（来源：千龙网） 每逢周五，本版摘选两篇论文。《新京报·书评周刊》在图书评介的基础上拓展了"学术评议和文摘"这 一全新的知识传播工作，筹备"新京报中文学术文摘服务所"，与期刊界一道服务中国人文社会科学事 业。每一期均由本所初选，由相关学科领域的专家学者担任评议人参与推选。我们希望将近期兼具专业 性和前沿性的论文传递给大家，我们还希望所选论文具有鲜明的本土或世界问题意识，具有中文写作独 到的气质。 此为2026年第1期（总第16期）。第一篇论文的作者彭必生讲述了中国古代自动机的文本、技术。在我 们印象中，自动机好像只是现代世界的产物，若是从动力角度来看，限定由电动机或内燃机驱动的自动 装置，确实为现代所独有。但其实有关自动机的想象、印象源远流长，是我们观念世界的一部分，甚至 部分在技术上也有实物。作者回顾了中国古代自动机的智慧及其向理性方向的演进，当然争议也伴随其 中，或关于文本版本的，或关于技术可行性的，直至近代巨变，由自我演进之路转向革新。第二篇论文 的作者施畅考察了数字年代有关面容的风险。面容为我们所有，它经过媒介传播，由他人接收并解读。 而随着AI（人工智能）演进，AI换脸等技术已经逐渐铺开，当人们仍将 ...

Group 1 - In 2025, major asset performance showed significant differentiation driven by three forces: financial cycle downturn, global order restructuring, and deepening industrial revolution. Gold prices surged over 60%, A-shares entered a structural bull market, while the bond market faced fluctuations and real estate prices continued to adjust [1][75]. Group 2 - The macro asset allocation framework for 2026 indicates a complex transition period. The strategic layer suggests maintaining a defensive stance due to the ongoing financial cycle downturn. The tactical layer anticipates a combination of economic recovery and financial easing, providing opportunities in commodities and equities [2][76]. Group 3 - The outlook for 2026 suggests that the stock market may experience a profit-driven rally supported by improved economic fundamentals and ample liquidity. Key focus areas include technology sectors like AI and semiconductors, high-quality export sectors, and renewable energy benefiting from anti-involution policies [3][77]. Group 4 - The bond market is expected to see a wide range of fluctuations with a moderate upward trend in yields, projected to be between 1.6% and 2.1%. The key factors influencing this include a rebound in PPI and alleviation of the "asset shortage" issue, with potential trading opportunities arising from small rate cuts early in the year [3][43][77]. Group 5 - In the commodities market, structural differentiation is expected to continue, with basic metals like copper and aluminum benefiting from global fiscal expansion and liquidity easing. Traditional energy sources like oil may perform relatively poorly due to the financial cycle downturn and supply pressures [3][59][77]. Group 6 - Gold is anticipated to enter a phase of high-level consolidation, with long-term support from weakened dollar credit and central bank gold purchases. However, short-term volatility may arise from price corrections and geopolitical factors [3][68][77].

Core Insights - The video generation sector is experiencing significant participation from various players, including major internet companies and startups, indicating its potential as a high-quality market [1] - The industry is still in the early stages of rapid technological iteration and product exploration, suggesting ongoing innovation and development [1] - Competition within the industry is accelerating progress, enhancing video generation technology to better meet user needs and penetrate more application scenarios [1]

Core Insights - The 2025 Sustainable Global Leaders Conference will be held from October 16 to 18 in Shanghai, focusing on AI development, investment logic, and globalization opportunities for Chinese companies [1][3]. Group 1: AI Development and Investment - AI is currently in the infrastructure investment phase, with investments concentrated in three main areas: models, computing power, and data [3]. - Applications such as intelligent Q&A, AI programming, video generation, and intelligent driving are beginning to emerge [3]. - The transition from application to full penetration of AI across industries will take considerable time, and while there may be industry adjustments, there are no signs of a bubble similar to the 2000 internet bubble [3]. Group 2: Commercialization and Technology - Both technological barriers and commercialization capabilities are equally important for investment, and a flywheel effect must be established [4]. - The flywheel effect in intelligent driving illustrates that increased deployment leads to richer data, which enhances model intelligence, further driving user preference and deployment [4]. Group 3: Market Dynamics and Future Outlook - The era based on traffic has ended, and the current focus is on technology, necessitating a shift in investment logic [5]. - The Hong Kong capital market is becoming increasingly important for technology companies, and mergers and acquisitions will hold greater value in the tech sector compared to the traffic era [5]. - Over the next 10 to 20 years, a new wave of globalization led by Chinese companies will emerge, characterized by leveraging digital and AI infrastructure while emphasizing localization [6]. - China's strong R&D capabilities and the large number of graduates each year will enable Chinese companies to master core technologies and brand identities in overseas markets [6].

