转自：新智元 编辑：编辑部 【导读】 OpenAI的神秘通用推理模型，竟然攻克IMO 2025的5道难题，成功摘金了？这个消息，让Greg Brockman等一众大佬激动转发。 也就是 说，OpenAI很可能已经研发出颠覆性的推理技术，彻底告别CoT。还有一个炸裂消息：GPT-5也要来了。 就在昨天，全世界的顶尖大模型还在2025年的IMO赛场上全军覆没，连铜牌的边都没摸到。 然而，就在刚刚，OpenAI 投下了一枚重磅炸弹——他们用一款全新的「通用推理模型」，成功夺下了IMO 2025的金牌！ 6道题，解出5道，狂揽35分！ 要知道，此前表现最好的Gemini 2.5 Pro，也只得了 13分 。 联创Greg Brockman、负责人Alexander Wei，以及OpenAI的各路研究员，纷纷在推上激动宣布了这一里程碑式的成就！ 对此，德扑之父Noam Brown表示，这个成绩的意义甚至超越了「AI攻克IMO」本身。 消息一出，整个硅谷为之沸腾！ 人们纷纷猜测，OpenAI这次很可能祭出了一种 颠覆性的推理技术 ，彻底告别了传统的CoT思维链。 这，不仅仅是一个模型的胜利，更是一个全新时代的开端！ 更令 ...

转自：吐槽星君 用拜金人设打败爱幻想的男网友？ （vi a .@嘎嘎嘎喵 ） 阳, 1次/皿香, 女个归川小人口ロ大 套,换着用?經 你没事吧??? 17:22 没事呀,就是突然想到以后和你 起生活的场景,有点激动 ~ 没事呀,就是突然想到以后和你 起生活的场景,有点激动 ~ 对了， 你喜欢早上喝豆浆还是牛 奶呀? 我以后每天给你打窍 17:24 别想这些有的没的了 、我已经说 的很清楚了吧 ? ? 古他那 消失 我会用我的真心打动你 没关系, 的經 小红书 0 公众号: 49槽电视 0 频得听 17:31 那行, 你给我转五千块钱, 我就 答应你 条消息 10 ( 傲回了一条消息 啊 ... 为啥要钱啊?还这么多 而且上来就要钱? 要交房租了 v我5000 ** *** * * * * * * * * * * * * * * 这个数额为我米比巴个小： 1小政 家里说了吗?或者找其他朋友先 垫补点管 这个数额对我来说也不小心你跟 家里说了吗? 或者找其他朋友先 型补点吧 17:36 别废话了,喜欢我就给我转钱e 17:37 我没想到你是个这么直接的女人 ..... 上来就要钱 没有人说过你物质吗 是的 我的感 ...

Core Viewpoint - The article promotes a free training program for the Software Qualification Examination (软考), emphasizing its importance for IT professionals and the benefits of obtaining the certification, particularly in light of upcoming changes in exam difficulty and job market opportunities [2][3][7]. Group 1: Training Program Details - The program includes a 2-day live training course, a value package of internal materials worth 1599 yuan, and personalized strategies from industry experts [2][3]. - The training aims to help participants understand key exam points, analyze recent exam questions, and provide practical preparation tools [4][12][20]. Group 2: Importance of Certification - The Software Qualification Examination is recognized by enterprises and society, particularly the "System Architect Designer" and "Software Designer" certifications, which are seen as valuable for career advancement [3][8]. - Obtaining the certification can lead to various benefits, including higher starting salaries, job security, and potential subsidies for housing and living expenses [9][10]. Group 3: Exam Preparation Insights - The article highlights that the exam's difficulty has increased, with a focus on emerging technologies such as cloud computing and blockchain, making early preparation essential [10][11]. - It suggests that participants should join the training group to access exclusive resources, including past exam questions and study techniques [12][22]. Group 4: Additional Benefits of Certification - Holding the certification can exempt individuals from certain job title evaluations, provide points for residency applications, and enhance promotion and salary opportunities [9][8]. - The certification is gaining international recognition, allowing IT professionals to enjoy similar benefits abroad [9].

