Core Viewpoint - The FDA approval of the Jewel Patch-WCD wearable defibrillator marks a significant advancement in providing a non-invasive solution for high-risk patients facing sudden cardiac arrest (SCA) [1][3]. Group 1: Product Features - Jewel Patch-WCD is designed with a low-profile, waterproof patch that allows for continuous ECG monitoring without disrupting daily activities, including sleep and showering [3][6]. - The device is intended for temporary high-risk patients, including those recovering from myocardial infarction or with severe cardiovascular diseases, with an estimated 500,000 patients in the US and Europe benefiting annually [6][9]. - The device has demonstrated high compliance, with an average daily wear time exceeding 23 hours during clinical trials, successfully identifying and treating arrhythmias without any reported deaths or serious complications [6][9]. Group 2: Technology and Innovation - Jewel Patch-WCD utilizes a machine learning-based arrhythmia recognition system, achieving a low inappropriate shock rate of 0.36 per 100 patients per month, significantly better than many similar devices [7][9]. - The device integrates with the Jewel mobile app for near real-time data transmission to healthcare providers, facilitating remote monitoring and personalized cardiovascular care [7][9]. Group 3: Company Background - Element Science, founded by Dr. Uday Kumar in 2011, focuses on developing next-generation wearable technologies for high-risk cardiac patients, particularly in preventing SCA [8]. - The company has raised $145 million in Series C funding, positioning itself as one of the fastest-growing startups in the wearable medical device sector [8].

Core Findings - An international research team, including Israeli researchers, discovered that psychological stress during pregnancy can "reprogram" key molecular pathways in fetuses, affecting newborns' stress response systems in different ways based on gender [1][2] - The study involved over 120 mother-infant pairs recruited in Germany from 2016 to 2018, focusing on pregnant women with high perceived stress levels, which often go unmonitored [2][3] - The research highlighted significant molecular changes in female infants, particularly the near-complete disappearance of choline tRNA fragments, which regulate genes responsible for synthesizing acetylcholine, a neurotransmitter [1] - Male infants exhibited higher levels of acetylcholinesterase, an enzyme that breaks down acetylcholine, indicating an imbalance in their stress response systems from birth [1] - Machine learning algorithms were utilized to analyze choline tRNA fragment characteristics in newborn girls, achieving up to 95% accuracy in determining exposure to maternal stress during pregnancy, paving the way for early diagnostic and intervention tools [1] Research Implications - The study emphasizes the importance of maternal mental health during pregnancy, particularly for women who perceive high stress but are not diagnosed with depression or anxiety, as their stress can still impact both themselves and their newborns [2] - The findings suggest potential avenues for early diagnosis and support for pregnant women experiencing high levels of stress [2]

当前，量子计算领域蓬勃发展，却仍面临"它到底有什么用"的本质问题。在本文作者来看，在这样的环境下，正是大力推动量子算法的时刻，应降低对量子 算法原有要求，寻找可验证且实用的算法，呼吁理论家积极探索，推动量子计算突破瓶颈。值得一提的是，本文得到了理论物理学家John Preskill的推 荐："如果你对量子计算感兴趣，我强烈推荐加州理工学院学生robbieking1000的这篇文章，呼吁采取'更务实（scrappier）的方法'来寻找新的应用。" 本文来自微信公众号：返朴 （ID：fanpu2019），作者：robbieking1000，翻译：一二三，题图来自：AI生成 量子计算正处在一个奇特的阶段。技术层面上，经过数十亿美元投资和数十年的研究，实用的量子计算机正逐步接近实现。但令人尴尬的是，如今人们对量 子计算最常提出的问题，仍然和20年前一样：量子计算机到底能做什么？诚实的回答暴露了房间里的大象：我们至今也没有完全的答案。对于像我这样的理 论家来说，这既是一种挑战，也是一种行动的召唤。 技术动能 假设几十年后我们仍未拥有实用的量子计算机，原因会是什么？不太可能是因为遇到了无法逾越的工程障碍。量子纠错的理论基 ...

