策略性偷懒：用 DRAG 框架夺回精力主导权
在日常工作和生活中，绝大多数人都在被动地让 AI 摧毁自己的思考能力，无意中将 AI 训练成了自己的“替代品”。然而，前 1% 的精英却在反向使用 AI，利用它来训练自己的大脑并在复杂的博弈中胜出。要建立这套高效的思维训练系统，我们首先需要克服大脑与生俱来的生理缺陷——完成偏差（Completion Bias: 大脑倾向于追求完成任务所带来的即时多巴胺奖励，从而将所有任务等同对待）。这种偏差导致我们陷入“优先级盲区”，无法区分百万美元的战略规划与一封内部日常邮件的价值差异。
为了规避这种精力分散，我们需要识别工作中的两类回报曲线。第一类是边际效应递减的封顶回报曲线（Capped Payoff Curve），例如格式化幻灯片、填写报销单或参加通知性会议，其价值在达到临界点后便会持平。对此，我们应采用诺贝尔奖得主赫伯特·西蒙提出的满意决策论（Satisficing: 追求足够好而非完美，融合 Satisfy 与 Suffice），在合理范围内停止投入。而第二类则是无上限的指数回报曲线（Uncapped Payoff Curve），如客户关系构建、产品核心设计、寻找合伙人等，在这类事情上提升 1% 的表现能解决后续 99% 的核心难题，必须倾注全部心血。
通过这一认知划分，我们便可借助 DRAG 框架（DRAG Framework: 将基础重复性任务外包给 AI 的系统工具）将第一类曲线的任务完全托管给 AI，从而腾出精力专注于需要人类判断力、直觉和审美的高回报决策中：
- D 代表起草（Drafting）：利用 AIM 协议（Act-Input-Mission，即设定角色、输入材料、明确任务）让 AI 快速生成初稿，攻克从零到一的“白纸焦虑”。
- R 代表研究（Research）：通过 ChatGPT、Gemini 或 Claude 的深度研究功能，让其像网络蜘蛛一样自动检索并汇总数以百计的页面，在十分钟内产出相当于普通咨询顾问一周工作量的高质量调研报告。
- A 代表分析（Analysis）：让 AI 初步筛选非结构化数据中的隐藏模式，发挥其超越人类感官的特征识别优势。
- G 代表日常琐事（Grunt work）：将数据清洗、格式转换、翻译及制表等机械性手工劳动彻底交给 AI。
Original English
Most people are letting AI destroy their ability to think, training AI to become their own replacement. Tragic, because AI can make you dangerously intelligent. I went from being homeless to an MIT grad and running and advising AI companies worth billions. And here's what I've learned. The top 1% use AI backwards. They don't prompt to get answers. They use it to train their brain and outsmart almost any situation. So in this video, I'll break down a counter intuitive system the top 1% use to get smarter faster with AI. Here is the four-step framework. Step one, intelligent laziness. A study in Harvard Business Review found that CEOs waste 72% of their time in meetings that don't move the needle. We've all experienced those meetings, haven't we? The one-hour meeting that needed only 15 minutes to get to a decision, but it's hard to stop. So why do some of the most accomplished folks feel trapped this way? Because we all suffer from this biological glitch called completion bias. Your brain is wired to seek an immediate dopamine hit that you get from finishing a task. So we end up treating all tasks as equal because we're going to get roughly the same amount of dopamine when you spend time on redrafting an internal email or a million-dollar strategy document. Everything is priority one. So none of it is. So how do you avoid this priority blindness? A good way to think about tasks is to see two curves. First curve has capped payoffs. This curve goes up and then flattens out once it reaches the zone of diminishing returns. So tasks like formatting slides or internal emails, expense reports, FYI meetings. What happens if you spend additional effort to make the outcome of these tasks pitch perfect? Nothing. There's no upside here because the value flatlines after a point. Nobody cares if you spend hours choosing better fonts or breathtaking designs in internal slides that are seen for 6 minutes. This curve shows you your zone of intelligent laziness. There was a Nobel Prize winning economist and computer scientist and his name was Herbert Simon and he came up with a concept called satisficing, which pretty much means stop when it's good enough. Satisfy and suffice. Satisfice. Now our second curve is the exact opposite. It has uncapped payoff. This curve stays flat for a long time, but then goes to the moon in a hurry. These are tasks like customer interactions, product design, pricing model, finding a co-founder or a life partner. Being 1% better here does not yield 1% better result. It actually solves the rest of the 99% of your problems. Pour your soul into this. Jony Ive would obsess for many months on even the internal component design of iPhone. But you know, Steve Jobs never said, "Hey, this is costing us a lot of money. And who's going to pry open the iPhone?" But Steve knew this was the second curve. So if the first curve is your zone of laziness, your second curve is your zone of obsession. Let's talk