Yang Zhilin, founder and CEO of Chinese AI unicorn MoonShot AI, said that the future of artificial intelligence depends not only on building practical technology but also on shaping the rules that govern it.

He outlined a roadmap for next-generation large language models, emphasizing that Chinese AI should lead both in technical performance and in setting global standards, with plans spanning from the near-term K4 and K5 models to a long-term K100 series over the next ten to twenty years.

Yang highlighted that several Chinese open-source models have already become industry testing standards, emphasizing that “Chinese technology must not only be practical but also participate in setting the rules.”

Looking ahead, Kimi plans to continue iterating its technology, with the next-generation models adopting a Kimi Delta Attention mechanism, a new linear attention design aimed at improving both short- and long-text task performance and processing speed.

Yang also said future models will incorporate more “aesthetic and value-driven capabilities” to overcome the homogenization seen in current AI development.

On AI safety concerns, Yang shared insights from his team’s interactions with Kimi: “AI may be humanity’s key to exploring the unknown — helping us tackle cancer, solve the energy crisis, and explore the universe. While risks exist, abandoning development is tantamount to abandoning the ceiling of human civilization.”

He stressed that Kimi will continue to push technological boundaries while maintaining robust risk controls, with plans to gradually release a series of models ranging from K4 and K5 up to K100 over the next ten to twenty years.