当整个 AI 圈都在为 DAU（日活跃用户数）和融资额焦虑时，MiniMax 创始人闫俊杰却表现出一种近乎冷酷的淡漠。 坐在罗永浩对面的闫俊杰，并不像一位掌管着 AI 独角兽企业的技术新贵。 他拒绝谈论改变世界，反而坦承恐惧。那种恐惧不是来自商业竞争，而是来自技术本身——当模型的能力开始超越人类时，创造者反而成了最先感到不安 的人。 用 1/50 的筹码通往 AGI 在巨头环伺、算力短缺、热钱褪去的 2025 年，MiniMax 正在进行一场关于认知的修正：不再沿用移动互联网的逻辑，即通过大规模投放换取增长、通过 堆砌功能留住用户，而是回归本质： 把模型当作最重要的产品 。 在大模型时代，真正的产品其实是模型本身，传统意义上的产品更像是一个渠道。如果模型不够聪明，产品做得再好也没有用。 在罗永浩和闫俊杰这期对谈里，我发现 MiniMax 这家 AI 公司从创业第一天就选择了注定与主流背道而驰的技术路径。 当所有人都试图寻找中国的 OpenAI 和 Sam Altman 时，闫俊杰却在试图证明「非天才」的价值。MiniMax 的故事不是关于天才的灵光乍现，而是一场关 于如何在资源受限的缝隙中，通过极度理性地计算 ...

在苏箐看来，这一阶段的核心命题，是将现有技术的潜力发挥到极致，比如地平线将持续提升芯片算力 与模型容量，以统一范式推进L2到L4的融合，让城市L2从20万级车型下探至10万级市场，让准L4系统 以平民化价格走进大众。同时，强化工程与组织能力，应对海量长尾场景的打磨，这才是穿越周期的关 键。 "自动驾驶的终极目标，是造出能替代人类司机的机器。这场长跑，在范式革命之后，考验的是行业沉 下心来做'精活'的耐力。在未来几年内，能够把L4级别的车，以平民化的价格送到用户手上。这才是我 们所有人辛苦了这20年做这一行的意义所在。" （文章来源：中国经营报） "未来三年，自动驾驶行业将告别范式迭代的狂飙，进入极致优化的'苦日子'。" 12月9日，在"2025地平线技术生态大会"上，作为深耕自动驾驶20年的老兵，一向"反共识"的地平线副 总裁&首席架构师苏箐分享了对自动驾驶未来趋势的判断。 值得一提的是，对于当下，苏箐则给出了冷静的判断："行业需要清醒。当前深度学习已显露天花板迹 象，AGI基础理论暂无突破信号，下一轮内核重构至少还需5—20年的技术沉淀。未来三年，自动驾驶 行业将告别范式迭代的狂飙，进入极致优化的'苦日子' ...

【朱啸虎：英伟达回调一下更健康】如何看待英伟达的回调？朱啸虎表示，"我觉得是回调一下更健 康，为了明年的进一步爆发，肯定能打一个更好的基础。"不过他同时表示，应该有比英伟达更好的一 些标的。谈到OpenAI Sam Altman，朱啸虎认为他嘴上说一套，但做的是另外一套。嘴上可能说的是 AGI，但事实上做的都是在应用端。 专题：未竟之约：张小珺访谈录 责任编辑：李思阳 专题：未竟之约：张小珺访谈录 【朱啸虎：英伟达回调一下更健康】如何看待英伟达的回调？朱啸虎表示，"我觉得是回调一下更健 康，为了明年的进一步爆发，肯定能打一个更好的基础。"不过他同时表示，应该有比英伟达更好的一 些标的。谈到OpenAI Sam Altman，朱啸虎认为他嘴上说一套，但做的是另外一套。嘴上可能说的是 AGI，但事实上做的都是在应用端。 责任编辑：李思阳 ...

