henry 发自 凹非寺 量子位 | 公众号 QbitAI 如果没有PhD，是不是就和前沿AI研究没关系了？ 至少在 Noam Brown 看来，未必。 这位 OpenAI 研究员、o1的核心贡献者 ，刚刚分享了一串"非典型研究员"的经历。 有人没有论文、有人没读研、有人白天在麦肯锡上班，晚上在GitHub写研究。 还有人没事就在推特发帖、GitHub上提问的。 他们后来都去了哪？ OpenAI、DeepMind、Anthropic。 进OpenAI，并不一定要博士学位 在这篇长帖中，Noam分享了 Keller Jordan 、 Sholto Douglas 、 Andy Jones 、 Kevin Wang 等一批"非典型"研究员的经历。 总体看下来，他们都有一些相同的特质。比如，主动性极强（热爱）、公开研究，不闭门造车、工程能力在线、会PR、不执着于头衔。 接下来，我们具体来看。 Keller Jordan：套瓷研究+推特学术 首先被Noam分享的，是 Keller Jordan 。 他现在在OpenAI，从事预训练相关的研究。而他的研究生涯，某种程度上，起于一段典型的"套瓷研究"。 最终，他们与其他作 ...

Core Viewpoint - The article emphasizes the ongoing AI boom and invites individuals to join the company "Quantum Bit," which focuses on tracking AI advancements and has established itself as a leading content platform in the industry [1]. Job Opportunities - The company is hiring for three main directions: AI Industry, AI Finance, and AI Product, with positions available for both experienced professionals and fresh graduates [2][4]. AI Industry Direction - Responsibilities include monitoring innovations in infrastructure, such as chips, AI infrastructure, and cloud computing, as well as producing accessible interpretations of cutting-edge research and technical reports from major conferences [6][7]. - The company offers a dynamic work environment, opportunities for personal influence, and professional mentorship for newcomers [6]. AI Finance Direction - This role focuses on venture capital and financial reporting within the AI sector, tracking capital movements in the industry and producing analyses of investment trends and company strategies [9]. AI Product Direction - Responsibilities involve assessing AI applications and hardware, tracking new product releases across various platforms, and engaging with entrepreneurs and product experts in the AI space [10]. Company Growth and Impact - By 2025, Quantum Bit aims to have over 2.4 million subscribers on WeChat and more than 7 million users across all platforms, with a daily reading volume exceeding 2 million [12].

Core Viewpoint - The research from the Institute of Physics, Chinese Academy of Sciences, reveals a significant breakthrough in ferroelectric materials, specifically in the atomic-level "one-dimensional charged domain walls" within zirconia, laying a new physical foundation for next-generation artificial intelligence devices [1][4]. Group 1: Breakthrough in Ferroelectric Materials - The research team confirmed that the width and thickness of these domain walls are only the size of a single crystal cell, confined within a two-dimensional polar layer, achieving the physical limit of size [3][10]. - This discovery unveils the charge screening mechanism of oxygen ions' "self-balancing," breaking through the traditional storage density bottleneck of two-dimensional domain walls [3][22]. - The unique "polarization-ion" coupling transport characteristics of this one-dimensional structure open new physical pathways for constructing high-energy-efficient brain-like computing chips and AI devices [4][24]. Group 2: Characteristics of Ferroelectric Materials - Ferroelectric materials are defined as a class of crystalline materials with spontaneous polarization, where the polarization direction can be reversed by an external electric field [6]. - These materials can be visualized as filled with tiny "electrical compasses" that indicate the direction of charge separation rather than geographical north and south [6][7]. - The concept of ferroelectric domains is introduced, where these "compasses" align in groups to minimize energy, forming domain walls that separate different polarization regions [8][9]. Group 3: Unique Structure of Domain Walls - The research team discovered that in zirconia, the originally broad two-dimensional "walls" are compressed into atomic-scale one-dimensional "lines" due to the material's unique sub-cell layered structure [11][12]. - These one-dimensional structures are not ordinary "walls" but special charged domain walls, categorized as "head-to-head" and "tail-to-tail" [12][13]. - The stability of these high-energy structures, which are typically unstable, is maintained through the introduction of high concentrations of point defects acting as "charge glue" [29][30]. Group 4: Implications for Data Storage and Ion Transport - The theoretical data storage density using these atomic-level one-dimensional domain walls can reach 20TB per square centimeter, equivalent to storing 10,000 HD movies on a device the size of a postage stamp [24]. - The material exhibits superior ionic conductivity at room temperature, outperforming traditional solid electrolytes like yttria-stabilized zirconia (YSZ), transforming it into a "highway" for ion transport [22][23]. - The research highlights a precise "charge compensation mechanism" that allows the one-dimensional domain walls to exist stably while facilitating efficient ionic conduction [36].

