编者按：人工智能模型的可靠性取决于其底层数据的质量。然而，目前有一个新的悖论，模型越智能，数据可能变得越糟糕。本文来自编译，希望对您有所 启发。 神译局是36氪旗下编译团队，关注科技、商业、职场、生活等领域，重点介绍国外的新技术、新观点、新风向。 人工智能承诺带来更智能、更快速、更高效的未来，但乐观表象之下潜藏着日益恶化的隐忧：数据本身的问题。我们常讨论算法，却鲜少关注支撑算法的基 础设施。事实上，创新速度永远无法超越其输入数据的质量，而当前这些输入数据正显露疲态。当根基开始动摇，即便是最先进的系统也会失灵。 十年前，规模与精度尚能并行不悖。但如今，这两个目标往往背道而驰。隐私法规、设备授权限制及平台新规，使得获取高质量的第一方数据比以往任何时 候都更困难。为填补缺口，市场充斥着看似合法实则虚假的循环利用、伪造或推断信号。 [图片来源：Eugene Mymrin/Getty Images] 1. 当数据量沦为噪音 多年来，行业普遍认为数据越多洞察越精深。数据量象征着实力，输入越多意味着智能越强。但如今数据过剩已沦为干扰噪音。为维持规模，部分供应商采 用了填充数据或虚假信号，使系统看似健康，实则侵蚀了数据的可靠 ...

什么构成不良数据？治理框架的重点领域是什么？以及如何在大型组织中驾驭人员、流程和技术方面的 细微差别？ 一 什么构成"坏数据" 脏数据是指不完整、不准确、过时或重复的信息，它会对组织造成严重破坏。这是一个代价高昂的问 题，它会滋生不信任、浪费资源并损害决策。尽管数据质量至关重要，但它却常常被忽视，从而导致严 重的业务中断和机会损失。 1.脏数据造成的误判影响 数据质量差的后果十分严重且影响深远。研究表明，数据质量差每年给企业造成数百万美元的损失。销 售团队浪费时间和金钱去追踪无效线索，财务部门在报告中出现错误，营销活动也因为目标受众错误而 效果不佳。 更令人担忧的是，基于错误数据做出的决策可能会使整个公司偏离正轨，导致错失良机、资源错配和战 略失误。 例如，医疗机构如果患者记录不准确，就会造成严重后果。由于数据过时或不匹配而导致的错误诊断， 不仅会危及患者安全，还可能引发法律责任。在金融等行业，数据驱动的风险评估指导着数十亿美元的 投资，因此容错空间更小。 2.错误数据极其普遍 尽管风险显而易见，但脏数据问题几乎在每个组织中都持续存在，其根本原因通常是数据治理实践不 善、系统孤立以及缺乏责任感。 太多公司 把 ...

根据美国顶级统计专家的一份新报告，美国数据机构急需特朗普政府和国会的帮助，以确保在危机日益 加深之际履行基本职责并恢复公众信心。 由美国统计协会（ASA）牵头的一项研究指出，这些机构正受困于能力脆弱和信任受损， 并需要更多 资金和人员。报告列举了自去年的初版报告发布以来变得更加严峻的挑战，当时特朗普总统尚未重返白 宫。 就在这场闹剧上演的前一天，这项研究背后的统计专家曾发布一份中期报告，称他们确信数据是可以信 任的，没有迹象表明行政部门进行了干预。特朗普针对劳工统计局的举动迫使他们迅速重新思考。该文 件被修改为：总统的行为"通过指责统计机构负责人过去存在政治操纵行为，破坏了对未来的信任"。 劳工统计局（BLS）、经济分析局（BEA）和普查局等政府部门负责发布涵盖经济及许多其他主题的各 类数据，这些数据对政策制定者、投资者、公司以及广大公众的决策至关重要。长期存在的问题如预算 缩减和调查回复率下降，以及近期对其独立性和完整性的威胁，使得它们的工作变得更加艰难。 该组织的新报告引用了一项调查，发现公众对联邦数据的信任度从6月的57%下降到了9月的52%。 这份于周三发布的报告称，"必须立即采取行动，以阻止联邦统计 ...

在 Meta 豪掷 143 亿美元入股竞争对手 Scale AI 时，这家由谷歌前工程师创立、员工仅为对手十分之一的公司，已悄然实现了年营收超 10 亿美元的业 绩，且从未接受外部投资。 AI 竞技场上，聚光灯总在追逐着 OpenAI、Google 等发布下一个万亿参数模型的明星。而决定模型"思维"与"品格"的训练数据，则像被遗忘的地基。 硅谷正上演一幕对比鲜明的戏剧：一边是 Meta 豪掷 143 亿美元收购数据标注公司 Scale AI 近半股份，使其创始人亚历山大·王成为硅谷红人。 另一边，是其低调的对手 Surge AI：成立近五年没有任何融资、过去两年几乎不发新闻稿、员工仅为对手十分之一，却悄悄实现了超过 10 亿美元的营 收，在财务上已超越获得巨资的 Scale AI。 这次故事的主角轮到了 Edwin Chen。 Surge AI 的创始人兼 CEO Edwin Chen 是一位美籍华裔，曾在 Massachusetts Institute of Technology（MIT）学习数学与语言学。毕业后，Edwin 踏入职场 —— 他曾在包括 Google、 Meta Platforms（前身 F ...

