Solving Real Problems with AI

A few recent reflections on using AI:

  1. As Andrew Ng put it: In the AI era, not understanding AI is the new illiteracy.
  2. The deeper I go, the clearer it becomes: AI’s power to unlock human potential and boost efficiency is immense—and we’re likely tapping only a tiny fraction of it. Like the brain, we’re still using just a sliver of what’s available.
  3. Don’t treat AI as mere entertainment. Its real impact emerges in boring, repetitive work—much like seasoned short-video creators who thrive on daily iteration.
  4. Use it daily—not abstractly, but to solve actual tasks on your desk. Let it force you to confront real problems head-on.
  5. Early internet adopters mostly saw it as a toy: chat rooms, games, browsing. Similarly, many now view AI as a novelty. But those who seize its transformative potential are the ones who deliberately reshape their thinking—and tackle real problems with it—even when the process feels tedious. That’s where the real dividends of the AI shift accrue.
  6. Kai-Fu Lee announced his next book will be written entirely with AI. That’s not hype—it’s confidence grounded in practice.
  7. Every domain, every workflow, every routine—there’s a path to reimagine it with AI.

AI and Subject-Based Learning

A group member asked: How can AI support sixth-grade learning?
That’s an excellent, grounded question.

First: Use AI by forcing yourself to use it.
Why force? Because truly disruptive tools demand a shift in mindset—not just new features, but new habits of thought.
The core logic isn’t “AI as fun toy,” but “AI as reliable problem-solving partner.”

How to start?
Anchor it in concrete use cases.
The question above—How can AI help in sixth grade?—is perfect. It’s specific, urgent, and tied to real life.
Traditionally, students might Google an answer. AI does that too—but better: you can ask follow-ups, clarify ambiguities, and keep probing until you truly understand. Compared to search engines, AI doesn’t just retrieve—it teaches.

How to “force” yourself?
If AI can solve it, don’t do it manually.

Practical steps:

  1. Pick a tool.
    Abroad: ChatGPT. In China: DouBao (available on app stores). Like ChatGPT, DouBao offers multiple AI agents—ready to “connect instantly.”

  2. Apply it to real learning tasks.
    Stuck on a math word problem? Ask DouBao for step-by-step reasoning—not just the answer. Confused by a science concept? Prompt it to explain like you’re 11, then give an example. Drafting an essay? Paste your rough version and ask for concise, grade-appropriate edits.

The core skill? Learning to ask well.
Many users quickly anthropomorphize AI—treating it like a person they shouldn’t “bother.” That’s a trap.
Remember: It’s a tool—not a judge. There’s no shame, no impatience, no mockery. If your first prompt falls flat, refine it. Ask again. Try a different angle. Your hesitation is human; the AI has zero ego.

Used intentionally, AI transforms education—especially at this age. It builds not just knowledge, but intellectual stamina: the habit of questioning, testing, revising, and owning understanding.

Here are subject-specific ideas (generated and refined with ChatGPT):

  • Chinese Language
    • Reading & writing: Paste a passage and ask ChatGPT to generate comprehension questions—or summarize its main idea in simple terms. For essays, submit a draft and request line-by-line suggestions for clarity and flow.
    • Vocabulary & idioms: Enter an unfamiliar term; get definitions, usage notes, and three original sentences at a sixth-grade level.

  • Math
    • Concepts & problem-solving: Input a problem (“A train leaves Beijing at 60 km/h…”), and ask for both the solution and the underlying logic—broken into clear, teachable steps.
    • Practice: Request 5 similar problems on fractions or geometry—with answer keys and brief explanations.

  • English
    • Grammar & vocabulary: Ask for the rule behind “has gone” vs. “went,” plus two custom examples.
    • Reading & writing: After reading a short text, ask targeted questions (“What’s the author’s main argument?”); for writing, request alternatives to overused words (“very,” “nice”) or help converting passive to active voice.

  • Homework Support
    • Submit a homework question + your attempted solution. ChatGPT can validate reasoning, spot gaps, and suggest next steps—without giving away the answer.

Polya’s Problem-Solving Method

Dr. Li summarized George Pólya’s classic How to Solve It with one insight: Every new problem is just a variation of an old one.

He calls this the “Polya Learning Method.”

Here’s how it works in practice:
My child loves writing numbers on the whiteboard—but struggles with “8.”
The traditional approach? Hand-over-hand guidance.
The Polya way? First, observe: What do we already know? “8” looks like two connected “o” shapes.
Then verify: Does the child already write “o” confidently? Yes.
So we say: “Let’s draw one ‘o’, then another right below it.”
Instant click. Done.

That moment wasn’t magic—it was pattern recognition + prior knowledge activation.

We extended it: When you face any new challenge, ask: What parts of this are already familiar?
Even deeper: What universal building blocks underlie it?
All digits and letters, for instance, are combinations of lines and curves.
When “p” felt awkward, we broke it down: “Draw a straight line (‘1’), then add the right half of an ‘o’.” Again—clarity.

The real win? The child internalized the mindset, not just the skill. That kind of self-guided decomposition now shows up everywhere—from puzzles to coding exercises.

Self-guided prompts (adapted from Pólya):

  1. Understanding the problem
    • Can you restate this in your own words?
    • Do you know the meaning of every term?
    • What’s given? What’s being asked?

  2. Devising a plan
    • Have you seen something like this before?
    • Can you split this into smaller, solvable pieces?
    • Is there a theorem, formula, or simpler case that fits?

  3. Carrying out the plan
    • Is each step clear and justified?
    • What’s needed at each stage?
    • If one path fails, what’s your backup?

  4. Looking back
    • Does your answer make sense?
    • Can you solve it another way?
    • Could this method apply elsewhere? Why does it work?

  5. Concrete prompts for deeper thinking
    • What’s the unknown?
    • Is any information irrelevant?
    • Can you sketch it? Write an equation?
    • Try a simpler number—what happens?
    • Can you find a counterexample?
    • How is this different from last week’s problem?
    • Can you work backward from the answer?
    • Does symmetry or an unconventional angle help?
    • What assumptions are hidden?
    • Could induction clarify it?
    • Can you explain your steps to a classmate?
    • Where else might this logic apply?

WeChat Reading’s “AI Ask Books”

WeChat Reading recently launched “AI Ask Books”—and it’s impressive.
It directly tackles AI’s biggest weakness: hallucinated or unverifiable answers.
Here, every response cites exact passages from books in WeChat Reading’s library—high-quality, vetted content. Each answer includes source links. For anyone who values precision over flair, it’s a revelation.

This is AI applied well: narrow scope, clear constraints, high-signal output.
It exemplifies today’s dominant domestic AI startup logic: Find a real, bounded problem—and augment it thoughtfully with AI.

AI-Generated Books

I ordered AI Teen: The Mars Survival Challenge on JD.com.
It’s billed as China’s first sci-fi novel co-written by a 9-year-old and a large language model.
I haven’t received it yet—but the signal matters.

Kai-Fu Lee says his next book will be AI-authored, end-to-end.
Wanweigang (in his Season 6 “Get” column) reports he now relies on AI for research, drafting, and fact-checking. His workflow is fully AI-integrated—not to save time, but to enable depth. He admits: If banned from AI, I couldn’t do my job.

In just months, for some professionals, AI has shifted from “nice to have” to non-negotiable infrastructure.