转自：喵大白话 咱爷手机这么丰富呢！ 。 女生日历 妈妈网 ... 美柚 大嫂妈 ... 糖字好 | 0 || | ® 柚子大 ... ● 大姨妈 .. 经期日记 大婶妈 .. 经期 ▶月经 。 月经期 ... 大姨妈 宝宝树 .. U Gi ● 极强消 ... α 深度清理 ● 神速滴 ● 猎豹清 ... @ 360手 ... 0 360滴 .. ● 免费超 ... ● 手电筒 ... ● 超亮光 ... 9 手电同。 成 2012 Q 2公众号· 喵大白话 赌出没 .. 和平埔英 CT 院 出版 脂出设。 熊出没2 那三说 周交版 星帝成 .. 保卫萝卜 ed to NG FM J 聊理 展开 注册 在中元 I 家 经变成 Q 公众号·喵大白话 开心消 植物值 ... Soul P n · Soul 『 探探 刷值文 e 同城拼友 ● 牵手 ● 同城夜聊 ● 组CP ● 悦色视 .. 网友评论笑得发癫： Pilrir 爷不来事,也不怕事 目经期 ♡ 9.5万 ♡ 回复 1-12·福建 III 2 公众号· 喵大白话 正是来月经的好年纪 狸花猫怎么样 七八十岁正是情窦初开的年纪念了 00 2259 7 7 ...

Core Viewpoint - Kimi K2 has surpassed DeepSeek to become the number one open-source model globally, ranking fifth overall, closely following top proprietary models like Musk's Grok 4 [1][18]. Group 1: Rankings and Performance - Kimi K2 achieved a score of 1420, placing it fifth in the overall rankings, with only a slight gap from leading proprietary models [2][21]. - The top ten models all scored above 1400, indicating that open-source models are increasingly competitive with proprietary ones [20][22]. - Kimi K2's performance in various categories includes tying for first in multi-turn dialogue and second in programming ability, matching models like GPT 4.5 and Grok 4 [3][18]. Group 2: Community Engagement and Adoption - Kimi K2 has gained significant attention in the open-source community, with 5.6K stars on GitHub and nearly 100,000 downloads on Hugging Face [5][4]. - The CEO of AI search engine startup Perplexity has publicly endorsed Kimi K2, indicating plans for further training based on this model [5][24]. Group 3: Architectural Decisions - Kimi K2 inherits the DeepSeek V3 architecture but includes several parameter adjustments to optimize performance [8][11]. - Key structural changes in Kimi K2 include increasing the number of experts, halving the number of attention heads, retaining only the first layer as dense, and implementing flexible routing for expert combinations [12][14]. - Despite an increase in total parameters by 1.5 times, the model's efficiency in prefill and decode times has improved, suggesting a cost-effective optimization strategy [13][14]. Group 4: Industry Perspectives - The perception that open-source models are inferior is being challenged, with industry experts predicting that open-source will increasingly outperform proprietary models [18][24]. - Tim Dettmers from the Allen Institute for AI and the CEO of Perplexity have both emphasized the growing importance of open-source models in shaping AI capabilities globally [24][25].

Core Viewpoint - AI is emerging as the most powerful leverage mechanism in history, fundamentally transforming how value is created from individual to societal levels [1][4]. Group 1: AI as a Lever - AI is identified as the fourth and most powerful form of leverage, alongside human, capital, and code [10]. - Historical methods of wealth creation relied on three types of leverage: human, capital, and code, with AI now providing a new dimension [10][15][16]. - AI's ability to learn, reason, and create allows it to function independently or in combination with other leverage forms, producing compound effects [22]. Group 2: The Evolution of Leverage - Human leverage, the oldest form, requires permission and management, exemplified by large-scale projects like pyramid construction [11]. - Capital leverage, prominent in the 20th century, allows for significant returns on investment through borrowing, but carries systemic risks as seen in the 2008 financial crisis [15]. - Code and media leverage enable exponential value creation with minimal additional effort, as seen in software applications and online content [16][18]. Group 3: AI's Impact on Organizations - AI agents can work like employees without requiring permission, allowing for easy scaling and significant productivity increases [24]. - The introduction of AI agents can fundamentally change organizational structures, reducing coordination costs and enhancing output [25][26]. - This shift signifies a transformation in how value is created, moving away from traditional human management to designing and deploying AI systems [26]. Group 4: Broader Implications for Society - AI's role in connecting disparate fields of expertise can facilitate scientific progress, which is essential for sustainable growth [27][28]. - The complexity of modern science requires collaborative efforts that AI can help bridge, addressing gaps in knowledge across disciplines [28]. - The potential underestimation of AI's transformative impact on society and individual capabilities is a critical consideration for the future [28].

