The iFlytek-Huawei collaboration is geared towards LLM training and offers different models, each with parameter counts to suit customer needs.
Photo by Fan Jianlei
By YU Hao
China's leading voice recognition and AI company iFlytek, has unveiled an upgraded iteration of their AI-powered large language model (LLM), Spark Desk V2.0, a collaborative effort with Huawei, described as an all-in-one solution to AI model training.
LIU Qingfeng, chairman of iFlytek, reckons Spark Desk is now comparable to ChatGPT
Spark Desk has undergone improvements in multi-modal capabilities, particularly in image description and analysis, and text generation from images. Early alpha testing indicates that up to 30 percent of the code generated by the assistant can be used directly. iFlytek claims this will result in a 30-percent increase in coding efficiency.
During the launch, Spark Desk V2.0 crafted an image inspired by a celebrated poem. iFlytek has developed several applications based on these functionalities and integrated them into their AI learning pad, an intelligent tutoring device.
The collaboration between iFlytek and Huawei is geared toward LLM training, dialogue and task sequencing. iFlytek wants to create a platform of diverse datasets and knowledge to construct tailored AI models. The platform offers different models, each with parameter counts to suit customer requirements.
While the release of Spark Desk V1.0 led to a 10-percent rise in iFlytek stock, the launch of V2.0 triggered an 8.63 percent decline. Factors influencing market sentiment might include LIU's reduction in his own holdings, a decision attributed to debt repayment.
Additionally, major shareholder China Mobile has plans to reduce its stake. There is little scope for optimism in iFlytek's most recent financials.
iFlytek's aspirations are ambitious, but investors are less and less interested in comparisons with ChatGPT. They are interested in immediate commercial viability or at least signs that things are heading in that direction. That means pleasing numbers in financial reports, not record deficits.