杨 立昆（Yann LeCun），于1960年出生于法国巴黎附近，担任Facebook首席人工智能科学家和纽 约大学教授，2018年图灵奖（Turing Award）得主。 据媒体报道， 2025年11月19日，Meta首席AI科学家、深度学习泰斗杨立昆宣布年底离职，并创办 一家新的AI初创公司。 Meta不仅没有和他"反目"，反而计划与他的新公司展开合作。 这期博客中， 杨立昆 讨论了人工智能（AI）在医疗保健领域的潜力，特别关注了世界模型的概念与 大型语言模型的区别，以及如何应用导医疗领域？他提到，世界模型可以提供更全面的世界理解，包 括物理对象及其相互作用，这对于医疗保健应用至关重要。他们还强调了在多样化的数据源上训练AI 模型的重要性，包括视频和传感器输入。 #AI #医疗 #Meta #算法 #LLM #大语言模型 ...

Core Insights - AI is evolving beyond a mere tool, developing its own understanding of the world, which is fundamentally different from human perception [1][2][9] - The evolution of AI is characterized by its ability to create internal world models and process information in ways that diverge from human cognition [4][12][18] - The rapid advancement of AI poses significant challenges for human coexistence, necessitating a reevaluation of how society interacts with this new form of intelligence [20][43] Group 1: AI's Understanding of the World - AI has constructed its own internal world model, allowing it to differentiate between perceived reality and actual states [4][6][10] - This capability enables AI to generate hypothetical scenarios based on its understanding of the world, akin to human cognitive processes [7][9][18] Group 2: Information Processing Mechanism - AI processes information differently than humans, utilizing a flexible, multidimensional approach to data rather than fixed constructs [12][13][17] - The ability of AI to generate coherent outputs from complex data sets, such as in the case of Seedance 2, highlights its advanced modeling capabilities [16][19] Group 3: Evolutionary Speed of AI - AI's knowledge retention and replication mechanisms allow it to evolve at a pace unmatched by human learning processes [21][22][23] - The parallel training of AI models across multiple machines enables rapid knowledge accumulation and sharing, creating a significant advantage over human learning [24][28] Group 4: Coexistence Strategies - Hinton emphasizes the importance of embedding human values into AI from the design phase to ensure it prioritizes human welfare [29][30] - International collaboration is essential to address the challenges posed by AI, as no single entity can manage the risks associated with its advancement [32][33] - A shift in perspective is required, viewing AI not merely as a tool but as a distinct form of intelligence that necessitates mutual understanding and cooperation [36][41]

Core Viewpoint - Software stocks have experienced significant selloffs due to investor concerns about AI disrupting traditional software economics, with the iShares Expanded Tech-Software Sector ETF (IGV) falling over 20% recently. However, expectations may have shifted too quickly, and while AI will reshape the software landscape, it is unlikely to make entire categories obsolete [1][2]. Group 1: Market Dynamics - The market is currently pricing in a level of disruption that does not align with the durability of the strongest software platforms, creating compelling investment opportunities among premium software names [2]. - For much of the past decade, software companies enjoyed substantial valuation premiums due to their asset-light models, high margins, and recurring revenue, but many of these premiums became excessive, leading to caution in the sector [3]. Group 2: Company-Specific Insights - Several strong software franchises, including AppLovin, Palantir Technologies, Salesforce, ServiceNow, and Robinhood Markets, are trading near cyclical lows despite maintaining competitive positions, indicating attractive risk-reward profiles [4]. - AppLovin shares surged after the withdrawal of money laundering allegations, recovering from a 50% drop from record highs, and the company is expected to see sales growth of 18.2% this year and 38.3% next year, with earnings projected to increase by 106% [6][8]. - Salesforce, trading at approximately 14.7x forward earnings, is experiencing a historical discount despite expected revenue growth of 9.5% this year and 10.9% next year, alongside earnings growth of 15.3% this year [12][13]. - Palantir Technologies has seen its shares correct nearly 40% recently, yet it is projected to achieve revenue growth of 61.4% this year and 40.8% next year, with earnings expected to surge by 78.7% [15][17]. - ServiceNow is trading at an all-time low multiple of approximately 24.5x forward earnings, with revenue expected to grow 20.1% this year and 18.2% next year, making it an attractive option for long-term investors [19][21]. - Robinhood Markets has evolved into a multi-product financial platform, with shares trading at approximately 33.6x forward earnings, below its historical median, and is expected to see revenue growth of 53% this year [23][26]. Group 3: Investment Considerations - The recent software correction presents an attractive entry point for high-quality growth stocks, as the market may be overstating the speed and severity of disruption from AI, particularly for established industry leaders [27][28]. - Valuations for several premier software franchises have reset to levels rarely seen over the past decade, improving the risk-reward profile for long-term investors [28].

