Credit: Visual China

BEIJING, July 4 (TMTPost) – 01 Wanwu Technology, a Beijing-based AI big model startup created by Kai-Fu Lee, has recently launched its official website.

The startup company aims to develop an AI 2.0 platform, focusing on self-researched large models. Within three months, the company has conducted internal testing of a 10 billion-parameter model and are currently expanding it to a range of 30 billion to 70 billion parameters. The startup boasts a team of renowned industry experts, including former vice presidents from Alibaba and Baidu, the former algorithm leader of Didi Chuxing and Baidu, as well as former executives from Google China.

According to public information, the official name of the company is Beijing 01 Wanwu Technology Co., LTD. Ma Jie, the former vice president of Baidu and the person in charge of metaverse product XiRang, holds a 99% stake in the company. Sinovation Ventures (Beijing) Enterprise Management Co., Ltd. holds the remaining 1% stake.

While Ma holds a majority stake of 99% and serves as an executive director of the company, he is not the CEO. Sinovation Ventures acts as a co-founder of the company, with Lee not solely dedicated to the company but rather balancing his involvement between the company and funding projects.

During the 2023 Zhongguancun Forum, Lee mentioned the global spread of the generative AI fever, represented by GPT-4 and other significant models, signifying the arrival of the AI 2.0 era. On March 19 of this year, he announced his venture into the AI big model field and personally led Project AI 2.0.

The company aims to be a global enterprise dedicated to constructing a new AI 2.0 platform and AI-first productivity applications. “Driven by a technology vision and embracing Chinese engineering excellence, the company seeks to attract world-class talent to join its innovative endeavors,” said Lee.

After three months, 01 Wanwu Technology, the seventh company incubated by Lee’s Sinovation Ventures, officially went online.

In order to develop AI 2.0 big language models, 01 Wanwu has opted to develop its own models and establish an experimental platform. It has built thousands of GPU clusters specifically for training, fine-tuning, and inference processes. This approach allows them to increase the number of effective data parameters and leverage high-quality data with greater density, enabling the construction of continuous, controlled, and scientifically trained capabilities.

The technical experts at the company come from diverse backgrounds, with a significant proportion of them possessing PhD qualifications. Within just three months, they have successfully conducted internal testing of models with a scale of 10 billion parameters and are actively expanding the capabilities to encompass models ranging from 30 billion to 70 billion parameters. Additionally, the company plans to release an open-source version of their model in the future. This initiative aims to attract exceptional talent to collaboratively enhance the model and foster the growth of a global AI 2.0 developer ecosystem.

Lee emphasized that Sinovation Ventures focuses on three key aspects in the AI 2.0 era, including intelligent applications, platforms, and infrastructure. Through active support for hard-tech entrepreneurs in the field of AI 2.0, it aims to help China continue to thrive in the global AI landscape throughout the forthcoming decade.