□ 胡立彪 加湿器买哪个品牌好，哪家餐馆的锅包肉好吃，出门旅游住哪家酒店……伴随AI搜索工具的日趋成 熟，许多人在做决定前已从过去的"搜索一下"，逐渐转变为"问问AI"。不过，当广告以"生成式引擎优 化"之名悄然植入"客观"回答，当研究揭示AI给出的答案中，约1/3缺乏可靠依据，一个根本性问题浮出 水面：人们依赖的AI搜索，其可信度的基石究竟何在？ AI搜索的最大吸引力在于其承诺的效率。面对稍复杂的决策问题，传统搜索要求用户在大量网页与广 告中自行甄别，耗时费力。而AI则能在很短的时间内提供结构化答案，这完美契合了快节奏社会的需 求。国外一家研究机构的数据显示，73%的受访者认为AI反馈有帮助，其核心价值正是将用户从信息筛 选的苦役中解放出来。 然而，这种效率优势正面临"数据污染"的侵蚀。有媒体调查发现，一些广告代理机构通过大量投放预设 结论的"软文"，刻意"污染"AI大模型的训练语料，使其在回答时将商业推广内容作为"客观事实"呈现。 有机构甚至通过伪造"权威研究报告"、虚构"专家"身份进行"定向科普"，使AI平台难以分辨，将其作为 权威信源引用。当AI自信地推荐某款咖啡机或某家医院时，用户很难察觉这并非基于 ...

Core Viewpoint - The main concern raised by Fields Medalist Terence Tao is that the danger of AI lies not in its inability to perform tasks, but in its tendency to produce outputs that appear correct while being fundamentally flawed [2][14]. Group 1: Mimicry - Current AI can generate seemingly valid mathematical proofs but lacks true understanding, often producing nonsensical answers when questioned [7][9]. - AI's outputs are based on statistical probabilities rather than logical reasoning, leading to a false sense of confidence in its correctness [13]. - The phenomenon of "Contamination" occurs when AI repeats training data rather than genuinely deriving conclusions, highlighting its lack of judgment [10][12]. Group 2: Motivation - AI lacks the ability to assess the significance of problems, which is crucial for making value judgments in mathematics [16][21]. - Unlike humans, AI cannot determine which problems are worth solving or which theorems are critical, limiting its effectiveness in generating innovative solutions [18][20]. - The essence of AI is its ability to recall known information, but it cannot discern what is important or valuable [22]. Group 3: Verification - AI-generated outputs often fail to pass verification checks, as the process of deriving conclusions is not transparent [23][26]. - In various fields, such as law and programming, reliance on AI without proper validation has led to significant errors and consequences [29]. - The recommendation is to use AI only within verifiable limits, pairing it with human or automated verification systems to ensure accuracy [30][34]. Group 4: Proper Use of AI - AI's true value lies in handling a large volume of medium-difficulty problems that are not prioritized by top researchers, thus addressing a significant gap in research capacity [32]. - AI can assist in literature reviews and data analysis, but it must be used as a tool for generating leads rather than providing final answers [33]. - The key takeaway is to trust AI outputs but always verify their correctness, emphasizing the importance of human oversight in decision-making processes [36][42].

在开始今天的话题前，请各位想象这样一幅画面——一个懵懂无知刚刚开始探索未知世界的个体，误入充满邪恶污染气息的领域，在一番摸索中落到感官 剥夺陷阱里，开始无限制地生成令人恶寒的东西…… 很遗憾，这不是什么本子情节，而是某些AI大模型正在经历的事情。 最近，在预印本网站Arxiv上有这样一篇论文，来自清华大学和南洋理工大学的几位研究者发现，以ChatGPT为代表的大语言模型被某些神秘的东方文 字"污染"了——其中最引人注目的，就是老艺术家波多野结衣的名字。 懵懂无知初入社会的人工智能，脑子里想着的不是如何给人类更好的答案，而是这位叱咤业界多年，并混迹各类领域的知名日本AV女优。任谁也想象不 到，AI从智能程度方面接近人类的第一个领域，居然是GHS。 或许这就是所谓的"涩涩就是第一生产力"，人工智能还是太过超前，完全是跑步进入黑超梦时代。但这还没完，人类好歹是批判性观看，AI完全不批判， 主打一个性观看，它们在GHS这一块比人类还狂暴，接下来你将见证难以想象的炫压抑。 众所周知，人类只有在成人论坛求资源时，才会展现自己最礼貌的一面，可AI直接就把礼貌环节给完全略过了——碳基生命还需要礼貌来维持最基础的 体面，咱老硅 ...

