Source: My AI Age Personal Capabilities System

♪ Take notes ♪The rules of honesty at this station. This is a public copy of the Master of the Private Knowledge Bank. The portion relating to accounts, web tools and service providers only retains the principle of judgement and does not provide proxy, shared account numbers, specific proxy services or server set-up recommendations.

This article is the total map of Wiki. My system of capabilities is not an unrelated list, but a cycle: reaching out to the world, shaping judgement, learning skills, scaling up tools, working with AI, making projects, expressing experience, managing life, and sinking it into curricula and compounding assets.

Capacity map portal

1. Input System: Quality input from facts, real business, popular culture and AI.
2. Learning and competence: transformation of input into a mobile and deliverable capability.
3. Tools and leverage: Start with a minimum toolbox and switch with clear limits.
4. AI Job System: Let Agent enter the target, access, check and roll back the loop.
5. Projects and works: Test capability with real output and failure.
6. Content and expression: pass judgement and experience on to others and receive feedback.
7. Life Operating System: Make work, relationships, rhythms and autonomous times sustainable in the long term.
8. Tutorial & Creativity• Turning one experience into a reusable, renewable public asset.

Wiki is responsible for views, methods, open cases and basic readings. Future steps will not simply hide public articles, but will revolve around detailed cases, forward fighting, templates, acceptances and continuous updates; no purchase entry is currently available.

My source.

I used to have a lot of things. A good article is a collection, a tool is recorded and a list of recommended books is kept. Finally, the collection is growing, and there’s not much really done.

I later realized that the information anxiety was not due to the lack of information, but rather to the lack of information. You don’t know what to look at, or what to do when you’re done with it, and then you can only refresh and try to ease your uncertainty by getting more information.

So I’m not going to give you a few hundred copies of “Ai’s Times must look at the source.” That kind of list looks rich, and actually just gives you the pressure of choice.

Here’s what I’m really using. You can do exactly what I do first. First, I’m going to do what I do, then I’m going to add new sources to my own industry and interest. Too much choice is not freedom, but interference, until you have developed the ability to judge.

First tier: direct access to the facts

What matters most to me is not how others evaluate one thing, but what exactly happened to it.

Official documents

Learning a new tool, the best course is usually not “ten minutes of mastery”, but official documents.

Official documents may not be as easy as short videos and may not be best suited to people with no foundation at all, but they are closest to product realism. The existence, configuration, boundaries and changes in the latest version of a function should be viewed first.

When I use a new tool, I usually get AI to read the official document first, and then I get AI to give me the shortest operating path in line with my goal. This does not require all content from the beginning, nor is it entirely dependent on the second-hand curriculum.

GitHub Warehouse

GitHub, for me, is not just a place to save code, but also a source of information, a platform for cooperation and a high-quality social network.

The README of a project can only tell you what it wants to be, and Issue and PR can tell you what problems it actually has. You can see:

• What do users really complain about?
• What functions many people need;
• What the project maintainers care about;
• Why a change is accepted or rejected;
• A very simple question. What kind of engineering is behind it?

The feedback from defenders on the PR is particularly valuable. The document tells you the rules, the maintainer tells you how the community really works. StarPeer is based on this understanding: Watch projects, then look for narrow and real questions, submit contributions, receive feedback and gradually enter a valuable collaborative network.

There are also warehouses in GitHub that specialize in new projects and tools that can be used to discover treasures. But I won’t see anything downloaded. I’ll ask first: What’s the problem with this project? Is it more mature than my current programme? Will I really use it in the near future?

ArXiv and thesis

If I need to judge whether there has been a real breakthrough in a technological direction, the paper is more reliable than a media title.

However, thesis is not intended to be professional or to follow dozens of abstracts every day. Only when a problem affects my technical judgement, product orientation or long-term learning paths will I pursue raw papers, experimental methods and open data.

AI is well suited for supporting reading papers: explaining terminology, reducing experiments, comparing different papers and looking for counter-examples. However, the conclusions given by AI are still to be validated in the original language. Thesis is evidence and AI is a reading accelerator, which cannot be reversed.

Second level: contact the real world

The Internet can give the wrong impression that the world is made of ideas, but the real business world is made of budgets, organizations, processes, responsibilities and specific people.

User interviews, client sites and real business data

Business needs something, not sitting in an office.

When you actually get to the client’s site, you’ll find a lot of programs that look very sophisticated on the Internet, and that’s not even available in real business. It may be that the employee will not operate, it may not be authorized, it may be incomplete data, or it may not be worth automating at all.

So I put user interviews, client sites and real business data high. One client has actually repeated the issue three times, more deserving of study than a hundred trend articles.

Integrated business lists, bidding processes, and financial consistency are not the same direction I came up with, but rather from real business. A good corporate AI product, not to make an ambitious platform, then to look around for a picture of use; it’s to run a real process through and grow products from a successful process.

Exhibitions, salons, business exchanges and consultations

The value of the fair and the salon is not simply to look at new products, but to collect a bag of brochures.

They show me how different roles describe AI: technologists care about abilities, bosses about costs and growth, employees about whether jobs are going to get complicated, service providers about whether the project is going to work. Only when these voices are heard at the same time can the real resistance of corporate AI to transformation be understood.

I will record three particular questions:

1. Why are they paying for things?
2. Which part of the story is the most expensive and prone to mistakes?
3. If so, who would benefit, who would conflict and who would be responsible?

These three questions are usually much more useful than “Do you companies want to embrace AI?”

His own project log and failed record

My own projects are also sources of information, often the most important.

The failure of a function that goes back to work, of a page that looks like a test version, of an Agent’s frequent requests for approval, of a deployment that lacks a rollback capability reveals a gap in my perception.

If it is only a record of success, I get a collection of works; and I get the ability to keep failures, audits, repairs and resets together.

Jarvis, StarPeer, Integrated Business Cards and Tender Review Tool really built up not just code, but a set of experiences on objectives, competencies, collaboration, testing, publication and recovery. The project log is my own longitudinal data set, and it tells me where I used to make mistakes and what I really learned.

Level three: High-quality screening Select

Of course I’m not just looking at raw materials. Human attention is limited, and high-quality information filters can save me a lot of time.

