把学习变成反馈回路:七条原则如何连成一套方法
先做一个真实任务,让问题暴露缺口;再用反馈纠错、用提取验收、用连接积累可复用的知识。
- 学习从真实任务开始:行动会把模糊的“不会”拆成下一项可学习的具体缺口。
- 搜索或问 AI 前先留下一个假设和一次测试;达到时间或安全边界后及时求助。
- 反馈用来纠错,闭卷提取用来验收,两者共同把“看懂了”推进到“能独立使用”。
- 被提取、应用并连接到旧经验的知识,才会降低下一轮学习成本。
你可能遇到过这种情况:一段教程播放时,每一步都能跟上;关掉视频,面对空白文档,手却停住了。书里的观点读起来很熟,合上书后,只剩下一句模糊的“我大概懂了”。
Hades 在 X 长文《The Principles of Learning Faster》中写了七条学习原则。他用两个人做对比:两人都花了 100 小时学习同一项技能,一个看完课程、读完书,仍然难以使用;另一个很早就开始做,犯错、取得反馈,最后把技能变成了职业能力。
差距来自学习过程中发生了什么。信息进入眼睛,只完成了接触;学习还需要把信息变成可以独立调出、用于解决问题、并能在下一次迁移的知识。
七条原则组成一条学习循环
原文依次谈到:在行动中学、缩短反馈回路、搜索前先挣扎、按需学习、用提取代替熟悉感、先连接再收集,以及学习会复利。把它们拆开记,很容易得到七句口号。沿着一次真实任务看,它们其实首尾相接。
你先选择一个略微超出现有能力的任务,动手后暴露缺口;不要立即搜索,先形成自己的判断;再针对缺口补少量知识,回到任务中重做;从结果、用户、老师或测试那里取得反馈;最后合上资料,复述这次学到了什么,并把它接到已有经验上。下一轮任务开始时,你已经拥有更多可调用的模式。
点选任一步查看该做什么、留下什么记录、何时进入下一步;或先闭卷回忆,再逐项核对。
真实任务
- 现在做什么
- 写明本轮结束时必须出现的可观察结果。
- 留下什么
- 一条具体完成标准。
- 何时前进
- 产出已经明确,可以立即开始动手。
展开八步行动提示
- 真实任务写明本轮结束时必须出现的可观察结果。一条具体完成标准。产出已经明确,可以立即开始动手。
- 先尝试先猜一个原因,只改变一个变量并观察结果。一个假设、一次操作和实际结果。完成一次有效测试,或达到时间与安全边界。
- 暴露缺口把“不会”缩小成当前阻塞任务的具体问题。失败位置、现象和已经排除的可能。能够提出一个范围清楚、可以搜索或请教的问题。
- 定向补充只补刚才暴露的知识,不扩张成新的课程清单。答案、依据和下一次检查方法。新信息已经能翻译成一个可执行改动。
- 回到任务把新信息放回原任务,再做一次。改动前后可观察结果的差异。结果发生变化,或出现了更具体的新缺口。
- 取得反馈比较目标、当前结果与下一步调整。偏差是什么,以及下一次具体改什么。反馈已经指向一项新的可执行动作。
- 闭卷提取合上资料,用自己的话重建问题、错误与证据。一段脱离原材料仍能复述的解释。没有页面提示也能说出方法和检查顺序。
- 连接旧知找一个旧案例,比较相同点、不同点与适用边界。一个可复用模式和一个不能套用的条件。这个模式已经能帮助下一轮更快定位问题。
判断这条循环有没有转起来,可以问两个问题:我能用已经知道的东西做什么?我多快能发现自己错了? 前一个问题把学习推进行动,后一个问题决定纠错速度。
这也解释了为什么“再看一门课”经常没有带来相应进步。课程可以增加输入,却未必触发尝试、反馈、提取和连接。循环没有转动,时间只积累在消费端。
先做:让问题决定学习顺序
原文把 Learn by Doing 和 Learn Just in Time 分成两条,实际使用时它们紧密相连。
一个人想做产品,常会列出一张宏大的预习清单:编程、设计、营销、销售全部学完再开始。六个月后,每门都懂一点,产品仍然不存在。反过来,先做一个最小版本,学习顺序会迅速变得具体:页面打不开,就补部署;用户看不懂,就改文案;没人愿意付钱,就去验证需求和报价。
写作也一样。读十本写作书,很难提前知道自己的文章究竟会卡在选题、结构、论据还是句子。第一篇真实文章会把这些问题逐一暴露。销售也是如此:背完框架仍可能在客户说“太贵了”时停住;真实对话会给出下一项练习——如何追问预算、价值和替代方案。
