The godmother of AI on spatial intelligence and a human-centered future.
要点 · TL;DR
AI 总体有益，但需负责任地引导。 AI is a net positive but requires responsible stewardship.
世界模型和空间智能是 AI 的下一个前沿。 World models and spatial intelligence are the next frontier for AI.
当前 AI 缺乏创造力和抽象能力，需要超越规模的创新。 Current AI lacks creativity and abstraction; more innovation is needed beyond scaling.
核心观点 · Key points
AI 对人类是净利好，但它是双刃剑；我们必须负责任地行动。 AI is a net positive for humanity, but it's a double-edged sword; we must act responsibly.
大数据、神经网络和 GPU 这三要素是现代 AI 的黄金配方。 The trio of big data, neural networks, and GPUs is the golden recipe for modern AI.
世界模型和空间智能是具身 AI 及其他领域缺失的关键。 World models and spatial intelligence are the missing keys to embodied AI and beyond.
除了 Scaling，我们还需要更多创新；当前 AI 缺乏创造力、抽象能力和情商。 We need more innovations beyond scaling; current AI lacks creativity, abstraction, and emotional intelligence.
Marble 通过提示生成真正的 3D 世界，可用于视觉特效、游戏、机器人和治疗。 Marble generates truly 3D worlds from prompts, enabling applications in VFX, gaming, robotics, and therapy.
每个人在 AI 中都有角色；人的尊严和自主权必须保持核心地位。 Everyone has a role in AI; human dignity and agency must remain central.
反共识 · Contrarian takes
AGI（通用人工智能）更多是营销术语而非科学术语。 AGI is more a marketing term than a scientific one.
苦涩教训可能不完全适用于机器人，因为数据和物理限制。 The bitter lesson may not fully apply to robotics due to data and physical constraints.
当今 AI 在数椅子或推导牛顿定律上还不如幼儿或牛顿。 AI today cannot match a toddler in counting chairs or Newton in deriving laws.
直到 2017 年左右，公司才敢自称 AI 公司。 Calling yourself an AI company was avoided until around 2017.
世界模型比视频生成更深层；它们支持 3D 推理和交互。 World models are deeper than video generation; they enable 3D reasoning and interaction.
机器人将像自动驾驶汽车一样经历 20 年旅程，而非速胜。 Robotics will take a 20-year journey like self-driving cars, not a quick win.
本期章节 · Chapters（共 34）
引言与嘉宾背景Introduction and Guest Background
AI 对人类的影响AI's Impact on Humanity
历史背景：AI 寒冬与转折Historical Context: AI Winter and the Shift
突破性洞察：ImageNet 与数据The Breakthrough Insight: ImageNet and Data
AI 的乐观与责任Optimism and Responsibility in AI
AI 与 ImageNet 的历史The History of AI and ImageNet
早期职业生涯与 AI 寒冬Early Career and AI Winter
数据问题与 ImageNetThe Data Problem and ImageNet
2012 年突破与深度学习The 2012 Breakthrough and Deep Learning
两块 GPU 与计算规模Two GPUs and the Scale of Compute
高能动性与 AI 术语High Agency and the Term AI
其他早期历史Other Early History
研究者代际与 AI 文化Generations of Researchers and AI Culture
AGI 定义与进展On AGI Definition and Progress
未来 AI 组件与创新需求Components for Future AI and Need for Innovation
AI 无法复现历史突破AI's Inability to Replicate Historical Breakthroughs
世界模型介绍Introduction to World Models
空间智能的背景与动机Background and Motivation for Spatial Intelligence
赞助商插播Sponsor Break
苦涩教训与机器人学Bitter Lesson and Robotics
对大脑的敬畏Awe for the Brain
Marble 产品发布Marble Product Launch
Marvel 介绍及其能力Introduction to Marvel and its capabilities
应用与目标Applications and goals
暴露疗法与 MarbleExposure therapy and marble
与其他视频模型的区别Difference from other video models
Marble 作为平台Marble as a platform
团队与资源Team and resources
祝贺发布Congratulations on launch
创始人历程与建议Founder journey and advice
人才竞争与速度Competition for talent and speed
职业建议与无畏精神Career Advice and Fearlessness
斯坦福 HAI 与以人为本的 AIStanford HAI and Human-Centered AI