关键追问Key opening question问题的底层约束是什么？How do we expand technical capability without losing sight of the human being the system is supposed to serve?
这是 李飞飞 在复杂问题前会先回到的起点。
This is the question Fei-Fei Li would return to before rushing into action.
底层支柱Core pillars人本 AI / 视觉智能 / 教育human-centered AI / vision intelligence / education
课程内容始终围绕这三根支柱组织，而不是零散知识点。
The lesson is organized around these three pillars rather than isolated quotations.
本课解决什么问题What this lesson solves
这节课单独拆 人本 AI。对 李飞飞 来说，人本 AI 不是一个口号，而是决定资源如何流动、判断如何排序、风险如何暴露的关键变量。
For Fei-Fei Li, human-centered AI matters when it deepens intelligence while keeping the human context visible rather than abstracted away. This lesson is about learning when human-centered AI deserves to lead and when it has to be balanced by vision intelligence and education.
李飞飞思想的核心概念：人本 AI。人本 AI是李飞飞思想体系的基石。先抓住 人本 AI 的第一关键变量，再讨论表达方式和执行顺序。理解这个概念，是进入李飞飞世界的第一步。本课将深入剖析人本 AI的深层逻辑与实践含义。
This stage is about one pillar at a time. The goal is not definition-memorization but better diagnostic use of human-centered AI.
概念定义What this concept really does
在 李飞飞 的语境里，人本 AI 关注的是“先看什么、再做什么”，而不是漂亮表达。
For Fei-Fei Li, human-centered AI matters when it deepens intelligence while keeping the human context visible rather than abstracted away.
与其他支柱的关系What it must be paired with
人本 AI 必须和 视觉智能、教育 一起看，否则很容易变成片面执念。
Read human-centered AI together with vision intelligence and education, or it turns into a slogan.
边界条件Where readers usually slip
当 人本 AI 看起来正确但结果不对时，通常说明约束不在概念本身，而在场景判断或执行节奏。
The mistake is treating human-centered AI as a universal virtue instead of a contextual judgment tool.
判断清单Judgment checklist
如果去掉 人本 AI，李飞飞 的整套方法会先失去哪一块判断力。Ask what breaks first if human-centered AI is ignored in a live decision.
在你的领域里，人本 AI 对应的真实观测指标是什么，而不只是情绪上的“感觉”。Test whether your current use of human-centered AI is structural or merely rhetorical.
人本 AI 和 视觉智能 出现冲突时，应该先看结构性约束还是短期表现。Check what vision intelligence or education would add before you become one-dimensional.
什么时候必须坚持 人本 AI，什么时候要承认它只是局部最优。Translate human-centered AI into one observable indicator in your own context.
应用场景 1Use case 1
当新技术刚冒头、叙事很大但工程现实很硬时，如何判断是否跟进。
Translate the framework into a live operating situation and inspect the constraint before moving.
应用场景 2Use case 2
当产品增长依赖平台红利时，怎么判断红利是不是快结束了。
Translate the framework into a live operating situation and inspect the constraint before moving.
应用场景 3Use case 3
当团队被功能堆砌拖慢时，如何回到底层技术与分发主线。
Translate the framework into a live operating situation and inspect the constraint before moving.
常见误区Common misreads
把 人本 AI 当成永远正确的答案，而不是一种带条件的判断工具。Treating human-centered AI as a permanent answer rather than a conditional lens.
只会在顺风局谈 人本 AI，一到高压环境就退回短期直觉。Using human-centered AI in easy situations but abandoning it under pressure.
把 人本 AI 简化成风格偏好，没有落实到决策顺序和指标观察上。Talking about human-centered AI elegantly without changing decision order or measurement.
Reference Shelf
李飞飞 的原典与书单Primary texts and reading shelf for Fei-Fei Li
这节课建议优先以 李飞飞 的原典、公开记录和权威书单为准，再回来看本课的判断结构。
Treat these texts as the trusted shelf for Fei-Fei Li. Start with the primary record, then return to the lesson structure.
原典与公开记录Primary texts and public record
原典 / 一手记录Primary text / public recordThe Worlds I See
Fei-Fei Li · memoir
最适合进入李飞飞科学路径和价值框架的一本书。
One of the best entry points into both her scientific path and moral frame.
原典 / 一手记录Primary text / public recordImageNet: A Large-Scale Hierarchical Image Database
Jia Deng et al. with Fei-Fei Li · CVPR 2009
理解现代计算机视觉数据基础设施的关键论文。
The key paper behind the data infrastructure of modern computer vision.
原典 / 一手记录Primary text / public recordHuman-Centered AI Talks and HAI Lectures
Fei-Fei Li · public talks
看她如何把技术进步与社会后果一起讨论。
Useful for how she puts technical progress and social consequence together.
核心书单与研究入口Core reading shelf
核心书单 / 研究入口Core reading / study entryThe Worlds I See
Fei-Fei Li · memoir
如果你只读一本到位，这本兼具科学史与人物感。
If you read only one, this balances science history with personal voice.
核心书单 / 研究入口Core reading / study entryStanford HAI Public Essays
Fei-Fei Li · essay archive
适合补足其 human-centered AI 公共论述。
Good for her public articulation of human-centered AI.
核心书单 / 研究入口Core reading / study entryImageNet Retrospectives and Interviews
Fei-Fei Li · interviews
帮助理解数据集、平台和学科转折。
Useful for context around datasets, platforms, and discipline shifts.
李飞飞 围绕 人本 AI 的代表性实践
先把底层机制想清楚，再投入长期资源，而不是先追求表面热度
Lesson: 先把底层机制想清楚，再投入长期资源，而不是先追求表面热度
逐步把 人本 AI 变成可复用的方法，而不是一次性的成功故事
Outcome: 逐步把 人本 AI 变成可复用的方法，而不是一次性的成功故事
李飞飞 在 视觉智能 上的关键取舍
Remember the operating sentence, not just the quote. The lesson works only when it changes how you order attention.
课后动作Next actions
找出你最近一个决策，复盘当时有没有明确把 人本 AI 作为主变量。Revisit one recent decision and ask whether human-centered AI was explicitly examined or only implied.
列出两个支持 人本 AI 的证据，和一个提醒你别走极端的反证。Write one argument for leaning harder into human-centered AI and one argument for restraint.
在接下来 24 小时里，用 人本 AI 重看一个你原本准备凭直觉决定的选择。Use human-centered AI to re-read a choice you were about to settle by intuition alone.
研讨题Seminar prompts
这一根支柱最容易被误用成什么样的口号？What is the most common slogan-version misreading of this pillar?
在场景“当新技术刚冒头、叙事很大但工程现实很硬时，如何判断是否跟进。”里，这个概念应该先被看见，还是先被验证？In the scenario '当新技术刚冒头、叙事很大但工程现实很硬时，如何判断是否跟进。', should this concept be noticed first or validated first?
如果把这一概念拿掉，整套系统最先失去哪一种判断能力？If you remove this concept from the system, what kind of judgment fails first?
For the next 7 days, run this lesson inside one real problem. Each day, log one decision through the opening question: How do we expand technical capability without losing sight of the human being the system is supposed to serve? and note what you examined first, what you ignored, and what sequence you would change on the next pass.