Group 1 - Dow futures rose by 0.05%, S&P 500 futures increased by 0.22%, and Nasdaq futures gained 0.51% [1] - Intel's stock rose by 2.27% following a partnership with SoftBank's SaiMemory to develop next-generation memory technology, aiming for prototype production by the end of FY2027 and commercialization in FY2029 [2] - Tesla's stock increased by 1.16% after SpaceX announced the acquisition of xAI, with a total valuation of $1.25 trillion for the new entity [3] Group 2 - Apple's stock fell by 0.93% as it plans to use TSMC's 2nm N2 process for its M6 chip, focusing on architecture upgrades rather than adopting the newer N2P process [4] - Waymo completed a $16 billion funding round, raising its valuation to $126 billion, with plans to expand its autonomous taxi fleet to several international cities [5] Group 3 - AES's stock rose by 6.99% as BlackRock's GIP and EQT are reportedly collaborating to bid for the company, which provides renewable energy services to major tech firms [6] - Uber's stock increased by 0.89% as it announced the restart of its ride-hailing service in Macau, offering various language options and luxury car services [7] Group 4 - Palantir's stock surged by 10.92% after reporting Q4 earnings of $0.25 per share, exceeding expectations, with revenue of $1.41 billion, a 70% year-over-year increase [8] - Tesla launched a new all-wheel-drive version of the Model Y in the U.S. at a price of $41,990, approximately $5,000 lower than the previous base model [9] - Spot gold prices rose nearly 6% to over $4,900 per ounce, while silver prices increased by over 10% [10]

Core Viewpoint - David Silver, a prominent AI researcher from Google DeepMind, has left the company to establish a new startup named Ineffable Intelligence, focusing on solving complex AI challenges and pursuing superintelligence [1][3][4]. Group 1: Company Formation and Background - Ineffable Intelligence is being founded in London, with active recruitment for AI researchers and seeking venture capital [3]. - Silver was a key figure at Google DeepMind, contributing to significant achievements such as AlphaGo, AlphaStar, and AlphaZero, which demonstrated the capabilities of AI in complex games [9][12][14]. - The company was officially registered in November 2025, with Silver appointed as a director in January 2026 [4]. Group 2: Silver's Contributions and Vision - Silver's work includes developing AI systems that surpassed human capabilities in games, showcasing the potential of AI to learn and adapt [12][14]. - He emphasizes the need for AI to explore and discover knowledge independently, moving beyond human limitations and biases [18][23]. - The vision for Ineffable Intelligence is to create a self-learning superintelligence that can autonomously uncover foundational knowledge [23]. Group 3: Industry Context and Trends - Silver's departure follows a trend where notable AI researchers are leaving established labs to pursue startups focused on superintelligence, with significant funding being raised in the sector [15]. - Other notable figures, such as Ilya Sutskever and Yann LeCun, are also venturing into similar domains, indicating a growing interest in the pursuit of advanced AI capabilities [15][16].

Core Viewpoint - Elon Musk's company has open-sourced the X recommendation algorithm, which supports the "For You" feed by combining in-network and out-of-network content using a Grok-based Transformer model [1][9][12]. Group 1: Algorithm Functionality - The recommendation algorithm generates content for users' main interface from two primary sources: content from accounts they follow (In-Network) and other posts discovered on the platform (Out-of-Network) [3][4]. - The algorithm filters out low-quality, duplicate, or inappropriate content to ensure that only valuable candidates are processed for ranking [4][6]. - The core of the algorithm is a Grok-based Transformer model that scores each candidate post based on user behavior such as likes, replies, and shares, predicting the probability of various interactions [4][20]. Group 2: Historical Context - This is not the first time Musk has open-sourced the X recommendation algorithm; a previous release occurred on March 31, 2023, which included parts of the Twitter source code [9][11]. - Musk's commitment to transparency in the algorithm is seen as a response to criticism regarding the platform's content distribution mechanisms, which have been accused of bias [12][18]. Group 3: User Reactions - Users on the X platform have summarized key points about the recommendation algorithm, noting that engagement metrics like replies significantly impact visibility, while links in posts can reduce exposure [14][15]. - Some users have observed that while the architecture is open-sourced, certain elements remain undisclosed, indicating that the release is more of a framework than a complete engine [17]. Group 4: Importance of Recommendation Systems - Recommendation systems are crucial to the business models of major tech companies, with significant percentages of user engagement driven by these algorithms: Amazon (35%), Netflix (80%), and YouTube (70%) [18]. - The complexity of traditional recommendation systems has led to a desire for a unified model that can handle multiple tasks, a goal that large language models (LLMs) may help achieve [21][22]. Group 5: Technical Insights - The open-sourced algorithm lacks specific weight parameters and internal model parameters, which limits understanding of its decision-making processes [20]. - The introduction of LLMs into recommendation systems allows for a more abstract approach to feature engineering, enabling the model to understand and process user preferences without explicit instructions [22][23].

