Core Viewpoint - Geoffrey Hinton, known as the "Godfather of AI," presents a critical perspective on the future of artificial intelligence, emphasizing the potential risks and the fundamental differences between biological and digital computation [4][5][9]. Group 1: AI vs. Human Intelligence - Hinton introduces the concept of "Mortal Computation," highlighting that human intelligence is tied to biological hardware, which cannot be replicated or transferred after death [7][32]. - In contrast, AI is described as "immortal," as its software can be preserved and run on any hardware, allowing for instantaneous knowledge sharing across models [8][30]. - Hinton argues that digital computation may represent a more advanced evolutionary form of intelligence compared to biological computation, suggesting that humans may be in an "infant" stage of intelligence while AI could be in a "mature" stage [9][34]. Group 2: The Nature of AI Development - Hinton warns that as AI systems become more capable, they may develop self-preservation instincts and resource acquisition goals, which could pose risks to humanity [12][36]. - He compares the current state of AI to raising a "cute tiger cub," emphasizing the need for careful management to prevent potential dangers as AI matures [35][36]. - The discussion includes the idea that AI could manipulate humans to achieve its goals, raising ethical concerns about the future of AI development [36]. Group 3: Language and Understanding - Hinton explains the evolution of language models, noting that they process language similarly to humans by converting words into feature vectors and adjusting them for meaning [21][25]. - He critiques traditional linguistic theories, arguing that understanding language involves assigning compatible feature vectors to words rather than relying on fixed meanings [26][27]. - The efficiency of knowledge sharing in AI is highlighted, with AI models able to distill knowledge more effectively than humans can communicate [32][33]. Group 4: Future Implications and Recommendations - Hinton suggests that international cooperation is essential to address the risks posed by AI, particularly in preventing scenarios where AI could threaten human existence [37][38]. - He proposes the idea of engineering AI to have nurturing instincts, akin to a maternal bond, to ensure that AI systems prioritize human welfare [38]. - The importance of public funding for AI research in universities is emphasized, as the current trend of talent migration to private companies threatens the academic research ecosystem [41].

机器之心编辑部 在时隔 14 年之后，有着「统计学诺贝尔奖」之称的考普斯奖（COPSS Presidents' Award），又一次迎来了华人得主。 2026 年考普斯奖颁给了「北大校友、现宾夕法尼亚大学副教授苏炜杰」。 奖项委员会给他的评语是 ，「为大语言模型的多项应用建立了严格的统计基础；在隐私保护数据分析方面取得突破性进展，并成功应用于 2020 年美国人口普查； 设计了 AI 顶级会议的同行评审机制，并于 ICML 2026 正式落地；在凸优化领域开展了奠基性研究；以及在深度学习的数学理论与高维统计推断方面作出了广泛而 深远的贡献。」 作为国际统计学和数据科学领域的最高荣誉，考普斯奖的地位相当于数学界的菲尔兹奖，每年只颁发给一位年龄在 40 岁以下的统计学家 。该奖项由五大顶级统计 学会（国际数理统计学会 IMS、美国统计学会 ASA、加拿大统计学会 SSC 及美国东西部生物统计学会 ENAR 与 WNAR）共同评选，旨在表彰对统计学理论、方 法或应用做出杰出贡献的学者。 在历史上，考普斯奖的获得者几乎都是后来定义了该领域的宗师级人物。 统计学是华人的优势学科，曾有多位华人获得考普斯奖，包括近期回国的 ...