Core Insights - The article provides a comprehensive analysis of the intersection of reinforcement learning (RL) and visual intelligence, focusing on the evolution of strategies and key research themes in visual reinforcement learning [5][17][25]. Group 1: Key Themes in Visual Reinforcement Learning - The article categorizes over 200 representative studies into four main pillars: multimodal large language models, visual generation, unified model frameworks, and visual-language-action models [5][17]. - Each pillar is examined for algorithm design, reward engineering, and benchmark progress, highlighting trends and open challenges in the field [5][17][25]. Group 2: Reinforcement Learning Techniques - Various reinforcement learning techniques are discussed, including Proximal Policy Optimization (PPO) and Group Relative Policy Optimization (GRPO), which are used to enhance stability and efficiency in training [15][16]. - The article emphasizes the importance of reward models, such as those based on human feedback and verifiable rewards, in guiding the training of visual reinforcement learning agents [10][12][21]. Group 3: Applications in Visual and Video Reasoning - The article outlines applications of reinforcement learning in visual reasoning tasks, including 2D and 3D perception, image reasoning, and video reasoning, showcasing how these methods improve task performance [18][19][20]. - Specific studies are highlighted that utilize reinforcement learning to enhance capabilities in complex visual tasks, such as object detection and spatial reasoning [18][19][20]. Group 4: Evaluation Metrics and Benchmarks - The article discusses the need for new evaluation metrics tailored to large model visual reinforcement learning, combining traditional metrics with preference-based assessments [31][35]. - It provides an overview of various benchmarks that support training and evaluation in the visual domain, emphasizing the role of human preference data in shaping reward models [40][41]. Group 5: Future Directions and Challenges - The article identifies key challenges in visual reinforcement learning, such as balancing depth and efficiency in reasoning processes, and suggests future research directions to address these issues [43][44]. - It highlights the importance of developing adaptive strategies and hierarchical reinforcement learning approaches to improve the performance of visual-language-action agents [43][44].

Group 1 - The core viewpoint of the report indicates that overseas AI applications have accelerated significantly this year, driven by high demand and rapid growth in large model usage and revenue levels [1] - From the demand side, token consumption continues to grow at a high speed, while large model calls and revenue levels are increasing rapidly [1] - On the supply side, general applications based on LLM models, such as AI search, AI coding, and agents, have seen an initial explosion, alongside continuous iterations of multimodal model capabilities, with image and video generation applications showing potential for breakout success [1] Group 2 - Various vertical applications in marketing, customer service, recruitment, education, healthcare, and legal sectors are emerging continuously [1] - The report maintains a positive outlook on supply chain opportunities benefiting from overseas computing power demand and suggests focusing on domestic cloud and internet companies with AI infrastructure, model capabilities, and application scenarios [1] - Investment opportunities are highlighted in the areas of coding, agents, and the implementation of image/video generation applications [1]

Group 1 - The core viewpoint is that artificial intelligence (AI) technology is experiencing explosive growth and has become a new focus of international competition and economic development [1] - The "AI+" initiative proposed by the Central Economic Work Conference aims to cultivate future industries, with ongoing support in the government work report [1] - The demand for computing power and terminal applications is rapidly increasing, driving performance growth in listed companies within the AI industry chain [1] Group 2 - Major AI computing power companies like Haiguang Information and Inspur Information reported significant net profit growth of 52.87% and 28.55% year-on-year, respectively, with Q1 profits increasing by 75.33% and 52.78% [1] - AI storage company Zhaoyi Innovation saw a staggering net profit increase of 584.2% last year, with a 14.57% rise in Q1 [1] - In the smart wearable sector, Hengxuan Technology's net profit surged by 272.5% last year, with a remarkable 590.22% increase in Q1 [1] Group 3 - The domestic AI industry is in a rapid development phase, showing comprehensive progress in models, computing power, and applications, supported by policy initiatives [1] - Domestic top AI models are now competitive with overseas counterparts, and the supply and demand for AI computing hardware are both strong [1] Group 4 - The Chinese server market's key downstream sectors include internet, communication, and finance, which are expected to drive demand for computing power [2] - AI is becoming a core competitive advantage for major internet companies, leading to increased R&D and AI service demand for computing infrastructure [2] - Domestic AI computing chips are transitioning from usable to highly usable, with downstream clients actively collaborating with local chip manufacturers [2] Group 5 - The domestic AI industry is expected to maintain a rapid growth trend, with domestic large models quickly breaking performance barriers and sustained high demand for computing power [3] - The AI application layer is developing simultaneously, with industry application capabilities positioned in a leading global tier [3] - The future development of the domestic AI industry has vast potential, driven by technological iterations in AI software and hardware systems [3]