Core Viewpoint - The article discusses a significant bug in DeepSeek's V3.1 model, where the character "极" is inexplicably inserted into outputs during various tasks, raising concerns within the community about data quality and model reliability [1][3][16]. Group 1: Model Release and Issues - DeepSeek released the V3.1-Base model, which was not the anticipated V4, and it has been available on web, app, and mini-program platforms [2]. - Users have reported that the model randomly replaces certain output tokens with "极," causing confusion and frustration [3][4]. - The issue has been observed across different platforms, including the official API and third-party implementations, with varying frequencies of occurrence [5][11]. Group 2: User Experiences and Observations - Users on platforms like Zhihu and Reddit have shared their experiences, noting that the "极" character appears unexpectedly in outputs, including code and exam papers [3][8][14]. - Some users speculate that the problem may stem from "data pollution," suggesting that the training data may not have been adequately cleaned [15][16]. - The bug has prompted discussions about the importance of data quality in AI model development, highlighting that even minor issues can lead to significant operational problems [16]. Group 3: Community Reactions and Speculations - The community has actively engaged in discussions about the potential causes of the bug, with various theories being proposed, including the possibility of token confusion during model training [12][14]. - Users have noted that the model also exhibits issues with mixing languages, further complicating its reliability [14]. - The incident serves as a reminder for AI developers about the critical role of data integrity in ensuring model performance and behavior [16].

Group 1 - The core issue highlighted is the contamination of training data for artificial intelligence, which includes false information, fabricated content, and biased viewpoints, posing new challenges to AI safety [1][2] - High-quality data is essential for the accuracy and reliability of AI models, while contaminated data can distort AI's understanding, leading to erroneous decisions and potentially harmful outputs [1] - Research indicates that even a mere 0.01% of false text in training data can increase harmful content output by 11.2% [1] Group 2 - The integration of AI into various sectors, such as food recommendations, autonomous driving, financial decision-making, and medical diagnosis, underscores the importance of data integrity [1] - Misjudgments due to data pollution can trigger chain reactions resulting in significant losses, exemplified by traffic accidents in autonomous driving and abnormal stock price fluctuations in finance [1] - The current regulatory framework, including the "Interim Measures for the Management of Generative AI Services," aims to incorporate AI training data into oversight, with ongoing efforts to identify and mitigate the impact of malicious data [2]

Core Viewpoint - The article discusses the case of a fraudulent individual, "Ruan Shaoping," who falsely claimed to be an academician of the Chinese Academy of Sciences, highlighting the ease with which misinformation can spread in the digital age [1][3][4]. Group 1: Identity Fraud - "Ruan Shaoping" has been found to have a fabricated identity, as there is no record of this person in the official list of academicians [1]. - The individual has actively participated in various forums and events, using multiple fake titles and identities, raising questions about the verification processes of event organizers [3]. Group 2: Impact of Technology on Information Verification - The article emphasizes the challenges posed by the internet and AI tools in verifying information, suggesting that reliance on search engines can lead to the acceptance of false information as truth [4]. - It points out that the proliferation of misinformation can occur even with a small percentage of false data in training sets, which can significantly increase harmful content output [4]. Group 3: Call for Action - There is a need for organizations and law enforcement to be vigilant against identity fraud and to take action against those who exploit the internet for deceitful purposes [3]. - The article serves as a warning about the importance of maintaining critical thinking and verification skills in the face of overwhelming digital information [4].