这里的“做”需要满足两个条件。任务要产生可观察的结果;难度只比当前能力高一点。任务过于简单,暴露不了新缺口;跨度过大,学习者只会同时遇到几十个问题,无法判断先修哪一个。
一项汇总 225 项本科 STEM 研究的主动学习元分析发现,主动学习条件下的考试和概念测验成绩平均提高 0.47 个标准差。这个结果支持主动参与的方向,但场景局限在本科 STEM 课堂,不能把它换算成“任何技能都能快 47%”。
再查:先留下自己的判断
原文第三条是 Struggle Before You Search。它在 AI 时代尤其容易被误解。
代码报错时,把整段错误交给 AI,复制修复方案,问题可能在一分钟内消失。下一次出现相似错误,你仍然不知道错误信息指出了什么、自己改过哪一项、哪个假设被否定。任务完成速度提高了,诊断能力没有同步增长。
先尝试的作用,是留下一个可以比较的思考轨迹。读错误信息,猜一个原因,改一个变量,观察结果。随后再看 AI 或搜索给出的答案,你能比较“我原来怎样判断”“答案增加了哪条证据”“我的推理在哪一步偏离”。外部答案由此进入自己的模型,而不是只成为一次粘贴操作。
Manu Kapur 对九年级数学学习所做的两项随机研究,为这种顺序提供了一个有边界的例子。学生先尝试解决新问题、随后接受教学,在概念理解和新问题迁移上优于先教学再练习的组;两组的程序性知识都较高。这种方法被称为生产性失败。研究中的“失败”经过设计,后面紧跟教学。它不鼓励学习者无限期卡住。
实用做法是设置一个尝试窗口:低风险问题先独立处理 5—10 分钟,至少写下一个假设和一次测试;达到时间上限、出现安全风险,或连续试验不再产生新信息时,立即求助。先尝试制造学习材料,及时求助完成纠错。
反馈纠错,提取验收
Shorten the Feedback Loop 讨论的是纠错。原文举了两名写作者:第一人一个月写完十篇才发布;第二人每写一篇就发布,观察读者在哪里离开、哪些观点被分享,再写下一篇。两人都完成十篇,第二人经历了十轮外部校正。
练习次数相同,反馈回路数量可能完全不同。重复一个动作 100 次,如果始终不知道错在哪里,熟练的也可能只是错误动作。
不过,“越快反馈越好”不能当成无条件规律。一项覆盖 435 项研究、994 个效应、超过 61,000 名学习者的反馈元分析得到平均效应 d = 0.48,同时发现结果高度分散;反馈包含的信息会显著影响效果。只说“做得不好”几乎无法指导下一次行动。有效反馈至少要帮助回答:目标是什么,当前结果偏在哪里,下一步改什么。不同任务也可能适合立即或延迟反馈。
Retrieval Beats Recognition 负责另一件事:验收知识能否在材料消失后被调出。
重读时,页面上的句子会替你提供线索,于是大脑产生“我知道”的熟悉感。合上书,用自己的话写出三个要点,线索消失,缺口才会出现。讲课视频看完后不翻笔记解释一次,学完概念后用它解决一个新例子,都在做同一件事:要求大脑重建,而不是再次识别。
一篇针对真实课堂的提取练习系统综述筛选近 2,000 篇摘要,纳入 50 个实验、总样本 5,374。49 个可计算效应中,28 个显示中等或较大收益。研究者也提醒,各实验的学科、测验形式、反馈和实施方法差异很大。提取练习有效,具体做法仍需适配任务。
反馈和提取不能互相替代。反馈从外部指出偏差;提取从内部检查你能否重新生成知识。一次完整练习最好同时拥有两者:做完获得纠错信息,隔一段时间再闭卷复述或重做。
连接旧知,复利才会发生
原文最后两条是 Connect Before You Collect 和 Learning Compounds。
孤立事实只能回答它原本对应的问题。建立连接后,一个概念会获得更多入口。学过心理学的人再看营销,会注意动机、注意力和社会认同;学过营销再做销售,会理解为什么同一个卖点对不同人效果不同;写作训练又会迫使这些模糊认识变成可以解释的结构。
这也是经验丰富的人能更快理解新事物的一部分原因。他看到一个新商业模式,会把它与过去见过的获客方式、成本结构、失败案例和用户行为对齐。新信息不必从零开始组织,已有模式提供了落点。
“学习会复利”因此是一个有条件的比喻。收藏数量本身不会复利。知识经过提取、应用和连接,才可能降低下一次理解的成本。连接还可能出错:旧经验会让人过早套用熟悉框架,忽略新问题的差异。每次连接都应补一个问题——这次情境与旧案例哪里相同,哪里不同?
