android updates and security patches [10]. - This move does not necessarily mean a complete abandonment of AOSP, as it still plays a crucial role in maintaining Google's ecosystem across various devices [11][13]. Group 3: Historical Context - Over the past 15 years, Google's strategy has consistently involved "damaging AOSP to benefit GMS," with many core applications being removed from AOSP and made available through Google Play [11][12]. - Despite the reduction of local applications in AOSP, Google has continued to update it, recognizing the importance of AOSP-based devices in sustaining its ecosystem [12][13]. Group 4: Competitive Landscape - The competition in the operating system market is fundamentally about monopolization, and Google is unlikely to overlook the implications of AOSP's potential closure [14].

Core Insights - The emergence of DeepSeek AI has significantly impacted the AI industry, leading to rapid innovation and application across various sectors, with expectations of a breakthrough in AI penetration within 18 months [2][4][29] - The company, Zhongke Wenge, has successfully developed its own AI models, including the Yayi model, which has contributed to substantial revenue growth, with over half of its income directly linked to this model [3][5][25] - The shift in investment attitudes towards AI startups has transitioned from a focus on technology to practical applications, highlighting the growing importance of AI in the market [4][5] Group 1 - DeepSeek AI has reached the top of the iOS free app charts in both China and the US, showcasing its rapid adoption and influence in the global AI landscape [2] - The integration of DeepSeek into Zhongke Wenge's X-Agent platform allows clients to quickly develop industry-specific AI applications, reducing technical barriers and development time [2][14] - The company has expanded its business into various sectors, including finance, healthcare, and energy, demonstrating its versatility and adaptability in the AI market [3][21] Group 2 - The training cost for DeepSeek's models is significantly lower than that of competitors, with estimates suggesting costs are 1/10 to 1/20 of ChatGPT's, which encourages broader participation in AI development [6][10] - The company emphasizes the importance of both open-source and closed-source models, advocating for a balanced approach to innovation and intellectual property protection [7][10] - The AI industry is expected to experience explosive growth by 2025, driven by advancements in technology and increased market demand for AI applications [29][30] Group 3 - Zhongke Wenge's decision to develop the Yayi model was a pivotal moment, aligning with the broader AI transformation initiated by the launch of ChatGPT [24][25] - The company has achieved a near 100% renewal rate with key clients, indicating strong customer loyalty and satisfaction [17] - The integration of AI into decision-making processes is highlighted as a critical area for future development, with a focus on dynamic and real-time data analysis [28]

Core Viewpoint - DeepSeek has launched a time-limited discount for its API services, significantly reducing the prices for its models during off-peak hours, aiming to enhance user experience and encourage usage [1][2]. Pricing Structure - The API pricing for DeepSeek models is measured in "million tokens," where a token represents the smallest unit of natural language text. The main models are DeepSeek-Chat (V3) and DeepSeek-Reasoner (R1) [2][4]. - During standard hours (08:30-00:30 Beijing time), the input price for V3 is ¥0.5 per million tokens, and for R1, it is ¥1. The output prices are ¥8 for V3 and ¥16 for R1. In the off-peak hours (00:30-08:30), the input price for both models drops to ¥0.25, and the output price is reduced to ¥4 [2][3][4]. Model Specifications - Both DeepSeek-Chat and DeepSeek-Reasoner have a context length of 64K tokens. The maximum reasoning chain length for R1 is 32K, while both models have a maximum output length of 8K tokens [3][5]. API Service Resumption - After a 19-day suspension due to server resource constraints, DeepSeek has reopened API recharge services, allowing developers to continue using their existing balance [4][6]. Open Source Initiative - DeepSeek has initiated an "Open Source Week," where it plans to release five software libraries over five days, aiming to accelerate its technology ecosystem and share advancements in artificial general intelligence (AGI) with the global developer community [6][8]. - The first open-sourced library, FlashMLA, is optimized for Hopper GPU, while DeepEP, a communication library for MoE model training, has also been made public [7][8]. Industry Trends - The trend towards open source in the AI model sector is gaining traction, with DeepSeek's success prompting other companies to reconsider their strategies. The debate between open and closed source models continues, with notable shifts in attitudes among major players [10][11][12]. - DeepSeek's emergence has highlighted the viability of the open-source model, demonstrating that it can be a strategic path for rapid market capture and technological innovation [14].

作者：MD 出品：明亮公司 近日，美国知名播客Invest Like the Best再次访谈了Andreessen Horowitz的联合创始人 Marc Andreessen，在访谈中，Marc和主播 Patrick 深入探讨了AI正在重塑技术和地缘政治的重大变革 ，并 讨论了 DeepSeek 的开源人工智能以及其大国技术竞争中的意义，此外，他们还分享了对全球权力结 构演变的看法，以及风险投资行业整体的转型。 「明亮公司」借助AI工具第一时间整理了访谈中的核心内容，全文内容详见文末"原文链接"。 以下为访谈内容（有删节）： 谈DeepSeek、AI赢家和输家 Patrick ：Marc，我认为我们必须从最核心的问题开始。你能否谈谈你对DeepSeek的R1的看法？ Marc ：这里面有很多维度。（我认为）美国仍然是人工智能领域公认的科学和技术领导者。 DeepSeek中的大多数想法都源自过去20年，甚至令人惊讶的是80年前在美国或欧洲进行的工作。神 经网络的最初研究早在20世纪40年代的美国和欧洲研究型大学中就已展开。 因此，从知识发展的角度来说，美国仍然遥遥领先。 但DeepSeek对这些知识完成了非 ...