Tencent's internal large model layout has seen another key adjustment. According to "Intelligent Emergence," Xu Can, a senior researcher at Tencent Yuan, has officially been transferred to the WeChat Business Group (WXG), joining the WeLM large model team, focusing on participating in post-training and Agent development. Xu Can is an AI expert with a background from Microsoft Asia Research Institute. He created the WizardLM project as the first author and proposed the Evol-Instruct method. He has deep expertise in synthetic data, reinforcement learning feedback, and automatic evaluation.
This personnel change comes at a time when WeChat AI is accelerating its implementation. In June, the native AI assistant of WeChat, "Xiaowei," has started a small-scale gray-scale test, supporting voice or text control of WeChat functions and directly calling mini programs to complete services.
At the same time, the WeLM team is advancing the research and development of a highly sparse MoE architecture, significantly reducing inference costs while maintaining the total parameter scale, making it suitable for WeChat's massive user scenarios. Xu Can's experience in post-training and complex task execution will help WeLM enhance task planning and tool calling capabilities, strengthening the technical foundation of the WeChat AI ecosystem.
Amid the pressure on the growth of the Yuanbao App, WeChat is leveraging WeLM and "Xiaowei" to carve out an independent path in the competition for C-end AI entry points.