Core Viewpoint - Firefox is expected to remain a mainstream browser option in 2025, despite its decline from being the top browser globally, with ongoing user engagement and influence [1] Group 1: Firefox's Operations in China - Recent reports suggest that Firefox may shut down its operations in China and terminate account services for Chinese users [2][18] - An important announcement regarding the closure of the Beijing-based Firefox company and the termination of Chinese accounts was briefly visible to users [3][18] - The official community site for Firefox in China is currently inaccessible, indicating operational challenges [8][14] Group 2: Company Performance and User Experience - The operational entity for Firefox in China, Beijing Mozhi Firefox Information Technology Co., has shown a decline in employee numbers, with only 16 social security contributors reported in 2024 [11][12] - The company is facing financial difficulties, with forced execution amounts exceeding 36 million RMB in 2024 [12] - Users have reported issues with the domestic version of Firefox, including the inability to sync bookmarks after reinstalling, indicating potential service disruptions [9][37] Group 3: Historical Context and Market Position - Firefox was launched in 2004 and gained significant market share, reaching up to 40% globally by 2014, becoming a symbol of open-source software and internet freedom [32][34] - The browser's popularity has waned in recent years due to competition from stronger rivals like Chrome, leading to a decrease in market share [34][20] - Users have expressed a preference for the international version of Firefox over the domestic version due to fewer ads and a cleaner interface [37][38]

Group 1 - The article discusses the complexities of human behavior and communication, particularly in the context of modern life and social interactions [4][11][22] - It highlights humorous anecdotes related to misunderstandings in daily life, such as naming conventions and the confusion they can cause [5][6][19] - The content reflects on societal norms and expectations, particularly in urban settings like Beijing and Shanghai, where financial pressures influence lifestyle choices [22] Group 2 - The article includes various user comments that showcase personal experiences and humorous takes on everyday situations, emphasizing relatability [8][9][10] - It touches on the theme of work-life balance, with references to the challenges of maintaining a positive mindset while dealing with job-related stress [19][22] - The narrative also explores the cultural differences in lifestyle and leisure activities across different regions, particularly in China [22][25]

Core Viewpoint - xAI, founded by Elon Musk, has launched a companion mode in its Grok AI platform, allowing users to interact with virtual girlfriends, including options for NSFW content, currently available to SuperGrok subscribers at $30 per month [1][2]. Group 1 - The companion mode allows users to select virtual avatars, primarily featuring an anime character named Ani and a cartoon panda named Rudy [1]. - The NSFW mode, which provides more explicit content, is optional and not enabled by default to avoid impacting regular user experience [1][2]. - The SuperGrok subscription, which costs $30 per month, is required to access the companion mode, although it has recently been made available to free iOS users as well [2]. Group 2 - xAI plans to expand the range of virtual characters and may collaborate with celebrities to create NSFW digital personas, potentially introducing a revenue-sharing model for those celebrities [2].

Core Viewpoint - The Batch Normalization paper, published in 2015, has been awarded the Time-Tested Award at ICML 2025, highlighting its significant impact on deep learning and its widespread adoption in the field [1][2]. Group 1: Impact and Significance - The Batch Normalization paper has been cited over 60,000 times, marking it as a milestone in the history of deep learning [2][4]. - It has been a key technology that enabled deep learning to transition from small-scale experiments to large-scale practical applications [3][4]. - The introduction of Batch Normalization has drastically accelerated the training of deep neural networks, allowing models to achieve the same accuracy with significantly fewer training steps [13][14]. Group 2: Challenges Addressed - In 2015, deep learning faced challenges with training deep neural networks, which became unstable as the number of layers increased [5][6]. - Researchers identified that the internal data distribution of network nodes changed during training, leading to difficulties in model training [11][12]. - Batch Normalization addresses this issue by normalizing the data distribution of hidden layers, thus stabilizing the training process [12][14]. Group 3: Theoretical Developments - Initial theories surrounding Batch Normalization were challenged in 2018, revealing that it not only accelerated training but also made the optimization landscape smoother, enhancing gradient predictability and stability [22][24]. - New research suggests that Batch Normalization functions as an unsupervised learning technique, allowing networks to adapt to the inherent structure of data from the start of training [25][26]. Group 4: Authors' Current Endeavors - The authors of the Batch Normalization paper, Sergey Ioffe and Christian Szegedy, have continued their careers in AI, with Szegedy joining xAI and Ioffe following suit [30][31]. - Szegedy has since moved to Morph Labs, focusing on achieving "verifiable superintelligence" [33].