Scale AI 联合创始人兼 CEO 王亚历山大分享他从编程竞赛到融资 1 亿美元的经历，他对 AI 的愿景，以及为何从 MIT 辍学创业。
Alexander Wang, CEO and co-founder of Scale AI, discusses his journey from coding competitions to raising over $100 million, his vision for AI, and why he left MIT to start the company.
要点 · TL;DR
AI 的变革性堪比计算机的出现，比互联网更重要。 AI is as transformative as the advent of computing, bigger than the internet.
数据是机器学习的关键瓶颈；Scale AI 为 AI 系统提供数据精炼基础设施。 Data is the key bottleneck for ML; Scale AI refines data for AI systems.
自动驾驶汽车将比人类更安全，并在 10 年内实现无方向盘合法上路。 Self-driving cars will be safer than humans and legal without steering wheels in under 10 years.
核心观点 · Key points
AI 比互联网更重大，堪比计算技术的诞生。 AI is bigger than the internet, comparable to the advent of computing.
数据是机器学习的关键瓶颈；Scale AI 提供数据精炼基础设施。 Data is the key bottleneck for machine learning; Scale AI provides data refinement infrastructure.
自动驾驶汽车将比人类更安全，但需要激光雷达和摄像头共同作用以增强鲁棒性。 Self-driving cars will be safer than humans, but require both lidar and cameras for robustness.
机器学习将增强而非消除工作岗位，正如自动取款机与银行柜员的例子所示。 Machine learning will augment jobs, not eliminate them, as seen with ATMs and bank tellers.
AI 的可解释性很重要，但当前系统已存在与传统软件类似的尾部风险。 Explainability in AI is important, but current systems already have tail risks similar to traditional software.
反共识 · Contrarian takes
AGI 还很遥远；仅凭无限算力无法产生通用智能。 AGI is far off; infinite compute alone won't produce general intelligence.
中国在 AI 方面进步真实且迅速，但美国在多数领域仍领先。 China's AI progress is real and fast, but US still leads in most areas.
AI 监管是必要的，但当前美国政府缺乏监督代码的技术专长。 Regulation of AI is needed, but current US government lacks technical expertise to oversee code.
尽管存在监管障碍，自动驾驶汽车将在 10 年内实现无方向盘合法上路。 Self-driving cars will be legal without steering wheels in under 10 years, despite regulatory hurdles.
表单处理和放射学等枯燥的 AI 应用将产生巨大的经济影响。 Boring AI applications like form processing and radiology will have huge economic impact.
本期章节 · Chapters（共 26）
引言与背景Introduction and Background
Scale AI 的愿景Vision for Scale AI
机器学习 vs AIMachine Learning vs AI
示例：自动驾驶Example: Autonomous Vehicles
Scale 在数据标注中的角色Scale's Role in Data Annotation
自动驾驶恶作剧Prank on Self-Driving Cars
跨公司数据共享Data Sharing Across Companies
Calm 广告与介绍Calm ad and intro
数据孤岛与竞争Data silos and competition
基础设施与传感器之争Infrastructure and sensor debate
数据清洗流程Data Cleaning Pipeline
自动驾驶时间线Self-driving car timeline
自动驾驶未来与赌博Self-driving future and gambling
中美 AI 竞赛China vs US AI race
ML 系统的责任与可靠性Responsibility and Reliability of ML Systems
信任 ML 决策与可解释性Trusting ML with decisions and explainability
被替代的恐惧与 AI 高标准Fear of replacement and higher standards for AI
AI 的边界案例与监督Edge cases and oversight of AI
无限计算与通用 AIInfinite compute and general AI
下一个大型窄 AI 项目Next big narrow AI projects
AI 自适应学习与人脸识别Adaptive learning with AI and facial recognition