但如果把问题继续往深处追问：推理能力的本质，真的只是多算几步吗？ 谷歌、芝加哥大学等机构的研究者最近发表的一篇论文给出了一个更具结构性的答案，推理能力的提升并非仅源于计算步数的增加，而是来自模型在推理 过程中隐式模拟了一种复杂的、类多智能体的交互结构，他们称之为「思维社会」（society of thought）。 过去两年，大模型的推理能力出现了一次明显的跃迁。在数学、逻辑、多步规划等复杂任务上，推理模型如 OpenAI 的 o 系列、DeepSeek-R1、QwQ- 32B，开始稳定拉开与传统指令微调模型的差距。直观来看，它们似乎只是思考得更久了：更长的 Chain-of-Thought、更高的 test-time compute，成为最常 被引用的解释。 简单理解就是，这项研究发现，为了解决难题，推理模型有时会模拟不同角色之间的内部对话，就像他们数字大脑中的辩论队一样。他们争论、纠正对 方、表达惊讶，并调和不同观点以达成正确答案。人类智能很可能是因为社交互动而进化的，而类似的直觉似乎也适用于人工智能！ 通过对推理输出进行分类，以及结合作用于推理轨迹的机制可解释性方法，研究发现，诸如 DeepSeek-R ...

美股大型科技股盘前涨跌互现，Meta涨0.4%，微软涨0.2%，苹果涨0.1%，谷歌A持平，亚马逊跌 0.2%，英伟达跌0.3%，特斯拉跌0.6%。 ...

特别提示 微信机制调整，点击顶部"仪器信息网" → 右上方"…" → 设为 ★ 星标，否则很可能无法看到我 们的推送。 近 日 ， " 2 0 2 6 年 全 球 品 牌 价 值 5 0 0 强 " 榜 单 (Gl o b a l 5 0 0 2 0 2 6 ) 由 英 国 品 牌 评 估 机 构 " 品 牌 金 融"（Br a n d Fi n a n c e） 在 瑞士达沃斯世界经济论坛 上发布，苹果、微软、谷歌、亚马逊位居 前四，英伟达升至第五名。 每年 ， Br a n d Fi n a n c e会对全球规模前5 0 0 0的品牌进行估值 ， 并发布1 0 0多份报告 ， 对 各 行业和各国家 、 地区的品牌进行排名 。 品牌价值被理解为品牌所有者通过在公开市场上许可 该品牌所获得的净经济收益 。 值得一提的是，继2 0 2 3年上榜（No . 4 6 9）之后，时隔两年， 赛默飞在今年重新以排名4 5 4的 成绩回到榜单之上，同时也成为榜上唯一以科学仪器为主营业务的品牌 。 此外， 美的、海尔、 罗氏、日立、 西门子、通用电气（GE）、飞利浦、雅培 这些涉及仪器 业务的知名品牌也在5 0 0强榜单 ...

2026年12月10日，新德里国际会展中心人头攒动。微软CEO萨提亚·纳德拉站在聚光灯下，宣布将向印度投资175亿美元建设AI基础设 施。 就在同一天，亚马逊也承诺，将在印度投入350亿美元。 这是两个截然不同的印度——一个是硅谷眼中的"全球最大数字市场"，另一个是仍在为基本民生挣扎的发展中国家。当这两个印度相 遇，会发生什么？ 短短几个月内，谷歌、Meta、OpenAI、Anthropic等硅谷巨头纷纷加入战局。总计675亿美元的承诺投资，将在未来五年内涌入这个拥有 14亿人口的国家。 | | Investment Size | Time | Data Center | Data Center Capacity | | --- | --- | --- | --- | --- | | | | Horizon | Location | | | Google | $15B | Through | Visakhapatnam | gigawatt-scale | | | | 2030 | | | | Microsoft | $17.5B | Through | Hyderabad | | | | | 2029 | ...

