Key Takeaways NVDA closed at a record $154.31, overtaking MSFT with a $3.763 trillion market capitalization. Despite $8B in export losses, NVDA rallied nearly 80% from April lows on robust AI chip demand. NVDA forecasts $5B in robotics revenue in fiscal 2026 with multi-trillion-dollar opportunities in the future.NVIDIA Corp. (NVDA) — the undisputed global leader of the generative artificial intelligence (AI)-powered graphical processing units (GPUs) — regained the crown of the world’s most valuable compan ...

Everyone's betting on AI getting smarter. The amazing thing is they can reason. We're just at the beginning of the reasoning AI era.Smarter models, sharper intuition, superintelligence. I think we'll get superintelligence, and I would guess that it will be a continuation of this trend that humanity has been on for 100 plus years. Fueling explosive new demand for compute.The amount of computation necessary to do that reasoning process is 100 times more than what we used to do. And companies going all in, spe ...

以下文章来源于3D视觉之心 ，作者3D视觉之心 3D视觉之心 . 3D视觉与SLAM、点云相关内容分享 点击下方 卡片 ，关注" 自动驾驶之心 "公众号 戳我-> 领取 自动驾驶近15个 方向 学习 路线 评估协议的关键局限性 三维计算机视觉领域高度关注于捕捉场景的几何和视觉外观，以及理解其内容。近年来，三维高斯溅射（3D Gaussian Splatting, 3DGS）因其独特的能力——能够以一种紧凑的形式联合编码场景的几何、外观和理解属性 （该形式可以有效地从二维带位姿的图像中优化得到）——已成为最理想的三维表示方法。此外，视觉-语言推 理代表了三维场景理解最具前景的方向，因为它将场景的视觉和几何属性与我们用来定义、描述和推理概念的语 言连接起来。因此，本文专注于利用 3DGS 进行视觉-语言场景理解。 语言高斯溅射（Language Gaussian Splatting, LGS）最相关的方法可分为三类。前两类方法首先使用视觉-语言基 础模型（例如 CLIP）从所有训练图像中提取二维特征。第一类随后执行基于梯度的单场景优化，将特征向量分 配给每个三维高斯基元（primitive），并优化它们，使其渲染 ...

企业部署 AI 不再是试验项目，而是战略行动。预算已经常态化、模型选择多元化、采购流程标准化、AI 应用开始系统落地。尽管产业需求和 企业需求碎片化，但这正是企业拥抱的方向。一些关键厂商正在脱颖而出，企业也越来越多选择成品应用以加速落地。 市场形态愈加接近传统软件，但变化节奏与复杂性却完全不同——这是 AI 的特有节奏。 2025年，AI大模型在企业场景的落地走到哪了？ 过去一年，AI在企业中的地位发生了根本性转变。它不再是创新实验室里一场场孤立的试验，也不仅是技术部门热衷的"新玩具"，而是真正走入了核心业 务系统，成为IT和经营预算中不可或缺的一部分。 这是一场静悄悄却迅猛的演进：AI模型变得更多样，采购流程愈发严谨，企业不再"自己造轮子"，而是开始像采购传统软件那样，有条不紊地选择、部 署、评估人工智能服务。技术领导者们正变得越来越成熟——他们明白，不同模型适配不同任务，用例碎片化是常态，而高质量的AI原生应用，正在快 速超越传统软件厂商。 近日，A16z发布了一份主题为《AI技术在企业场景落地》的调研报告，报告基于与20多位企业买家的深度访谈和100位CIO的调研，全面回顾了企业在 2025年如何部署、 ...

KRIS-Bench团队 投稿 量子位 | 公众号 QbitAI 人类在学习新知识时，总是遵循从"记忆事实"到"理解概念"再到"掌握技能"的认知路径。 AI是否也建立了"先记住单词，再理解原理，最后练习应用"的这种知识结构呢？ 测评一下就知道了！ 东南大学联合马克斯·普朗克信息研究所、上海交通大学、阶跃星辰、加州大学伯克利分校与加州大学默塞德分校的研究团队，共同提出了 KRIS-Bench （Knowledge-based Reasoning in Image-editing Systems Benchmark）。 首创地 从知识类型的视角 ，对图像编辑模型的推理能力进行系统化、精细化的评测。 借鉴布鲁姆认知分类与教育心理学中的分层教学理念，KRIS-Bench让AI在事实性知识（Factual Knowledge）、概念性知识（Conceptual Knowledge）与程序性知识（Procedural Knowledge）三大层面上，逐步接受更深入、更复杂的编辑挑战。 基于认知分层的三大知识范畴 KRIS-Bench在每个类别下又细化出7大推理维度、22种典型编辑任务，从 "物体计数变化"到"化学反应预测 ...

Reasoning models are systems that can execute more complicated tasks through a step-by-step logical thought process. Mistral's new model "is great at mathematics [and] great at coding," according to Mensch. "We're announcing in a couple of hours our new reasoning model, which is very much competitive with all the others and has the specificity of being able to reason in multiple languages," CEO Arthur Mensch told CNBC's Arjun Kharpal onstage during a fireside chat at London Tech Week. French founder of arti ...

