E08. Gaoliming: AI Is the Oxygen of the New World

🎙️ Episode Preview

AI is not a tool. It is a shift in "oxygen concentration" — an environmental transformation, the opening act of a social revolution.

Once oxygen saturates the air, the old ecosystem will inevitably be rewritten.

In this episode, we are joined by Gaolimin to discuss: How do you identify the true "throughline" of an era? Why does a six-orders-of-magnitude change filter out most noise? How do you "avoid hard problems" and cultivate good habits for the new era, building "scaffolding" for your personal life? Entrepreneurial vitality, personal habits, corporate structures, value anchors — in the new world, all must be redefined.

This episode is not about "how to view AI." It is about how to spark your own vitality in a new world.

👤 Guest Bio

Gaolimin: First-generation Chinese internet entrepreneur, founder of Securities Star, and seasoned investor.

Spanning the search era, the mobile internet era, and the current AI era, he has long focused on underlying technology evolution, founder quality, and the core variables behind civilizational acceleration. He has been called "the sweeper monk of AI."

Follow his WeChat public account: "Avoid Hard Problems" (不解难题).

🕒 Selected Timestamps

01:22 Why could we identify OpenAI as a "six-orders-of-magnitude" signal as early as 2020?

05:00 "Normalcy is risk": Why must investing and entrepreneurship seek cracks at the "extremes"?

07:10 Civilization cannot pause: Why is the "Peach Blossom Spring" mentality a "loser's mentality"?

14:14 Warning: Are you doing "anaerobic work" in an "aerobic era"?

16:44 Why should an AI company with more than 5 people be viewed with suspicion?

23:13 Strong Take 1 (Coding): Why is 99.999% of a model's value in coding?

28:40 Strong Take 2 (RL): How can AI "build big companies in small markets" and mass-produce technology monopolies?

31:40 Strong Take 3 (Memory): Why does value lie not in workflow, but in "intent interception"?

35:16 Strong Take 4 (Leverage): The future is not 1-to-1 Copilot, but "remote pilot" mode.

41:35 Strong Take 5 (Infra): Blockchain was never built for C-end users. It was built for "machines."

47:51 Why does AI serve machines rather than humans? The relationship between humans and models is "rendering," not "true or false."

54:56 How can individuals keep pace with the times? Drucker: Seize external change and stand on the "steepest slope."

01:04:56 "The art of the negative": 8 PM bedtime, 4 AM wake-up, no dinner — how to build "scaffolding" for your life?

📚 References

Moore's Law: Computing power roughly doubles every 18–24 months, the foundation of exponential technological growth over the past 50 years. First proposed by Intel co-founder Gordon Moore in a 1965 paper.

Taste (judgment / aesthetic): The ability to "choose the right problem," more consequential than diligence. Physicist Chen-Ning Yang emphasized in multiple public lectures: two students of equal ability diverge because of "taste in problem selection."

RL (Reinforcement Learning): Through trial-and-error and feedback, models self-optimize strategies in an environment. The concept was systematized in the classic textbook by Sutton & Barto.

AlphaZero: Learned chess entirely through self-play without human data, a canonical case of "using compute to create technology monopolies." Published by DeepMind in a 2017 paper.

Token: A digital unit recording value, rights, or execution authority, a potential "currency form" for machine-to-machine transactions. The concept was systematically elaborated in the 2014 Ethereum Whitepaper.

Pre-training: Foundational training of models on large-scale corpora, equivalent to "general capability building." In management studies, Drucker used a similar metaphor to note that knowledge workers must periodically "retrain themselves."

Context: The core input determining a model's comprehension ability, foundational to the Attention mechanism. The concept originates from the 2017 Transformer paper Attention Is All You Need.

Rendering: Generative models do not "answer questions" but "render a version of the world" according to human expectations. Their generative nature derives from GAN, Diffusion, LLM, and other architectures.

🎵 Music

Jordan Critz - Beau Et Rapide (Piano)

🎤 Production Team

Host | Jinjian Zhang

Produced by | Oasis Capital

Editing & Production | Shengdu Studio Podcast Workshop

💬 Join the Conversation

If you view the next three years of development as a "quantum leap," which critical capability do you most want to strengthen: Taste (judgment), learning velocity, habit systems, or the ability to create high-quality Context?

Leave a comment below, or scan the QR code to join the group chat 👇

Or search WeChat ID VB20240606 to add our assistant

Disclaimer

All investment-related content in this podcast is for exchange and sharing purposes only, for reference, and does not constitute any market prediction, judgment, or investment or advisory recommendation. Thank you for your interest in this podcast's original content! If you repost or reference content from this podcast, please indicate the source. Contact Oasis Capital for prior consent before reposting.

View the full episode transcript on Xiaoyuzhou