9x founder left the US to return to China, creating AI that ’caused a stir’

Zhilin, 34, caused a stir in Silicon Valley after the startup Moonshot AI he founded announced the Kimi K3 last week. This is the world’s largest open weight model (publicizing the parameters used in the training process, allowing others to download and modify) with 2.8 trillion parameters, expected to be fully released on July 27.

Theo Business InsiderKimi K3 can process hundreds of pages of text with just one command line, suitable for the task of analyzing long documents and large code bases. Its programming capabilities and agents are competitive with leading models from US companies such as OpenAI and Anthropic, but at a much lower cost, marking a new step for China in the AI ​​race.

Zhilin was born in Shantou, a coastal city in Guangdong. According to Australian Financial Reviewhe belongs to a generation of students with a talent for mathematics who were selected through China’s fiercely competitive Informatics Olympiad system, then entered the computer science major at Tsinghua University, the top school in the country.

He then moved to the US to complete a doctorate in machine learning at Carnegie Mellon University under the guidance of two famous AI researchers Ruslan Salakhutdinov and William Cohen. During his PhD, he interned at Google Brain and Meta, and published influential research on large language models (LLM) and long-context learning techniques with Turing Award winners Yoshua Bengio, a professor at the University of Montreal, and Yann LeCun, former chief AI scientist at Meta.

In addition to working in the laboratory, Yang Zhilin also plays drums in a university rock band and is a fan of the band Pink Floyd. He shared the similarities between innovation and rock music: “When it first appeared, rock was a completely new genre, representing extremely different ideas. During that period, the spirit of innovation and doing unique things was very strong.”

“He is truly excellent,” Salakhutdinov commented about Yang in a post on X on July 17, adding that Yang was sought after by many large technology companies in the US after graduating but was still determined to start a business in China.

Back home, Zhilin contributed to several major AI projects, including Huawei’s PanGu model and Beijing AI Institute’s large multimodal model Wu Dao. He also co-founded Recurrent AI, a startup that uses artificial intelligence to analyze sales conversations, helping businesses improve performance.

One of his principles in AI development is to aim for scale. “If the problem can be solved by scaling, don’t solve it with a new algorithm. The value of a new algorithm is to allow more efficient scaling,” he said.

Theo SCMPat the end of 2022, right after ChatGPT launched, he went to the US to learn about this application and recruit talent for his own project. Yang told Chinese media that he felt it was time to “ride the wave” of generative AI.

In March 2023, Yang Zhilin co-founded Moonshot AI with two old friends at Tsinghua University, Zhou Xinyu and Wu Yuxin. He raised $60 million and built a team of about 40 AI experts in just the first three months of establishment.

The company’s Chinese name means “Dark Area of ​​the Moon”, reminiscent of Pink Floyd’s 1970s album. According to some sources, meeting rooms in the company are also named after famous bands such as Rolling Stones or Led Zeppelin.

MoonshotAI team develops many AI systems and architectures such as Transformer-XL, RoPE, Group Normalization, ShuffleNet, MuonClip and Mooncake. However, the company first made a splash when it equipped the Kimi K2 platform model with an extremely large context window, up to one trillion parameters, to process long documents. They also launched the AI ​​assistant Kimi and expanded into programming, research, and automated agents, attracting funding from Alibaba, Tencent and many other large Chinese investors.

Now, Kimi K3 is bringing Moonshot AI and founder Yang Zhilin to the world stage. According to Guillermo Rauch, CEO of cloud computing company Vercel, Kimi K3 marks “the first time an open model has surpassed a proprietary model” in a comprehensive review of web engineering. Meanwhile, Ethan Mollick, a professor at the Wharton School of the University of Pennsylvania, said that the product “comes closest to today’s leading group of advanced models”.

After Kimi K3 made a splash, some famous people in the US technology field expressed regret that Zhilin did not stay here to work. Billionaire Vinod Khosla blames the harsh immigration policy of President Donald Trump’s administration, saying the US is “losing outstanding talents from other countries”.

Last year, the US announced a fee of $100,000 for businesses sponsoring new H1B visa applications for foreign workers. The regulation was later challenged by a federal judge and remains in litigation. In May, a memo from the US Citizenship and Immigration Services (USCIS) showed that people who previously applied for green cards in the US may now have to leave the country while the application is processed. In July, the government issued new regulations on the maximum length of stay for student visa holders.

According to Salakhutdinov, current immigration procedures to the US can make many people afraid and insecure. However, he said, Yang Zhilin was always determined to return to his hometown to build a startup.

“I remember he once said that if he didn’t try starting a business, he would regret it for the rest of his life. I respect that, and he was right,” he wrote on X.