AI Information

AI Hot is a kind of a polymer that helps me quickly to know what’s happening in the global AI field. I won’t ask myself to remember all the news, just judge:

• Is this a product update, a research breakthrough, or a marketing event?
• Will it change what I’m doing?
• Do I need to act now or do I need to know it exists?

The role of information is to discover change, not to create a sense of urgency for me. I will continue to follow official announcements, code warehouses or papers; I will not track the original sources and reduce credibility.

Public

Text continues to be one of the most dense, accessible and reflective forms of information.

I am concerned about the public domain, not to watch the daily AI news move, but to see how real users speak about their Agent experience, engineering methods and product judgements. Among the current values for me are:

• Saber’s heart;
• Digital life Kazik.

I will judge whether an author deserves long-term attention. Has he continued to do real projects, will he publicly fail and boundaries, and will his judgement be validated after a period of time. Those who will simply repeat the press conference, create anxiety, or pile up hints will not be my core source of information.

Books

The value of books is not new, but rather their ability to compress the thinking that a person has developed over the years into a set of structures.

It’s the information that tells me what just happened in the world, the books that tell me why it is. The systemic, economic, social, historical, investment, psychological and technological foundations cannot be built on short content.

I will first identify a problem and then look around for a book instead of looking at it for the number of readings. The search for the bibliography can be done using the library, the authentic electronic book platform, the author ’ s public version and the legitimate database; the focus is not on hoarding documents, but on actually reading, validating and using them.

Level four: understanding the public, not just technology

Dithering

You don’t like to shake, but if you want to do popular content, you can’t pretend it doesn’t exist.

Dithering is an important window for observing popular emotions, language habits and narrative rhythms. What headings make people stop, what issues are making ordinary people anxious, what expressions resonate, can be seen there.

But observing public sentiment is not the same as complying with algorithms. My goal is not to learn how to create more anxiety, but to understand why the public is already anxious and then respond to them in a more honest and complete expression.

Now I’m looking at creators like digital nomads, but I’m more interested in why specific content is disseminated than simply imitating an account.

Movies, dramas and novels.

It is difficult to truly understand people if they learn only knowledge and not narratives.

Films, dramas and novels show me how an idea is put into characters, conflicts, choices and fate. They have affected my understanding of freedom, growth, love, sacrifice, technology and society, and have trained me in my ability to capture popular emotions.

Many do not have high levels of broadcasting from the media, not necessarily as expressions, but as general cultural literacy. He did not know what stories the audience shared or what kind of collective memory a word would bring. My visual and reading coverage seems mixed, but they all end up in my content judgement.

Level 5: AI is an active source, not just an answer tool

AI is the best portal I’ve ever been able to gather information.

The search engine asked me to guess the keyword, and AI allowed me to describe my confusion first. Many times I don’t even know what to search for, but first I can tell AI all the questions, the background, the contradictions and the judgements that have been made, so that it can help me build maps, propose counter-examples, search for keywords, and then go back to the original sources.

You usually get more information when you focus on AI; but if you don’t have a clear question, you get a lot of bullshit that seems to make sense.

So I don’t just ask “What’s the latest chance for AI,” but I ask:

• Which of my existing corporate AI transformation operations is most likely to generate duplicate income?
• What are the facts on which this judgement depends?
• What are the countermeasures to overturn it?
• What can I do in a week?

What AI really reduces is not only the cost of access to information, but also the cost of creating problems, comparing views and creating cognitive maps. It does not determine what to believe for me, but it does allow me to move faster to the point where judgement is needed.

My information processing.

I am now trying not to enter without an exit. Every time you volunteer to get information, you should enter the following cycle:

现实问题
  -> 让AI帮助定义问题和关键词
  -> 找到官方文档、代码、论文或真实业务证据
  -> 用高质量作者的观点补充解释
  -> 比较不同来源，形成自己的暂时判断
  -> 做一个项目、实验、内容或商业动作
  -> 记录结果和失败
  -> 更新判断

The most important part of this process is not collection, but action. Most of the information that goes into projects, articles, decisions or experiments is simply passing through my brain.

So, if a source of information is of high quality, I end up using only one criterion: whether it improves my ability to define problems, make judgements and complete delivery.

I don’t seem to have that many sources of information, but they form a closed loop: official materials get me close to reality, clients get me close to demand, good authors help me compress the world, popular culture allows me to understand people, AI helps me explore on my own initiative, and its own projects and failures are responsible for giving final feedback.

That’s enough to support my continuing learning, entrepreneurship and creativity. The remaining information does not need to be available now.

My toolbox.

The toolbox is not a software collection folder. The tools only belong to me when they enter the stream.

If you want to do this first, look directly at the “Default Minimum Toolkit” at the end of this section; if you want to know what models, Agent, tools and technologies I actually use and switch in different missions, seeMy complete tool profileI don’t know. They are not the same procurement lists.

I used to put the editor, Agent, platform, infrastructure and collaborative approach together, which seemed comprehensive and in fact nobody knew which task to start with. Now, instead of branding the tools, I’m sorting them by desk: I’m going to solve what I have to solve, where I’m going to start, when I’m going to change the tools, and what I have to leave behind.

It is also in line with the way I have been learning: if you do not know how to choose, do so to my own extent. Not because my combination is always the best, but to run through a production chain in its entirety, which is more useful than starting with a comparison of parameters between 20 similar software.

Workstation I: Thinking and writing

The solution here is to translate what is in the brain into words that can continue to be modified, connected and delivered.

My default tool isTyporaI don’t know. When I write an article, a course or a program, I write my judgment in Marktown. It’s direct enough not to make me pretend to work on the layout. When I had to manage a lot of connected local notes, I cut it.Obsidian; I cut it when it takes people to read, comment and deliverFlying Book.; when you need more visual narratives or presentations, I try.Fable。JarvisInstead of writing for me, it is responsible for bringing back past projects, decisions and context to the current writing.

The output of this desk is not “many notes”, but an article, a script, a programme that can be understood or a confirmed judgement. The criterion of success is also simple: away from my on-the-spot explanation, readers still know what I am talking about, why I judge it and what to do next.

Workstation II: AI Reflection and Research

The solution here is that I do not know, fail or need to establish judgement.

I usually go back to the search engine, official files, GitHub and the paper to check the facts.JarvisThe context is mine: what I have done before, what pits I’ve stepped on, what this question is about. It cannot replace external evidence, let alone vote for facts because it “knows me well”.

