一水 发自 凹非寺 量子位 | 公众号 QbitAI 我并不认为Scaling Law已经完全结束了 。 正当学生Ilya为Scaling Law"泼下冷水"时，他的老师、AI教父Geoffrey Hinton却毅然发表了上述截然相反的观点。 这一场面一出，我们不禁回想起了两件有趣的事。 一是Ilya几乎从学生时代起就坚信Scaling Law，不仅一抓住机会就向身边人安利，而且还把这套理念带进了OpenAI。 可以说，Ilya算是Scaling Law最初的拥趸者。 二是Hinton后来在回顾和Ilya的相处时，曾大肆夸赞Ilya"具有惊人的直觉"，包括在Scaling Law这件事上，Hinton曾坦言： 当时的我错了，而Ilya基本上是对的。 比如Transformer确实是一种创新想法，但实际上起作用的还是规模，数据的规模和计算的规模。 但是现在，这对师徒的态度却来了个惊天大反转。 所以，这中间到底发生了什么？ Scaling Law不死派：Hinton、哈萨比斯 其中，最大的挑战无疑是数据缺失问题。 大部分高价值数据都锁在公司内部，免费互联网数据已基本耗尽。 而这个问题将由AI自行解决，即模型通过推 ...

一部纪录片，在YouTube上仅仅上线四周，就突破了2亿人观看！ 诺奖得主Demis Hassaibs亲自力荐：想知道一个通用人工智能实验室幕后是怎么运作的吗？是什么造就了AlphaFold这样的诺奖级获奖项目？一定要看这部 片！ 另外，这部纪录片的配乐，也是一流。 人类，正亲手创造第二种智慧 当人类第一次意识到，自己或许正在创造一种不再依附于血肉、不再受限于寿命、不再困于经验的智慧。这个瞬间，足以撕裂时代。 令人感动的是，纪录片《The Thinking Game》并不是一部炫技的科技宣传片，它更像一部时代的自白书。 它从人类与AI的第一次朴素笨拙的交流开始——「你，能学会思考吗？」 真正令人震撼到，不是AI给出的回答，而是提问的人类。这群人被同一个执念牵引—— 是的，这部名为《思考游戏》的纪录片，并不是一部普通的片子，它由AlphaGo原班团队历时五年贴身拍摄。 短短四周内，它就如一场风暴，席卷全球。 可以说，它绝不仅仅是一部电影，而是一次对AGI科学盛典最核心地带的闯入。 如果智能可以被创造，那么人类理解自身的方式，将被彻底改写。 当还不是诺奖得主的Demis Hassabis说出这句话「Trying ...

【导读】2025年底，当人类都在憧憬和等待一个全知全能的AI之神时，谷歌DeepMind却泼了一盆冷水！ 12月19日，谷歌DeepMind抛出了一个让人细思极恐又脑洞大开的新观点： 如果所谓的AGI（通用人工智能）并不是一个超级实体，而是「凑出来」的呢？ 论文地址：https://arxiv.org/abs/2512.16856 在人工智能发展的宏大叙事中，我们长期被一种单一的、近乎宗教般的想象所占据：通用人工智能（AGI）将以一个全知全能的「超级大脑」形式降临。 这种叙事深深植根于科幻文学与早期AI研究的土壤中，导致当下的AI安全与对齐研究主要聚焦于如何控制这个假设中的单体化存在。 而且包括人工智能教父Hinton等人都试图将人类价值观植入这个大脑，仿佛只要解决了这个超级单体的「心智」问题，人类的安全便有了保障。 然而，DeepMind这篇在2025年末发布的重磅论文《分布式AGI安全》犹如一道惊雷，彻底颠覆了这一根深蒂固的假设。 这种「单体AGI」假设存在巨大的盲区，甚至可能是一个危险的误导！ 它忽视了复杂系统演化的另一种极高可能性的路径，也是生物界和人类社会智慧产生的真实路径：分布式涌现。 这不仅仅是 ...

Core Insights - The public's perception of AI is significantly lagging behind its actual advancements, with a gap of at least one generation [2][5][41] - AI is evolving at an exponential rate, with predictions indicating that by mid-2026, AI models could autonomously complete tasks for up to 8 hours, potentially surpassing human experts in various fields by 2027 [9][33][43] Group 1: AI Progress and Public Perception - Researchers have observed that AI can now independently complete complex tasks for several hours, contrary to the public's focus on its mistakes [2][5] - Julian Schrittwieser, a key figure in AI development, argues that the current public discourse underestimates AI's capabilities and progress [5][41] - The METR study indicates that AI models are achieving a 50% success rate in software engineering tasks lasting about one hour, with an exponential growth trend observed every seven months [6][9] Group 2: Cross-Industry Evaluation - The OpenAI GDPval study assessed AI performance across 44 professions and 9 industries, revealing that AI models are nearing human-level performance [12][20] - Claude Opus 4.1 has shown superior performance compared to GPT-5 in various tasks, indicating that AI is not just a theoretical concept but is increasingly applicable in real-world scenarios [19][20] - The evaluation results suggest that AI is approaching the average level of human experts, with implications for various sectors including law, finance, and healthcare [20][25] Group 3: Future Predictions and Implications - By the end of 2026, it is anticipated that AI models will perform at the level of human experts in multiple industry tasks, with the potential to frequently exceed expert performance in specific areas by 2027 [33][39] - The envisioned future includes a collaborative environment where humans work alongside AI, enhancing productivity significantly rather than leading to mass unemployment [36][39] - The potential transformation of industries due to AI advancements is profound, with the possibility of AI becoming a powerful tool rather than a competitor [39][40]

