Summary of Key Points
The release of Kimi K3 has suddenly made the Chinese AI company, Yuezhiànmiàn (Moon’s Dark Side), a sensation in Silicon Valley, turning it into a focal point in the global AI open-source competition. However, Kimi’s success is not an overnight achievement; it represents the culmination of talent and technical experience accumulated by China’s previous generation of AI companies (such as the “AI Four Dragons”). The news also raises three critical questions: Why would Meituan Longzhu invest in Yuezhiànmiàn, which still didn’t have a product at that time? How do the experiences of two generations of AI professionals converge on the large-model track? And how can Chinese AI companies catch up with the forefront when their computing power lags behind that of foreign companies?
1. Kimi Is Not a “Miraculous Solution”; It’s the Result of a Decade of AI Progress
You can view Kimi’s success as a “relay race” where the previous generation of AI professionals passed on their “batons” (technology, experience, and talent) to Yuezhiànmiàn:
- The Legacy of the Previous Generation: The “AI Four Dragons” (such as SenseTime and Megvii) have worked hard in areas like computer vision and speech recognition, nurturing a large number of professionals skilled in model development and data handling, as well as gaining valuable experience in avoiding common pitfalls.
- The Transfer of Knowledge within the Core Team: Key members of Yuezhiànmiàn, including Yang Zhilin (who conducted significant research at OpenAI) and Tang Jie (a renowned scholar in AI), along with many others from the Innovation Works AI Engineering Institute (which has trained numerous AI professionals), have brought their technical aspirations and experiences from the past decade to Kimi’s development. In other words, Kimi wasn’t created out of thin air; it’s a reapplication of the foundations laid by the AI industry over the years.
2. Meituan Invested in Yuezhiànmiàn (Without a Product) Because of the Team and Industry Trends
Meituan Longzhu invested in Yuezhiànmiàn not because it had a product ready, but based on two key factors:
- The Team’s Strength: The core team members are seasoned professionals with expertise in large-model technology, academic research, and practical application. Even without a product, they were likely to make significant progress.
- Industry Trends: Large models were seen as the future direction of AI, and as an investor, Meituan recognized the potential of this field. Kiwi (the investor behind Meituan Longzhu) personally experienced the rise of the “AI Four Dragons” and understood the value of these AI professionals, which led her to take the lead in the investment.
It’s similar to investing in a chef—not based on whether they have prepared a dish yet, but on their cooking skills and potential.
3. The Convergence of Two Generations of AI Professionals: Old Experience + New Track = New Opportunities
How did the talents from the previous generation of AI companies (the “AI Four Dragons”) integrate with the new generation of large-model teams? It’s about adapting old experiences to new challenges:
- The Value of Past Experience: Skills such as data handling and model optimization are still crucial in the era of large models. For example, the knowledge gained from handling data issues in the past can be applied to making models more efficient.
- New Models Require These Skills: Large models rely on a combination of data, algorithms, and models, and the experiences of the previous generation can be highly valuable. For instance, professionals from the “AI Four Dragons” can apply their expertise in computer vision to optimize large-model training or use their knowledge of speech recognition in dialogue systems like Kimi.
4. Overcoming Computational Power Gaps with “Soft Power”
While there is a gap in computational power (e.g., the number of GPUs) between China and foreign countries, this doesn’t mean China can’t catch up. Several “soft powers” can help bridge this gap:
- Algorithm Optimization: More sophisticated algorithms can reduce the need for additional computing resources. By optimizing code, we can train models with fewer GPUs, similar to packing goods more efficiently so that smaller vehicles can carry more.
- Open Source Ecosystem: By making Kimi’s technology open source, more developers can contribute to its improvement, accelerating progress and attracting more participants in China’s AI development.
- Advantages of Chinese Language Data: Large models require extensive data for training, and China has a wealth of Chinese-language data (articles, conversations, scenarios), which can help models better understand the needs of Chinese users. This could give Chinese models an edge over their foreign counterparts.
- Customized Applications in Vertical Fields: We can develop tailored models for specific industries like healthcare and education based on local requirements. For example, creating medical AI for Chinese doctors or educational AI that fits Chinese students’ learning habits.
In summary, even if China lacks in hardware, it can compensate with software innovation, algorithms, an open-source ecosystem, and localized applications. Kimi’s success demonstrates that China’s AI progress is the result of a decade of accumulation. In the future, even with computational power disparities, by leveraging talent, open-source collaboration, and local insights, China can still have a significant role in the global AI landscape. The story of Yuezhiànmiàn reflects the “steady growth” of China’s AI industry.