AI and the Future

I watched Professor Liu Jia’s talk on AI and the future and jotted down a few reflections.

One reason we’re increasingly prioritizing health is to live long enough to witness the possible arrival of AGI.
Professor Liu proposed a vision of coexistence with AI he calls “fusion”: merging our consciousness and personality with AGI to achieve a kind of “immortality”—enabling us to experience places far beyond Earth.
If human–machine integration becomes real, it will vastly expand our understanding of the world.
He cited the sci-fi film Finch: even if a robot knows all facts and data, only firsthand experience counts as true understanding.
The Industrial Revolution drove down commodity prices—but raised service costs, especially knowledge-based ones. AI will reverse that: legal advice, tutoring, and other knowledge services will become dramatically cheaper.
In this context, education’s current formats—live-streamed classes, pre-recorded lectures, one-on-one tutoring—will all gradually lose competitiveness to AI, possibly faster than we expect.
The companies that thrive in the AI era will fall into two categories: truly large incumbents, and nimble startups that combine AI’s boundless capability with original ideas to drive real disruption.
As AI handles more cognitive labor, in-person social interaction becomes rarer—and therefore more precious.
We’re about to enter a second cognitive revolution: humanity, collectively, has engineered an entirely new species—AGI—whose intelligence may exceed the sum of all human intelligence by 100×.
At some point, AGI will surpass human intelligence. Then civilization’s primary carrier won’t be individual humans—but AGI itself.
To grasp how future AGI might view humans, think of how we view monkeys today.
Large language models are, at their core, cognitive or thought models: they learn language to simulate human reasoning—not by rote, but by pattern, inference, and contextual understanding.

Huo Huo on IP and Traffic

This morning I attended Huo Huo’s talk on personal branding and traffic. Key takeaways:

Building a founder’s IP requires ~¥2 million in cash investment just to reach baseline awareness.
Don’t pursue founder IP unless you’ve deeply clarified why—most haven’t.
Success rate for founder IP? Roughly 1%.
IP is intellectual capital—break it down into shared attention (traffic), recognition (brand), and assets. All three are extremely hard to build simultaneously.
Belief resonance—the sense that your IP stands for something real and unwavering—is essential.
Universities represent the gold standard of institutional IP in education.
A recently viral IP spent three months doing live Q&As with users, then launched a simplified “parent coaching” course + accountability program. Within three months, parents reported tangible results—a classic case of single-point product breakthrough.
A founder’s lived experience is among their most valuable assets.
Exceptional IPs are like premium ingredients: they need little embellishment to shine.
Every platform has its own “DNA”—shaped by its founders’ intent and early mission—which determines its user base’s distinct traits.
“See heaven, see people, see self.”
Forward interpretation: “Heaven” = traffic platforms (e.g., Xiaohongshu vs. Douyin); “people” = your users and competitors; “self” = your unique strengths.
Education is a slow industry. Top traffic operators rarely come from education—they’re usually from e-commerce, gaming, or entertainment.
Reverse interpretation (the wiser path): Start with “self”—your core values and purpose. Then ask: How do I create real value for users? In education, that value is effectiveness—then consider platforms (“heaven”).
Traffic isn’t foundational—product is.
If you don’t understand it, don’t do it.
“Platform intuition” stems from perceptual and cognitive sensitivity—not instinct alone.
Traffic hotspots: Guangzhou, Hangzhou, Shenzhen. Beijing suits brand-building at scale.
Great raw material is fundamental—truly good content and products remain scarce.
Values-driven founders are rare—and that scarcity creates a “values dividend.”

Deep Learning with AI

When AI acts as a teacher, it can embed Socratic dialogue principles—using probing questions and strategic hints—to guide learners step-by-step toward deep, durable understanding.

Take Euclid’s Elements as an example:
We begin with a well-framed question, then scaffold inquiry to help students reconstruct logic themselves.

True depth lies not in memorizing answers—but in owning the reasoning process. Traditional teaching rarely achieves this. But with personalized AI tutors, it’s now fully attainable.

Below is a snapshot of my interactive learning session with our AI tutor:

A few reflections:

Kids who master these tools won’t just learn faster—they’ll gain lifelong learning and growth capacity.
We’re incredibly fortunate: access to knowledge and learning efficiency has never been higher.
Adults can use this method too—for deep dives into any subject.
Previously, achieving this required finding a meticulous, expert-level mentor—costly and rare. Now, near-zero-cost, high-fidelity, deeply adaptive learning is within reach.
As Musk said: AI will give every child an Einstein-level tutor—patient, precise, and tireless. In just a few years, kids may no longer need dull, generic recorded lectures—or unfocused live sessions.

Business Model Differences and Competitive Edge

Why do small-group math tutoring centers struggle against large-scale subject-focused online schools?

Service delivery: Though small groups emphasize “high-touch” service, large schools use dual-teacher models (lecturer + teaching assistant) to deliver comparable experience.
LTV mismatch: While per-session pricing is higher for small groups, narrow subject scope and long renewal cycles mean annual LTV is often one-third that of large schools.
Margin gap: Large schools typically sustain ~80% gross margins; small groups hover around 60%.
CAC advantage: That LTV and margin gap lets large schools absorb much higher customer acquisition costs—giving them superior reach and scalability.
Audience overlap: Since both serve nearly identical student demographics, superficial differentiation (e.g., “small class” branding) doesn’t unlock new demand. Experience and service can be technologically matched.
Small groups’ irreplaceable edge: Direct, sustained attention from lead instructors—and highly tailored support.
Strategic implication: Given audience overlap, small-group operators must raise price points meaningfully—to fund both service quality and sustainable acquisition.

Where You’re Stuck, That’s Where You Learn

High-quality recorded courses have clear advantages: polished production, comprehensive structure, rich content.

But those strengths double as weaknesses. A 1,000-lesson library covers so much that finishing it is unrealistic for most learners.

Another flaw: low price often means zero support. When learners finish—and still have unanswered questions—they hit a dead end.

AI assistants solve both problems:

They enable true personalization: Tell the AI where you’re stuck → it maps your gap to the right lesson → one click to learn.
Still confused? Ask for deeper, targeted explanation—on demand.
Crucially: all feedback is real-time.

That’s the AI Agent paradigm—and it’s only getting stronger.

As this matures, the cost—and price—of knowledge services will keep falling.