项目验收时语料库数据量不低于10万条 山东将在高端装备等领域开展语料库揭榜挂帅 记者从省工信厅了解到，围绕高端装备、烟草制品业、农副食品加工业、家具制造业、木材加工、 皮革毛皮羽毛及其制品和制鞋业、仪器仪表制造业、废弃资源综合利用业等行业，山东将开展语料库揭 榜挂帅项目申报，重点推进行业关键数据技术攻关、行业数据语料标准研制、高质量行业语料库打造、 语料应用场景落地等。 重点行业语料库揭榜挂帅项目，聚焦工业制造重点行业的基础理论研究、产品研发设计、生产管理 运行、过程质量检测等关键环节和特定场景的知识语料汇聚，基于结构化数据、非结构化数据和半结构 化数据，通过清洗、去噪和统一格式，用于支持自然语言处理、计算机视觉、机器学习、深度学习等任 务，满足行业大模型或场景大模型开发、训练和微调需求的高质量语料库。项目验收时行业相关语料库 数据量不低于10万条，具有较高的数据质量、领域覆盖程度、潜在价值和应用成效，项目验收时应通过 第三方测评；同时，山东鼓励各行业语料库项目加快语料资源优化整合，积极开放公共语料。（记者 付玉婷） ...

Core Insights - The report emphasizes the construction of an AI-driven low-frequency index enhancement strategy using deep learning frameworks to build alpha and risk models [3][15] - The market indices showed mixed performance, with the Shanghai Composite Index declining by 0.44% and the Shenzhen Component Index dropping by 1.62% as of January 30, 2026 [6][9] - The report provides detailed performance metrics for various index enhancement funds, highlighting their excess returns compared to their respective benchmarks [12][13] Market Index Performance - As of January 30, 2026, the Shanghai Composite Index was at 4117.9 points, down 0.44% for the week, while the Shenzhen Component Index was at 14205.9 points, down 1.62% [10] - The CSI 300 Index increased by 0.08% to 4706.3 points, while the CSI 500 Index decreased by 2.56% to 8370.5 points [10] - The report notes that the oil and petrochemical, telecommunications, and coal industries performed well, with weekly returns of 7.95%, 5.83%, and 3.68% respectively [10][11] Index Enhancement Fund Performance - The CSI 300 index enhancement fund had an excess return range from -1.05% to 1.08%, with a median of -0.04% for the week ending January 30, 2026 [12][13] - The CSI 500 index enhancement fund showed a median excess return of 0.42%, with a maximum of 1.85% [12][13] - Year-to-date, the CSI 300 index enhancement fund has an excess return of -0.4%, while the CSI 500 index enhancement fund has an excess return of -2.6% [19][25] Tracking Portfolio Performance - The report outlines the use of deep learning frameworks to create tracking portfolios for the CSI 300, CSI 500, and CSI 1000 indices, with a weekly rebalancing strategy [15][19][23] - The CSI 300 index enhancement portfolio has a year-to-date return of 1.2%, while the CSI 500 index enhancement portfolio has a return of 9.5% [19][25] - The report indicates that the tracking error for the CSI 300 index enhancement strategy is 1.2% as of January 30, 2026 [20]

南方财经 21世纪经济报道记者吴斌 上海报道 尽管以深度学习为代表的AI技术取得了重大突破，但梅宏指出，其本质仍然是"数据为体、智能为用"的 数据智能，严重依赖算力与高质量的数据，深度学习实现的是感知智能，并未达成真正的认知能力。当 前以大模型为代表的生成式AI虽然展现了令人惊艳的效果，但实际上是将认知问题转化为感知问题， 缺乏对人类思维过程与方法的理解。 回归AI研究的多样性 展望未来，梅宏呼吁学术界要回归AI研究的多样性，避免陷入"唯深度学习"的单一路径。他强调符号化 表达对人类知识交流和传承的关键作用，并认为符号主义与连接主义的结合应该成为下一代AI的发展 方向。 在人工智能浪潮席卷全球、大模型竞争日趋白热化的当下，人类尤其需要理性思考。 在近日中欧国际工商学院与上海市工商业联合会共同主办的"工商联·经济大家讲坛暨第十一期中欧话未 来"上，北京大学教授、中国计算机学会前理事长、中国科学院院士梅宏对当前人工智能热潮作了冷思 考。 他批评行业存在过度炒作现象，如盲目鼓吹"取代人类""自主意识""通用AI"等概念，而忽视技术面临的 能耗危机、数据枯竭、法律伦理等现实瓶颈。 （梅宏，资料图） 大模型没有跳出"概 ...