Core Viewpoint - The article emphasizes the ongoing AI boom and invites individuals to join the company "Quantum Bit," which focuses on tracking AI advancements and has established itself as a leading content platform in the industry [1]. Group 1: Job Opportunities - The company is hiring for three main directions: AI Industry, AI Finance, and AI Product, with positions available for both experienced professionals and fresh graduates [2][4]. - Positions are open for various levels, including editors, lead writers, and chief editors, with a focus on matching roles to individual capabilities [6]. Group 2: Job Responsibilities - **AI Industry Direction**: Responsibilities include tracking innovations in infrastructure, such as chips, AI infrastructure, and cloud computing, as well as interpreting technical reports from conferences [6][7]. - **AI Finance Direction**: Focuses on venture capital, financial reports, and capital movements within the AI industry, requiring strong analytical skills and a passion for interviews [11]. - **AI Product Direction**: Involves monitoring AI applications and hardware developments, requiring a keen understanding of product experiences and market trends [11]. Group 3: Benefits and Growth - Employees will have the opportunity to engage with industry leaders, participate in significant tech events, and receive mentorship from senior editors [6]. - The company offers competitive salaries and comprehensive benefits, including social insurance, meal allowances, and performance bonuses [6]. Group 4: Company Growth Metrics - By 2025, Quantum Bit aims to have over 2.4 million subscribers on WeChat and more than 7 million users across all platforms, with a daily reading volume exceeding 2 million [12].

Group 1 - Meta is elevating its focus on computing power to an unprecedented strategic level [39] - The company has undergone organizational restructuring, increasing its executive team from 148 to 167 members, with more individuals reporting directly to CEO Mark Zuckerberg [2][8] - The new project "Meta Compute" aims to establish tens of gigawatts of computing infrastructure within this decade, potentially expanding to hundreds of gigawatts [12][39] Group 2 - The leadership of "Meta Compute" includes Santosh Janardhan, a long-time Meta employee, and Daniel Gross, a recent hire with a strong AI background [10][25] - Meta's capital expenditure is projected to soar to $70 billion to $72 billion in 2025, nearly doubling from approximately $39 billion in 2024 [50] - The company has committed to investing over $600 billion in data center expansion in the U.S. by 2028, which supports the ambitious plans of "Meta Compute" [52] Group 3 - Meta is integrating AI chip capabilities by acquiring talent from the startup Rivos, which specializes in high-performance AI acceleration chips [41][47] - The company is also developing new AI models, with the first models expected to be delivered internally and launched in Q1 2026 [55][58]

Core Viewpoint - The article discusses the significance of Gwynne Shotwell, the President of SpaceX, as a key figure behind the company's success, especially in light of the upcoming IPO that aims for a valuation of $1.5 trillion and over $30 billion in financing [2][55]. Group 1: Gwynne Shotwell's Role and Characteristics - Gwynne Shotwell is recognized as the "Adult in the Room" at SpaceX, effectively managing the company alongside Elon Musk [6][5]. - She is described as bold, passionate, and direct, which allows her to maintain a productive relationship with Musk despite his challenging personality [10][11]. - Shotwell has a strong technical background, holding degrees in mechanical engineering and applied mathematics, which contributes to her credibility in the aerospace industry [15]. Group 2: Key Contributions to SpaceX - Shotwell played a crucial role in saving SpaceX during its early struggles in 2008, particularly after the failure of the Falcon 1 rocket launches [28][29]. - She successfully negotiated a $1.6 billion contract with NASA for cargo transport to the International Space Station, which was pivotal for SpaceX's survival [32][34]. - Her ability to understand Musk's vision and provide constructive feedback has been essential in aligning the company's goals and strategies [40][41]. Group 3: Leadership and Team Dynamics - Shotwell's leadership style is characterized by her loyalty to Musk while also advocating for the team, ensuring a balance between company demands and employee welfare [49][50]. - She emphasizes teamwork and collective achievement over individual recognition, often attributing SpaceX's success to the efforts of all employees [52]. - The dynamic between Musk and Shotwell is likened to a partnership where Musk's visionary ideas are complemented by Shotwell's operational stability, akin to a ship's sail and rudder [54]. Group 4: Market Impact and Future Outlook - As SpaceX approaches a critical juncture in the capital markets, Shotwell's presence is seen as a stabilizing factor that reassures investors and stakeholders [55].