Group 1 - The article provides a step-by-step guide for selecting currency pairs based on different trading styles, emphasizing the need for alignment between trading style and currency characteristics [1][2][5]. - Day trading is characterized by short holding periods, relying on technical analysis for small profit margins, with recommended pairs including EUR/USD and USD/JPY due to their high liquidity and low spreads [2][4]. - Swing trading focuses on capturing medium-term trends over days to weeks, with suitable pairs like AUD/USD and GBP/USD driven by fundamental factors such as commodity prices and policy expectations [5][7][8]. Group 2 - Carry trade involves long-term positions to earn interest rate differentials, favoring high-yield currencies while managing exchange rate risks [12][15]. - Event-driven trading capitalizes on market reactions to economic data releases, with pairs like USD/JPY and USD/TRY being highlighted for their volatility during such events [4][22]. - Algorithmic trading strategies include statistical arbitrage, monitoring price discrepancies between currency pairs, and utilizing low-latency execution for optimal performance [14][15][26]. Group 3 - The article outlines a five-step self-assessment method for traders to evaluate their risk tolerance, time commitment, and tool compatibility when selecting currency pairs [17][19][21]. - Recommended currency pairs for different trading styles include EUR/USD and XAU/USD for day trading, AUD/USD and GBP/USD for swing trading, and AUD/JPY and USD/ZAR for carry trading, each with specific risk management parameters [21][22][23]. - Common pitfalls for novice traders include mismatching trading styles with currency pairs, overlooking overnight costs, and overtrading less liquid pairs [24][25][26].

Core Viewpoint - Gold exhibits a unique price fluctuation mechanism influenced by multiple factors, including currency pricing systems, macroeconomic risks, market structure evolution, supply-demand elasticity, and technical reinforcement mechanisms. Group 1: Currency Pricing System Linkage - The international gold price is negatively correlated with the US dollar index, where a 1% increase in the dollar index raises gold purchasing costs, suppressing investment demand [1]. - Major central banks' balance sheet expansions directly elevate gold price benchmarks, with each additional $1 trillion in quantitative easing raising gold valuations by 8%-12% [2]. Group 2: Macroeconomic Risk Matrix - The forward price of gold is determined by the nominal interest rate minus inflation expectations, with gold prices reaching a historical peak of $2075 per ounce when the real yield on US Treasuries fell below -1% [3]. - A 10-point increase in the global geopolitical risk index results in a 3.2% increase in average monthly gold holdings, evidenced by events like the Crimea crisis and the Russia-Ukraine conflict [4]. Group 3: Market Structure Evolution - Emerging market central banks have increased gold purchases for 13 consecutive years, with global official reserves rising by 1136 tons in 2022, accounting for 23% of annual supply [5]. - An increase of 100,000 open contracts in COMEX gold futures raises price volatility by 1.8 basis points, with significant spikes in implied volatility during events like the Silicon Valley Bank incident [6]. Group 4: Supply-Demand Elasticity - The average extraction cost of the top ten gold mines has risen to $1250 per ounce, with newly discovered reserves declining by 15% year-on-year [7]. - India and China account for 55% of global physical gold demand, with a 40% surge in imports during festive seasons, despite India's recent increase in import tax to 15% [8]. Group 5: Technical Reinforcement Mechanisms - Algorithmic trading strategies hold over 30 million ounces of gold, with momentum factors contributing over 35% to price volatility, triggering significant buy orders upon breaking key price levels [9]. - A 50% year-on-year increase in Google searches for "gold investment" correlates with a 68% probability of gold price increases in the following 30 days [10].

Core Insights - The article discusses the rise of algorithmic power in decision-making processes and its implications for human autonomy, as highlighted in Marek Kowalkiewicz's book "Algorithmic Economy: The Intelligent Evolution of Business Logic and Human Life" [1][2] Group 1: Algorithmic Power and Its Impact - The introduction of algorithms in predicting student grades during the pandemic led to significant protests, revealing how algorithmic decisions can undermine individual efforts and fairness [2] - Approximately 36% of students received grades lower than teacher expectations, with 3% scoring two grades lower than anticipated, showcasing the potential pitfalls of algorithmic assessments [2] Group 2: Definition and Characteristics of Algorithmic Economy - Algorithmic economy is defined as an economic model driven by algorithms, focusing on automated decision-making, data analysis, and resource optimization [3] - Companies can create "agents" that utilize various AI models to shift from human management to algorithmic management, increasing the influence of algorithms on resources and decisions [3] Group 3: Principles for Algorithm Governance - Companies should establish a human-centered collaboration model, ensuring that human intelligence complements AI capabilities, particularly in emotional and ethical considerations [3][4] - Maintaining curiosity and innovation is essential for companies to guide AI algorithm iterations and explore new opportunities [4] - Continuous interaction with customers and stakeholders is crucial for refining algorithms, emphasizing transparency in algorithmic logic to prevent unfair practices [4] Group 4: Human-Machine Relationship - The book redefines the human-machine relationship, suggesting that algorithms should empower human decision-making rather than dominate it, akin to the ancient "two-wheeled chariot" analogy [6] - The balance between efficiency and humanity is vital for preserving human agency in the face of advancing technology [6]

