iFlytek Claims Its Large Language Model Outperforms ChatGPT in Three Key Areas

On May 6th, iFlytek, a Chinese information technology company, showcased its large language model’s seven core competencies and shared application results in four major industries: education, office, automotive, and digital employees. iFlytek’s chairman, Liu Qingfeng, claimed that their large language model named SparkDesk has outperformed ChatGPT in three key areas: text generation, question and answer, and mathematical ability. He also expressed confidence that iFlytek’s model would surpass ChatGPT entirely by the end of October.

Liu stated that after analyzing ChatGPT’s 48 primary task directions and the needs of over four million developer teams on iFlytek’s open platform, they concluded that general artificial intelligence should possess seven core competencies. These competencies include text generation, language understanding, question and answer, logical reasoning, mathematical ability, code ability, and multimodal ability, among others.

The cognitive model presents new opportunities for addressing the most pressing human needs, such as changing how we acquire and distribute information, innovating content production, interacting naturally with virtual assistants, and subverting traditional programming modes. “Our aim is to empower every creative person to become an entrepreneur in the digital economy,” Liu stated.

During the press conference, iFlytek demonstrated the seven core competencies of SparkDesk through on-site testing. The model was presented with a prompt: “Sun Wukong, one of the main characters in the 16th-century Chinese novel Journey to the West, and Ultraman, the main hero of the Ultraman series, have snuck into the venue. They both want to be your spokesperson and get into a heated argument. Please create a story based on this information.” The model quickly generated a vivid story, with different results each time it was regenerated. “This variability is also a key feature of our model,” Liu remarked. The successful test underscores the model’s potential in natural language processing and content creation.

Furthermore, SparkDesk can perform a variety of tasks, such as composing emails, creating schedules, drafting press releases, generating English propaganda, and correcting English grammar errors. Its versatility as an AI assistant was demonstrated during the press conference. Liu emphasized that when it comes to generating long text, the model is significantly ahead in China and surpasses ChatGPT in Chinese language abilities. He added that the iFlytek team is continually improving the model’s capabilities to meet the evolving demands of users.

SEE ALSO: iFlytek Denies Layoffs Rumor, Plans to Recruit Over 1,000 Employees in 2022

iFlytek also presented the application results of SparkDesk in various fields, including education, office, automotive, and digital employees. Examples of applications showcased include AI-assisted composition correction that mimics the work of teachers, virtual oral teachers that can converse with students, automatic generation of meeting minutes based on handwritten notes, more natural and anthropomorphic communication, smarter and safer driving experiences, and assistance with repetitive work for employees.