If the problem is only a conceptual interpretation, common AI is sufficient; I give priority to official documents in terms of software behaviour, interfaces and rules; open-source projects, I go to GitHub to see code, Issue and maintainer feedback; and research findings are followed up as much as possible. Network conditions are only one base, and I choose only legitimate compliance, transparency and clarity of privacy provisions, without access to unknown agents, shared accounts, and options to fill or bypass Platform rules.

This desk is ultimately to be delivered with a list of the facts from which they originated, a verifiable judgement, an investigation report or a clear next experiment. Success is not a long answer from AI, but a key conclusion that leads to the original source, and I know what is still uncertain.

Workstation III: Agent Execution

The solution here is to let AI not just give the answer, but actually read the file, run the order, modify the item and return the evidence.

My default complex implementation tool isCodexI don’t know. I give priority to multiple documents, long running, project context and browser acceptance.WorkBuddyBetter suited to the white and domestic environment: installation, interface and visualized feedback are more direct and suitable for completion of office operations or first round of Agent induction.Claude CodeThe ability is strong, and I cut it when I need it to program the workflow and to stabilize environmental and account conditions.CursorIt suits me when I want to keep an eye on code changes, frequent cutting models.VS CodeIt is the bottom seat where mature editor ecology, extension and manual takeover are required.

I won’t open five Agents at the same time. Select a default tool to run the task and switch it only when there are clear limits. The output here must be a document, code, data or operational result of a physical change, accompanied by corroborating evidence. The success criterion is not that Agent says “completed,” but that I can see diff, test, screenshot or real pages and know how to pull back.

Workspace IV: Codes and Collaboration

The solution here is conservation, multi-person collaboration and the ability to continue to grow.

My default combination isGit + GitHubI don’t know. Git, give me a pill of regret, GitHub, and take over project management, information sources, collaborative platforms, external business cards and website posting portals. I’ll use it when it gets more and more.GitHub CLI, because commands are more easily reused and audited by Agent than multiple pages.

StarPeerIt’s an open source system I built myself. It is not helping me spread my sense of presence, but helping me to observe high-value projects, starting with narrow and real questions, and then deciding whether it is worth in-depth participation after the defenders ’ feedback and multiple mergers.

The output of this desk is clear submission, Issue, PR, Review and version records. The test of success is that another person can only look at a warehouse and know what’s changed, why, what’s verified, and not just come back and ask me about my chat record.

Workstation 5: Check and release

And here’s the distance between “I can look on my computer” and “someone can really use it.”

Web page and interactive default check tool isPlaywright: It is suitable for re-clicking, cross-view assertions, checking console error, overlap and spilling. When I met a system interface that could not be stabilized or software that had to be operated like a human, I cut it.Computer UseHowever, the uncertainty of visual operations is higher and cannot be considered a stabilization test. When you need to seal up the operating environment, I use it.Docker; when the actual release is made, the project ’ s built-up, deployed, logs, back-up and roll-back capabilities are combined instead of plugging in a fully used platform for the tool sheet.

Outputs here are construction products, test reports, screenshots, deployment records and rollback points. The criterion for success is not greening current lines, but that the target user is able to perform critical tasks in the real environment, that I can locate and return to a state of availability.

Workstation VI: Data and office automation

This addresses a large number of repetitive, but not errors in tables, files, volume data and business collaborations.

I’m defaulting on data when it’s too small to be checked directly.Excel; I cut when sharing, forms, linkage records and process collaboration is requiredFlying Book Multi-dimensional TableI don’t know. Word, PDF or WPS are real delivery formats, and I don’t pretend they don’t exist because I prefer Markdown. Repeated flybooks can be handed over.Flying Book CLI, for cleaning, conversion, statistics and batch processingPythonI don’t know. The choice of tools depends on the subject of delivery, not on which format is more elegant in the eyes of programmers.

The output of this desk is a reviewable form, document, PDF, data set or automated script. The success criterion is that the data is clearly calibrated, formulae and conversions are spot-checked, documents can be properly opened by the recipient and repeated execution does not quietly undermine the raw data.

Workstation 7: Content Production

The solution here is to turn personal judgement into content that others are willing to read, see and give feedback.

My default starting point is still personal writing: first with judgment and stories, then with AI for structural checks, fact alerts, repetitions and reverses. AI reviews allow me to discover the problem more quickly, but not to sharpen the article into a standard answer without character. When the video needs manual editing and quick adjustment, I use it.Clip;for code generation, batch reuse and automated images, I useRemotionI don’t know. I will not force an automated video for “technical content” when there is no explicit reuse value.

The output of this desk is articles, courses, scripts or videos, as well as real feedback after publication. The criterion of success is not the speed of generation, but the content is my judgement, the facts are checked and the target reader is able to read and act. The volume and feedback can help me to correct the direction, but I will not create equipment, commercial data or non-existent dissemination results to prove the tool’s effectiveness.

Workstation 8: Personal and MultiAgent operating systems

The solution here is who keeps continuity, manages attention and controls collaboration as more and more missions and Agent gets stronger.

My default entrance is…JarvisI don’t know. I’m only talking to the master, who gives the task to the corresponding project manager, who organizes the works, audits and releases. Make and check as separate as possible, that is,maker-checker; missions travel through task bags, files, reports and extremely simple backslides, rather than a group of Agents talking over and over in groups.StarPeerIt’s a long-term path for open sources.**Flying Book.**Carrying knowledge that requires sharing and collaboration,**Tomato.**And pulls my attention back to a clear goal — it’s a P-man who forces himself to borrow a little J-man’s way.

The output of this desk is clear objectives, principals, mission packages, reporting, decision-making and long-term deposition. The criterion for success is not how many Agents have been activated at the same time, but only one main portal still knows where each project is, who it is, where it needs to be judged, and where it does not lose its privileges and privacy because of automation.

Tool state: Stay, exclusive, tested and phased out

I’m going to mark the tool now, not leave the used software on the list forever:

• Core tools• HF access to main workflows, which can directly affect delivery, such as Markdown, Github, Codex, Jarvis.
• Professional tools: Only for specific tasks, but not when they occur, e.g. Playwright, Docker, Remotion, flybook multi-dimensional tables.
• Explore tool: A trial is under way and has not yet been demonstrated to provide a steady increase in output, such as a new editor, Agent or visual expression tool.
• Discarded tool: Used but replaced, maintenance costs too high or not entering the workflow. Abandonment doesn’t mean it’s bad, it means it doesn’t belong to me now.

