Chinese platform company Tencent has unveiled its self-developed large language model (LLM) artificial intelligence (AI) called 'HunYuan.' Although it was relatively late to enter the market compared to competitors, Tencent claims that its performance surpasses that of the existing OpenAI models.


According to local media such as China Business Journal on the 7th, Tencent publicly demonstrated 'HunYuan' in Shenzhen and launched the service. Tencent explained that all related technologies were developed in-house, and HunYuan has over 100 billion parameters and used more than 2 trillion 'tokens.' Tokens are the smallest units of text used in AI training, which can be considered as words or characters.


[Image source=Reuters Yonhap News]

[Image source=Reuters Yonhap News]

View original image

This is comparable to the number of parameters used in OpenAI's GPT-3 (175 billion) and Meta's LLaMA 2 (70 billion). Although GPT-3 has more parameters, HunYuan has the advantage of supporting both Chinese and English.


In particular, Tencent explained that it has significantly improved the issue of 'hallucination' (AI fabrications). The research team stated, "HunYuan has 30% fewer hallucinations compared to LLaMA 2," adding, "Even the most advanced AI models tend to lie easily and manipulate facts when uncertain." They further noted, "The problem of hallucination is more severe in fields requiring multi-step reasoning," emphasizing, "HunYuan was trained using an algorithm for 'truth-seeking' rather than memorizing questions and answers."



Tencent operates China's largest social networking service (SNS), Weibo, and the largest messaging platform, WeChat. WeChat has 1.33 billion monthly active users and 115 million paid video subscribers. It is anticipated that combining emerging trends and new slang with HunYuan will enable the rapid provision of diverse services.


This content was produced with the assistance of AI translation services.

© The Asia Business Daily(www.asiae.co.kr). All rights reserved.

Today’s Briefing