} ?>
(Yicai) Jan. 7 -- Kai-Fu Lee, a pioneer in artificial intelligence and the founder of 01.AI, has denied a report that Alibaba's cloud computing arm is in talks to buy the Chinese AI startup’s large language model pre-training team.
01.AI is not selling its pre-training team, infrastructure, and computing resources to Alibaba Cloud, Lee, former head of Google China, said on WeChat yesterday. Beijing-based 01.AI had revenue of more than CNY100 million (USD13.6 million) last year, Lee added.
Sources close to the matter previously told Yicai that Alibaba Cloud and 01.AI had agreed on a price for the team, with the deal involving around 60 staff dedicated to LLM pre-training. It would not cover 01.AI's other businesses, including its to-business operations in China and to-customer operations overseas, they noted.
Alibaba Cloud and 01.AI said on Jan. 2 that they will link arms on setting up an Industrial Foundation Model Joint Laboratory, aiming to hasten the transition of foundation models to practical use and further enhance the ecosystem for industrial large models.
The tie-up serves as a means for Alibaba Cloud to integrate 01.AI's technology, business, and talent in the to-business fields through a partnership rather than an acquisition.
This year will be a critical one for Chinese LLMs, Lee pointed out.
Several senior executives recently left 01.AI's pre-training team, with its leader Huang Wenhao joining TikTok's Chinese owner ByteDance in August, while core member Li Xiangang left around the same time to return to realtor Ke Holdings.
After Huang and Li's departure, 01.AI appointed a researcher to head up the pre-training team. LoRA, short for Low-Rank Adaptation, is a method for fine-tuning LLMs by reducing trainable parameters while minimizing the loss of model performance.
The researcher graduated from China's Beihang University and Kent State University in the United States and is based in North America. He has worked as a principal researcher at Microsoft's Azure AI.
Several members of Alibaba Cloud's pre-training team also left previously, with the firm eager to find a strategic AI leader, sources said. The company has also consistently prioritized high-performance small models as a key strategic direction, with LoRA being one of the important methods to achieve its goals, they added.
Editor: Martin Kadiev