Group 1 - The core point of the article is that Elon Musk announced that the recommendation algorithm of the X platform is being replaced by a lightweight version of Grok [1] Group 2 - The transition to Grok's lightweight version indicates a strategic shift in the platform's approach to content recommendation [1] - This change may impact user engagement and content discovery on the X platform [1] - The move reflects ongoing efforts to enhance the platform's functionality and user experience [1]

Core Insights - The article highlights the transformation of the power transmission inspection process in Shanghai through the integration of innovative technologies, particularly drones, which have significantly improved operational efficiency and safety [1][2]. Group 1: Technological Advancements - The introduction of drone technology in 2015 has allowed for various applications in power line inspections, including fire-spraying, laser ranging, and autonomous patrols, leading to zero faults during major power supply protection tasks [2]. - The establishment of the Shanghai Eagle Eye Drone Intelligent Inspection Center has enabled a combination of drone patrols and manual inspections, enhancing the overall reliability of the power grid [2]. Group 2: Operational Efficiency - The digital operation and maintenance monitoring center has been developed to integrate various digital devices, optimizing system processes and achieving online monitoring of all 500 kV and most 220 kV transmission lines in Shanghai [2]. - The monitoring center has successfully provided early warnings and addressed over 10,000 defect hazards in the past three years, resulting in a continuous decline in power line tripping incidents and generating an economic benefit of approximately 340 million yuan [2]. Group 3: Future Innovations - Future innovation plans include the development of flexible materials, power meteorological forecasting, and power data analysis, reflecting a strong commitment to ongoing technological advancement in the industry [3].

Core Viewpoint - MHmarkets aims to enhance forex trading efficiency through innovative technologies, focusing on smart algorithms, real-time data analysis, and user experience optimization. Group 1: Technology Innovation - Smart algorithms optimize trading strategies, improving market adaptability and data analysis accuracy [4][7] - Automation systems and low-latency technology ensure efficient trade execution [5][10] - Artificial intelligence and blockchain drive technological innovation, enhancing market competitiveness [7][15] Group 2: Data Analysis and Market Prediction - Big data analysis allows for more accurate market trend predictions, reducing uncertainty in decision-making [8][9] - Real-time data analysis improves market transparency and optimizes trading decisions [8][9] Group 3: User Experience Enhancement - User interface improvements and security measures enhance overall user experience [6][12] - Simplified trading processes increase operational efficiency for users [12] Group 4: Security Measures - Advanced data encryption techniques ensure user information security during transactions [13] - Account security management focuses on password complexity and two-factor authentication [13][15] - Continuous updates to security protocols help prevent hacking attempts [15] Group 5: Customer Support and Service - Multi-channel customer communication enhances customer experience and satisfaction [17] - 24/7 technical support ensures timely resolution of customer issues [18] - Personalized customer service strengthens loyalty and satisfaction [18] Group 6: Future Directions - Focus on the potential applications of AI and blockchain in forex trading [19] - Strategies for optimizing user experience and expanding into global markets [19][20]

李武军教授为通讯作者，硕士生林昊（已毕业 ，现工作于阿里巴巴）、吴轲、李杰为共同第一作者，博士生李俊为参与作者。 训练成本高昂已经成为大模型和人工智能可持续发展的主要障碍之一。 大模型的训练往往采用多机多卡的分布式训练，大模型的分布式训练挑战巨大，即使硬件足够，不熟悉分布式训练的人大概率（实验中验证有 64%-87% 的概率）会因为超参数设置（模型怎么切分和排布、数据怎么切分和排布等）不合理而无法成功运行训练过程。 此外，不熟悉分布式训练的人在碰到大模型训练慢时容易只想到增加 GPU 硬件等 横向拓展（scale-out）方法，而忽略了分布式训练算法的 纵向拓展（scale- up）作用。 论文被 CVPR 2025 录用为 Oral（所有投稿论文的 0.7%，所有录用论文的 3.3%）。 方法简介 实际上，分布式训练算法会极大地影响硬件的算力利用率。高效能分布式训练算法具有高算力利用率。用同样的硬件算力训练同一个模型，高效能分布式训 练算法会比低效能分布式训练算法速度快，最高可能会快数倍甚至数十倍以上。 也就是说，训练同一个模型，高效能分布式训练算法会比低效能分布式训练算法成本低，最高可能会节省数倍甚至数十 ...