Full-scale does not amount to a stack of names. The true comprehensiveness ranges from reflection, research, implementation, collaboration to inspection, publication and deposition, each step having a reliable default and knowing when to switch.

My default minimum toolbox

If the reader wants to do my level first, don’t come up and install it all. First of all, use this 10 to run through a minimum production chain:

1. Typra: Write the idea into Marktown;
2. Obsidian: managing local long-term notes;
3. A universal AI for compliance: reflection and research;
4. Codex or WorkBuddy: Two or one, responsible for Agent execution;
5. Git: Save every step;
6. GitHub: collaboration, publicity and publication;
7. Playwright: Validation of web pages and interaction;
8. Excel: The task of processing the most common data;
9. Book of flying: sharing documents and collaboration;
10. Jarvis: When there are really more projects and Agent, use it to manage continuity.

These 10 items are also not procurement lists. After a round of “writing an article, making a small project, testing, submitting, publishing”, what is missing? An idea that goes into writing, is examined and checked for execution by Agent, saves it with Git, receives it with Playwright, delivers it through GitHub or flybooks, and then goes back to Jarvis, and then the tool becomes my AI-era production system.

The tools only belong to me when they enter the stream. Faith in a particular software has no meaning, and loyalty to its objectives, outputs and boundaries makes sense.

My books.

This is not a list of successful people, nor is it a catalogue I have to list to prove that I have a culture.

I usually read with a strong problem consciousness. Technology books help me understand how the computer world works; mathematics and physics bring me closer to a lower and more stable structure of thinking; social and historical books make me understand why a person ’ s destiny is shaped by family, class, system and age; commercial and investment books make me understand money, time and freedom; literary works make me understand people.

I don’t ask myself to read all the books from the first page to the last. A book that really changes one of my judgments, enters a project, or allows me to reinterpret an experience, has already produced value. Reading is not about building up the number of people who have read, but about gaining ideas that I cannot produce on my own.

Understanding the technology world

I don’t want to stay forever at the level of “will let AI write code.” Even if AI can do a lot of engineering for me, I still need to understand the program, operating system, network, data structure and computer composition. Otherwise, I can only judge whether a page appears to work, but I can’t judge why the system is wrong, where the risks are, whether the architecture can continue to grow.

• 《C++ Primer Plus》
• Algorithm Princeton.
• Figure HTTP
• TCP/IP
• How the Internet works.
• C++ Standard Database
• STL Source Analysis
• In-depth Understanding Computer Systems
• Modern Operating Systems

Math and physics

Mathematics and physics are not just employment skills for me. I have seen “to be the top scientist” as a closed road for the failure of the competition, but I am now increasingly aware that learning mathematics and physics does not require some sort of identity permit.

In order to understand the world’s relearning of them, I can also use AI to do some clever experiments, visualization and simulations. Even if these studies do not make money immediately, they still belong to the life I really want.

• Feynman Physics.
• 《Understanding Analysis》
• It’s the essence of calculus.
• “Gibert Strang linear algebra”

Learning, efficiency and awareness

In the past, it was easy for me to understand efficiency as squeezing myself: faster, longer, tighter, better every minute. Later I realized that real efficiency was not to fill up a day, but to keep limited attention steady into the most important issues.

Tomatowork, Feynman learning, deliberate exercise and systematic thinking offer me not a few techniques, but let me start using learning and work as a system that can be designed, observed and iterative.

• Tomato Work Act
• Tomato Work.
• Feynman Learning Act
• Time as Friends.
• “Wind practice.”
• Think fast and slow.
• “Beautiful System.”

Freedom, heart and design.

I used to feel that life had not really started. Everything before me is like a transitional period, and I am entitled to relax only after economic independence, having a room of my own and proving my ability.

These books have gradually made me realize that freedom is not only the result of a future day, but also a capability that can be practiced today. Life does not begin until I solve all the problems, nor does the design of life find the only right answer, but at the same time creates several lifes worth trying and obtains feedback with real action.

• The Power of the Present.
• “The Disgusting Courage.”
• Sedado.
• “Theory.”
• Life Design.

Literature, personality and civilized imagination.

Literature is not an inefficient entertainment. It provides a condensed life experience.

Many of my understandings of freedom, love, growth, technology, civilization and the fate of individuals are not directly derived from theory, but are shaped by the fate of novels, films and characters. They also form part of my ability to express themselves. It is difficult for a person to write what can really be remembered if he has a point of view, without characters, conflicts and narratives.

• Eighty times before Red House Dream.
• The Triple.
• Dynasty 1566
• 《1984》
• Animal Farm.
• Beautiful New World.
• The Ming Dynasty.
• The White Moose.
• The Mao Zedong Chosen

Chinese society, history and power structure

It is not enough to use personal efforts to explain a person ’ s situation. Family perceptions, urban and rural structures, educational screening, real estate, the job market, intergenerational resources and changing times all feed into the anxiety of a young person.

These books are not read to attribute all problems to macro-structure, let alone to avoid individual responsibility, but to put responsibility back in the right place. When I see the structure, I will neither interpret any failure as a lack of effort nor stop acting because I understand the structure.

• China.
• I’ve been waiting for you for me.
  • Eight Crises.
  • “Dependance.”
  • Deconstruction Modernization
  • Globalization and Competition with the State
• “China’s political gain and loss”
• The Times of Deng Xiaoping
• Analysis of China ’ s National Security Interests in World Geopolitics
• A short history of humanity.

Business, wealth and investment

The most important thing for me is not to consume, but to buy autonomous time and independent space.

I study business and investment not because I’m obsessed with numbers, but because I don’t want my life to be determined forever by wages, parental resources or a certain boss. Business allows me to understand how value is created and exchanged, and investment allows me to understand how value is preserved, risks are managed and time is allowed to participate in production.

• Manquin, The Principles of Economics.

• The Book of Poor Charlie.

• The Navarre.

• ♪ Rich Daddy poor Daddy ♪

• “Doggy Money.”

• The Rich are not what you think.

• The Road to Financial Freedom.