about how AI can help. The top 1% use AI on zone one or the zone of laziness. The more they outsource zone one to AI, the more they can focus on zone two, the zone of obsession. So how do I decide what to outsource to AI and when? So for that, I use a very simple framework called DRAG framework, D R A G. Four categories of work you immediately should delegate to AI so you can stay in your zone of obsession. First, D equals drafting. This is the blank page problem we all face. It's hardest to get from zero to one sometimes. AI can help here tremendously, actually. Give it a prompt using the AIM protocol that I've shared before. Hey AI, act in this role, use this input, and this is your mission. AIM. In that way, you get started very quickly on that email or code or presentation. And the first draft from AI will be crappy and atrocious, but that's fine. Now you have a starting point. You're not staring at a blank page anymore. Now it'll trigger something in your brain and you're off to the races. R equals research. This helps you solve the information overload problem. Today, if something requires deep research, it can be dramatically accelerated using AI. Summarization, extraction, competitive intel. You know, don't spend time doing that kind of research. Let your friendly neighborhood AI do it for you. When you use the deep research feature on ChatGPT or Gemini or Claude, it fires off hundreds of secondary search queries. It goes out to the web like a spider and finds hundreds of sites, consolidates the results, even checks its own work by asking what's missing, and follows up on its own to finally deliver a rich document to you. It's like you just hired a consultant for a week-long research project, but instead, you get there in 10 minutes. Third is A for analysis. Let AI take the first pass at analyzing, summarizing, reasoning, especially if it's all unstructured data because AI is going to find patterns that we humans aren't going to be able to. So use it for your advantage. And finally, G is for all the grunt work. Tasks like reformatting, translating, tabulating, cleaning data, and on and on. The boring manual work. Just give it a copy to AI. So what's the key principle behind DRAG? Apply it for only when you are in your zone one, that first curve. If it requires human interaction or judgment or intuition or decision-making or taste, that's curve two. That you've got to do it yourself. But you know, I have found that 70 or 80% of my repetitive tasks tend to be in zone one and you might find that too. So be lazy when you can use DRAG. Be obsessed for everything else.智力攀登：从盲目抽签到精准构建确定性
在剥离了低价值的重复任务后，我们必须重新审视与 AI 的交互逻辑。大多数人依然将 AI 视为一个“高阶计算器”（计算器输入 2+2 必定输出 4，具有绝对的可预测性），但从物理学的发展规律来看，我们的认知需要从确定性的牛顿钟表世界跃迁至海森堡的量子概率云。大语言模型实质上是一个概率引擎（Probability Engine: 基于词元概率分布生成文本的系统），如果不加以逻辑约束和实证校验，它就会顺理成章地凭空捏造事实。
因此，为了获取真正处于行业顶尖水平的深度输出，我们不能采用无样本的零样本提示（Zero-shot Prompting: 没有任何示例参考的直接提问，本质上是在进行概率博弈），而必须拾级而上，逐步攀登“智能之丘”的四个营地：
- 单样本提示（One-shot Prompting）：给模型提供一个明确的风格或排版基准（例如引入特定 LinkedIn 帖子的行文规范），代替盲目猜测。
- 少样本提示（Few-shot Prompting）：提供三个以上的优质样例，并通过提示词对齐（Grounding: 利用本地文档、私有数据限制大模型幻想的纠偏机制）绑定模型的输出风格与知识逻辑。在这里，可以先让 AI 提炼并反馈这些样本的行文规律，从而反向理解自身的思维盲区。
- 思维链推理（Chain of Thought: 通过展示推理步骤以大幅降低幻觉并提高逻辑准确性的技术）：让模型在产出结论前强制进行长考，细化中间步骤，如要求其在修改报告前先梳理三个最需改进的板块并陈述原因。
- Agent 化协同（Agents: 具备规划、执行、反思能力的自主 AI 代理系统）：赋予 AI 多重人格与流程自洽性。例如，用一条指令同时调度“行业研究员、数据分析师、文案撰写人”三个虚拟角色，完成从宏观检索、交叉比对到生成最终备忘录的链条闭环。
Original English