Core Insights - The article discusses the significant leap in reasoning capabilities of large models over the past two years, highlighting the advancements made by models like OpenAI's o series, DeepSeek-R1, and QwQ-32B in complex tasks such as mathematics and logic [1][2] - It emphasizes that the improvement in reasoning ability is not merely due to increased computational steps but rather stems from a complex, multi-agent-like interaction structure termed "society of thought," where models simulate internal dialogues among different roles to arrive at correct answers [2][3] Group 1: Reasoning Mechanisms - The research indicates that reasoning models exhibit higher diversity of perspectives compared to baseline models, activating a broader range of features related to personality and expertise during reasoning tasks [2][3] - Controlled reinforcement learning experiments show that even with reasoning accuracy as the only reward signal, base models spontaneously increase dialogic behaviors, suggesting that socialized thinking structures enhance exploration of solution spaces [3][4] Group 2: Dialogic Behaviors - The study identifies four types of dialogic behaviors in reasoning trajectories: question-answer sequences, perspective shifts, viewpoint conflicts, and viewpoint harmonization, which collectively enhance cognitive strategies [7][8] - The Gemini-2.5-Pro model's evaluations show high consistency with human scoring, indicating reliable identification of these dialogic behaviors [9][13] Group 3: Social Emotional Roles - The analysis categorizes social emotional roles in reasoning trajectories into 12 types, which are further summarized into four high-level categories, demonstrating a balanced interaction among roles rather than isolated usage [10][22] - The Jaccard index is used to measure the co-occurrence of roles, revealing that models like DeepSeek-R1 organize different roles in a more coordinated manner during reasoning processes [10][22] Group 4: Cognitive Behaviors - The study identifies four cognitive behaviors that influence reasoning accuracy, including information provision, information inquiry, positive emotional roles, and negative emotional roles [11][12] - The consistency of the Gemini-2.5-Pro model's evaluations with human scoring reinforces the reliability of these cognitive behavior classifications [13] Group 5: Experimental Findings - The findings demonstrate that even with similar reasoning trajectory lengths, models exhibit a higher frequency of dialogic behaviors and social emotional roles, particularly in complex tasks [16][23] - Experiments show that guiding dialogic features positively impacts reasoning accuracy, with a notable increase from 27.1% to 54.8% in a specific task when dialogic surprise features are positively reinforced [24][29] Group 6: Reinforcement Learning Insights - A self-taught reinforcement learning experiment indicates that dialogic structures can spontaneously emerge and accelerate the formation of reasoning strategies when only correct answers are rewarded [30]

Core Viewpoint - Google and Apple are facing legal repercussions for secretly recording users' conversations without consent, leading to a combined settlement of $163 million to resolve the lawsuits [1]. Group 1: Apple - Apple has agreed to a $95 million settlement for a class-action lawsuit that accused the company of eavesdropping on users who did not activate Siri with the prompt "Hey, Siri" [1][4]. - Users who purchased Apple devices between September 17, 2014, and December 31, 2024, and experienced unintended Siri activations are eligible for compensation, capped at $20 per device, with a maximum of five devices per person [7][8]. - Apple reported a net income of $93.74 billion in the last fiscal year, indicating that the settlement amount represents approximately nine hours of profit for the company [8]. Group 2: Google - Google has reached a tentative $68 million settlement related to a lawsuit claiming that Google Assistant recorded users without the activation phrase "OK Google" [4][15]. - The settlement is part of a lawsuit filed in 2019 and is pending approval from a federal judge [4][15]. - The class-action suit against Google includes all users in the U.S. who purchased a Google device and had Gmail accounts linked to Google Assistant-enabled devices between May 18, 2016, and December 16, 2022 [15]. Group 3: User Experience and Advertising - Users reported receiving targeted advertisements for brands they discussed in conversations that were recorded, such as Olive Garden and Air Jordan [3][9]. - The lawsuits allege that recorded discussions were shared with third-party businesses, leading to these targeted ads [9]. Group 4: Company Responses - Both Apple and Google have denied any wrongdoing regarding the allegations made in the lawsuits [5]. - Apple has since implemented a policy requiring users to opt in before their recorded audio can be used to improve Siri's functionality [5].