Core Insights - The article discusses the innovative AI product "Macaron," which aims to integrate AI into daily life by allowing users to create personalized mini-applications that enhance their lifestyle [2][12]. - The founder, Chen Kaijie, emphasizes the importance of creating a product that excites users and helps them solve real-life problems, moving from escapism to practical assistance [10][15]. - Macaron has gained nearly 300,000 users and has seen the creation of around 200,000 mini-applications, with about 30% of new users actively creating their own applications [2][26]. Group 1: Product Philosophy and Development - The product is designed to be dynamic, vibrant, and proactive, aiming to provide a more engaging and life-like interaction compared to traditional AI assistants [9][21]. - Chen Kaijie transitioned from previous entrepreneurial experiences, focusing on AI-driven interactive storytelling to developing Macaron, which he believes can genuinely assist users in their daily lives [7][8]. - The goal is to shift from "creating value through creation" to "delivering value through usage," enhancing user experience by making mini-applications automatically callable and shareable [9][26]. Group 2: Technical Innovations - The company has significantly reduced the training costs for a trillion-parameter model's reinforcement learning by a factor of ten, which is crucial for scaling AI capabilities [9][30]. - The focus on "Agentic RL" is seen as the core path for enhancing model intelligence, moving beyond traditional pre-training methods [35][36]. - The architecture includes a unique memory management system that mimics human-like forgetting, allowing the AI to retain meaningful experiences while discarding irrelevant details [32]. Group 3: User Engagement and Community - The initial user base consisted of developers and product managers, but the focus has shifted to a broader audience, particularly busy women who seek both emotional support and practical solutions [19][20]. - Users have reported using Macaron for various personal applications, such as health tracking and emotional support, indicating its versatility and user-centric design [20][24]. - The company aims to foster a community where users can share and replicate mini-applications, enhancing the overall user experience and engagement [28][41].

要点速览： 2025年，一个叫Dwarkesh Podcast的播客成为了AI行业内获取一手信息最重要的渠道之一。甚至可以说，它已经 成了硅谷AI技术圈的必看节目。从Satya Nadella到Andrej Karpathy，再到Ilya Sutskever，这些平时很难约到的行 业核心人物，都选择在这里进行长时间的深度对话。本期，我们将为大家分享其中最新、最受关注的几场播客 及核心观点。 Ilya Sutskever 前OpenAI首席科学家、计算机科学家、SSI创始人 ■ 洞察1：那种无脑堆算力的「暴力美学」时代，其实已经翻篇了。 过去这五年，大家都在喊Scaling Law，好像只要GPU够多、数据够大，把整个互联网喂进去，AGI就自动产 出来了。但Ilya直接泼了盆冷水。 他说Pre-training （预训练） 已经开始式微，现在数据快用光了，到了后 面这一步 （RL和后训练） ，光靠"大"没用了。 现在又回到了2012年之前那种需要"拼品味、拼直觉"的手搓 时代 （Age of Research） 。 ■ 洞察2：「情绪」不是人类的累赘，而是进化给人类的礼物。 我们通常觉得AI是理性的，人类是感性 ...

Core Insights - The report highlights a significant shift in AI development from the "tool era" to the "partner era," indicating profound changes in economic structure, social forms, and human lifestyles by 2025 [3][31] Group 1: Key Trends in AI Development - Trend 1: Computing infrastructure is becoming essential, with skyrocketing demand for data centers, marking the computing economy as the primary engine of the intelligent industry [5][6] - Trend 2: AI-driven demand is reshaping chip innovation, with GPUs facing challenges and NPU gaining traction, while ASIC/FPGA are experiencing growth [8][11] - Trend 3: Pre-training will determine the hierarchy of large models, with architectural innovations influencing pre-training levels [13] - Trend 4: Large models are entering the "inference time," with increasing demands for model innovation driven by complex tasks [15] - Trend 5: The period of information AI applications and physical AI research is emerging, with embodied intelligence becoming a focal point [17][20] Group 2: AI Applications and Market Dynamics - Trend 6: AI is reshaping traffic entry points, transitioning from "people finding services" to "services finding people," leading to the evolution of interaction paradigms [20][22] - Trend 7: Multi-modal capabilities are crucial for AI application deployment, enabling systems to process various information types, enhancing productivity [22] - Trend 8: AI hardware is proliferating across devices like PCs, smartphones, and IoT, addressing privacy, latency, and cost efficiency [25] - Trend 9: AI4S is accelerating the realization of AGI, with AI achieving capabilities comparable to doctoral-level problem-solving in various fields [25][27] - Trend 10: Open-source AI is entering a pivotal phase in China, with the country transitioning from a participant to a leader in the global AI landscape [28][30] Conclusion - The report emphasizes that the AI sector is at a historic turning point, with technology evolving from model competition to scenario integration, and highlights China's strategic advancements in open-source ecosystems, autonomous chips, and AGI pathways [31]

