AI Photo: VCG

Major Chinese technology firms have recently ratcheted up efforts to upgrade artificial intelligence (AI) language models in a race to compete with OpenAI Inc's ChatGPT. 

Chinese voice-recognition company iFlytek reportedly said its SparkDesk aims to surpass ChatGPT in Chinese language possibly by the fourth quarter of 2023, while Baidu said its AI model 'Ernie 3.5' has now achieved broad enhancements.

At the ongoing 2023 Mobile World Congress (MWC) Shanghai, Jiang Tao, senior vice president of iFlytek, said the company aims to exceed ChatGPT's capabilities in Chinese language while reaching the same level with the chatbot in English in the fourth quarter of the year, domestic news outlet nbd.com.cn reported on Thursday.

On May 9, iFlytek introduced its AI large language model SparkDesk through a live demonstration in Hefei, East China's Anhui province, and made upgrade in June to be internet connected and provide real-time information, according to Jiang.

He said the company would plan to upgrade SparkDesk on August 15 to advance its coding capacity, including multi-modal perception and enhanced ability to produce text and images.

The significant application potential across a wide range of sectors has led to Chinese firms and institutions focusing on their own language models one after another. In addition to iFlytec, Chinese tech giants such as Alibaba, Tencent and Baidu have launched plans to develop a ChatGPT-like chatbot, saying services will be embedded into each company's applications.

In March, Chinese search engine giant Baidu took the stage in Beijing to showcase the company's new large language model named 'Ernie Bot'.

Just three months after the release of Ernie Bot, Baidu's large language model Ernie 3.5 has achieved broad enhancements in efficacy, functionality, and performance, Wang Haifeng, CTO of Baidu, told the Global Times on Thursday.

"These improvements are evident in creative writing, question and answer, reasoning, and code generation, as well as in training performance and inference performance," Wang said.

Global Times