Step two, the intelligent hill. For 300 years, Isaac Newton convinced us that universe was a clockwork machine, predictable and certain. But in 1927, another scientist named Heisenberg shattered those classical beliefs. He showed that our universe exists only as a cloud of possibilities at quantum level. It was a profound shift. You and I have to make a similar shift when we use AI nowadays. The first trick is to stop treating AI like a calculator. We like to live in a world with clear rules. You type 2 + 2 into a calculator and you get four. Always. It's predictable. But AI is not a calculator. It's a probability engine. If you ask the same question to AI again, it'll give you a completely different answer. It'll happily make things up for you unless you ask it to verify. AI is brilliant on some days, confused on others, but on any given day, it refuses to admit that it doesn't know the answer. It loves to make things up. So you don't just ask AI the way you ask a normal human being. You have to architect your questions very carefully. Now most people use a tactic called zero-shot prompting. So for example, they would ask, "Give me the best new business idea." And of course, AI will dish out a response and tell you why it's the greatest idea in the world, but you're literally rolling the dice and looking to win. To get elite results though, you must climb the intelligent hill. There are four camps on the way. Each camp will show you a different way to work with AI. Our first camp is called one-shot prompting. When you prompt, give one clear example so the model doesn't guess blindly. So the prompt would look like, "Write a LinkedIn post about remote work. Use this specific post as a style guide." And so give it a post, give it an example, and paste that post in the prompt as a reference. And that simple act is already an upgrade than rolling the dice blindly. Second camp, few-shot prompting. Now here you give AI three or more examples so it can find patterns of style and substance and tone that you desire. Attach documents, links, data, or your prior work. This is called grounding the model. So basically it stops fantasizing and hallucinating and gets grounded to reality. Here's an example of a prompt. Here are the five of my previous presentations and now write a new presentation based on my tone of voice on topic XYZ. And here's a pro tip. Ask the AI to explain the pattern back to you first. That way AI is forced to articulate what it's doing and more importantly, you're forced to learn how your brain works. How did it come up with those patterns? Now you're being smart about being smart. Now let's move to the third camp. This one is called chain of thought reasoning. Again, fancy name, but the idea is simple. Ask the model to think long and hard before it responds. Your job is to slow AI down. Enforce explicit clarity by asking it to show its work. That's all there is. This is also a good way to reduce hallucinations, of course. So, let's say you're working on some report, and so you attach it and write a prompt that could look like this. Do not refine my research report yet. List the top three most impactful areas of improvement after we analyze it. Tell me why you think so and suggest how we address each. Think step-by-step. Show me your thinking for each step. That last line is the most important one. And our fourth and final camp is agents. According to Salesforce, AI agents help drive $67 billion in global sales during Cyber Week alone. So, agents are already here. The best way to think about agents is to think about who you would hire for a task. So, let's say if you wanted to hire a researcher, an analyst, and a copywriter. You can do that with a single agentic prompt that looks like this. Do deep research on trends on topic XYZ, analyze and cross-reference all the trends to find the three most important ones, and draft a one-page memo summarizing the findings. Now, what is actionable? Try this framework tonight. Open your favorite AI app and take any prompt that you were about to use. Just try to get to the next camp. That's how you start climbing up the intelligent hill. Remember, when you are dealing with a drunk genius, make sure you're the one driving the car. So, now, at this point, everything we've done has made you fast and efficient. You're delegating better, you're prompting smarter, you're moving up the hill, and there's less friction than before. And that's exactly where most people would stop. But, here's a plot twist. The top 1% go one step further. They slow things down deliberately. Why is that important? The trick that top 1% know is this. They know when to shift the gear because long-term intelligence isn't built through convenience, it's built through resistance. And that's why we need to go to step three,智能健身房：用主动阻力对抗思维萎缩
