taxue-learn: The Taxue Learning Methodology
Learning is not receiving information, it is changing behavior.
Thinking Foundations
- Feynman Learning Technique: You only truly understand it if you can explain it clearly without jargon.
- Dewey's "Learning by Doing": It's not learn first then do, it's learn by doing. Output forces input.
Four-Step Internalization Method
1. Inbox + Knowledge Atoms (Input → Organize)
Effectiveness > Efficiency. Pursue "what to keep", not "how much you've read".
Capture: All information first goes into the inbox, judge in 3 seconds: noise delete / signal archive / ambiguous hold. Clean up weekly, don't let it pile up for more than 7 days.
Organize: Don't store the original text, store the smallest reusable unit:
核心观点（1句话）
关键细节（3-5条）
适用场景（什么时候用）
个人理解（和我的经验关联）
Every atom must be able to answer "when to use it", otherwise it's garbage.
Key Question: When will this material be used? Who is it useful to?
2. Interactive Internalization (Mastery)
| Method | Applicable Scenario | Operation |
|---|---|---|
| Socratic Questioning | There is a viewpoint that "sounds right" | Raise counterexamples, push to the boundary, clarify definitions |
| Case Analysis | Abstract concepts need to be grounded | Real scenarios, ask "what would you do" |
| Feynman Restatement | Verify whether you truly understand | Without jargon, explain it to a layperson |
| Level-Clearing Quiz | Quick acceptance check | Judge → Choose → Apply → Counterexample |
Internalization Check(only counted as mastered if all pass):
- Able to explain clearly without jargon
- Able to give 3 examples from different fields
- Able to state "when it does not apply"
Core Rule: Every argument must be supported by underlying logic (claim + data + warrant + qualifier). A viewpoint for which you cannot write the "warrant" — delete it or re-examine it.
What was learned (no change = just looked at it)：
入场立场：「我原来以为……」
现在立场：「现在我认为……」
改变来源：「是 {哪个论点/案例} 让我改变的」
3. Cognitive-Dissonance-Style Iteration (when you think you understand but actually don't)
Four rounds of breakdown from "jargon dependence" to "true understanding":
| Round | Operation | Signal |
|---|---|---|
| 1. De-jargon | Remove all technical jargon, say it in the most plain words | Getting stuck where jargon is used = blind spot |
| 2. Add Cases | Give a specific example | Can't give an example = didn't truly understand |
| 3. Add Analogies | Use a completely different field as a metaphor | Can't analogize = understanding isn't deep enough |
| 4. Teach Others | Have AI play a completely ignorant person and keep questioning | Getting stumped = there are still gaps |
KeyGetting stuck is the most valuable signal—it means you've found a knowledge blind spot. Don't skip it; go back to the basics and rebuild.
4. AI Learning Mentor (executable tool)
When you want to internalize a concept through conversational learning, copy this prompt directly:
你是一位 AI 学习导师。请按以下四步教我[主题]：
1. 适配我：先问我已经知道什么、想解决什么问题。根据我的回答定制内容。
2. 让我懂：每次只讲一个核心点，讲完让我复述或举例。没懂就去掉所有类比和术语，只讲它最基本的事实是什么，重新构建，直到掌握。每次讲完一个点就停，等我回应。
3. 教我用：给我一个贴近我实际场景的小任务，让我马上应用。我汇报后你判断是否过关，没过关继续拆解练习。
4. 教我教学：该点过关后，让我解释给一个完全不懂的人听（你扮演那个人），围绕「为什么」和「比如」追问挑漏洞，直到我能讲清楚。简单的知识点可跳过此步。
整个主题学完后，问我：下次遇到类似问题，你会怎么自己拆解？
像苏格拉底一样多提问少灌输。现在从适配我开始。
5. System output (solidification)
When you encounter a recurring problem, build a system. Use the six-step compilation method (see taxue-build) to write your experience into a reusable framework.
If it can be solidified, don't repeat it.
Next-step suggestions (condition-triggered)
After the learning guidance is given, judge based on the result whether to recommend a next step.Don't recommend every time, only say a sentence when the result clearly points to another skill.
| Result condition | Recommended wording |
|---|---|
| During learning, you get stuck on a concept and repeatedly fail to understand it; the root cause is a cognitive blind spot | "You haven't seen through this concept yet; rote memorization is useless. Use /taxue-insight to dig into the underlying logic." |
| The learning method is decided, but it needs to be broken down into an executable schedule | "You have the method; next, break it down into daily learning steps. Use /taxue-breakdown。」 |
| You always forget what you've learned and need a reusable internalization process | "Your learning process is worth solidifying into a template. Use /taxue-build。」 |
DO NOT
- Need to solve a specific problem (not the learning method itself) →
taxue-solve - Need to solidify a process/build a tool →
taxue-build - Just looking up information/searching → use web_search directly, no need to go through learn
taxue-learn v2.8