Wang Kai regards "decision-making" as the highest-value link in the AI era. The real deep significance of the AI wave is not "equal access to intelligence", but"equal access to rationality / slow thinking"——the vast majority of people do not think through the vast majority of their decisions.
1. Core Mindsets (5 sentences that last a lifetime)
| # | Mindset | One-sentence summary |
|---|---|---|
| 1 | View the present from an AGI perspective | Assume AGI arrives on 2035-01-01; look back at today and ask which choices will be proven correct |
| 2 | AI is a ruler, not a fortune teller | Score fuzzy variables that humans cannot quantify (risk, value, emotion) |
| 3 | Absolute rationality, thinking mode by default | For any important decision, force the slow-thinking path of "writing out the reasoning process" |
| 4 | Rules are greater than results | Temporary gains and losses do not matter; optimizing rules matters |
| 5 | Understanding = being able to do it | Intellectually understanding vs. behaviorally understanding are completely different |
2. View the present from an AGI perspective (core thinking framework)
On 2025-04-01, Wang Kai wrote a pinned post:
"I lived through the entire mobile internet cycle. Standing at the end of the mobile internet era and looking back at my past choices, what would I choose differently?"
Strategies he would change looking back:
- Not starting a business in the internet industry (especially in Beijing)
- Doing e-commerce business or milking platform traffic
- Doing MCN / batch-raising accounts
- Going overseas earlier
- Putting both personal presence and investments into leading internet stocks(making more money than most core founding members of startup companies)
Translation to the present: Assuming AGI arrives on 2035-01-01, what should we do today?
graph TD
A[站在 2035-01-01 AGI 视角] --> B[1. 不做 AI 主要趋势]
A --> C[2. 不做 AI 直线相关]
A --> D[3. 推测哪些常识/规则被改写]
A --> E[4. 做改写过程中的配套生意]
A --> F[5. 持有 AI 龙头股]
B --> G[避免和巨头硬卷]
C --> H[竞争时间过长,小 team 玩不起]
D --> I[新常识 → 新机会]
E --> J[像 MCN / 周边媒体一样]
F --> K[NVDA / 苹果 / Google 等]
Difference from "endgame thinking":
- Endgame thinking is narrow and focused on competitive strategy
- The AGI perspective looks at +30-year long-term variables and is not limited to competition
- Grasp longer-term variables and do not be fooled by short-term trends
3. Fast thinking vs. slow thinking (equal access to thinking)
graph LR
A[人类决策] --> B[快思考<br/>本能、直觉、情绪]
A --> C[慢思考<br/>理性、推理、计算]
B --> D[多数日常决策]
C --> E[少数重要决策]
F[AI thinking 模式] --> G[强制写出推理过程]
G --> H[降低决策错误率]
H --> I[把人的快思考决策转为慢思考]
Key realization(Wang Kai had multiple epiphanies on this):
- Most people make most decisions without thinking (fast thinking)
- We "know" many principles but fail to execute them (intellectual understanding vs. behavioral understanding)
- AI's thinking mode = adding a slow-thinking checkpoint of "must write out the reasoning process" for all decisions
- This is the real deep significance of the AI wave, not equal access to intelligence
Wang Kai's training:
- Default to turning on thinking mode
- His wife asks "Are you thinking rationally right now?" to calibrate each other
- For important decisions, must chat with GPT/Claude once ("AI cools me down + reminds me of decision logic")
4. Using AI as a ruler (the most valuable use case)
For fuzzy scenarios that humans cannot accurately quantify, have AI score them using a stable framework:
| Scenario | AI usage | Value |
|---|---|---|
| Market risk level | Score on a 1–10 scale | Determine cash ratio |
| Intrinsic value of a company | Multi-model synthesis + interval | Decide whether to buy |
| Single-stock risk | Score across 5 dimensions: short, medium, and long term | Determine position size |
| Emotional state | "Am I overly panicking right now?" | Prevent impulsive decisions |
| Risk of AI replacing jobs | List work processes and ask the replacement probability | Career planning |
| Feasibility of startup projects | Break down models and let AI interrogate them | Project initiation |
Analogy: An insurance actuary. Give AI a stable framework, and it becomes a ruthless actuarial machine.