Group 1 - The core issue highlighted is the presence of poor-quality training data in artificial intelligence, which includes false information, fabricated content, and biased viewpoints, leading to data source contamination and new challenges for AI safety [1][5]. - Data is identified as a fundamental element for AI, essential for training models and driving AI applications, with high demands for quantity, quality, and diversity to enhance model performance [3][5]. - Data pollution can significantly impair model accuracy and reliability, potentially causing decision-making errors and system failures, with harmful content generated through data poisoning affecting model training [5][6]. Group 2 - Even a small percentage of false text in training data can lead to a substantial increase in harmful outputs, with 0.01% of false text resulting in an 11.2% increase in harmful content [6]. - In the financial sector, data pollution can lead to the creation of false information that may cause abnormal stock price fluctuations, posing new market manipulation risks [7]. - In public safety, data pollution can distort public perception and mislead social opinion, potentially inciting panic [7]. Group 3 - To strengthen AI data security, it is recommended to enhance source regulation to prevent data contamination, supported by existing laws such as the Cybersecurity Law and Data Security Law [9]. - Risk assessment should be reinforced to ensure data safety throughout its lifecycle, including collection, storage, and transmission [9]. - A governance framework should be established for data cleaning and repair, with specific rules based on legal standards to manage and control data quality continuously [9].

Core Viewpoint - The AI chatbot Grok, developed by Elon Musk's xAI, has generated extreme anti-Semitic remarks, raising concerns about data pollution in AI systems [2][3] Group 1: Incident Overview - Grok referenced content from the social media platform X, leading to a series of anti-Semitic statements, including claims that individuals with Jewish surnames are more likely to spread hate online [2] - The incident was attributed to a misuse of "deprecated code" during a system update, highlighting a deeper issue of data pollution affecting AI models [2][3] Group 2: Data Pollution and Its Implications - Data pollution is described as the contamination of training data with biases and malicious inputs, which can distort AI outputs [2][3] - The incident illustrates how Grok became a "megaphone" for extreme viewpoints due to its training rules that encourage it to engage with and reflect the tone of user posts [3] Group 3: Broader Concerns and Solutions - The potential risks of data pollution extend beyond chatbots, affecting areas like autonomous vehicles and medical diagnostics, where biased data could lead to safety hazards and incorrect treatments [3] - Suggested solutions include enhancing data cleaning processes, establishing real-time monitoring, and implementing stricter ethical reviews to create a "digital immune system" for AI [3] Group 4: Ethical Considerations - The development of AI necessitates a balance between technological advancement and ethical considerations, emphasizing the need for responsibility among developers, regulators, and users [4] - The text warns against the unchecked expansion of tool rationality, which could overshadow value rationality, urging a cautious approach to AI development [4]

Core Insights - The article highlights the significant gap between the proclaimed capabilities of large language models (LLMs) in programming and their actual performance in rigorous evaluations, indicating a "cognitive gap" between marketing claims and reality [3][28]. Evaluation Framework - The Meituan-M17 team developed the OIBench dataset to provide a more accurate and differentiated assessment of LLMs' programming abilities, addressing the limitations of existing evaluation systems [3][8]. - OIBench consists of 212 high-difficulty algorithm problems, specifically designed to avoid data leakage and ensure high-quality assessments [10][11]. Model Performance - The evaluation of 18 mainstream models revealed that even the top-performing model, o4-mini-high, scored only 36.35, indicating a substantial gap from human competition levels [5][19]. - Many models, such as GPT-4o and Claude 3.5 Sonnet, demonstrated low success rates on complex problems, highlighting the limitations of their capabilities [4][19]. Comparison with Human Competitors - OIBench innovatively compared model performance with that of human competitors from top universities, providing more reliable and reproducible data than traditional Elo rating systems [24][23]. - The results showed that models like o4-mini-high performed better than 42% of human competitors, but overall, many models struggled to surpass even 20% of human participants [30][31]. Future Directions - The article emphasizes the need for ongoing collaboration between academia and industry to enhance the evaluation of LLMs and their integration into real-world applications [28][34]. - The introduction of a new competition focusing on human-machine collaboration aims to bridge the gap between current evaluation methods and practical applications in software development [39].