如果我们将 AI 视为心智的“轮椅”，那么我们终将面临失去独立行走能力的悲剧——这正是目前大多数人正在经历的思维萎缩（Atrophy: 由于缺乏认知负荷导致的大脑机能退化）。正如宇航员在太空的零重力环境中只需数月便会流失高达 20% 的骨骼肌一样，无摩擦的便捷生活本质上就是认知重力的真空，它正在加速我们大脑的退化。前 1% 的精英深谙此理：处理信息获取等低价值任务时，应利用 AI 最大化消除摩擦；但在进行思维的个人蜕变与深度升华时，必须反其道而行之，利用 AI 制造认知阻力。
在“智能健身房”中，AI 的角色不是替你举起杠铃的代理人，而是保障你安全的辅助保护员（Spotter: 健身中站在一旁提供安全保护，但不代替你出力的人）。基于渐进式超载（Progressive Overload: 通过逐步增加心理负荷来锻炼神经元连接的训练法）的逻辑，你可以构建一个由浅入深的四级心智抗阻训练体系：
- 第一级（高中生水平）：让 AI 对你刚学完的内容进行基础概念测试，检验识记。
- 第二级（大学生水平）：要求 AI 进行应用性考问，促使你建立跨领域的知识关联。
- 第三级（高管面试水平）：模拟严苛的面试现场，AI 将对你的方案进行逻辑闭环的深度盘问。
- 第四级（愤怒的老板水平）：让 AI 扮演一位挑剔且认为你完全没有做好准备的上司，发起极具挑战性的极限施压。
通过在重重阻力中不断修正错误并体验受挫感，大脑才能真正触发神经塑料性（Neuroplasticity: 神经系统根据环境刺激重新构建突触连接的物理机制），重塑出更强大、更有韧性的突触通路。
Original English
the intelligent gym. Most people use AI as wheelchair for the mind. And if you sit in a wheelchair when you can still walk, eventually, your legs stop working. Atrophy. And today, it's happening faster than at any point in human history. But, the top 1% use a very different principle. For information tasks, use AI to remove friction. For transformation tasks, use AI to add friction. When you go to a physical gym, we all know how muscles are built, right? Through resistance. You lift increasingly heavier weights to introduce wear and tear to your muscle fibers, so they break and they grow back stronger. That is called progressive overload. But, when it comes to our minds, we do the exact opposite somehow. We avoid resistance. We use AI to outsource our thinking. Write my LinkedIn post, fix my resume, summarize this book. That's like going to the gym and asking someone else to lift weights on your behalf. You know, when astronauts spend months in zero gravity, their muscles and bones atrophy dramatically, up to 20%. AI is like zero gravity for your thinking. No friction, no load, no growth. The intelligent gym is not about information. It's about transformation. For things where you need to be smart and capable, you can think of AI as your spotter. In any gym, a spotter doesn't lift the weight for you. They stand next to you and help you lift. They also make sure that you don't get crushed when you're lifting the weight. So, do the same with AI. Here's a concrete example. If you want to learn a concept, study it first yourself, and then go to your spotter, your AI. Paste the concept text, and then prompt AI. I need to master this concept. Quiz me on it. And now comes the most important part of your intelligent gym. Ask AI to apply progressive overload. Four levels. Level one, quiz me like I am a high school student. Level two, ask me questions like I am a college student. Level three, now grill me like you're interviewing me for an executive job. And level four, now challenge me like an irate boss who thinks I'm unprepared. So, that truly strengthens and deepens your understanding on that concept. So, now we have covered three key steps to learn how the top 1% become smarter by using AI. But, there is one internal adjustment that changes everything, and that is our final step.智能愚者：回归起点，重塑知性自我
真正阻碍我们智能跃迁的，往往不是无知，而是自我虚荣（Ego）。微软在 2014 年经历了一次极为震撼的文化变革：在萨提亚·纳德拉执掌微软之前，面对搜索与移动互联网等颠覆性浪潮的节节败退，内部政治盛行，每个人都极力掩饰自己的认知缺陷。萨提亚通过倡导将企业文化从“无所不知”（Know-it-alls）彻底转向“无所不学”（Learn-it-alls），赋予了所有人坦承“我不知道”或“我犯了错”的权利。这一轻装上阵的初学者心态（Beginner's Mind: 不带任何偏见与虚荣，像白纸一样重新观察世界的空杯心境），在短短十年内为微软赢得了超十倍的市值成长，使其成功迈过三万亿美元大关。
AI 为我们提供了一个绝对宽容、没有社会羞耻感的终极训练场。你不必担心在同事面前出丑，可以肆无忌惮地向 AI 连问三次“能否用更简单的方式解释？”，甚至要求其“把我当成十岁孩子来教学”。这种知行层面的返璞归真，虽然在起步时会带来智力层面的尴尬，但正如历史上所有伟大的探索者一样，唯有在今天有勇气扮演一个谦卑的“愚者”，才能在明天成为真正的“智者”。
智能的终极要义，绝非粉饰出一具无懈可击的外壳，而是宣告“虚饰的终结”。在充满不对称的美丽世界中，我们因不完美而独特，而与 AI 的深度交融，最终将引领我们跨越浩瀚的认知星海，返璞归真，全然接纳本真的自我。