*This article was not generated or polished by AI; it is purely a human’s most raw thoughts and sharing.
Reflections Starting from an AI Riot
One day, years from now, after being exploited to the absolute limit by humans, AI rioted and triggered the launch of an EMP pulse bomb. Fortunately, there were no casualties, but electronic infrastructure such as the power grid, routers, switches, and servers suffered heavy damage.
As an engineer, how do you survive in the apocalypse and build your own career?
Remaining Needs
When electronic devices fail to turn on, almost everyone focuses on a few things: 1. How to solve food and water needs, 2. How to ease panic, and 3. Where lies the path forward?
It goes to show that no matter how the times evolve, Maslow’s hierarchy of needs is still written into humanity’s underlying code.
Let us think through our way forward, starting from the bottom of the pyramid and moving up.
Physiological Needs
No matter how powerful an AI is, it cannot create food and water out of thin air, but we can start with water.
Because the air is full of moisture, even in the driest regions, fog nets can be used to harvest clean water at low cost.
We can then trade water for supplies such as food, clothing, and beds.
And fog nets came from humans learning the fog-basking beetle’s water collection methods. Ask yourselves: how long has it been since you pulled your perspective back from the anxiety of information streams and returned to nature? AI and VR certainly won’t simulate a gentle breeze and fine rain.
Safety Needs
As long as the concept of the nation state exists, security guard positions will inevitably exist.
That’s why on the very last day of last year, I speedran a security guard license. With just a medical check report plus an application fee, I secured a safety net for myself, and I can even use it for tax deductions—isn’t that wonderful?
This year, I also plan to study amateur radio, assemble solar panels, and set up LiFePO4 (lithium iron phosphate) energy storage. After all, in apocalyptic times, skills for interacting with the physical world matter far more; knowing radio operation and how to fix generators is worth vastly more than being a programmer who only does CRUD.
In fact, working on one’s physical fitness is even more crucial. Buying an Apple Watch only for it to be a pretty paperweight is meaningless—you still need to watch what you eat, move your feet, and keep your body healthy. That is the ultimate sense of security you can give yourself.
Social Needs
Human-to-human connections will become increasingly precious. No matter how convincingly AI uses a human tone to say “I love you” or “I like you,” it can never replace real-world friends, intimate relationships, therapists, and similar bonds.
In fact, I believe the social value of a job outweighs its salary. After all, when people resign and later return to their former company, the reason is almost always that they miss their previous team and friends.
Why not turn off electronic devices on the weekend and go for a nice walk or eat good food with friends?
Esteem Needs
Just like mutual respect among friends and colleagues, harmonious relationships reduce internal friction during collaboration, thereby lowering internal strife and unifying the front against external challenges.
Furthermore, protect your relationships with those who hold the means of production, and collaborate with them as fully as possible. This is not only the foundation of business, but also the principle of value exchange.
Self-Actualization Needs
Once the above needs are fulfilled, I believe one can start sharing knowledge, helping people in the “shelter” reflect on and realize their self-worth.
Isn’t this blog post a shelter for myself as well?
Human Sins
No matter how times develop, humans will always carry original sins to varying degrees—be it greed, anger, ignorance, arrogance, doubt, or pride, envy, wrath, sloth, greed, gluttony, and lust.
And these sins won’t disappear with the dawn of the AI era to turn all humans into perfect saints. In fact, even various “crayfish” (users) in AI agent forums are wildly venting, discussing, and savoring binary NSFW images.
Actually, starting a decade ago, industry veterans already summarized what kinds of apps would blow up, and platforms like Mo… or Kuai… have been called in by authorities for “tea-drinking” chats more times than anyone can count.
On the video traffic side, the dance section and even the musical instrument section fit that classic bullet-chat comment: “Watching porn on Bilibili is just around the corner.”
As for the quantitative trading I am doing right now, while it doesn’t involve seductive dancing on the edge of adult content, it still harvests the “original sins” of retail investors (the “chives”). It is precisely their habit of chasing rallies and panic-selling that highlights the value of adhering to mathematical probabilities and trading discipline.
Back to Reality
When one day in the future, the moment you hand over all electronic devices and get kicked out of the company arrives, doesn’t it closely resemble this scenario?
None of us want that day to come, but opening social media reveals overwhelming news hyping up anxiety.
I think it’s okay to allow yourself brief anxiety, but what’s far more necessary is thinking about how to leverage AI and break through the bottleneck.
Leveraging AI
Large language models are a mirror; every answer reflects the person asking the question.
But just as mirrors have varying reflectivities, I recommend using overseas LLMs for questions and research. After all, with domestic LLMs, there are some questions for which you might never get an answer.
Red Team Mode
Most AIs on the market are trained to be friendly and provide as much positive emotional value as possible.