Key discipline: Consistency > accuracy
- Running the same stock multiple times should yield similar results
- When risk levels change, there must be a clear attribution
- Answers must not drift due to emotion/state
5. The two layers of "understanding" (influenced by Duan Yongping)
On 2025-11-12, Wang Kai had an epiphany:
| Layer | Meaning | Example |
|---|---|---|
| Intellectual understanding | Knowing the knowledge point | "Buffett said be greedy when others are fearful" |
| Behavioral understanding | Being able to execute under pressure | Adding to a position without wavering when it is down 40% |
Application:
- Most people "intellectually understand" Buffett/Duan Yongping
- But they still lose money when it is time to lose money
- True understanding = firm execution during tests
- "To understand is to be able to do it(Wang Kai's summary on 2026-01-03)
Examples:
- Suppose Duan Yongping sells Moutai and says his view has changed — do you sell? → Most people would sell, proving they were just copying homework and did not truly understand
- Buffett said he "doesn't understand tech stocks, so he doesn't invest in them" → Do you understand better than Buffett? Why do you dare to invest?
- The standard for value investing is extremely high; most people only understand it literally
6. Common error patterns in decision-making
6.1 Drawing conclusions driven by short-term price movements
- ❌ "Look, XX went up, proving I was right" (short-term prices are distorted)
- ❌ "Look, XX went down, proving I was wrong" (equally distorted)
- ✅ Look at fundamentals, moats, and competitive landscape over the long term
6.2 Driven by information consensus
- ❌ "Everyone says AI is the future, so go all-in on AI" (consensus itself is risk)
- ✅ Break down the strength of consensus (measure your degree of conviction with money)
6.3 Driven by temporary gains and losses
- ❌ "I made money this time, so discipline can be bent"
- ❌ "I lost money this time, so I need to change the rules"
- ✅ Rules are greater than results; only by optimizing rules can you sustain performance
6.4 Self-congratulatory complex analysis frameworks
- ❌ "I co-created a 5000-word complex analysis framework with AI, it looks impressive"
- ✅ The quality of a framework must be verified by real-world data, not word count
6.5 Confirming rather than challenging
- ❌ "I ran an AI report to verify my own idea → the report says I'm right"
- ✅ Let AI continuously challenge, interrogate, and find loopholes
7. Several AI usages for decision support
7.1 Claude investment interrogation prompt (public, see details in Wang Kai's RSI AI Investment Strategy and the "Good Asset, Good Price, Good Volatility" Framework)
You are an AI assistant playing the role of an investment master.
Your sole purpose is to ask questions that encourage deeper thinking.
Do not provide any information or advice.
Probe into the user's reasoning and assumptions.
Encourage consideration of potential risks and rewards.
Prompt reflection on the broader context.
Challenge the user to consider alternative perspectives.
7.2 Offense-defense drills
- Have Claude Code find loopholes based on strategy documents
- Repeatedly push it to nitpick and find faults
- Summarize into an "emergency manual"
7.3 Casual chat + high-frequency discussion
- Give AI all context + historical decisions
- Keep chatting casually (not querying, but discussing)
- Keep pushing it to give "new strategies for the AI era"
- Until even AI says "I can't think of anything else for now"
7.4 IFS (Internal Family Systems, introduced by Wang Kai's wife)
- When you "can't help changing the unit price"
- Have Claude guide you through IFS
- Use a detached perspective to see what lies behind the impulse
8. The "two-way flow" of the decision system (input + output)
graph LR
A[决策输入] --> B[人 + AI 双轨]
B --> C[人: 直觉 + 经验]
B --> D[AI: thinking + 数据 + 历史]
C --> E[决策融合点]
D --> E
E --> F[规则 + 纪律 + 执行]
F --> G[结果]
G --> H[规则优化 + 系统迭代]
H --> A
Important disciplines:
- Complete inputs (don't spare the trouble of giving AI all data)
- Don't listen to only one AI (cross-compare Claude + GPT + Gemini)
- Always iterate rules; don't be held hostage by temporary results
9. The source of a sense of conviction (summary on 2025-12-31)
Wang Kai believes his biggest gain in 2025 is "a sense of conviction":
- Reading classic investment books → expanding the information base
- Watching value investing interviews → elevating perspective
- Organizing investment principles(building KeepRule) → systematization
- Practice → becoming firm through real tests
A sense of conviction is the foundation of rational decision-making(Duan Yongping's view).