Core Viewpoint - The article discusses the introduction of MIRA, a universal base model designed for medical time series data, which addresses the challenges of irregular and heterogeneous medical data, aiming to enhance predictive capabilities in healthcare AI [5][25]. Group 1: Medical AI Landscape - Large Language Models (LLMs) and Computer Vision (CV) are transforming the healthcare industry, enabling AI to interpret CT images and write medical summaries [1]. - A critical missing piece in medical AI is the ability to understand the "dynamic evolution of life," which is essential for capturing the continuous trajectory of vital signs [2][4]. Group 2: Challenges in Medical Time Series Data - Traditional deep learning models rely on idealized assumptions of uniform data sampling, which do not hold true in real-world medical scenarios, particularly in Intensive Care Units (ICUs) where vital signs are recorded at irregular intervals [9][10]. - The characteristics of medical time series data include irregular time intervals, heterogeneous sampling rates, and data missing due to non-standard clinical workflows [12]. Group 3: MIRA Model Introduction - MIRA is built on 454 billion medical data points and aims to overcome the limitations of traditional models by learning physiological dynamic patterns across various scenarios and modalities [5][25]. - MIRA employs two core technologies: Continuous Time Rotational Position Encoding (CT-RoPE) for understanding historical data and Neural ODE for predicting future states [13][18]. Group 4: Experimental Validation - MIRA demonstrates zero-shot transfer capabilities, outperforming some supervised models in out-of-distribution tests, indicating its ability to learn general physiological signal changes [21]. - MIRA shows high robustness in handling sparse data, maintaining performance even with only 30% of observation points, unlike traditional models that rely on interpolation [23][24]. Group 5: Future Implications - The introduction of MIRA marks a significant step towards a "universal base" era in medical AI, allowing hospitals to quickly develop high-precision customized models with minimal local data [25].

衡宇 发自 太湖边 量子位 | 公众号 QbitAI 就在昨晚，具身智能创业公司 魔法原子，宣布将登上今年央视春晚舞台 。 几乎在官宣新闻发布的同一时间，魔法原子联合创始人顾诗韬在与我们的交流中透露， 魔法原子将在今年冲击IPO 。 魔法原子相对还比较年轻，我们是在2024年的时候刚成立。 现在正在加速上市的进程，正严格按照最快的时间表在排期。比较期待2026年能在二级市场看到我们的消息。 魔法原子 成立于2024年1月 ，旗下产品包括Magic Bot机器人系列及Magic Dog四足机器人系列。 据介绍，Magic Bot可完成复杂动作并具拟人社交能力；Magic Dog能耐受-20℃至55℃极端环境，凭精准定位穿梭复杂场景。 沟通过程中，顾诗韬同步透露了一些数据和此前未曾公开的信息。 她告诉我们， 魔法原子即将发布新产品 ，除此之外，还传递了关于魔法原子"现在"和"未来"的更多消息。 顾诗韬 ：长期具身智能之战当前还在商业化落地的非常早期的阶段，未来的具身智能之争， 最终会变成一个资源之争，资金之争和人才之争 竞争越来越激烈的前夜，我们要做两件事。 第一件事情是 快速去跑通自己商业化的闭环 ，需要从经营 ...