Core Insights - The automotive industry is experiencing a trend of functional homogenization due to the widespread adoption of large model capabilities, shifting competition from "model scale" to "data quality, scene granularity, and depth adaptation" [1] - Companies need to build an integrated closed-loop capability of "scene-data-model" to achieve "model as application," creating differentiated experiences in real-world usage scenarios [1] Industry Trends - There is an increasing willingness among users to pay for comfort hardware such as smart seats and smart audio systems, with experience depth becoming a key value anchor for future smart cockpits [1] - The interaction paradigm of smart vehicles is transitioning from "passive response tools" to "proactive cognitive partners," moving beyond user-triggered commands to proactively predicting and providing services based on integrated sensor data, user behavior, and contextual needs [1] Future Directions - By 2025, smart cockpits are expected to shift from "digital redundancy" to "pragmatism," with a rational transformation in smart cockpit interactions, leading to a "rebalancing" of touch controls and physical buttons [1] - The competitive focus of future smart cockpits will return from "breadth of functionality" to "depth of experience" [1]

Core Insights - The rise of AI is fundamentally transforming the e-commerce landscape, shifting traffic from traditional search engines to AI-driven platforms like ChatGPT and Doubao [2][3][4] - Major e-commerce players are adapting to this trend, with companies like OpenAI and Shopify collaborating to integrate shopping experiences directly into conversational interfaces [3][4] - In China, Doubao, supported by ByteDance, is creating a closed-loop e-commerce ecosystem with Douyin, indicating a significant shift in consumer shopping behavior [5][6] E-commerce Traffic Shift - AI is changing the way consumers shop, moving from traditional search methods to conversational interactions, as highlighted by Shopify's partnership with OpenAI [3][4] - ChatGPT has surpassed 700 million monthly active users, becoming a new entry point for e-commerce, competing with established platforms like Google and Amazon [3][4] - Traditional search engines are experiencing a decline in usage, while AI models are gaining traction, which could disrupt the advertising revenue model of companies like Google and Baidu [4][9] AI Integration in E-commerce - Doubao allows users to ask for product recommendations directly within the app, linking to Douyin's shopping platform, thus streamlining the shopping process [5][6] - Alibaba's recent launch of the Qianwen app aims to compete in the AI-driven e-commerce space, although it has yet to achieve significant user engagement compared to Doubao [6][10] - The integration of AI in e-commerce is expected to enhance user experience by providing personalized recommendations based on historical behavior [9][12] Future of Search Engines - Experts predict that traditional search engines may gradually decline as AI-driven conversational interfaces become the norm for information retrieval and shopping [7][9] - Companies like Baidu are developing AI products to adapt to this shift, with their Wenxin assistant showing significant user growth [10][11] Data Quality and Trust Issues - The effectiveness of AI in e-commerce is heavily dependent on the quality of data used, with concerns about the reliability of information generated by AI models [13][16] - The challenge of misinformation and the need for accurate data integration are critical for the successful implementation of AI in commercial settings [14][17]

Core Insights - The article discusses the launch of the Bee project by Tsinghua University and Tencent's Mixuan team, aimed at bridging the performance gap between fully open-source multimodal large language models (MLLMs) and their closed or semi-open counterparts [2][5][26]. Group 1: Background and Motivation - The current MLLM landscape exhibits a three-tier structure: (1) top-tier closed-source models (e.g., Gemini 2.5, GPT-5), (2) semi-open models with private data (e.g., Qwen2.5-VL), and (3) significantly underperforming fully open-source models [5]. - The core bottleneck is identified as the "data quality gap" rather than model architecture [2][10]. Group 2: Key Contributions of the Bee Project - **Honey-Data-15M**: A high-quality SFT dataset comprising 15 million samples, enhanced through a dual-layer Chain of Thought (CoT) approach [6][16]. - **HoneyPipe & DataStudio**: An open-source, end-to-end data enhancement pipeline that provides a transparent and reproducible methodology for data cleaning and CoT augmentation [6][12]. - **Bee-8B**: A new 8 billion parameter model trained on Honey-Data-15M, achieving state-of-the-art (SOTA) results in various benchmarks, rivaling or surpassing mainstream semi-open models [6][21][26]. Group 3: Data Quality Issues - Existing open-source datasets suffer from two main issues: pervasive noise (e.g., factual inaccuracies, mismatched images) and a lack of complex reasoning data [11][14]. - The Bee project emphasizes that the most viable path for the open-source community is to focus on "data quality" rather than merely increasing "data quantity" [11][26]. Group 4: HoneyPipe Process - The HoneyPipe process involves a meticulous "filter-enhance-validate" workflow that produces high-quality datasets [15][18]. - The process includes three stages: noise and irrelevance filtering, short CoT enhancement and validation, and long CoT enhancement for complex queries [18]. Group 5: Performance of Bee-8B - Bee-8B demonstrates superior performance across various benchmarks, including MathVerse and LogicVista, where it achieved scores of 67.0 and 61.3, respectively, outperforming semi-open models [28]. - In general VQA tasks, Bee-8B achieved excellent SOTA scores in multiple benchmarks, including MMStar and CountBench [28]. Group 6: Conclusion - The Bee project effectively addresses the core data quality issues hindering the development of fully open-source MLLMs, advocating for a methodology that prioritizes data quality over sheer volume [26].

