IFLYTEK Releases Spark V3.0 Cognitive Model Liu Qingfeng: There are Four Differences with GPT-4

On October 24th, NetEase Technology reported that at today's 6th World Sound Expo and 2023 Global 1024 Developers' Festival, iFlytek Chairman Liu Qingfeng and Research Institute Dean Liu Cong released iFlytek Spark Cognitive Model V3.0, stating that its overall performance surpasses ChatGPT, and its performance in the medical field surpasses GPT-4

On October 24th, NetEase Technology reported that at today's 6th World Sound Expo and 2023 Global 1024 Developers' Festival, iFlytek Chairman Liu Qingfeng and Research Institute Dean Liu Cong released iFlytek Spark Cognitive Model V3.0, stating that its overall performance surpasses ChatGPT, and its performance in the medical field surpasses GPT-4.

According to Liu Cong, the iFlytek Spark V3.0 cognitive model has been upgraded and improved in multiple aspects to better meet user needs. The AI persona has been upgraded, and through the virtual personality function, users can define their AI persona, including personality, emotional understanding, expression style, etc., in order to achieve dialogue with different characters.

It is understood that the iFlytek Spark V3.0 cognitive model has also been upgraded and optimized in areas such as mathematical automatic extraction of rules, small sample learning, code project level understanding ability, multimodal instruction following and detail expression. These upgrades have made the Spark Model more widely used, not only for tasks such as Q&A and dialogue, but also in fields such as intelligent programming assistants and technology literature models.

It is reported that in terms of application, iFlytek has released the Antelope Industrial Model for the industrial field; In the field of scientific research, we have collaborated with the Literature and Information Center of the Chinese Academy of Sciences to create a large model of scientific and technological literature, which can help researchers quickly understand the latest papers and improve research efficiency. In addition, applications such as AI psychological partners and heuristic English AI Q&A aids have been launched to meet the needs of different users.

In comparison with other large models, Liu Qingfeng pointed out that there are four differences between domestic large models and GPT-4. Firstly, in terms of computing power, GPT-4 has a parameter count of over 500 billion, while iFlytek Spark has a parameter count of 18 billion; Secondly, in terms of algorithms, GPT-4 has a more advanced algorithm architecture and algorithm capabilities; Thirdly, in terms of application scenarios, GPT-4 has a wider range of application scenarios and landing capabilities; Finally, in terms of data, GPT-4 has a richer and more diverse range of data sources and types. (Jiang Miao)