By CUI Peng

Tencent pushed its large language model (LLM) to the center of an already very crowded stage on Monday. 

Tencent Cloud’s LLM will cater to industries ranging from finance and media to travel and education.

According to TANG Daosheng, CEO of Tencent Cloud, a barrage of big-hitting clients have already been signed up, including the China Media Group, Shanghai University, and Fujian Big Data Group.

VP of Tencent Cloud WU Yunsheng said the company hoped to solve real-world problems with the LLM. Clients, he said, found LLM services expensive, but he claimed the price was justified. Wu said his LLM could quickly generate customized models by incorporating customer data. The LLM itself is not so expensive, the algorithm is. 

Using LLM in real business comes with a host of problems including costs and security. LLMs need computing resources. Compared to CPU servers, GPU servers have low stability, and training requires thousands of GPUs. It’s an intrinsically unstable setup. 

Meanwhile, the data from the Internet is a constant flux of fragmented information. Data quality is often poor and the costs of training a LLM become high.