Dynamic circle formed for AI's adoption

ByteDance's Doubao large model team has open-sourced a new generation of MoE(Mixture of Experts) training optimization technology called COMET, achieving a 40 percent reduction in model training costs and an average 1.7-fold increase in training efficiency, a breakthrough that could potentially reduce the demand for high-end Nvidia GPUs.

DeepSeek has led the trend among Chinese companies in using algorithmic innovation to break through US chip restrictions, creating conditions for the widespread application of AI in China.

Historically, the United States has held a leading advantage in computing power and algorithms, while China has excelled in application scenarios and data scale. Now, Deep-Seek has disrupted this dynamic by using innovative algorithms to overcome the computing power constraints, unleashing China's advantages in application scenarios and data scale, a process that has accelerated thanks to innovations like Doubao. This means China can take the lead in applying AI across numerous industries and is forming a virtuous cycle of research, development, and application.