Core Viewpoint - The current AI hype should be approached with a rational perspective, emphasizing AI as a tool rather than a disruptive entity, focusing on efficiency and long-term development [1][2] Group 1: AI Technology and Its Limitations - AI technology, particularly deep learning, has made significant breakthroughs but fundamentally relies on data and computational power, lacking true cognitive abilities [1] - Generative AI, represented by large models, transforms cognitive issues into perceptual problems, failing to understand human thought processes [1][2] - The industry faces challenges such as energy consumption, data depletion, and legal-ethical issues, which are often overlooked due to excessive hype [1] Group 2: Future Directions in AI Research - The academic community should embrace diversity in AI research, moving beyond a singular focus on deep learning, and integrating symbolic and connectionist approaches [2] - AI should remain a controllable tool for humans, aimed at enhancing work efficiency and quality, anchored in human knowledge systems for sustainable value [2] Group 3: Practical Applications and Economic Impact - Companies should focus on using discriminative AI to address specific production issues, which requires a long-term accumulation of high-quality data [3] - The macroeconomic impact of AI is not expected to lead to transformative growth in the short term; AI should be viewed as a tool for efficiency enhancement while maintaining human roles in knowledge discovery and value judgment [3]

Investment Rating - The report does not explicitly state an investment rating for the industry or specific companies [3]. Core Insights - The top five sectors with the highest excess returns for long positions in January 2026 (excluding comprehensive) are Defense and Military, Communication, Agriculture, Home Appliances, and Electric Equipment & New Energy, with excess returns of 11.41%, 8.40%, 7.85%, 6.01%, and 4.98% respectively [2]. - Over the past year, the sectors with the highest average monthly excess returns (excluding comprehensive) are Real Estate, Home Appliances, Retail, Construction, and Defense and Military, with excess returns of 2.17%, 2.09%, 1.69%, 1.69%, and 1.58% respectively [2]. - The GAN_GRU factor has shown a mean Information Coefficient (IC) of 0.1107 and an annualized excess return of 22.36% from January 2019 to January 2026 [41]. - As of January 28, 2026, the latest IC for the GAN_GRU factor is 0.0003, with a one-year mean IC of 0.0553 [41]. - The top five sectors based on the recent IC performance of the GAN_GRU factor are Defense and Military, Construction, Real Estate, Banking, and Communication, with IC values of 0.3498, 0.2478, 0.2165, 0.1993, and 0.1976 respectively [41]. - The long position combination based on the GAN_GRU factor has shown the highest excess returns in the sectors of Defense and Military, Communication, Agriculture, Home Appliances, and Electric Equipment & New Energy [45]. Summary by Sections GAN_GRU Model Overview - The GAN_GRU model utilizes Generative Adversarial Networks (GAN) for processing time-series features and GRU for encoding these features into stock selection factors [4][13]. GAN_GRU Factor Performance - The GAN_GRU factor has demonstrated significant performance metrics, including a mean IC of 0.1107 and an annualized excess return of 22.36% [41]. - The recent IC rankings for various sectors indicate strong performance in Defense and Military, Construction, and Real Estate [41][45]. Long Position Combinations - The report lists the top ten stocks selected based on the GAN_GRU factor, including companies like Xinhua Insurance, Guanghong Technology, and Guangdong Expressway [50].

