Tencent, the Operator of China’s Super App WeChat, remains reserved amid the Generative AI Surge. Since launching its self-developed Large Language Model (LLM) Hunyuan in September, the tech giant has maintained a relatively low profile in its AI ventures.
This approach starkly contrasts with that of its competitor ByteDance, which is actively carving a niche in the generative AI space. ByteDance is rolling out a variety of consumer-facing products, aiming to capture new users in this burgeoning field, expected to be the next dominant user interface.
This conservative stance could pose a strategic risk for Tencent, potentially causing it to fall behind in the generative AI era as users increasingly flock to novel technologies and products. Furthermore, Tencent’s main foray into generative AI, its Hunyuan LLM, is a lackluster product, offering little in terms of differentiation from existing similar offerings.
Officially launched in September, the Tencent Hunyuan model is a pre-trained language model based on the Transformer architecture. It boasts a scale of over a hundred billion parameters and a pre-training corpus of over 2 trillion tokens.
Like many other similar models in China, it is capable of understanding and generating human language text, with support of multiple languages. It also provides developers with an array of API interfaces and tools for integration into various applications.
According to Tencent, over 50 Tencent businesses and products, including Tencent Cloud, Tencent Advertising, Tencent Games, Tencent Financial Technology, Tencent Meeting, Tencent Docs, WeChat Search, and QQ Browser, have already integrated the model for testing and have achieved preliminary results at the time of launch.
In October, Hunyuan model completed an upgrade and released a text-to-image function to the public. This new function is a Chinese-native text-to-image model, capable of creating artwork directly from any Chinese poetry or idioms entered by the user.
The model assists developers in writing code. It defaults to using Python, but it can also use other languages. The coding capabilities, according to testers, are not as impressive as other models as of October. According to the company, Tencent Cloud’s AI Code Assistant integrated the Hunyuan model, and the adoption rate by programmers reached 30%.
Two months after its release, it seems Hunyuan was used mostly by Tencent with over 180 internal Tencent businesses having already integrated the model. External customers from various industries such as retail, education, finance, medical, media, transportation, and government are using the model API through Tencent Cloud, says the company without naming any companies.
A Tencent official told Chinese media that the Hunyuan model has a high training cost. But for Tencent, the effectiveness is more important than the cost. Financial and resource constraints are not a significant concern for the tech giant.