iFlytek Unveils Upgraded Spark Cognitive Large Model V3.0 at World Sound Expo
Hefei, China – The 6th World Sound Expo and the 2023 iFlytek Global 1024 Developer Festival kicked off in Hefei on October 24, showcasing the latest advancements in artificial intelligence. During the event, iFlytek, a leading Chinese AI company, unveiled its upgraded Spark Cognitive Large Model V3.0 as scheduled.
The new model, which according to iFlytek Chairman Liu Qingfeng has surpassed both ChatGPT and GPT-4 in the medical field, marks a significant milestone in the development of large AI models. Liu stated that the era of large models has just begun and emphasized the importance of collaboration with developers and partners to achieve common evolution.
Since the release of the Spark model in May, iFlytek’s open platform has seen a significant increase in developer teams, with 1.43 million teams joining, representing a year-on-year increase of 331%. During the expo, iFlytek announced partnerships with industry leaders and jointly released 12 industry-specific large models. They also revealed plans to collaborate with Huawei on building the “Flying Star One” computing power platform based on the Ascend ecosystem, which will support the training of larger-scale models.
The Spark Cognitive Large Model V3.0, which was released in October, demonstrated superior capabilities compared to ChatGPT. Chinese language abilities exceeded those of ChatGPT, while English abilities achieved equivalent results in benchmarking tests. With 48 tasks successfully completed, the model showcased its potential in various industries. Since its public launch in September, iFlytek Spark has gained 12 million users.
The National Economic Research Institute of the Development Research Center of the State Council evaluated the Spark Model V3.0 and found that it surpassed ChatGPT in comprehensive capabilities, particularly excelling in the medical, legal, and education sectors. Liu Qingfeng confidently announced that the Spark model’s medical capabilities have exceeded GPT-4.
To enhance the user experience, iFlytek added a virtual personality function to Spark Model V3.0. By simulating personal traits, emotional understanding, and expression style, the model offers a more personalized AI persona. Currently, the “Xinghuo Buddy” virtual personality application is available to users.
iFlytek’s large-scale models are making significant strides in empowering industries. The Antelope Industrial model, based on the Spark model, has gained momentum in just over a month since its release, attracting a substantial number of users and serving numerous enterprises.
In the TO B market implementation, iFlytek provides both off-the-shelf common products and industry-specific solutions. For example, the Spark Code large model powers iFlyCode, an intelligent programming assistant that has already collaborated with institutions such as JD Cloud and iSoftStone.
The impact of large models is particularly evident in sectors like education and healthcare. At the expo, iFlytek unveiled the Spark AI learning machine equipped with the Spark Big Model V3.0. With enhanced interactive English AI Q&A functionality and personalized solutions, this innovation is expected to revolutionize learning experiences. In the medical field, iFlytek introduced the Spark Medical model, along with the supporting “iFlytek Xiaoyi” APP and mini-program. The medical model has undergone rigorous evaluations and surpassed GPT-4 in various medical-related tasks.
Looking ahead, iFlytek aims to benchmark GPT-4 in the first half of 2024. The company’s continued investment in large-scale general artificial intelligence cognitive models, combined with its thriving ecosystem and independent computing power base, provide confidence in achieving this goal. With over 5.53 million developers in its team and 12 million Spark users, iFlytek is well-positioned to make strides in the field of AI.
Despite the success, Liu Qingfeng emphasized the gap between Chinese large models and GPT-4 in complex knowledge reasoning, rapid learning with small samples, ultra-long text processing, and cross-modal unified understanding. To truly empower industries with cognitive large models, Liu stressed the importance of scientific progress and a commitment to seeking the truth from facts.