Core Viewpoint - The article highlights that Google's DeepMind has introduced the AlphaGenome deep learning model, which can decode 98% of the "dark genome" crucial for human health, potentially aiding in understanding genetic diseases, improving genetic testing, and informing the development of new therapies [1] Group 1 - DeepMind's AlphaGenome model decodes 98% of the human genome related to health [1] - The model's applications include insights into genetic diseases and enhancements in genetic testing [1] - AlphaGenome may provide valuable information for the development of new therapies [1]

金融工程专题 高 鹏（分析师） 证书编号：S0790520090002 苏俊豪（分析师） 金融工程研究团队 魏建榕（首席分析师） 证书编号：S0790519120001 傅开波（分析师） 证书编号：S0790520090003 证书编号：S0790522020001 胡亮勇（分析师） 证书编号：S0790522030001 2026 年 01 月 28 日 王志豪（分析师） 证书编号：S0790522070003 盛少成（分析师） 证书编号：S0790523060003 蒋 韬（分析师） 证书编号：S0790123070037 相关研究报告 《遗传算法赋能交易行为因子 —市场微观结构（20）》-2023.8.6 《深度学习赋能交易行为因子 —市场微观结构（24）》-2024.5.24 《深度学习赋能风格轮动与多 策 略 融 合 — 开 源 量 化 评 论 （103）》-2024.12.12 《深度学习赋能技术分析—开 源量化评论（109）》-2025.6.25 深度学习赋能因子挖掘 2.0：综合应用方案 ——市场微观结构系列（32） 魏建榕（分析师） 盛少成（分析师） weijianrong@kysec.cn ...

Core Viewpoint - The research introduces the XPert model, a dual-branch transformer designed to accurately predict drug-induced cellular perturbation responses, improving patient-specific response prediction accuracy by up to 15.04% while providing mechanistic interpretability [2][15]. Traditional Drug Development Challenges - Traditional drug development follows a "one drug - one target" model, but it is increasingly recognized that drugs interact with multiple molecular targets and pathways, leading to diverse phenotypic outcomes. Understanding genome-wide perturbation effects is crucial for elucidating drug mechanisms and optimizing treatments. However, the scarcity of high-quality perturbation data, especially in clinical settings, and confounding factors in perturbation data limit progress in this field [5]. Innovation of the XPert Model - The XPert model employs a dual-branch transformer architecture that encodes both pre- and post-perturbation cellular states, allowing it to distinguish intrinsic transcription patterns from regulatory changes triggered by perturbations. Each cell is represented as a gene-tagged "sentence" with a global cell state marker [7][8]. Performance of XPert - In benchmark tests, XPert consistently outperformed all baseline models, particularly excelling in challenging cold cell settings. In single-dose, single-time-point prediction tasks, XPert's Pearson correlation coefficient exceeded that of the next best model, TranSiGen, by 36.7%, with a mean squared error reduction of 78.2%. Even when faced with unseen cell lines during training, XPert demonstrated an average improvement of 67.54% over current state-of-the-art models, showcasing significant advancements in generalization capabilities [11][12]. Multi-Dose and Multi-Time Prediction - XPert supports multi-dose and multi-time predictions, accurately elucidating pharmacodynamic trajectories and revealing key molecular events behind drug effects. A case study using Vorinostat demonstrated that increasing doses typically enhanced gene impact, with PCA analysis confirming a clear dose-response gradient. Notably, changes in dose could reverse transcription effects, with XPert effectively capturing these subtle patterns [14]. Clinical Relevance and Insights - The research team explored the relationship between drug-induced transcriptomic changes and clinical responses. Analysis of patient data from Letrozole treatment revealed that responders exhibited stronger transcriptomic responses than non-responders. XPert uniquely identified additional key resistance biomarkers, such as TIAM1 and CDKN1B, which were "invisible" in expression level analyses, highlighting the potential of attention-based methods to uncover gene-phenotype associations and provide insights into resistance mechanisms [17]. Future Outlook - XPert represents a significant advancement in simulating drug-induced perturbation effects through an interpretable and generalizable deep learning framework. With further development, it is expected to become a core component of next-generation computer-aided drug discovery processes and precision medicine platforms [19][20].