Summary of Key Points
Through a week of interviews in Silicon Valley, the author discovered a significant "temperature difference" between AI in Silicon Valley and other regions of the United States. The paths taken by China and the US in developing AI are difficult to replicate due to local environmental differences. The value of AI may not necessarily lie in star companies; instead, it could be found in small businesses that are closely integrated with real-world scenarios. AI is changing the way people work—eliminating waiting times and overcoming bottlenecks, with high-quality feedback being more important than mere commands. The cutting-edge AI developments in Silicon Valley evolve every two to three months, so it's crucial to focus on steady progress rather than sudden breakthroughs.
Detailed Analysis
1. Silicon Valley AI ≠ US AI: The Internal Gap is Greater Than Expected
Silicon Valley is like an "AI pressure cooker"—professionals there discuss AI as a "social infrastructure" and a "new species," even researching ways for AI to improve itself automatically. However, outside of the Bay Area, most companies in the US still view AI as a tool for increasing efficiency. Ordinary people are concerned about job losses, while some universities see it as a potential threat, and others are slowly defining its boundaries. In contrast, although there are differences in the depth of AI adoption across cities in China, they all use similar models (such as DouBao and DeepSeek). Outside of Silicon Valley in the US (for example, people in New York or Los Angeles who are not in the tech industry), ChatGPT is often seen merely as a tool similar to Google. It's more the people in lower-tier Chinese cities who have developed more innovative uses for AI.
In one sentence: AI development in the US is "layered"—labs, investors, universities, and ordinary users are all on different timelines.
2. Different Paths for China and the US in AI Development
The past practice of copying each other's approaches in the internet era no longer applies to the AI revolution:
- US Approach: Focusing on models, computing power, and self-evolution, turning small problems into standardized software solutions (which can be valued at hundreds of millions of dollars) while relying on software subscription habits and capital support.
- Chinese Approach: Using inexpensive open-source models, the robust hardware supply chain in Shenzhen, and engaging in practical tasks in factories and stores (such as installing cameras for monitoring hundreds of stores) to solve complex local problems.
The bottlenecks on both sides are different: US investors don't understand the intricacies of the Chinese hardware supply chain, while Chinese teams face challenges with data compliance and labor costs when working in the US. While the surface aspects of products can be copied, the underlying payment methods, supply chains, and organizational practices remain unique.
3. A New Form of AI Value: Not Always in Star Companies
Silicon Valley tends to measure the value of AI companies by their valuation (e.g., a small software company valued at hundreds of millions of dollars), but this can lead to bubbles. In China, the value of AI is often found in less well-known applications:
- Large companies create internal teams to use AI for optimizing processes (such as procurement or store monitoring), which are then integrated into their main operations without forming separate AI companies.
- Small teams of three to five people can succeed with simple projects like store monitoring and even thrive without external investment.
These small businesses may not make it onto the unicorn list, but they represent a crucial aspect of AI's implementation in China. Focus on real changes that bring value to customers, rather than just the amount of funding raised.
4. AI Changing Work: Eliminating Waiting Times, Shifting Bottlenecks, and the Importance of Feedback
AI is first eliminating waiting times—tasks that used to take a week can now be completed in one night, allowing for new ideas to emerge quickly. However, the total amount of work hasn't decreased; instead, bottlenecks have just shifted. For example, while text generation and code creation have accelerated, clinical trials and factory construction remain slow (the "barrel theory" applies: faster processes in some areas create new bottlenecks in others).
Who is more important? Those who can solve the slower parts of the process. For instance, even if AI generates 100 ideas, valuable feedback from the real world (such as customer purchases or complaints) is essential to improve its accuracy. Providing specific reasons for issues (e.g., "This product is about to expire") is more useful than just giving general feedback.
5. The Rapid Evolution in Silicon Valley: Updates Every Two to Three Months
The most challenging aspect of keeping up with developments in Silicon Valley is the rapid change in ideas from leading figures. What was discussed half a year ago (e.g., using multiple AI agents together) has now shifted to focusing on user feedback. Companies are reevaluating their strategies, such as using AI during job interviews to assess candidates' abilities without exposing them to core data.
The author warns against overestimating sudden announcements; instead, pay attention to the small, steady improvements that occur daily (e.g., changing the focus of AI interviews from preventing cheating to assessing skills). These gradual changes can significantly impact work methods over time.
Final Thoughts
The author is not pessimistic: China's rich range of AI applications, large data volume, and strong manufacturing capabilities are advantages that Silicon Valley should leverage. There's no need to chase the most advanced models; open-source models are often sufficient for many use cases. The key is to integrate AI into local contexts to achieve real benefits. While Silicon Valley provides a valuable perspective, it's important to apply these insights back to one's own work to identify areas where AI can improve processes.