Hunyuan, a large language model (LLM) that has more than 100 billion parameters and been pre-trained with over 2 trillion tokens, is now available for enterprises in China to test and build apps via the company’s cloud-computing arm Tencent Cloud, according to a Thursday announcement by Tencent vice-president Jiang Jie at the main forum of the company’s 2023 Global Digital Ecosystem Summit in Shenzhen.
A foundation model is a deep-learning algorithm, trained on mountains of raw data from the internet, that can be adapted to accomplish tasks in various AI applications. An early example of a foundation AI model is the LLM GPT-3 used by ChatGPT creator OpenAI.
Tencent’s Hunyuan provides a range of functions, including image creation, copywriting and text recognition, that can be applied in multiple industries such as finance, social media, e-commerce and video gaming.
This AI model is already integrated into the company’s various business units including Tencent Cloud, Tencent Games and WeChat. It is also being used in the company’s daily operations to help programmers write code and support customer service staff in answering user inquiries.
Tencent’s launch of Hunyuan reflects the Chinese tech industry’s strong interest in closing the gap with the West in building ChatGPT-like services, as the government sets out to implement a national standard for LLMs in line with efforts to regulate AI.
It also underscores the intense competition among China’s Big Tech companies in providing LLMs – the technology used to train AI chatbots – to help domestic enterprises become more efficient and push the large-scale modernisation of industries.
At its ecosystem summit that concludes this Friday, Tencent played up the advantages of Hunyuan over OpenAI’s GPT-3 and GPT-4 LLMs. Tencent’s Jiang said Hunyuan outperformed those models in abilities such as “handling complex tasks”.
Hunyuan, for example, was able to generate an agriculture-related research paper with more than 4,000 Chinese characters, a feat that OpenAI’s GPT-4 has not been able to accomplish, according to Jiang. Hunyuan was also doubling the speed of “mainstream open-source frameworks” in data training, he said without elaborating.
Online conferencing tool Tencent Meeting now features a Hunyuan-powered AI assistant that can generate summaries and to-dos based on conference recordings, while documentation tool Tencent Docs can now support dozens of text-creation scenarios “with just one click”, Jiang said.
Tencent on Thursday also unveiled Hunyuan Bot, powered by the firm’s foundation AI model, that is now available as a mini-program on its ubiquitous super app WeChat. Tencent’s AI chatbot, however, is still labelled as an “internal testing version” with functions that can only be accessed by invited users.
Jiang conducted a live demonstration at the summit of how the Hunyuan Bot can do self-introductions and solve simple maths problems.
Tencent, however, did not give details on when the chatbot would be open to the public, which resulted in many Chinese netizens expressing their disappointment online for the delay in trying out the country’s latest ChatGPT rival.
Although Tencent is promoting LLM enterprise use, senior company executives indicated that the tech giant expects to play a limited role in its industrial ecosystem, which now has nearly 11,000 partners around the world.
“We let partners play the leading role in building a new digital economy ecosystem,” said Tencent senior executive vice-president Dowson Tong, who is also chief executive at the company’s Cloud and Smart Industries Group, at the summit. He pointed out that Tencent will focus on product development, while its partners will work on the integration and sales of AI-powered services.