Tencent revealed that Hunyuan has now become the core technology underpinning over 50 of its various products and services. "By July, there are more than 130 large language models in China.
A war of a hundred models has begun," said Jiang Jie, Tencent's vice president. Hunyuan's introduction follows the recent unveiling of AI models by several Chinese tech companies, including Baidu Inc and SenseTime Group.
Tencent, China's most valuable internet company, disclosed that Hunyuan boasts over 100 billion parameters and was trained using more than two trillion tokens, which are commonly employed metrics to gauge the capabilities of AI models. In 2020, OpenAI's GPT-3 AI model had 175 billion parameters, while Meta Platform Inc's Llama 2 model boasted 70 billion parameters by 2023.
As per Reuters, Tencent has asserted that its own AI model, capable of conversing fluently in both Chinese and English, exhibits superior performance compared to OpenAI's ChatGPT, particularly in tasks involving the generation of lengthy text spanning thousands of words and the resolution of specific mathematical problems. Tencent has also noted that the company’s model encounters 30 per cent fewer instances of hallucination in comparison to Llama 2.
Moreover, the term "hallucination" in the context of AI refers to the situations where AI models generate inaccurate information but present it as if it were factual, a phenomenon commonly discussed by AI experts. (With inputs from Reuters)
. Read more on livemint.com