IFlytek Spark V3.0 Release: From Multiple Conversations to Active Q&A and then to Heuristic Q&A

On October 24th, at the 6th World Sound Expo and 2023 Global 1024 Developer Festival, iFlytek Chairman Liu Qingfeng and Research Institute President Liu Cong released iFlytek Spark Cognitive Model V3.0; Liu Qingfeng stated that the seven major capabilities of Spark 3

On October 24th, at the 6th World Sound Expo and 2023 Global 1024 Developer Festival, iFlytek Chairman Liu Qingfeng and Research Institute President Liu Cong released iFlytek Spark Cognitive Model V3.0; Liu Qingfeng stated that the seven major capabilities of Spark 3.0 continue to improve, surpassing ChatGPT as a whole and GPT-4 in the medical field, and upgrading heuristic dialogue and AI persona functions to create more personalized AI assistants.

Liu Qingfeng stated that for application scenarios, iFlySpark has further upgraded its intelligent programming assistant iFlyCode2.0 to the industry, and launched a Spark research assistant for researchers; Targeting the student community, we have launched AI psychological partners to care for the mental health of teenagers, and for the first time, we have released an inspiring interactive English AI Q&A learning aid and landed it in the iFlytek AI learning machine; For every family, release the iFlytek Spark Medical Big Model to create an AI health assistant for everyone. In addition, the iFlytek robot supercomputer platform is undergoing an upgrade, making its debut as the first humanoid robot with a large model and embodied intelligence.

According to iFlytek data, since the release of Spark on May 6th, iFlytek's open platform has added 1.43 million developer teams, a year-on-year increase of 331%. Among them, there are 178000 new model developers added, and iFlytek is using Spark's innovative application experience with over 100000 enterprise customers; IFlytek also collaborated with industry leaders to jointly release 12 industry models, accelerating the landing of the industry; On the other hand, iFlytek and Ascension Ecology jointly released the "Feixing 1" large model computing platform and launched larger parameter scale Starfire large model training for benchmarking GPT4, which will be achieved in the first half of 2024.

The big model is entering every one of our lives, "Liu Qingfeng said. Since the launch of the iFlytek Spark on September 5th, there have been 12 million users and higher expectations have been raised for the cognitive big model, such as" not only being able to answer questions but also being able to ask questions "and" not only having knowledge but also having personality ".

IFLYTEK stated that the iFLYTEK Spark Model V3.0 has added a virtual personality function, which can form an initial persona based on personality simulation, emotional understanding, and expression style, and then combine specific knowledge learning and dialogue memory learning to form a more personalized AI persona.

At the press conference, the iFlytek Spark Medical Big Model was officially released. Liu Qingfeng stated that medical treatment needs to be treated very rigorously. As early as 2017, iFlytek Intelligent Medical Assistant had already passed the national licensed physician qualification exam, ranking over 96.3% of human candidates. Firstly, they provide assistance to doctors in grassroots and level hospitals, helping them become AI diagnosis and treatment assistants for doctors. On this basis, they began to face families and create AI health assistants for everyone.

IFLYTEK stated that behind the ecology of Spark lies the foundation of independent innovation in computing power; The joint special forces of iFlytek and Huawei have achieved a doubling of the training and reasoning efficiency of the Spark model within six months; At the 1024 Global Developer Festival, iFlytek and Huawei jointly released a "Feixing 1" large model computing platform based on the Ascension ecosystem.

However, Liu Qingfeng also mentioned that domestic large models must face the gap with GPT-4. There is still a gap between domestic large models and GPT4 in terms of complex knowledge reasoning, small sample rapid learning, ultra long text processing, and cross modal unified understanding. "He said that only a scientific spirit of seeking truth from facts can truly achieve transcendence.