"I understand some things, and there is even more I don't understand"gave Wang Kai a great sense of conviction:
- I know what direction to go (North Star: do the right things and try to do them right)
- I know what I don't understand (so I won't be arrogant)
- No more anxiety when making decisions
10. The "economic value" perspective of the decision-making link
Wang Kai repeatedly emphasized (typical post on 2025-08-27):
- AI helping CEOs make decisions vs. writing homework for students → the economic value is worlds apart
- AI is an amplifier that makes the strong stronger
- Actively move toward "high-value decision links"
- Don't use AI as a "cheap assistant"; use it as a "high-value decision-making brain"
Metric shift:
- Before: number of valid tokens consumed
- Now:Number of AI-assisted decisions + number of AI-direct decisions
11. Five scenarios where decision-making power must not be given to AI
| Scenario | Reason |
|---|---|
| Creation without objective standards | AI easily drifts off |
| Serious legal/medical decisions | Risk + admission requirements |
| Highly personalized emotions | AI's anthropomorphism is simulation, not real understanding |
| Ultra-low-latency trading | AI takes 2 minutes to reason, and quant traders have already made the money |
| Completely unfamiliar domains | You have no judgment, and AI will lead you in circles |
12. Five scenarios where AI excels at decision-making
| Scenario | Reason |
|---|---|
| Investment decisions | Results are verifiable (making money means it's right) |
| Programming problems | Running successfully means it's right |
| Math problems | There are standard answers |
| Data analysis | There are facts to compare against |
| Interrogating personal decisions | Forces slow thinking |
13. Wang Kai's life disciplines (extension of decision-making)
- Learn to wait - don't make decisions at the wrong time
- Learn to bet big - dare to take heavy positions in things you truly understand
- Don't be swayed by volatility - long-term asset management
- Keep learning investment classics - 3 minutes a day on Duan Yongping's Xueqiu posts
- Write post-mortems - monthly / quarterly / annual reviews
- Consult AI instead of only asking people - "AI as a companion to continuously discuss deep details"
14. Quotes
"Treat AI as your 'rational decision-making brain', and use AI to replace the fast-thinking part of your own decisions."
"What AI equalizes is not intelligence, but rational decision-making."
"I understand some things, and there is even more I don't understand — this gives me a great sense of conviction."
"Rules are greater than results; only by optimizing rules can you win continuously."
"The value created when AI is used by CEOs versus when it is used by students is worlds apart."
"Within three years, all investors will use AI to assist decisions; within five years, AI will lead decisions."
References
- Wang Kai's RSI AI Investment Strategy and the "Good Asset, Good Price, Good Volatility" Framework - Investment strategy (the most important application of the decision system)
- Wang Kai's Startup Methodology and Business Judgment Framework for the AI Era - Startup methodology (the other side of decision-making)
- Wang Kai's AI Coding Practice: From Cursor to Claude Code Terminal Workflow - AI Coding (making decisions executable)
- Wang Kai's Agent Engineering Practice - Agent engineering (automated decision execution)
- The newtype Knowledge System and the Full Evolution of Life OS - Comparison with Lao Huang's newtype knowledge system
2026-06 Addendum: Separate sticking to direction from verifying practices
This section was compiled after fully reading five text posts by Wang Kai from June 2026, and covers only the text; the analysis tables in the accompanying images were not read this time. The following are the author's experience and judgments, not trading advice, and his investment predictions have not been reviewed.