Core Insights - The article discusses the breakthrough of the Being-H0.5 model in the field of embodied intelligence, addressing the challenges posed by data isolation and the "Matthew Effect" in the industry [1][3][39] - Being-H0.5 is the largest VLA model with 35,000 hours of training data, enabling cross-robot zero-shot skill transfer and showcasing remarkable generalization capabilities [2][3][30] Data and Model Development - The Being-H0.5 model integrates 35,000 hours of data, including 14,000 hours of robot data and 16,000 hours of human data, across 30 robot types, allowing for rapid adaptation and stable execution regardless of hardware configuration [2][8] - The UniHand-2.0 dataset, an iteration of UniHand-1.0, features over 35,000 hours of high-quality data, marking a significant advancement in cross-domain data integration [8][9] Training Paradigms - The model employs a human-centric learning paradigm, aligning human intent with robotic actions through a unified token sequence that captures physical interaction signals [20][39] - A unified action space framework is established to overcome the dimensional gap between heterogeneous hardware, facilitating joint training and knowledge sharing [16][17] Architectural Innovations - The Mixture-of-Flow (MoF) architecture allows for the decoupling of action experts, focusing on learning universal motion primitives while ensuring precise execution for specific robot types [22][23] - The model incorporates mechanisms like manifold-preserving gating and universal async chunking to enhance robustness and adaptability across different hardware [23][24] Performance and Validation - Extensive real-world testing on various robot types demonstrated that Being-H0.5 can perform complex tasks, achieving competitive success rates compared to specialized models [28][30][35] - The model's performance in quantitative evaluations shows it surpasses existing VLA models, achieving an average success rate of 98.9% in specific tasks [35][36] Open Source and Future Directions - The BeingBeyond team commits to a full-stack open-source approach, providing not only pre-trained models but also complete training frameworks and evaluation tools to foster community innovation [37][38] - The vision is to establish Being-H0.5 as a foundational infrastructure in the embodied intelligence sector, enabling rapid development without the need for extensive data collection [39]

Core Viewpoint - The article discusses the increasing integration of AI in the medical field, particularly focusing on the advancements made by Baichuan Intelligent in developing a reliable AI model for clinical use, addressing the challenges of trust and cost in medical AI applications [5][6][20]. Group 1: AI Adoption in Healthcare - Many doctors, especially younger ones, are beginning to embrace AI technologies in their practice, with Baichuan's professional model having around 100,000 doctor users [2]. - The medical community generally agrees on the potential of AI, but there are significant barriers to its clinical implementation, primarily trust and cost [4][5]. Group 2: Baichuan's M3 Plus Model - Baichuan's latest model, M3 Plus, has achieved a hallucination rate of 2.6%, the lowest in global evaluations, thanks to its unique six-source evidence technology [6][19]. - The model's success is attributed to Fact-Aware Reinforcement Learning, which incorporates medical facts into the training process to reduce hallucinations [12][46]. Group 3: Cost Reduction and Accessibility - M3 Plus has undergone extensive optimization, resulting in a 70% reduction in API call costs compared to its predecessor, making it more accessible for hospitals and doctors [21][47]. - The Gated Eagle-3 architecture enhances inference throughput by approximately 15%, further lowering the cost per request [22]. Group 4: Evidence Anchoring Technology - Baichuan has introduced "Evidence Anchoring" in M3 Plus, ensuring that every medical conclusion made by the AI is directly supported by original evidence from literature [32][46]. - This approach addresses common issues in AI-generated medical responses, such as incorrect citations and conflicting information, which have historically plagued the industry [25][30]. Group 5: Free Access Initiative - Baichuan has launched the "Haina Baichuan" free plan, allowing unlimited access to M3 Plus for institutions serving medical professionals, provided they display "Powered by Baichuan" [47][48]. - This initiative aims to prevent redundant technological development in the industry and facilitate real-world testing and iteration of AI applications in healthcare [54][56]. Group 6: Impact on Medical Professionals - The advancements in AI, particularly the reduction in hallucination rates, provide medical professionals with greater confidence in their decision-making processes [57]. - The article emphasizes the importance of translating these technological improvements into practical applications that benefit patients directly [61].