今天就跑一轮:30 分钟练习法
这套方法不要求重做整套学习系统。今天选一个真实问题,用 30 分钟跑一轮即可。
写下本轮真实任务,再按顺序勾选五步。进度只记录是否完成,不代表学习效果。
- 3 分钟:定义产出。 写下本轮结束时要出现的可观察结果,例如“让表单成功提交一条数据”或“写出一个读者能复述的开头”。
- 10 分钟:独立尝试。 不查答案,记录至少一个假设、一次操作和结果。
- 7 分钟:定向补充。 只搜索刚才暴露的缺口;使用 AI 时,先提供自己的判断,再要求它指出错误、证据和下一次检查方法。
- 7 分钟:回到任务重做。 把新信息转成新的操作,观察结果是否改变。
- 3 分钟:闭卷复述。 不看资料写下:问题是什么,我原来错在哪里,什么证据改变了判断,下次遇到相似问题先检查什么。
完成标准也很直接:你得到一个可观察结果,一条能指导下一次的反馈,以及一段脱离原材料仍能复述的解释。第一次没完成任务并不等于这一轮没有学习;只要缺口比开始时更明确,下一轮已经获得了准确入口。
Hades 的七条原则最终指向同一件事:让信息尽快进入行动,让行动产生反馈,让反馈经过提取与连接变成可复用的知识。学习速度由循环的质量和次数共同决定。
The Principles of Learning Faster.

Every day, you're learning something. From books. Conversations. Podcasts. Mistakes. Articles. Videos. The people around you.
But consuming more information doesn't necessarily mean you're learning more. Two people can spend the same 100 hours learning the same skill and walk away with completely different results.
- One watches every tutorial, reads every book, and still struggles to use what they learned.
- The other starts building, makes mistakes, gets feedback, and becomes good enough to turn that skill into a career.
The difference isn't always intelligence. It's how they learn.
Learning is one of the few skills that improves almost every other skill you decide to acquire.
Get better at learning, and you can become a better writer, founder, designer, engineer, investor, or creator faster. You can adapt as industries change, pick up new skills as opportunities arise, and spend less time consuming information you'll forget a week later.
Yet almost nobody is taught how actually to learn. We're taught to read, remember, take notes, and sit for exams. However, the real world values the ability to understand quickly, apply what you know, learn from feedback, and connect new ideas to what you already understand.
