Summary of the Analysis
This is a record of an informal conversation among industry veterans who have experienced multiple technological waves in the fields of portals, search engines, mobile internet, and short videos. The discussion is unfiltered and lacks the rhetoric typically used in roadshows or industry hype. It bypasses the popular topics being discussed online, such as large model parameters, multimodality, and agents, and presents several counterintuitive insights: AI is not the lightweight internet software revolution people think it is; at its core, it is a heavy industry that relies on tangible resources. The energy and manufacturing capabilities of the physical world represent the real limitations of AI. The current linear expectation of AI growing exponentially each year is unfounded and can easily lead to bubbles and unnecessary anxiety. AI will fundamentally change the value of human connections; instead of competing with AI for jobs, the ability to assume legal responsibility will become a scarce resource. When AI moves from the digital realm to the physical industry, China’s decades of built-up industrial infrastructure will become a unique competitive advantage. While there are no definitive answers to these assertions, they all challenge the myths commonly promoted in the AI community.
---
Simplified Explanation of Key Points
1. Don’t Let Parameters Fool You: AI Is a Heavy Industry That Consumes Energy and Hardware
The current focus on AI often revolves around comparing model parameters and algorithms, similar to comparing the horsepower of cars. However, no one realizes that even the most powerful car won’t go fast on a two-lane road without gas stations or sufficient fuel. The fundamental architecture of AI, the Transformer, involves processing vast amounts of data, which requires expensive hardware: chips, servers, data centers, cooling systems, and continuous power supply. The energy consumed to train large models is equivalent to what an average household would use over hundreds of years. Without adequate power, even optimized code won’t work effectively. The competition in AI is not just about programmers writing code in offices; it’s a global competition involving chip production, power grids, land resources, and manufacturing capabilities. Until affordable and controllable nuclear fusion becomes a reality, energy will be a limiting factor for the AI industry.
2. Don’t Believe the Myth That AI Will Replace Everyone
The idea that AI will replace all jobs at a constant rate is a misconception. Technological growth is never linear. We’ve seen this before with the internet and mobile phones, where initial rapid growth eventually hit a ceiling. Many AI startups use exaggerated growth forecasts to attract funding, leading people to worry about job losses. In the future, the value of connections will shift from practical to emotional, as AI will handle many tasks, leaving human relationships as a more valuable asset.
3. Building Connections Is Becoming Less Valuable
In the past, connections were valuable because professional skills were scarce. Knowing a doctor or lawyer provided immediate access to those services. However, AI will handle these tasks, reducing the need for such connections. In the future, the value of connections will lie in emotional bonds like family and friendship, rather than their practical use in obtaining resources.
4. Don’t Compete with AI for Jobs: Your True Value Lies in Legal Responsibility
AI can handle tasks like writing contracts and making investment decisions, but humans must bear the consequences of any mistakes. The ability to legally oversee these processes will be crucial. In the future, the real asset might be the human identity that can take responsibility for AI’s actions.
5. China’s Advantage Lies in Its Industrial Base
The competition in AI doesn’t just focus on models and algorithms. As AI moves into physical industries, China’s extensive industrial infrastructure and manufacturing capabilities will become a unique advantage that others cannot easily replicate.