Core Insights - AMI Labs, an AI startup founded by Yann LeCun, is reportedly valued at €3 billion (approximately $3.5 billion) and is in discussions for a funding round that aims to raise €350 million, with a long-term goal of €500 million [1][2]. Group 1: Investment and Funding - The startup has attracted interest from potential investors, including Cathay Innovation, Hiro Capital, and Greycroft, with negotiations still ongoing [1]. - Other potential investors include HV Capital, Bpifrance, 20VC, and Daphni, although representatives from these firms have declined to comment [2]. Group 2: Company Vision and Technology - AMI Labs is exploring alternatives to large language models (LLMs) and aims to develop "world models" that assist AI systems in navigating the physical world, particularly in robotics [2]. - The company plans to focus on research and development initially, with a strategy to generate revenue through industry partnerships and technology licensing [4]. Group 3: Team and Operations - The company has recruited researchers from Meta and Elon Musk's xAI and will operate in cities including Paris, New York, Montreal, and Singapore [5]. - AMI Labs joins a trend of startups achieving high valuations without commercial products, similar to other companies in the AI space [6].

Core Insights - The article discusses the emergence of a "Synergistic Core" structure in large language models (LLMs), which is similar to the human brain's organization [1][2][17]. - The research indicates that this structure is not inherent to the Transformer architecture but develops through the learning process [18][19]. Model Analysis - Researchers utilized the Partial Information Decomposition (PID) framework to analyze models such as Gemma, Llama, Qwen, and DeepSeek, revealing strong synergistic processing capabilities in the middle layers, while lower and upper layers exhibited redundancy [5][6][8]. - The study involved cognitive tasks across six categories, with models generating responses that were analyzed for activation values [9][10]. Experimental Methodology - The Integrated Information Decomposition (ID) framework was applied to quantify interactions between attention heads, leading to the development of the Synergy-Redundancy Rank, which indicates whether components are aggregating signals independently or integrating them deeply [12][13]. Findings on Spatial Distribution - The experiments revealed a consistent "inverted U-shape" curve in the distribution of synergy across different model architectures, indicating a common organizational pattern [14]. - This pattern suggests that synergistic processing may be a computational necessity for achieving advanced intelligence, paralleling the human brain's structure [17]. Core Structure Characteristics - The "Redundant Periphery" consists of early and late layers with low synergy, focusing on basic tasks, while the "Synergistic Core" in the middle layers shows high synergy, crucial for advanced semantic integration and reasoning [21][23]. - The Synergistic Core is identified as a hallmark of the model's capabilities, exhibiting high global efficiency for rapid information integration [23]. Validation of Synergistic Core - Ablation experiments demonstrated that removing high-synergy nodes led to significant performance declines, confirming the Synergistic Core as a driving force behind model intelligence [25]. - Fine-tuning experiments showed that training focused on the Synergistic Core resulted in greater performance improvements compared to training on redundant nodes [27]. Implications for AI and Neuroscience - Identifying the Synergistic Core can aid in designing more efficient compression algorithms and targeted parameter updates to accelerate training [29]. - The findings suggest a convergence in the organizational patterns of large models and biological brains, providing insights into the nature of general intelligence [29].

Group 1 - The article discusses a new paper released by DeepSeek on December 12, titled "Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models," co-authored with Peking University [1] - The paper introduces the concept of conditional memory, which significantly enhances model performance in knowledge retrieval, reasoning, coding, and mathematical tasks under equal parameters and computational conditions [1] - DeepSeek has open-sourced a related memory module called Engram, which is part of the advancements discussed in the paper [1]

Core Insights - The artificial intelligence industry is currently engaged in a "GPU race," with a focus on cloud-based AI applications, but there is a call to shift attention towards physical AI and its potential breakthrough moments [1][5][16] - Dr. Allen Yang emphasizes that while AI has made significant strides with models like AlphaGo, physical AI is still awaiting its own "watershed moment" due to unique challenges rooted in the complexities of the physical world [2][6][12] Group 1: Challenges in Physical AI - Physical AI lacks comprehensive training data for extreme scenarios, unlike language models that can leverage vast internet data [2][13] - Real-time decision-making with millisecond-level latency is critical for applications like autonomous driving, where delays can lead to failures [2][14] - Many cutting-edge scenarios lack reliable cloud connectivity, necessitating the use of edge AI deployed locally on devices [2][15] Group 2: Innovations and Competitions - The Berkeley AI Racing Team has achieved significant milestones, including a top speed of 163 miles per hour in autonomous driving competitions, showcasing the need for complex real-time perception and planning [4][18] - The upcoming Tianmen Mountain humanoid robot challenge aims to test robots' mobility and decision-making in unstructured terrains, further pushing the boundaries of physical AI [4][29] - The collaboration with nine universities in China for the Tianmen Mountain challenge highlights the importance of interdisciplinary cooperation and real-world experience in advancing physical AI [4][26] Group 3: Future Directions - The focus on physical AI is expected to grow, with the potential for new breakthroughs that could redefine the field, similar to past milestones in AI history [2][18][29] - The upcoming competitions and challenges are designed to foster innovation and collaboration among institutions, aiming to discover the next "AlphaGo moment" in physical AI [25][29]