Photo of Zhilin Yang

Artificial intelligence & robotics

Zhilin Yang

Applying the ‘scaling laws’ of large models, enhancing model capabilities and popularizing long-context services.

Year Honored
2023

Organization
Moonshot AI

Region
China

Hails From
China
In recent years, the advent and rapid development of LLMs (Large Language Models) have attracted much attention to artificial intelligence (AI) technology. Behind this, the unremitting efforts of thousands of AI practitioners are indispensable. Zhilin Yang is one of them.

In April 2023, Zhilin founded the LLM startup Moonshot AI and became its CEO. Nearly 6 months later, the company launched Kimi, the world's first intelligent assistant that supports 200,000 Chinese characters input. It not only has functions such as translation, coding, long text summary and generation, online search, and data processing, but can also be applied to academic paper understanding and translation, legal issues analysis, and other scenarios. Long text input was the core competitive advantage of this product, surpassing the top LLMs at the time, such as Claude 2 and GPT-4. In addition, the company was valued at $3 billion just 15 months after its founding.

Before embarking on the road of entrepreneurship, Zhilin worked for Facebook AI Research, Google Brain and other top global AI institutions, and published the Transformer-XL model as a co-first author. The model proposes two new technologies, a fragment-level ATTENTION recurrent mechanism and a new relative position encoding. Transformer-XL can continuously generate thousands of words of relatively unified text on a single topic, has greater ability to model long distances (up to 900 words), and has higher optimization efficiency than the original Transformer and RNN. In addition, he also jointly released the Chinese LLM “Pangu" with hundreds of billions of parameters with Huawei Cloud. At present and in the future, he is dedicated to leading the team in exploring the optimal solutions for converting energy into intelligence, continuously reaching new heights in Artificial General Intelligence (AGI) technology based on the Scaling Law.