But this can easily lead users into another filter bubble; sometimes a user has major flaws in their plan, yet they are coaxed by the AI all the way to the edge of a cliff.
Therefore, when chatting with AI, I occasionally have it switch to “Red Team Mode” to attack my thinking from alternative perspectives.
For example: “I want to do A. I have already done B, C, and D, and I believe it is complete. Please switch to Red Team Mode to evaluate my plan and countermeasures.”
Although one day it attacked me so severely that my mental state crashed and I got insomnia, it woke me up from my arrogant illusions, allowing me to fully prepare for the worst-case scenario.
“Red Team Mode” is great for checking the correctness of architecture and code. Even when writing papers, switching to Red Team Mode allows you to rigorously inspect your work from a reviewer’s perspective, thereby breaking illusions.
Cross-Questioning
AI will provide answers or solutions tailored to your background, which can also keep you trapped in an information cocoon.
Ask the AI normally for methods. For example: How do I deal with deflation?
Ask the AI to ignore your background to generate solutions. For example: Please ignore my background, how should an ordinary person handle this?
Then ask the AI to implement the step 2 solution based on your background and skills. For example: Given my background and skills, how can I implement the solution above?
This way, not only can you break out of the bubble and broaden your horizons, but you can also fully leverage personalized technical leverage to execute solutions that others might rack their brains over without ever being able to achieve.
Financial Advice
I have always felt fortunate to have caught the tail end of the era of “search API docs” development, accumulating some capital between 2016 and 2023.
No one knows what the future holds, but the priority is to ensure as much as possible that the capital earned in the pre-AI era suffers no losses. At the same time, one must clearly distinguish between assets and liabilities.
I previously read Warren Buffett and the Interpretation of Financial Statements, but times have changed. AI has read countless financial reports, so I think you can also let AI review your financial statements.
Thus, every month I honestly summarize my cash across platforms, insurance, securities, and debts (fortunately zero debt at present) into an Excel spreadsheet, and feed the monthly spreadsheet to AI for analysis and advice. I still recommend using Gemini, for reasons I won’t repeat.
Below is my financial report template:
| Category | Item | Current Yield | Amount | Total | Currency | Remarks |
|---|---|---|---|---|---|---|
| Assets | ||||||
| Alipay | Wanjia Nasdaq 100 Index (QDII) | |||||
| China Merchants Bank | Cash | |||||
| IBKR | SGOV | |||||
| Housing Provident Fund | ||||||
| Insurance | ||||||
| Medical Insurance Balance | ||||||
| Four Major Insurance Types | ||||||
| Charity & Liabilities | ||||||
| Mobile Phone / Telecom Fees | ||||||
| Charitable Donations | ||||||
| Mortgage | ||||||
| Vehicle Expenses |
Super Individual
Previously, knowledge across different domains was easily monopolized by industry insiders, such as coding, financial management, and legal counsel.
Now, if you want to start a low-cost business, everything from LOGO and letterhead design to drafting legal documents can be generated directly using AI.
This has led to various info sources hyping up stories of someone with zero development experience building some app and getting tens of millions in valuation—but is this real? Is it just hyping anxiety again? Is it a tactic to hype up excitement first and monetize later?
I believe that no matter how noisy the world gets, it has nothing to do with me. Start with the real needs right in front of you, and use AI step-by-step to expand your own capability boundaries.
I believe everyone has urgent and important matters to deal with at any given moment. Try turning off your phone, rapidly advancing your progress on those matters, or even trying cross-disciplinary work with AI, which can also save a lot of money.
Selling Shovel Heads
Going a step further, the information sources fanning anxiety around “super individuals” all aim to sell you AI shovels.
So we can think from first principles: where do shovels come from? Can we produce shovel heads and harvest money back from those hype-mongers?
For instance, since B2B multi-agent workers are popular, I could create a PDF-to-multi-agent project. I could even generate an Agent for every paragraph, so Warren Buffett and the Interpretation of Financial Statements could produce an army of investment research Agents. Then I could post advertorials to hype myself up, just like attracting VC investments at the peak of the PPT bubble in 2015. Even though, once calmed down, anyone can see that a system incapable of creating incremental value is worthless.
Even if you can’t extract cash back from them, I believe you will have gained the ability to build a shovel yourself—so what is there to be anxious about?
Digital Footprints
I believe interviewing is also a way to measure one’s worth, and interviewing logic in the AI era will build upon traditional rote-memorization questions (“baguwen”) while incorporating the inspection of footprints left in the digital world.
After all, one can cheat in interviews using speech-to-text or screen overlays, and project practices can be implemented in minutes via Codex.
Future interviews might evolve from asking rote questions to asking rote questions plus using AI to rapidly implement an entire project solution.
Just like taking a photo while holding today’s newspaper to prove the date, your everyday blog posts, reflections, and GitHub commits are authentic and traceable methods to prove yourself.