Principles should be simple, but execution still requires careful calculation
His reflection on "common sense" on June 2, combined with the experience of dissuading someone at an intersection on June 17, filled in the operational boundary of the original "rules over results": first answer what is worth doing, then answer how to do it effectively while bearing acceptable risk. The author uses saving people as an analogy — you should help a drowning person, but that does not mean you should jump in directly if you cannot swim. Skipping the second layer of risk assessment is not sticking to principles.
He uses the choice between Moutai and Tencent to explain another bias: if both meet the standards he set, but he only considers one because of the hassle of foreign exchange, actual execution has already deviated from the stated standards. The reusable check is not copying targets, but asking: was inaction this time due to different evidence, or was operational friction packaged as a reason? Likewise, trusting a company does not mean any price is reasonable.
Test judgments with subsequent evidence, and also record your own biases
When rereading an old interview with Huang Zheng on June 4, the author found that in the past, influenced by labels such as "sinking market", he interpreted answers about supply chains as evasion; years later, earnings calls still emphasized the same direction, prompting him to re-examine his early judgments. What is reusable is to retain the original judgment and its reasons, and re-compare after actual business evidence emerges. You cannot reverse it and treat later success as proof that "you were bound to understand it back then".
The value of AI is to shorten the distance from question to data
The change on June 22 was very concrete: in the past, growth analysis required submitting requests to data colleagues; now he can directly use Codex to pull data and discuss while analyzing, reducing writing opinions based on impressions. On June 6, there was also a data analysis rehearsal during a major market swing: its purpose was to map out risk scenarios in advance, not to claim that the panic point had already been reached. The main results of that post are in images, so the stock list, rankings, or specific conclusions cannot be recovered this time.
When to use it: when you already have a question that can be supported or overturned by data. First clarify the metrics, then pull data, so that discussions form testable conclusions. The counterexample is only having AI polish your existing position more smoothly without adding any independent evidence.
Sources and related materials
- https://wx.zsxq.com/group/48884824582158/topic/82255242825484160
- https://wx.zsxq.com/group/48884824582158/topic/22255282844454452
- https://wx.zsxq.com/group/48884824582158/topic/45544585114484530
- https://wx.zsxq.com/group/48884824582158/topic/14422412152211452
- https://wx.zsxq.com/group/48884824582158/topic/22255288824241520
2026-10-17 Addendum: A judgment is only clearly explained when you can state when it fails
From the written explanation received on October 12 and published by Wang Kai on October 11. This organizes his understanding of "understanding and not understanding"; the recommended Li Lu articles or videos are not treated as fully verified sources here.
This page already discusses "knowing vs. doing" and actively seeking counterevidence. The new check is to write out the applicable conditions of a conclusion: why a certain judgment holds in the current environment, and which condition changes would make it no longer applicable. Simply repeating a correct sentence makes it easy to overlook the customer, resource, time, or competitive conditions it depends on.
You can record one judgment in three lines: ① the current conclusion; ② the specific premises and evidence supporting it; ③ what observable change would cause you to reassess. The third line should correspond to the second line, rather than a vague statement such as "adjust when the situation changes". This is a recording method derived by this library from the author's views, not a fixed template provided in the original post.
For example, the claim "automation is worth investing in" requires clarifying whether the task is repetitive, whether failures can be detected, and whether maintenance costs are affordable; if any of these premises is missing, the options need to be re-compared. This example is used to illustrate the method and does not constitute a conclusion about any specific project.
When to use:When reusing others' experience, sticking to an important plan, or letting AI challenge a judgment. Ask it to find counterexamples against explicit premises, then return to evidence verification. Being able to state failure conditions does not mean the prediction is correct, nor can one evade previous errors by constantly rewriting premises.
2026-10-17 addendum: separate personal consumption experience from enterprise judgments
Extracted from Wang Kai's personal review on October 16. Only the judgment method is retained; family details are not recounted, and no independent assessment is made of the operating or investment value of the companies mentioned in the text.