智通财经APP获悉，开源证券发布研报称，为应对电网老旧与AI算力高耗能挑战，北美数据中心正加速 采用自备电源模式，柴油发电机作为核心备用电源需求刚性凸显。持续看好北美缺电背景下大缸径发动 机的增长机会，有望带动相关上中游企业的营收和利润显著增长。 开源证券主要观点如下： AI浪潮推升机柜功率密度，催生对高功率、高响应柴发的刚性需求。传统IDC主流单机柜功率为4- 8kW，而AIDC部署的高功率GPU/TPU使得单机柜功率逐步提升至20-100kW，未来或将超过600kW这使 得柴发容量和并机数量进一步增加。目前市场主流柴发功率仍集中在1.8-2WM，随AIDC功耗提升，2- 4MW功率柴发机型出货量有望提升。 外资订单排产长期饱和，供需紧平衡下柴发价格持续上行 外资订单排产长期饱和，柴发价格持续上行。当前全球大缸径发动机主要由康明斯、卡特彼勒、 MTU、三菱重工、科勒等外资品牌掌控，其中康明斯部分订单交付周期达12-18个月，卡特彼勒和MTU 订单排期到2026年。国内厂商抓住国产替代窗口期，通过灵活组织产能承接部分订单，但由于产能受 限，行业大功率柴发仍供不应求，柴发价格或将持续上行。 大缸径发动机关键组件 ...

When Jeff Bezos said that one breakthrough technology would shape Amazon’s destiny, even Wall Street’s biggest analysts were caught off guard. Fast forward a year and Amazon’s new CEO Andy Jassy described generative AI as a “once-in-a-lifetime” technology that is already being used across Amazon to reinvent customer experiences. At the 8th Future Investment Initiative conference, Elon Musk predicted that by 2040 there would be at least 10 billion humanoid robots, with each priced between $20,000 and $25,000 ...

Core Insights - The article discusses the challenges faced by AI models, particularly Anthropic's Claude, in playing the children's game Pokémon, highlighting a significant gap in AI capabilities compared to human players [2][3][8] - The performance of Google's Gemini model in successfully completing a Pokémon game is attributed to its superior toolset rather than inherent intelligence [5][8] - The article emphasizes the importance of long-term memory and continuous reasoning in AI, which are currently lacking in existing models [6][8] Group 1: AI Performance in Pokémon - Claude's attempts to play Pokémon resulted in numerous failures, including getting stuck for hours and making basic mistakes that a child would not [2][3] - In contrast, Google's Gemini 2.5 Pro successfully completed a Pokémon game, showcasing the impact of a more advanced toolset that enhances AI capabilities [5] - The differences in toolsets between Claude and Gemini highlight how essential external capabilities are for AI performance in complex tasks [5][8] Group 2: Limitations of AI Models - The article points out that AI struggles with tasks requiring sustained reasoning and memory over time, which are essential for success in games like Pokémon [6][8] - Despite advancements, AI models like Claude and Gemini still face significant challenges in executing long-term goals and maintaining context over extended periods [8][11] - The article notes that while AI can excel in specific tasks, such as exams and coding competitions, it still falls short in dynamic and open-ended environments like gaming [8][11] Group 3: Broader Implications for AI Development - The challenges faced in Pokémon are indicative of broader issues in the pursuit of Artificial General Intelligence (AGI), where AI models struggle with complex, multi-faceted tasks [11][24] - The article suggests that Pokémon has become an informal benchmark for evaluating AI capabilities, as it allows for long-term tracking of reasoning and decision-making processes [24] - The ongoing difficulties encountered by AI in games like Pokémon illustrate the limitations of current models and the need for further advancements in AI technology [24]

智通财经APP获悉，近期，投资者通过聚焦人工智能领域的小众股票收获颇丰。本周，全球部分大型科 技公司将发布的财报，或将成为投资者判断2026年是否继续采用该策略的重要依据。 过去三年里，"科技七巨头"——谷歌公司(GOOGL.US)、亚马逊公司(AMZN.US)、苹果公司 (AAPL.US)、Meta Platforms公司(META.US)、微软公司(MSFT.US)、英伟达公司(NVDA.US)和特斯拉公 司(TSLA.US)——在很大程度上引领股市走高。但这一趋势在2025年底发生逆转，华尔街对这些公司投 入数千亿美元开发人工智能以及这些投资何时能产生回报日益怀疑。 追踪该集团的指数于2025年10月29日收于纪录高位，此后七家成员公司中有五家股价下跌，跑输标普 500指数。在此期间，谷歌股价飙升近20%，亚马逊公司也录得上涨，这两只股票是仅有的赢家。 对此，交易员们纷纷涌入那些从大型科技公司获得大量资金支持的企业。自"科技七巨头"指数创下历史 新高后回落以来，存储芯片制造商Sandisk Corp(SNDK.US)股价已上涨逾130%，美光科技公司(MU.US) 上涨76%，西部数据公司(WDC.US ...