Core Insights - The report commissioned by Google DeepMind predicts that by 2030, the cost of AI compute clusters will exceed $100 billion, capable of supporting training tasks equivalent to running the largest AI compute cluster continuously for 3,000 years [3][5] - AI model training is expected to consume power at a gigawatt level, with the computational requirements reaching thousands of times that of GPT-4 [3][5] - Despite concerns about potential bottlenecks in scaling, recent AI models have shown significant progress in benchmark tests and revenue growth, indicating that the expansion trend is likely to continue [4][9] Cost and Revenue - The training costs for AI are projected to exceed $100 billion, with power consumption reaching several gigawatts [5] - Revenue growth for companies like OpenAI, Anthropic, and Google DeepMind is expected to exceed 90% in the second half of 2024, with annualized growth rates projected to be over three times [9] - If AI developers' revenues continue to grow as predicted, they will be able to match the required investments of over $100 billion by 2030 [19] Data Availability - The report suggests that publicly available text data will last until 2027, after which synthetic data will fill the gap [5][12] - The emergence of synthetic data has been validated through models like AlphaZero and AlphaProof, which achieved expert-level performance through self-generated data [15] Algorithm Efficiency - There is an ongoing improvement in algorithm efficiency alongside increasing computational power, with no current evidence suggesting a sudden acceleration in algorithmic advancements [20] - The report indicates that even if there is a shift towards more efficient algorithms, it may further increase the demand for computational resources [20] Computational Distribution - The report states that the computational resources for training and inference are currently comparable and should expand synchronously [24] - Even with a potential shift towards inference tasks, the growth in inference scale is unlikely to hinder the development of training processes [27] Scientific Advancements - By 2030, AI is expected to assist in complex scientific tasks across various fields, including software engineering, mathematics, molecular biology, and weather forecasting [27][30][31][33][34] - AI will likely become a research assistant, aiding in formalizing proofs and answering complex biological questions, with significant advancements anticipated in protein-ligand interactions and weather prediction accuracy [33][34]

AI & Games - Games serve as a valuable testing environment for AI development, including the company's work on AlphaGo & AlphaZero [1] - The company anticipates rapid advancements in AI through the addition of more games and challenges to the Arena [1]

Core Viewpoint - The article discusses the evolution of the human brain and its implications for artificial intelligence (AI), emphasizing that understanding the brain's evolutionary breakthroughs may unlock new advancements in AI capabilities [2][7]. Summary by Sections Evolutionary Breakthroughs - The evolution of the brain is categorized into five significant breakthroughs that can be linked to AI development [8]. 1. **First Breakthrough - Reflex Action**: This initial function allowed primitive brains to distinguish between good and bad stimuli using a few hundred neurons [8]. 2. **Second Breakthrough - Reinforcement Learning**: This advanced the brain's ability to quantify the likelihood of achieving goals, enhancing AI's learning processes through rewards [8]. 3. **Third Breakthrough - Neocortex Development**: The emergence of the neocortex enabled mammals to plan and simulate actions mentally, akin to slow thinking in AI models [9]. 4. **Fourth Breakthrough - Theory of Mind**: This allowed primates to understand others' intentions and emotions, which is still a developing area for AI [10]. 5. **Fifth Breakthrough - Language**: Language as a learned social system has allowed humans to share complex knowledge, a capability that AI is beginning to grasp [11]. AI Development - Current AI systems have made strides in areas like language understanding but still lag in aspects such as emotional intelligence and self-planning [10][11]. - The article illustrates the potential future of AI through a hypothetical robot's evolution, showcasing how it could develop from simple reflex actions to complex emotional understanding and communication [13][14]. Historical Context - The narrative emphasizes that significant evolutionary changes often arise from unexpected events, suggesting that future breakthroughs in AI may similarly emerge from unforeseen circumstances [15][16].