Core Viewpoint - The article discusses a study indicating that large language models (LLMs) can experience cognitive decline, referred to as "brain rot," due to prolonged exposure to low-quality internet content, similar to the effects observed in humans [4][10][12]. Group 1: Research Findings - The study conducted by Texas A&M University, the University of Texas at Austin, and Purdue University demonstrates that LLMs can suffer from cognitive degradation when trained on viral Twitter data characterized by short, engaging posts [4][6]. - Cognitive functions such as reasoning and long-term memory showed significant declines, with reasoning ability decreasing by 23% and long-term memory by 30% after exposure to low-quality data [14][15]. - The research established a "brain rot hypothesis," suggesting that continuous exposure to poor-quality text leads to a sustained decline in cognitive abilities of LLMs [12][29]. Group 2: Experimental Methodology - Researchers utilized a controlled experiment on real Twitter data, creating datasets based on engagement (M1) and semantic quality (M2) to assess the impact of low-quality content on LLMs [13][20]. - M1 focused on the popularity and brevity of posts, while M2 evaluated the sensationalism or superficiality of the content, with both methods indicating a negative correlation between data quality and cognitive performance [13][22]. Group 3: Implications and Recommendations - The findings highlight the necessity for regular "cognitive health checks" for deployed LLMs, emphasizing the importance of data quality in maintaining their cognitive capabilities [17][29]. - The study suggests that the effects of exposure to low-quality data are not easily mitigated through standard fine-tuning techniques, indicating a need for improved data curation practices [29].

Core Insights - The article emphasizes the importance of bridging the information gap between individual investors and professional institutions, a mission that Morningstar has pursued since its founding in 1984 [2]. Group 1: Data Quality and Investment Solutions - Morningstar has built one of the largest and highest-quality investment databases globally, covering approximately 800,000 investment products, with a strong focus on rigorous data quality checks [3]. - The company connects disparate data sources by acquiring firms like PitchBook, enabling a comprehensive view of both public and private markets [3]. - Morningstar offers unique analytical tools, such as medal ratings and sustainability ratings, to facilitate quicker decision-making from vast amounts of investment data [3]. Group 2: Services for Asset Managers and Institutional Investors - Morningstar provides professional research support for public funds and bank wealth management, offering independent and objective evaluation systems for investment decisions [10]. - The company assists institutions in constructing robust investment portfolios that align with long-term goals through macro and strategic research support [12]. - Insights into global product innovation trends are shared to help institutions develop forward-looking financial products that meet investor needs [13]. Group 3: Commitment to Client Interests - Morningstar aims to empower investment advisors to better serve their clients, believing that those who genuinely represent client interests will ultimately achieve market returns [7]. - The company maintains an independent viewpoint and cautious approach in its research, providing a foundation for investment advisors to guide clients towards long-term strategies amidst market noise [6]. Group 4: Company Overview - Morningstar, Inc. is a leading investment research firm with operations across North America, Europe, Australia, and Asia, providing financial information, fund, and stock analysis to various professional investors [17]. - As of June 30, 2025, Morningstar manages and advises on assets totaling approximately $352 billion across 33 global markets [20].

Core Insights - The ongoing government shutdown is complicating the Federal Reserve's ability to make informed decisions due to a lack of critical labor market and inflation data [1][3][10] - The Fed may need to pause its decision-making process until more reliable data becomes available, which could lead to increased market volatility [3][9][10] Labor Market Analysis - There are signs of a slowdown in the labor market, but the Fed is uncertain whether this trend is stabilizing due to missing data [2][4] - Alternative data sets are available, but they do not fully compensate for the absence of official data, leading to uncertainty in estimating payroll changes [10][11] Economic Impact - The government shutdown is estimated to create a quarterly GDP drag of approximately 10 basis points per week, equating to about $400 million per day in lost federal worker compensation [6] - The integrity of economic data, such as the Consumer Price Index (CPI), is being compromised, with a significant increase in the percentage of estimated goods in the CPI basket [4][5] Market Reactions - If the Fed decides to hold off on rate cuts due to the lack of data, it could negatively impact market sentiment, as investors have been pricing in expected cuts [8][9] - Despite the uncertainty, underlying earnings remain strong, with growth estimates for Q3 rising from 8% to 8.8%, indicating resilience in the market [14][16]