Chinese video gaming and social media giant Tencent Holdings has launched its large language model (LLM) called Hunyuan and announced its availability to the public through Tencent Cloud.
Hunyuan, a Large Language Model (LLM) with over 100 billion parameters and pre-trained using a vast dataset of over 2 trillion tokens, is currently accessible to Chinese enterprises for testing and app development through Tencent's cloud computing division, Tencent Cloud. This announcement was made by Tencent's Vice-President Jiang Jie during the primary forum of the company's 2023 Global Digital Ecosystem Summit held in Shenzhen.
Hunyuan has been integrated into testing for over 50 Tencent businesses, including Tencent Cloud, Tencent Advertising, Tencent Games, Tencent Financial Technology, Tencent Meetings, Tencent Docs, WeChat Search, QQ Browser, and other services and products.
Tencent's Hunyuan offers a diverse array of capabilities, encompassing tasks like image generation, copywriting, and text recognition. These functionalities can be applied across various industries, including finance, social media, e-commerce, and video gaming, among others.
Jiang highlighted that Hunyuan achieved the remarkable feat of generating an agriculture-related research paper containing over 4,000 Chinese characters. This accomplishment reportedly surpasses the capabilities of OpenAI's GPT-4. Additionally, Jiang mentioned that Hunyuan demonstrated double the speed in data training compared to "mainstream open-source frameworks," although he did not provide specific details about this performance improvement.
Tencent also introduced the Hunyuan Bot on Thursday, which is driven by the company's foundational AI model. This chatbot is now accessible as a mini-program within Tencent's widely used super app, WeChat. However, it's worth noting that Tencent's AI chatbot is currently designated as an "internal testing version," and its functions are limited to invited users only.
Tencent's introduction of Hunyuan reflects the Chinese tech industry's robust determination to narrow the gap with Western counterparts in developing services akin to ChatGPT. This pursuit aligns with the government's objectives of establishing a national standard for Large Language Models (LLMs) as part of its broader efforts to regulate artificial intelligence (AI) technologies.
In late August, China granted approval for its initial set of generative AI services to be used by the public. These services encompassed Baidu's Ernie Bot, SenseTime's SenseChat, as well as chatbots developed by state-supported startup Zhipu AI and Baichuan, a new venture founded by Sogou co-founder Wang Xiaochuan and backed by Tencent.