Core Insights - The current trend in the AI space is driven by the rise of Agents, with a potential next hotspot being Agent Infrastructure [1][3] - The number of Agents is expected to increase significantly, potentially reaching thousands of times the current number of SaaS applications [2] - The collaboration between Agents and humans is anticipated to shift, with Agents becoming more autonomous and capable of processing information at a higher bandwidth than humans [4][5] Group 1 - Agent Infrastructure represents a substantial market opportunity due to the need for restructured internet infrastructure to accommodate AI [3] - The interaction methods between humans and Agents differ significantly, with Agents capable of multi-threaded tasks and learning simultaneously while executing tasks [5][6] - A new mechanism is required to manage the state of multiple tasks executed by Agents, as they can handle numerous tasks concurrently [8][10] Group 2 - The concept of a "safety fence" is crucial for AI operations, ensuring that the impact of AI actions is contained within a controlled environment [10][11] - E2B is highlighted as a popular product providing a secure and efficient sandbox for code execution, significantly influenced by the Manus project [12][14] - Cloud service providers are expected to benefit from the increased demand for resources as more Agents operate in cloud environments [15][16] Group 3 - Browserbase is identified as a leading product designed specifically for AI, with a valuation of $300 million within a year [22] - The design of AI-specific browsers must consider continuous operation, feedback loops, and security measures to protect user information [24][27] - The architecture of AI browsers includes a Runtime layer and an Agentic layer, which are essential for effective interaction between AI and web content [32][33] Group 4 - The Agent Infrastructure market is expected to grow significantly, with opportunities in both environmental setups and tools for Agents [36][40] - The potential for AI to enhance efficiency in various sectors, such as sales and recruitment, indicates a large market for Browser Use applications [48] - Differentiation in Agent Infrastructure products is crucial, with a focus on finding unique scenarios and deepening product offerings rather than competing for a small market share [55][56]

Core Viewpoint - The article emphasizes the importance of foundational skills such as programming, mathematics, and physics for young people in the AI era, arguing that understanding these subjects is crucial for effectively utilizing AI tools and adapting to future job markets [16][25]. Group 1: Demis Hassabis and His Contributions - Demis Hassabis is a renowned AI scientist and entrepreneur, known for his early achievements in chess and his academic excellence, having graduated from Cambridge University at the age of 20 [4][7]. - He founded DeepMind in 2010 with the goal of using AI to solve complex scientific problems, leading to significant milestones such as the defeat of Go champion Lee Sedol by AlphaGo in 2016 [10][11]. - AlphaFold, developed by DeepMind, revolutionized protein structure prediction, reducing research time from years to minutes and contributing to the understanding of 2 billion proteins, earning Hassabis a Nobel Prize in Chemistry in 2024 [13]. Group 2: Recommendations for Young People - Young individuals are encouraged to focus on foundational subjects like programming, mathematics, and physics to fully grasp AI principles and develop a personalized AI capability [16][25]. - The article suggests that the ability to effectively utilize AI tools depends on a deep understanding of their underlying principles, similar to how a manager's effectiveness relies on their ability to leverage team members' strengths [17][18]. Group 3: AI in Education - The article introduces an AI-based college application tool called "Sweet Volunteer," which uses a data-driven approach to assist students in selecting their majors and universities based on their preferences and past admission data [19]. - This tool features a "reach, safe, and steady" strategy model, intelligent search capabilities, and personalized AI Q&A to provide tailored recommendations for students [19]. Group 4: Future Outlook - The article concludes that while the future holds uncertainties, the AI era presents numerous opportunities, and individuals must actively engage with AI to avoid being left behind [23][25].

Core Insights - The article discusses the transition from the "human data era" to the "experience era" in artificial intelligence, emphasizing the need for AI to learn from first-hand experiences rather than relying solely on human-generated data [2][5][10] - Richard S. Sutton highlights the limitations of current AI models, which are based on second-hand experiences, and advocates for a new approach where AI interacts with its environment to generate original data [6][7][11] Group 1: Transition to Experience Era - The current large language models are reaching the limits of human data, necessitating a shift to real-time interaction with environments to generate scalable original data [7][10] - Sutton draws parallels between AI learning and human learning, suggesting that AI should learn through sensory experiences similar to how infants and athletes learn [6][8] - The experience era will require AI to develop world models and memory systems that can be reused over time, enhancing sample efficiency through high parallel interactions [3][6] Group 2: Decentralized Cooperation vs. Centralized Control - Sutton argues that decentralized cooperation is superior to centralized control, warning against the dangers of imposing single goals on AI, which can stifle innovation [3][12] - The article emphasizes the importance of diverse goals among AI agents, suggesting that a multi-objective ecosystem fosters innovation and resilience [3][12][13] - Sutton posits that human and AI prosperity relies on decentralized cooperation, which allows for individual goals to coexist and promotes beneficial interactions [12][14][16] Group 3: Future of AI Development - The development of fully intelligent agents will require advancements in deep learning algorithms that enable continuous learning from experiences [11][12] - Sutton expresses optimism about the future of AI, viewing the creation of superintelligent agents as a positive development for society, despite the long-term nature of this endeavor [10][11] - The article concludes with a call for humans to leverage their experiences and observations to foster trust and cooperation in the development of AI [17]