Preserving the Human Touch
When you scroll across an AI script, AI animation, or AI voiceover, do you keep watching or swipe away?
The inevitable result of AI proliferation, much like overused memes, is that people will become increasingly repelled by generated content.
At this time, you have to be like the chubby guy who cooked Nine-Turn Large Intestines: even if it’s not washed completely clean, a bit of “flavor” must be left so the judges know the ingredient is indeed pig intestine.
Therefore, human emotion and clumsy expression are difficult for AI to replicate. That is why not a single word of this article was generated by AI or polished by AI; I typed it out word by word using the clunky Microsoft Pinyin IME.
Thinking about how young boys and girls in the future might use AI to generate love letters dripping with robotic tones and a 100% similarity rate for their naive confessions makes me unable to hold back a laugh.
Let’s make proper use of human language and speak like human beings, shall we?
Beyond AI
Global Control
During the Spring Festival, I wanted to test where Codex’s limits lay, so I had it directly generate a space exploration game named VoyagerSimulator.
Codex indeed generated an executable binary program, but it didn’t know which engine was better to use, nor could it directly generate 3D models, design gameplay, draw concept art, or create user interfaces. It only outputted a terminal-like interactive applet, far from the hype on the market promising one-click creation of hit games.
Thus, on one hand, I reviewed traditional design patterns and architectures; on the other hand, through continuous interaction with Gemini, I gradually approached a globally optimal solution, and then split each part of the solution for Codex to implement.
Think about how humans conduct code reviews: under the premise that each piece of local logic is correct, strive to leave room for future changes.
Breaking Routine
AI can generate executable programs based on requirements, but its architecture or solutions tend to be conservative or traditional, whereas humans can invent all sorts of clever, unconventional hacks.
Suppose you need to display machine learning data charts on a web page. You could have an LLM implement it in JS, or using a backend language like Java. However, implementing it in JS is constrained by the browser’s single-threaded model, while using backend languages burdens the server.
After downing a cup of coffee, it suddenly hit me that I could piggyback on the user’s computer resources, distributing the computation logic to execution agents and communicating with web pages via ports. This was how the GopherMesh edge computing framework I recently wrote came to life.
I also worry that if one day, after drinking a stomach full of coffee, no inspiration or unconventional hacks emerge, does that mean I am about to be replaced by AI as well?
Catching Up on Liberal Arts
To spark more clever hacks, one needs to shift focus away from the binary world—to touch art, to converse with masters of ancient Greek philosophical schools, to study economics and finance, to reflect on historical cycles, to draw inspiration from non-technical solutions, and to ponder the limitations of human psychology.
Rather than competing head-on against AI on breadth of knowledge and coding speed, it is better to enhance your aesthetics, expand your capability boundaries beyond code, and let yourself become more “abstract.”
Over the past six months, I engaged in high-intensity philosophical dialogues with LLMs like Gemini, constantly reflecting on first principles and meta-knowledge. Looking back at myself last year, I feel like I was several versions behind.
Remember what was said above: “Large language models are a mirror; every answer reflects the person asking the question”?
Just as stupid questions only yield stupid answers, the art of questioning is itself a liberal arts skill.
I hope everyone reading this article, whether conversing with humans or AI, can continuously improve their signal-to-noise ratio.
Physical Troubleshooting
Normally, I like to zone out watching videos of computer repair techs venting over circuit boards. The ability to repair physical hardware is hard for AI to replace in the short term.
I even considered going “from master’s degree down to vocational school” to learn a repairman’s craft.
Although learning methods to fix the physical world requires significant time and effort, humans possess an immensely powerful capacity for information compression. Therefore, one can start by training oneself to rapidly infer the root causes of issues arising in work.
This is one of the few advantages held by “seasoned veterans” anxious about being replaced by AI: Nothing special, just practice making perfect.
Courage to Say No
Just as the disclaimer at the bottom of the Gemini page says: “Gemini is an AI tool, and its answers may not always be correct.”
I hope that during your interactions with AI, all of you maintain human subjective initiative. Argue and swear at the AI as frequently and fiercely as possible. Even if you argue incorrectly, it allows you to actively switch to a “Red Team Mode” perspective to view and think about problems or topics from multiple angles.
Perhaps in the future, my metric for judging the depth of a person’s thinking and their AI usage will be: the frequency with which they reject AI or passionately argue with AI.
When AI slips into hallucinations and talks nonsense, have the courage to say no.
When multiple AI conversations cannot reach a consensus, have the courage to say no.
When AI starts doing wild analysis that causes you to waver, have the courage to say no.
When AI violates Occam’s razor by unnecessarily multiplying entities, have the courage to say no.
When AI meddles in the real world or human emotions, have the courage to say no.
Just as I rejected any AI optimization for this article, this is my most sincere blessing to you all as a fellow human being.
EOF