That's what this article is about. Seven principles for learning almost anything faster:
I. Learn by Doing
There's a strange trap that happens when you're trying to learn something new. Learning feels productive. You watch another tutorial. Read another book. Save another thread. Take another course. And because you're constantly consuming information, it feels like you're making progress.
But knowing how something works and being able to do it are two completely different things.
- You can read ten books about writing and still struggle to write a great article.
- You can watch fifty hours of coding tutorials and still have no idea how to build your first product.
- You can study sales for months and freeze the moment a real customer says, "It's too expensive."
At some point, more information stops being the answer. You have to do the thing. The first article teaches you things no writing book can. The first product you build exposes gaps no tutorial could predict. The first sales call gives you objections you never thought to prepare for.
That's because action turns abstract knowledge into real problems. And real problems tell you exactly what you still need to learn.
Instead of spending three months learning everything about marketing, try getting your first customer. You'll quickly discover which parts of marketing actually matter.
Instead of watching another ten hours of coding tutorials, try building something slightly beyond your current ability. Every time you get stuck, you've found the next thing you need to learn.
This creates a much faster cycle:

So whenever you catch yourself consuming more information, ask: "What can I do with what I already know?" Because at some point, the fastest way to keep learning is to stop learning and start doing.
II. Shorten the Feedback Loop
Two people can practice the same skill for the same amount of time and improve at completely different speeds. The difference is often how quickly they find out what they're doing wrong.
Imagine two people learning to write.
- The first spends a month writing ten articles before publishing anything.
- The second writes one article, publishes it, sees where people stop reading, notices which ideas get shared, gets feedback, and uses all of that information to write the next one.
After ten articles, both have technically done the same amount of work. But only one of them has gone through ten feedback loops.
This applies to almost anything.
A founder launches a feature, watches how users interact with it, and improves it. A salesperson hears the same objection three times and changes the pitch. A developer ships something, breaks it, fixes it, and understands the system better than before.

The shorter that cycle becomes, the faster you can correct your mistakes before they turn into habits.
That's why spending more time practicing doesn't automatically mean you're learning faster. If you repeat the same mistake 100 times without feedback, you're not getting 100 repetitions better. You're getting better at repeating the mistake.
So when you're trying to improve at something, don't only ask: "How much am I practicing?" Ask: "How quickly can I find out if I'm wrong?" Because the speed of learning is often limited by the speed of feedback.
III. Struggle Before You Search
We live in a time where almost every answer is seconds away.
- Don't know something? Google it.
- Can't solve a problem? Ask ChatGPT.
- Code doesn't work? Paste it into Claude.
This is incredibly useful. But it also makes it easy to skip one of the most valuable parts of learning, struggling with the problem yourself.
Imagine you're learning to code and run into an error. You immediately paste it into AI, copy the solution, and move on.
The problem is solved. But did you actually learn how to solve it?
Now imagine spending ten minutes trying first. You read the error. Form a hypothesis. Change something. Break something else. Try again.
Even if you eventually ask AI for the answer, something has changed. You now have your own ideas to compare against the solution. You understand where your thinking went wrong. The same applies almost everywhere.
- Before looking at the answer, try solving the problem.
- Before reading someone's analysis, form your own opinion.
- Before asking AI to explain something, try explaining it yourself.
The goal isn't to make learning unnecessarily difficult or avoid useful tools. It's to give your brain a chance to work on the problem before someone else does the thinking for you.
So when you get stuck, resist the instinct to search immediately. Give yourself a few minutes first. Think, guess, test, then search.
Because getting the answer faster and learning faster aren't always the same thing. Sometimes the struggle you keep trying to avoid is the exact part that teaches you the most.
IV. Learn Just in Time
A common mistake when learning something new is trying to prepare for every possible problem before you've encountered any of them.