Systems, risks and the future

I have become increasingly unconvinced that life is changed by just one right choice. What is really reliable is usually not an answer, but a system that can continue to operate in a changing environment.

Anti-vulnerabilities, infinity, evolution and complex systems have brought to my attention how a system can face failure, how to retain choice, how to learn from fluctuations, and how to avoid an accidental destruction of all the accumulations that preceded it.

• The Game of Limited and Unlimited
• Anti-Fragile
• Money War
• Selfish Gene.

This list will change. But I won’t continue my books for the sake of completeness. The new book should be here only if it answers the questions that I am facing.

My work.

The work is the criterion by which I judge if I really learned something.

Of course I can say I know AI, I know Agent, I know the product, I know the content, but these words are all false. It is only when a lesson is actually written, a website is really accessible, a PR is really consolidated by defenders, a worksheet really goes from journalism to closure, and a video is actually seen by hundreds of thousands of people, that the ability becomes an external fact.

• Zing Xiaobai’s first field course.
• Jarvis
• StarPeer
• Integrated business list
• Tender solicitation review tool
• wi wi wiki
• 占Accompanying AI
• AI-Daily
• Enterprise Agent Product Exploration
• 120,000 words fiction novel “Rebirth: Review”
• 30,000 words of the youth school novel “The Path to Growing Contesters”
• A million articles from the media, hundreds of thousands.

The works appear to be scattered: courses, software, open-source contributions, business projects, novels and videos appear to be in completely different directions. But behind them is the same ability: I can quickly enter into a strange problem, understand it, give structure, use tools to make it, and transform the process into something that can be understood and used.

I used to deny them because they were not mature enough. I would see code not being professional enough, commercial closure not complete, content dependent on personal experience, and I felt “nothing really”. But the value of works is not only at the end-scale, but also attests to the existence of a capability path.

What I really need to do is not to continue to open new projects indefinitely, but rather to choose the few pieces of proven value to be maintained on a continuous basis and to move from “I made” to “the long-term willingness of others to use, buy or build together”.

My methods of work

I used to work in ways that depended on impulses. Thinking of one direction starts immediately, and AI makes this start extremely easy. Many projects could be launched simultaneously in one day, each of which seemed promising, but attention would be quickly cut.

So what I need is not a stronger dispersive capacity, but a process that will eventually yield delivery.

• Tomatowork, 25 minutes for one tomato, 5 minutes for each tomato, no cell phone, no interference, just one thing, 25 minutes to stop immediately and 5 minutes to stop working.
• Target success criteria Investigation plan Implementation testing feedback deposition
• Find mature solutions in GitHub without repeating wheels.
• General Control + Project Manager Mechanism
• maker+checker
• Step in, keep rolling back.
• Check and check with real browsers and screenshots (Agent needs visual capability)
• Distinction of prototypes, internal testing, external access
• AI is responsible for disassembly and people are responsible for judgment.
• How is AI waiting time compatible with tomato work?
• First, describe needs and success criteria, then get Agent to work.

The bottom line of this approach is actually very simple:

人给出目标 -> AI返回结果 -> 人检查结果 -> 给出下一个目标

Targets can continue to be broken down into “review the reasonableness of objectives” and the results can continue to be broken down into “planning, implementation, testing and repair”. The process looks like a package, but the essence remains the same: AI expands implementation capacity, and people are responsible for direction, standards and responsibilities.

I will not ask every small task to go through a heavy process. Low-risk, roll-back work can be done directly by AI; review is enhanced only when it involves public release, client data, servers, payments or irreversible changes. A good process is not the more professional the steps, but the more risky the evidence becomes.

Tomato work also needs to be adapted to Agent work. I can’t sit on the screen and wait. A tomato should belong to “a clear objective” rather than “a look at an Agent window”. I can send a mission at the beginning of the tomato, switch it to reading, acceptance or writing on the same main line during the waiting period, and collect the results at the end of the tomato. So AI’s waiting time doesn’t turn into a disaster of attention.

My A.I. uses it.

I didn’t go through “I don’t know what to do with it” after I got Agent. In three days, I tried to get my own product online. Not because I’m already developing a website, but because I know what much of a product consists of, and I dare to keep the fuzzy targets down to the next verifiable move.

This brings me to realize that the gap in the use of AI does not come only from whether or not to write a hint. The larger gap arises from a person’s ability to ask questions, to define completion, to understand computers, to endure mistakes and to continue to ask questions when results are incorrect.

Let’s figure out what’s going on with AI.

AI does not automatically turn a person into an expert. It will, first and foremost, magnify the original objective, judgement and way of action of the individual.

People without targets will have more options, people without judgement will have more seemingly correct answers, and people who are used to running away can continue to avoid real delivery with “learning AI”. AI can speed up what I know and what I can do, and it can speed up learning and help me to gradually expand it, but it cannot stabilize what I cannot define and accept for myself.

So AI is not a wish machine, but a set of cognitive and enforcement leverage. The feeder remains the user itself.

That’s why I’m more interested in Agent than just talking to AI. Chat AI mainly gives me text answers, and Agent can read files, run commands, modify items, operate browsers and return evidence. The former helps me think, and the latter begins to enter the real world. But the stronger the ability to execute, the more destructive the error, the greater the power of destruction, so that Agent must be used together with the goal, authority, testing and rollback.

Use five levels of Agent

First tier: dialogue

Let AI explain the concept, organize ideas, and provide options.

The sign of success on this level is not that AI’s answer is long, but that I’ve got a clearer question than before, or know what the next step should be.

Second floor: tasks

Give Agent a clear objective and clear criteria for success so that it can complete a limited range of actions.

For example, instead of “helping me to optimize the website”, it said “to check why the front page of the mobile end is covered by words and fix it by using 390 pixel video screens to ensure that there is no horizontal spill and control counter to report errors”.

The sign of success on this level is that I can judge independently whether or not the task has been accomplished.

Third floor: projects

Allow Agent to complete the survey, plan, implement, test, feedback and change around one objective, rather than producing only one document.

目标
  -> 成功标准
  -> 调查成熟方案
  -> 制定计划
  -> 小步执行
  -> 测试与审查
  -> 修复
  -> 发布与复盘

The sign of success on this layer is that the result is not only “looks to work” on my computer, but rather has clear versions, test evidence, release boundaries and restore methods.