Financial Data and Key Metrics Changes - NVIDIA reported revenue of $44 billion, a 69% increase year-over-year, exceeding expectations despite a challenging operating environment [5] - Data center revenue reached $39 billion, growing 73% year-on-year [5] - GAAP gross margins were 60.561%, while non-GAAP gross margins would have been 71.3% excluding a $4.5 billion charge related to inventory write-downs [30][32] Business Line Data and Key Metrics Changes - Data center revenue was significantly impacted by new export controls, with $4.6 billion recognized prior to the controls and a $4.5 billion charge for inventory write-downs [6][30] - Gaming revenue reached a record $3.8 billion, increasing 48% sequentially and 42% year-on-year, driven by strong adoption of Blackwell architecture [21] - Networking revenue grew 64% quarter-over-quarter to $5 billion, with strong demand for NVLink and Spectrum X solutions [17][20] Market Data and Key Metrics Changes - China data center revenue was below expectations due to export licensing controls, with a meaningful decrease anticipated in Q2 [20] - Singapore accounted for nearly 20% of Q1 build revenue, primarily for US-based customers [20] - The AI market in China is estimated to be around $50 billion, which NVIDIA is currently unable to access due to export restrictions [6][60] Company Strategy and Development Direction - NVIDIA is focusing on AI factory deployments, with nearly 100 AI factories in progress, doubling year-over-year [12][13] - The company is committed to a robust product roadmap extending through 2028, with a focus on enhancing AI capabilities and infrastructure [10][32] - NVIDIA is exploring ways to comply with new export control rules while maintaining competitiveness in the AI market [7][36] Management's Comments on Operating Environment and Future Outlook - Management expressed concerns about losing access to the China AI accelerator market, which could materially impact business [7][36] - The company anticipates continued growth in AI demand, particularly in reasoning AI, which is driving significant increases in inference workloads [11][82] - Management expects total revenue for Q2 to be around $45 billion, with modest sequential growth across all platforms [31][32] Other Important Information - NVIDIA returned a record $14.3 billion to shareholders through share repurchases and dividends [30] - The company is investing heavily in onshore manufacturing and partnerships to strengthen its supply chain [44][45] Q&A Session Summary Question: How much of the inference demand is NVIDIA able to serve? - Management indicated they are on track to serve most of the inference demand, with Blackwell NVLink 72 being the ideal solution for reasoning AI [52][54] Question: What is the impact of the China export controls on future revenue? - Management confirmed a significant decline in China data center revenue is expected, with a total of $8 billion in H20 revenue lost for Q2 [58][60] Question: Are there more large GPU cluster investments expected? - Management noted that there are many AI factories being planned globally, indicating a strong demand for AI infrastructure [70][72] Question: What is the outlook for the networking business? - Management highlighted strong adoption of Ethernet solutions and improvements in utilization rates, particularly with Spectrum X [95][100]

大模型做数学题的能力很强，可是它们真的能够理解基本的数学原理吗？ 拿小学生的数学题进行测试，人类平均得分为93.30%，而大模型的表现让人意外： 闭源模型中Gemini2.0-Flash（49.77%）、Qwen-VL-Max（47.03%）、Claude-3.7-Sonnet（46.63%）的综合表现最佳，但仍未突破50% 准确率。 why？ 因为大模型可能并不能真正理解基本数学元素和视觉概念。 现有的视觉数学基准测试主要集中在知识导向的评估上，容易受到大型语言模型中预先嵌入的知识的影响。 上述结论来自达摩院推出的新基准 VCBench ——这是一个专为评估 具备显式视觉依赖性的多模态数学推理任务 而设计的综合基准。 VCBench团队 投稿 量子位 | 公众号 QbitAI 该基准主要面向小学 1-6 年级的数学问题，即 并不涉及复杂的数学或几何推理，但高度依赖于显式的视觉依赖性 的问题。 解决这种问题，需要模型识别和整合图像中的视觉特征，并理解不同视觉元素之间的关系。 △ 论文标题：Benchmarking Multimodal Mathematical Reasoning with Explicit ...

本文由匹兹堡大学智能系统实验室（Intelligent Systems Laboratory）的研究团队完成。第一作者为匹兹堡大学的一年级博士生薛琪耀。 当前文本生成视频（T2V）技术正在从注重视觉质量与模型规模的扩展阶段，迈向更关注物理一致性与现实合理性的推理驱动阶段。 物理规律作为建模现实世界的基本知识体系，是实现高质量视频生成的关键约束。提升大模型对现实物理动态的理解与遵循能力，成为推动 T2V 技术落地 的重要突破方向。 为推动物理一致性驱动的 T2V 生成研究，来自匹兹堡大学的研究团队提出了 PhyT2V 框架，并在最新论文中系统阐述了该方法的核心机制，该论文已被 CVPR 2025 接收。 论文标题：PhyT2V: LLM-Guided Iterative Self-Refinement for Physics-Grounded Text-to-Video Generation 论文地址： https://arxiv.org/abs/2412.00596 该方法不依赖模型重训练或大规模外部数据，而是通过引入大型语言模型引导的链式推理与迭代自我修正机制，对文本提示进行多轮物理一致性分析与优 化，从而 ...

Today I want to talk about building the future with design matters and want to talk about this kind of insights and future trends as well for this year. I want to focus on how we solve the customer's problem and less about ourself. So I want to start off with something huge because for a lot of us we know about AGI or artificial general intelligence.I think it's basically means that um we want to have AI that's uh achieving the the level of intelligence comparable to human and also maybe even surpass human ...