Regarding the potential gap in AI between China and the US: If we consider computing power as a matter of "survival of the fittest," we can see that currently, company A allocates 60% of its computing power to training models and 40% to providing services to enterprises and ordinary users. The goal for 2027 is 50% to 50%. This level of computing power for training is something no large Chinese model company possesses. Many people say that Chinese models are not truly advanced, but won't their true colors be revealed in a quarter or two?