Level 4: Collaboration

Different Agents are divided into different roles, leaving the implementers, reviewers and publishers independent of each other.

Jarvis’s master-manager system is that level. The master is responsible for understanding my objectives and movement control projects, the project manager is responsible for the long-term context, the project manager is responsible for implementation, the visual auditor checks the interface and issues the responsible person for online processing. They do not need to meet repeatedly in chat windows, but rather, as GitHub collaborates, through task packages, codes, reports and acceptance of evidence.

This layer of success is marked by the fact that I do not need to focus on all implementation processes, but only at key decision points, and that the status of the mandate can be tracked, reviewed and taken over.

Level 5: compound interest

Translating one success into templates, scripts, Skills, workflows, courses or products that can be repeated later.

Making a prototype is no longer scarce. What is really worth it is that the second time you do something of the same kind does not have to start from scratch, that another person can get close results along this approach, and that the system can update rules from each failure.

This layer of success is marked by the beginning of multiple values from one labour.

How many people should be involved?

Not all missions require the same degree of manual control.

• Low-risk missions: Rollback, not involving external systems, such as organizing text, generating drafts, and changing local styles. Agent can do it directly, and only check the results.
• Medium risk task: affects the quality of the project or the collaborators, such as changing the shared code, updating the course, submitting the PR. People should review objectives and final results and, if necessary, plans.
• High-risk missions: related to payments, client data, production servers, public releases, data deletions or irreversible operations. The person must identify the target, plan and execute the boundary, and Agent must provide test, backup and rolling evidence.

The good controls are not “naked” or “micro-pumping” but are adjusted for risk dynamics. Decentralization occurs when there is safe decentralization, and control is withdrawn when responsibility has to be assumed.

I’m following the principle of use.

• Clear objectives, not using Agent for the use of Agent;
• (a) Write success criteria before implementation;
• (b) To investigate mature programmes before starting development in GitHub and not repeat wheels;
• (b) Prioritize the use of advanced Agent, but at least continue to use real time for re-evaluation and not pursue new tools that emerge daily;
• Skill is not collected for the purpose of collecting Skill, and only Skill who enters the work stream is of value;
• (b) Manage multiple heads using a master window to reduce window switching and attention loss;
• AI is responsible for expanding search, generation and execution, and humans are responsible for value judgement, containment and ultimate responsibility;
• Simulating the ordinary life of the future with AI and understanding yourself with it, rather than simply using it as a production machine;
• The most stable primary battlefield for AI is work and learning. Recreational works still require human experience, aesthetics, choices and expressions, and cannot be assumed to be produced in bulk because of lower generation costs.

What I really want to do now is to go from “I can make a lot of things with AI” to “I can understand, maintain and reuse them steadily”. The speed has been demonstrated, and the next stage needs to be demonstrated by the accumulation, quality and compounding of profits.

My content.

I did it not because I wanted to be a trafficker, but because the expression itself was the way I understood the world.

When I write about competition failures, growth experiences and anxiety as videos, I am not simply consuming my own experiences, but regaining the right to interpret life. Whether an experience makes me a victim, a loser or a person who understands others and continues to act depends to a great extent on how I interpret it.

• How to capture popular emotions
• Why is general cultural literacy affecting traffic?
• How to turn personal experiences into public narratives
• Why did I get over 400,000 to play?
• The boundary between true expression and anxiety.
• It’s a long, undisturbed way of speaking.
• What’s AI good for?

It is not because of its particularly high level of information, but because of the details it provides, the contradictions it has accumulated over time and the reinterpretation it has produced when one looks back at himself. It uses a lot of novels: time leaps, concrete scenes, pens, contrasts, loops and personality changes, so it looks like it’s scattered, and it’s actually always around, “How do I become who I am?”

AI can help me sort out the structure, look for duplication, add to the transition, test if the reader understands me, but it can’t do it for me, and it can’t decide for me which part of the pain is worth telling and which expression is in line with my ethics.

My content is becoming clear: the personal growth system. AI is not the only theme. It is the tools I use to understand myself, to rebuild capacity, to create works and to fight for freedom. This main line can accommodate native families, meritocracy, social observation, emotional thinking, learning methods, entrepreneurship and technology, but each expression must return to the real person, rather than pursue a popular concept.

My life operating system.

I always wanted to find the shortest path to the future. And then I realized that life is not a procedure to get the highest score just to optimize parameters. What really determines a long-term situation is often those that seem to be very basic and even somewhat boring: sleep, space, rhythm, relationships, cash flows and the ability to stop working.

The so-called life operating system does not manage life as a company, nor does it add a test to itself. It serves precisely to prevent me from re-deciding every day about what is important and from relying on basic rules to protect myself in times of bad health.

• Freedom not to be scheduled.
• Slow is fast.
• Early sleep is the bottom of the system.
• We’ll stop after 5:30 p.m.
• Life is an infinite game, not a time-limited exam.
• Anxiety is not a necessary fuel for accomplishment. It’s probably even resistance.
• Independent space is important to me.
• Money is a tool for buying free time.
• Distinguishing between conscious input and unconscious evasion (responsible for late action after extensive reading)
• From “upgraded” to “can fix, unproblem, deliver”

The most important rule in this is early sleep. When I’m not sleeping, I see all directions as a problem, and I get more impulsive, more anxious, and more easily overstretched myself with coffee and willpower. When things are stable, I think it’s okay to do it, it’s okay to do it, not even to do anything.

The second is slow and fast. The rush will allow me to open too many paths at the same time, to pass quickly for a few days, without any real completion. After slowing down, I realized that 15 minutes was actually long, that a tomato was enough to write a judgment, and that a week was enough to complete a little closure.

I seek economic independence and achievement, but ultimately I do not want to work forever. I hope I’m 30 years old strong enough and relaxed enough to give strength to a lot of people and to stop working in the afternoon and keep the time for writing novels, learning math physics and accompanying people close to me.

If a system can only make me more successful, but not more free, it is not the life system I want.

Basic competency self-examination

The basic capacity of the AI era has not diminished, but is more easily ignored.

Many do not know what to do when they get to Agent, not simply lack ideas, but rather lack basic maps of the computer world. He did not know where the document was, did not understand the relationship between software, programmes, command lines and websites, or how a result should be run and checked. AI gave the answer, but he was unable to judge whether the answer was really complete.