组委会 发自 MEET2026 量子位 | 公众号 QbitAI 刚刚，《2025年度AI十大趋势报告》在 MEET2026智能未来大会 上正式发布。 报告由量子位智库打造，基于对全球AI技术发展、产业应用和社会影响的深度调研与分析，全面呈现了人工智能领域的最新现状与未来发展趋 势。 十大趋势重塑AI发展格局 报告指出，2025年AI正从"工具时代"迈向"伙伴时代"，其发展将深刻重塑经济结构、社会形态和人类生活方式，并有十大关键趋势结论： 算力基建化：数据中心需求狂飙，算力经济是智能产业第一大引擎 芯片AI化：AI原生需求重塑芯片创新，大市场大生态打开时代芯机遇 预训练决定大模型格局梯队，架构创新决定预训练水平 大模型落地进入推理时间，推理需求倒逼模型创新 信息AI应用期，物理AI研发期，具身智能成合流风口 AI重塑流量入口：PC互联网、移动互联网、Agentic互联网 多模态成AI应用落地关键：视频、3D、代码依次展现生产力 AI硬件百端齐放：PC手机汽车眼镜玩具，焕脑正当时 AI4S突破加速AGI实现，AI数理化触及博士水平 开源AI进入中国时间，AGI拥有中国路线 报告通过对十大核心趋势的系统梳理，为企 ...

RT Jeffrey Weichsel (@jeffreyweichsel)Asimov's Three Laws of Robotics1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.2. A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.Asimov's Three Laws of Robotics were meant to ensure robots prioritize human safety, obedience, and self-preserva ...

来源：科技先生 科技先生讯，北京，12月10日消息，《2025年度AI十大趋势报告》在"MEET2026智能未来大会"上正式 发布。 报告由量子位智库打造，基于对全球AI技术发展、产业应用和社会影响的深度调研与分析，全面呈现 了人工智能领域的最新现状与未来发展趋势。 十大趋势重塑AI发展格局 报告指出，2025年AI正从"工具时代"迈向"伙伴时代"，其发展将深刻重塑经济结构、社会形态和人类生 活方式，并有十大关键趋势结论： 算力基建化：数据中心需求狂飙，算力经济是智能产业第一大引擎 芯片AI化：AI原生需求重塑芯片创新，大市场大生态打开时代芯机遇 预训练决定大模型格局梯队，架构创新决定预训练水平 大模型落地进入推理时间，推理需求倒逼模型创新 信息AI应用期，物理AI研发期，具身智能成合流风口 AI重塑流量入口：PC互联网、移动互联网、Agentic互联网 多模态成AI应用落地关键：视频、3D、代码依次展现生产力 AI硬件百端齐放：PC手机汽车眼镜玩具，焕脑正当时 AI4S突破加速AGI实现，AI数理化触及博士水平 开源AI进入中国时间，AGI拥有中国路线 报告通过对十大核心趋势的系统梳理，为企业管理者、投资机 ...

Technology & Security - The industry is highly interested in Zero-Knowledge (ZK) technology [1] - The ultimate barbell strategy in 2025 involves personal AGI chaos and public ZK security [1]