Some thoughts from attending the Yunqi Conference:
1. Although the cost of inference has been greatly reduced (the call price of Tongyi Qianwen API on Alibaba Cloud Bailian has dropped by 97%), scaling laws are still in effect, which requires manufacturers to increase capital expenditures to solve the problem. Wu Yongming said that "the investment threshold for advanced model competition in the world will reach billions and tens of billions of US dollars." The industry still focuses on capital investment. In the future, who will win the big model will not only be determined by technology, but also by the computing power reserves accumulated by funds;
2. At present, the application of AIGC is still at the mobile app stage (from ChatGPT to domestic manufacturers, all use apps as the main path), but this should not be the greatest imagination of AI. Wu Yongming believes that "the greatest imagination of AI is to infiltrate the digital world, take over the digital world, and change the physical world", that is, to completely imitate human actions and think about transforming the world. This is the starry sea of AI;
3. After the rise of AI, the demand for cloud computing in society has risen (in the new computing power market, more than 50% of the new demand is driven by AI). This gives cloud computing vendors a great opportunity to "have funds to reserve computing power - provide customers with additional cloud usage - increase revenue and profitability - continue to protect the computing power moat". Under the dual pressure of large models and customer cloud usage, the competitive landscape of cloud computing vendors will change rapidly: funds are not omnipotent, but without funds, it is absolutely impossible. Xiaopeng Motors announced that it will invest 3.5 billion yuan in research and development every year, of which 700 million yuan will be used for computing power training, which is also one of the demonstrations.
4. The dispute between closed source and open source is not only a discussion of the so-called technical route, but also a choice between "little money" and "all money". Simply put, the former costs less, has clear income, and has relatively low computing power requirements, but the scene is relatively single, while the latter has large investment, high computing power, strong collaboration, high computing power requirements, and rich scenes (such as the ability to build a large local area network model).$Alibaba(BABA)$
Comments