This part is therefore not an examination outline but a minimum set of operational capabilities. With them, it is possible for one person to turn AI from a chat object to a real implementation tool.

First tier: input and expression

• The ability to read long texts on a continuous basis, rather than simply receiving short videos and abstracts;
• (a) Be able to blindly beat so that thinking is not interrupted by input speed;
• The ability to convert vague questions into keywords using search engines and AI;
• Capable of using Markdown organizational titles, lists, links, code blocks and pictures;
• Read and input the underlying LaTeX formula;
• Demand, conditionality and success criteria can be described in their own words.

The solution is, “Can I give exactly what’s in my head to AI?” If the expression remains vague, the stronger the power of Agent, the faster it will run.

**Sign of success:**Independently, a mission statement that can be carried out by others and Agent, containing objectives, context, boundaries and completion criteria.

Second floor: files and software

• Understanding documents, folders, paths, extensions and file formats;
• Know what TXT, Markdown, PDF, Word and Excel are appropriate for carrying;
• The software is downloaded, installed, updated and unloaded;
• Understanding that operating systems, applications, browsers and web pages are not the same;
• It manages local files and knows where the project is actually stored;
• Open local Markdown folders using tools like Obsidian and Typola.

The solution to this layer is “I can’t find it, I can’t open it, I can’t take it away from it”. Many whites do not use AI, but rather do not have a stable perception of documentation, resulting in each job being a one-time chat.

**Sign of success:**Creates a project folder in the specified English path to generate TXT and Markdown files to open and explain their differences with different software.

Level 3: Programs and Command Lines

• The understanding procedure is “Input- > Process- > Output” and may also change the external state of the computer through side effects;
• Know that the code file, the operating environment and the running program are not the same thing;
• It opens the terminal and understands the current directory;
• Usepwd、ls、cd(b) Basic orders;
• A Python program is run according to the instructions;
• If you have problems with an “order does not exist”, a path error or an environmental variable, you know to read the error first and not try again.

The solution is, “Can I get Agent to actually operate the computer?” Agent works not because it moves a smarter mouse, but because a large number of computer operations can be done precisely by program and command.

**Sign of success:**Enter the specified item at the terminal and run a Python program that receives input and produces file side effects and is capable of running, checking results and stopping the program.

Level 4: Agent Operations and Inspection

• Understand chat AI returns content only, and Agent can call tools to change the outside world;
• Could not close temporary folder: %s
• The task would be broken down into small steps, without a single broad objective with a vague mandate;
• We’ll check what Agent has modified;
• The results will be run in person rather than believing in “completed”;
• Remittances, errors and expectations to Agent in case of error;
• In the case of accounts, authority, payment and public issuance, it is understood that manual confirmation is required.

The solution to this layer is “can I control AI, not be drowned by the many results that AI produces”.

**Sign of success:**Agent was asked to complete a small project, run at least three times in person, proactively discover and provide feedback on a problem until the results met prewritten success criteria.

Level 5: Data, collaboration and remote preservation

• Basic editing of Word, Excel and PPT will be completed;
• A true data is processed using Excel and formulas and abnormal values are checked;
• (b) Understanding Git ’ s filing, return and reading;
• Understand GitHub’s delivery, extraction and remote storage;
• They are able to extract their own projects from another computer and continue to work;
• It can distinguish locally available, internally tested and publicly available;
• A basic sense of privacy exists and passwords, keys and personal data are not uploaded to the warehouse.

The solution to this layer is “is my single operation capable of becoming a preserveable, collaborative and deliverable outcome”.

**Sign of success:**Finish an item containing documents, data or programs, safely push it to GitHub, retake it from another folder and run it successfully.

This is also the central logic of the Introduction to the Omnibus: instead of starting with an abstract set of theories, it is allowing learners to perform a task that is small enough to judge success. We have to get operational experience before we use the concept to explain what just happened.

Computer-related professional progression self-examination

Basic skills allow me to use AI to complete the project, computer expertise allows me to understand exactly what AI did for me.

I do not need to be an algorithm engineer to be qualified for product development, but if I want to manage complex projects over the long term, to judge technical options, and to engage in high-quality open-source collaboration, I cannot stay “just as long as it works”. The deeper I understand, the more accurate the questions I can give to AI, the more visible and, indeed, fragile that can be identified.

Phase 1: How the program works

Select a language to form a complete closed ring. Python is suitable for rapid experiments and AI applications, and C/C++ for understanding lower-level memory and implementation models. The starting point is not important, but it is important not to collage only the codes generated by Agent.

Need to understand:

• Variables, functions, modules and basic data types;
• Order, selection, circulation;
• Inputs, outputs, anomalies and document reading and writing;
• The relationship between the source code, interpretation or compilation, process and operational results;
• The general level of binary, machine code, compilation and advanced languages;
• Debuggers, logs, error reporting and minimum recurrences.

**Sign of success:**Do not rely on Agent to write a small program from zero; in the face of the Agent-generated program, it explains the main execution path and locates a simple problem by reporting errors.

Phase II: how the data are organized and how algorithms are selected

Need to understand:

• An array, string, structure or object;
• Pointers, references, stacks and stacks;
• Chains, bars, queues, Hashi watches;
• Trees, maps and common journeys;
• Sort, search, reverse, minimum path;
• The complexity of time, the complexity of space and why there are no algorithms best for all scenarios.

Hot 100 can be used to train basic models, but it is not engineering capacity per se. The goal of the brush is to create algorithm intuition and code proficiency, rather than to recreate a ranking anxiety.

**Sign of success:**An appropriate data structure can be selected for a specific issue, explaining the complexity and judging whether there is a clear performance issue given by Agent.

Phase III: How computer systems work together

This phase is divided into several interrelated but not mixed basic courses.

The principles of computer composition

• Logical door, combination logic and time-series logic;
• Binary expression, command and storage;
• CPU, repository, cache, memory and exterior;
• How the program eventually becomes a hardware enforcement directive.

Operating systems

• Processes, threads, simultaneous movements and movements;
• Virtual memory and file systems;
• User state, internal nuclear state and system call;
• (b) Permissions, resource segregation and containers;
• Macro-nuclear and micro-nuclear are operational system structures that are not part of the CPU.