You want to build a product, so first you decide to learn programming, design, marketing, sales, and everything else you might eventually need. Six months later, you know a little about everything. But you still haven't built anything.
A faster approach is to reverse the process. Start with the problem. Then learn what the problem requires.
- If you're building your first product, you don't need to understand everything about programming. You need to know enough to build the next feature.
- If you're learning sales, you don't need to memorize every sales framework. Talk to ten potential customers. Their questions and objections will quickly show you what you need to improve.
- If you're learning to write, don't spend three months studying writing. Publish something. See where it fails. Then learn how to make the next one better.
This changes learning from an endless collection of information into a response to real problems. And there's another advantage, information becomes much easier to understand when you immediately have somewhere to use it. Problems have a surprisingly good way of designing your curriculum for you.
V. Retrieval Beats Recognition
There's a simple way to discover how much you've actually learned. Read a chapter of a book. Close it. Now explain the three most important ideas without looking.
Suddenly, something that felt completely clear while you were reading becomes surprisingly difficult to explain. That's because recognizing information isn't the same as being able to retrieve it.
When you reread a page, rewatch a video, or look through your notes, everything feels familiar. Your brain sees something it has encountered before and gives you the feeling,
"Yeah, I know this."
But familiarity can create an illusion of knowledge.
The real test comes when the information disappears and you have to reconstruct it yourself. That's why testing yourself can be more valuable than simply rereading something again.
- After finishing a book, write down what you remember.
- After watching a lecture, explain the idea without your notes.
- After learning a new concept, try teaching it to someone else.
You'll immediately discover the gaps that passive consumption was hiding. And once you know where those gaps are, you know exactly what to revisit. So don't measure learning by how familiar something feels.
VI. Connect Before You Collect
It's easy to mistake collecting information for building knowledge. You finish a book. Save another article. Bookmark a thread. Add a podcast to your queue. Your library keeps growing.
But your understanding doesn't necessarily grow with it. The most useful knowledge rarely exists in isolation. It becomes valuable when you can connect it to something you already understand.
- Learn psychology, and suddenly marketing makes more sense.
- Learn marketing, and you start seeing why certain sales strategies work.
- Learn sales, and you begin understanding why some products spread while others struggle.
- Learn writing, and you become better at explaining all of them.
Each new idea gives you another way to understand the ideas you already have. That's why two people can read the same book and get completely different value from it.
- One sees seven interesting ideas.
- The other connects those ideas to years of experience, previous books, conversations, mistakes, and problems they're currently trying to solve.
The information is the same. The number of connections isn't.
A smaller number of deeply connected ideas can be far more useful than thousands of facts you've collected but never learned to use.
VII. Learning Compounds
Learning gets easier the more you learn.
At the beginning of any field, almost everything is unfamiliar. Every concept requires an explanation. Every new idea feels like another isolated piece of information you have to remember.
But eventually, something changes. New ideas stop arriving alone. They start connecting to things you already understand. Your existing knowledge gives every new idea somewhere to land.
That's why an experienced founder can sometimes understand a new business model in twenty minutes while a beginner needs days.
They're not necessarily more intelligent. They're connecting the new idea to hundreds of patterns, mistakes, conversations, and concepts they've already collected. This creates a compounding effect.
What you learn today doesn't just give you knowledge today. It can make what you learn tomorrow easier to understand.
Over years, the difference becomes enormous. Someone who keeps learning isn't simply adding one new idea after another. They're building a larger network where every new idea can make the existing ones more useful.
The better you become at learning, the easier it becomes to get better at almost everything else.
Now here's what I actually want you to take away from all of this.
Learning faster is about getting better at turning information into understanding, and understanding into action. Do this long enough, and something interesting happens. Learning itself becomes easier.
Every skill you build makes another skill easier to understand. Every problem you solve gives you experience for the next one. Every useful idea becomes a foundation for ideas you haven't discovered yet.
Get better at learning, and you get better at becoming better. ❤️