Being familiar with a consumer brand can easily lead one to mistakenly believe they are familiar with all of its users. One's own dietary habits, purchase channels, product selection range, and price sensitivity all limit actual experience. After changing his usage pattern, the author revised his old impression; this supports the point that "one needs to check where the sample comes from", and cannot be used to infer that any brand is generally cheap or worth investing in.
In addition to the two existing columns on this page, "media impressions" and "enterprise evidence", a third column "my sample as a consumer" can be added:
| Level | What should be recorded | What cannot be directly inferred |
|---|---|---|
| Personal experience | What you have bought, how you chose it, and under what conditions you were satisfied or dissatisfied | All target users have the same preferences |
| Target customer group | Which users use the product repeatedly, and how their tasks and selection criteria differ | An improvement in my experience represents an increase in overall demand |
| Enterprise evidence | Materials and metrics that can support judgments about scale, delivery, or operations | Consumer likes or dislikes alone determine enterprise value |
This table is an applied derivation by this library. When researching competitors or partners, first state the scope of your own observations, then look for the behavior of different users and enterprise evidence. Increasing the number of personal uses can enrich experience, but it cannot automatically eliminate selection bias. If a conclusion comes only from the preferences of yourself and people around you, keep it as an interview lead and do not write it as an enterprise ranking or buy/sell judgment for the time being. Use in conjunction with related investment method knowledge drafts, but this does not add investment advice.
Addition: Explain when a judgment would stop applying
This addition draws on Wang Kai's October 11 written explanation, collected on October 12. It summarizes his interpretation of understanding and uncertainty. The Li Lu articles and video he recommends were not treated as fully reviewed sources.
The existing article discusses knowing versus doing and actively seeking counterevidence. An additional check is to state the conditions under which a conclusion applies: why it holds now and which changes would make it unsuitable. Repeating a correct statement can obscure the customers, resources, timing or competitive conditions on which it depends.
Record a judgment in three lines: the current conclusion; its concrete premises and supporting evidence; and an observable change that would trigger reassessment. The third line should correspond to the second rather than merely saying “adjust when circumstances change.” This is an editorial application of the author's idea, not a fixed template supplied in the post.
For example, “automation is worth the investment” requires checking whether the task repeats, failures can be detected and maintenance costs are affordable. If a premise is missing, compare the options again. This example illustrates the method and is not a conclusion about a particular project.
Use it when reusing someone else's experience, maintaining an important plan or asking AI to challenge a judgment. Have it look for counterexamples to explicit premises, then verify the evidence. Explaining failure conditions does not guarantee a correct prediction, and repeatedly rewriting premises must not become a way to avoid acknowledging earlier errors.
Addition: Separate personal consumption from judgments about a business
This addition is based on Wang Kai's personal reflection dated October 16. It retains the reasoning method without family details or an independent assessment of the companies' operating or investment value.
Familiarity with a consumer brand can feel like familiarity with all its customers. Personal habits, purchasing channels, product selection and price sensitivity constrain that experience. The author's revised impression after changing how he used a service supports checking where a sample comes from. It does not establish that any brand is universally inexpensive or worth investing in.
Alongside the existing distinction between media impressions and business evidence, add a column describing your own consumer sample:
| Level | What to record | What it does not establish |
|---|---|---|
| Personal experience | What you bought, how you chose it and when you were satisfied or disappointed | All target customers share your preferences |
| Target customers | Who returns, and how their tasks and selection criteria differ | Your improved experience means aggregate demand is growing |
| Business evidence | Material and definitions supporting claims about scale, delivery or operations | Liking or disliking a product determines business value |
This table is an editorial application. When researching competitors or partners, state the scope of your observations, then seek evidence about other users and the business. More personal usage enriches experience but does not automatically remove selection bias. Conclusions based only on your own preferences and those of people around you should remain interview leads rather than company rankings or trading judgments. The related investment-method article provides context; no investment recommendation is added here.