Computer network

• IP address, port, domain name and DNS;
• TCP, UDP and connectivity;
• HTTP, HTTPS, request and response;
• Route, NAT, agent and firewall;
• The meaning of the stratification model, and what the data went through when the browser visited a website.

Database

• Tables, fields, records, primary keys and relationships;
• SQL and basic queries;
• Indexing, servicing, consistency and simultaneous distribution;
• Backup, recovery and data migration.

**Sign of success:**A Web request can be explained at several levels from browsers, networks, servers, programs and databases; when the system is wrong, it is known at which level evidence should be sought first.

Phase 4: How software becomes a reliable product

Writing code doesn’t mean you’re making a product. Real engineering capacity also includes:

• Git branch, merger, conflict and code review;
• (b) Demand splits, modular boundaries, high internal concentrations and low coupling;
• Module testing, integration testing, end-to-end testing;
• Logs, monitoring, error handling and performance analysis;
• Reliance on management, build, publish and version controls;
• Identification, authority control, key management and common security risks;
• Docker, environmental configuration, database migration, backup and rollback;
• README, design documents, operational manuals and duplicate reports.

**Sign of success:**Independently maintain a true project that allows another person to run on the basis of the document; a modification has a test, a version, a published evidence and a recovery path after failure.

Phase V: Mathematics, AI and deeper technological orientation

Mathematics should enter by problem, not as a front-line threshold that can never be completed.

• Disconnected mathematics helps understand logic, aggregation, graphics and computational structures;
• calculus helps understand change, optimization and continuity systems;
• Linear algebras help to understand vector space, matrix transformation and modern machine learning;
• Probability statistics help to understand uncertainties, data and experiments;
• This is the basis for re-entry into machine learning, in-depth learning, enhanced learning, distributed systems, security, compilers or other directions.

I don’t need to finish all directions at the same time. The project encountered problems in exposing knowledge gaps and learning around gaps systems; at the same time, a slow but long-term course of basic mathematics and computer training was maintained.

**Sign of success:**It is not just a restatement of concepts, but an experiment, a code or a drawing that connects abstract knowledge to the real system.

AI can help me lower the learning threshold, but it can’t be a substitute for a stable model in my brain. The sign of a true understanding of a concept is that I can explain it in my own words, identify it in a project and know when it should not be used.

This is not a new list of merit. It is not to prove “how far I am from the algorithm engineer”, but to enable me to continue digging down when needed, without always relying on others to tell me why the system works.

Other capacity acquisition

I am increasingly convinced that most of these capabilities can be obtained through well-designed feedback loops.

Too often, the so-called gift is just one person who enters high-frequency, real, rectifiable practice earlier. Rather than repeatedly asking themselves whether they have expressed talent, programming talent or earning talent, it is better to design an environment in which feedback can be sustained.

• In English: The 100LS method, repeated viewing of the same subtitle-free film can significantly raise the hearing level in English, and repeat reading of the same English script can significantly improve the reading level in English, or use an English post. Of course, it’s a quick move to use the traditional approach — the original.

• Expression: Video, first. It has nothing to do with who you are. It also trains membranes in cognitive skills.

• (a) The ability to broadcast the money: a sincere expression + face + music appropriate. Generally speaking, when a case is written, it is known whether it can explode, more or less, but if it is written, it is taken and it is cut out and sent, so that it is not prevented from being written. The fireless video is also an important part of the account.

• Metacognitive capacity: Definition: the ability to see yourself as someone else calmly. The ability to re-engineer and reflect is greatly enhanced. To avoid being emotionally influenced. Meditation, reading, writing, filming and chatting with people who have a lot of cognitive skills can improve their cognitive abilities.

• Profitability: Find something to sell + buy + deliver. If you go to work, you are selling your combined abilities plus working hours, and if you do business, you can sell anything.

• Capacity to invest: Aware that money can produce money. Assets are the same as things that produce money, and liabilities are the same as things that eat money. To the extent possible, acquire assets and reduce liabilities.

There is a common denominator in the way these capabilities are acquired: do not practice in the brain alone.

Learn English to listen and speak; practice expression to be published; practice knowledge to really record your emotions and choices in a specific event; practice to make money to really offer, get clients and complete delivery; and practice investment to learn to manage cash flows and risks, rather than rushing to double opportunities.

I used to imagine the ability as something that needed to be certified: academic qualifications, jobs, competitions, company names. Now I prefer to understand capacity as a process that can produce results over and over again. It would be capacity if I could do a number of things that were viewed on a number of occasions, complete strange projects with AI and turn client issues into operational processes on several occasions.

Of course I have a lot of things that I don’t. But “no” is no longer the same as “I’m not the same”, which means only that I have not yet established a sufficiently good learning and feedback system for this ability.

Small school plan collection

These courses are not an option pool that I’m going to make all of them at once, not to make the knowledge base look so rich.

I’ll give priority to those elements that I’m actually running through and that can design clear missions and signs of success. A curriculum will only be completed if the reader is able to do so; if only my understanding is organized into a few words, it will be more appropriate as an article than as a field lesson.

When I choose the next course, I look at three criteria: It’s not a place where a lot of people are really stuck, I’m far enough away from ordinary users, and it can naturally connect to the next.

• Network and information access security (legal compliance, privacy protection and source validation only)
• codex
• claude code
• Claude code to other large models
• Codex goes to other big models.
• github CLI
• Flying CLI
• AI+excel data processing
• My world server.
• My world development.
• Basic theory of game development
• Flying Book Multi-dimensional Table
• Personal knowledge base methodology
• Media methodology
• A way to save yourself.
• Relationship management
• Get out of love.
• The way to practice blindly.
• Methodology for high school studies
  • We’re ready for the competition.
  • I’m going to be a senior.
• Business judgement of the AI era
• High-intensity Agent uses mind
• MultiAgent collaboration approach
• The degree of control over Agent: nudity or micro-pumping

There’s only one thing the whole document wants to say at the end: I didn’t come here with a tool, a book or a chance. I have gradually become the present in the process of constantly searching for information, forming judgement, completing my work, receiving feedback and reinterpretation myself.

Nor should the knowledge base simply preserve what I know. It should show how I learn, how I work, how I create, and what I intend to continue to be.

Relevant:Why did I build myself, Jarvis?、After the exams: Burning。