Underestimating Chinese AI Labs

In this episode, Stewart Alsop sits down with Joe Wilkinson of Artisan Growth Strategies to talk through how vibe coding is changing who gets to build software, why functional programming and immutability may be better suited for AI-written code, and how tools like LLMs are reshaping learning, work, and curiosity itself. The conversation ranges from Joe’s experience living in China and his perspective on Chinese AI labs like DeepSeek, Kimi, Minimax, and GLM, to mesh networks, Raspberry Pi–powered infrastructure, decentralization, and what sovereignty might mean in a world where intelligence is increasingly distributed. They also explore hallucinations, AlphaGo’s Move 37, and why creative “wrongness” may be essential for real breakthroughs, along with the tension between centralized power and open access to advanced technology. You can find more about Joe’s work at https://artisangrowthstrategies.com and follow him on X at https://x.com/artisangrowth.

Check out this GPT we trained on the conversation

Timestamps

00:00 – Vibe coding as a new learning unlock, China experience, information overload, and AI-powered ingestion systems
05:00 – Learning to code late, Exercism, syntax friction, AI as a real-time coding partner
10:00 – Functional programming, Elixir, immutability, and why AI struggles with mutable state
15:00 – Coding metaphors, “spooky action at a distance,” and making software AI-readable
20:00 – Raspberry Pi, personal servers, mesh networks, and peer-to-peer infrastructure
25:00 – Curiosity as activation energy, tech literacy gaps, and AI-enabled problem solving
30:00 – Knowledge work superpowers, decentralization, and small groups reshaping systems
35:00 – Open source vs open weights, Chinese AI labs, data ingestion, and competitive dynamics
40:00 – Power, safety, and why broad access to AI beats centralized control
45:00 – Hallucinations, AlphaGo’s Move 37, creativity, and logical consistency in AI
50:00 – Provenance, epistemology, ontologies, and risks of closed-loop science
55:00 – Centralization vs decentralization, sovereign countries, and post-global-order shifts
01:00:00 – U.S.–China dynamics, war skepticism, pragmatism, and cautious optimism about the future

Key Insights

1. Vibe coding fundamentally lowers the barrier to entry for technical creation by shifting the focus from syntax mastery to intent, structure, and iteration. Instead of learning code the traditional way and hitting constant friction, AI lets people learn by doing, correcting mistakes in real time, and gradually building mental models of how systems work, which changes who gets to participate in software creation.
2. Functional programming and immutability may be better aligned with AI-written code than object-oriented paradigms because they reduce hidden state and unintended side effects. By making data flows explicit and preventing “spooky action at a distance,” immutable systems are easier for both humans and AI to reason about, debug, and extend, especially as code becomes increasingly machine-authored.
3. AI is compressing the entire learning stack, from software to physical reality, enabling people to move fluidly between abstract knowledge and hands-on problem solving. Whether fixing hardware, setting up servers, or understanding networks, the combination of curiosity and AI assistance turns complex systems into navigable terrain rather than expert-only domains.
4. Decentralized infrastructure like mesh networks and personal servers becomes viable when cognitive overhead drops. What once required extreme dedication or specialist knowledge can now be done by small groups, meaning that relatively few motivated individuals can meaningfully change communication, resilience, and local autonomy without waiting for institutions to act.
5. Chinese AI labs are likely underestimated because they operate with different constraints, incentives, and cultural inputs. Their openness to alternative training methods, massive data ingestion, and open-weight strategies creates competitive pressure that limits monopolistic control by Western labs and gives users real leverage through choice.
6. Hallucinations and “mistakes” are not purely failures but potential sources of creative breakthroughs, similar to AlphaGo’s Move 37. If AI systems are overly constrained to consensus truth or authority-approved outputs, they risk losing the capacity for novel insight, suggesting that future progress depends on balancing correctness with exploratory freedom.
7. The next phase of decentralization may begin with sovereign countries before sovereign individuals, as AI enables smaller nations to reason from first principles in areas like medicine, regulation, and science. Rather than a collapse into chaos, this points toward a more pluralistic world where power, knowledge, and decision-making are distributed across many competing systems instead of centralized authorities.