Interface News Reporter | Yu Hao
Interface News Editor |
When the iFLYTEK Spark cognitive model was officially released on May 6 this year, iFLYTEK announced that iFLYTEK Spark will have three milestones of continuous upgrading within the year, and breaking through the open-ended question and answer, multi-round dialogue ability and mathematical ability are one of the milestones.
On June 9, iFLYTEK Xinghuo Cognitive Model V1.5 was released as scheduled, and the key ability of this update is to break through open-ended question and answer, multi-round dialogue ability and mathematical ability. According to Wu Xiaoru, President of iFLYTEK, version 1.5 of the Spark Cognitive Big Model has improved by 24% compared with the previous relatively weak knowledge Q&A, and has improved text generation, language understanding, logical mathematics ability and code ability by 7%, 8%, 10%, 9% and 8% respectively.
In terms of open-ended question and answer, due to the inherent mechanism of the large model, the update of new knowledge is a difficult problem, and the timeliness of the knowledge contained in the provided answers is often insufficient. Wu Xiaoru said that by combining the language understanding ability and summary expression ability of the large model with the class search plug-in, the Spark big model effectively solved the industry problems such as the difficulty of updating new knowledge and the easy "Zhang Guan Li Dai" of fact-based Q&A, and achieved a 24% improvement in knowledge Q&A ability.
As for the natural language comprehension ability related to the quality of knowledge questions and answers, Liu Cong, Dean of iFLYTEK Research Institute, chose to demonstrate the ability of iFLYTEK Spark Cognitive Model V1.5 with cases such as Chinese essay questions and writing civil dispute pleadings in the 2023 National College Entrance Examination Paper.
In order to prove the improvement of logical reasoning and mathematical ability, Liu Cong proposed some basic logic questions and college entrance examination data questions to iFLYTEK Xinghuo Cognitive Model V1.5, and the product also made corresponding answers.
At this press conference, iFLYTEK also released the Xinghuo APP, which is supported by iFLYTEK Spark Cognitive Model V1.5, which is launched 200+ assistants for high-frequency use scenarios such as life and work, and users can use products and services through PC, Android APP, WeChat mini program and other forms.
It is worth mentioning that the Spark APP and Mini Program can support pure voice dialogue, multi-modal input and other functions, which can realize barrier-free voice communication, support graphic and text recognition, mathematical formula recognition, and present in a variety of terminals in a variety of forms.
In terms of application landing, according to Liu Cong, iFLYTEK Xinghuo cognitive large model has been implemented in education, medical, industrial, office and other fields, including Xinghuo language APP, medical post-diagnosis rehabilitation management platform, antelope industrial Internet platform, iFLYTEK hearing smart screen and other products.
At the meeting, Liu Cong also demonstrated the ability of the above products to complete tasks in complex scenarios such as English speaking practice, medical diagnosis, and enterprise demand matching. The Xinghuo Language Partner APP can conduct open dialogue, situational communication, real-time oral error correction, and support video dialogue with virtual human teachers; Industrial AI "Antelope Machine" can match solutions, service providers, experts and other resources for enterprise needs; The post-medical rehabilitation management platform can automatically analyze according to the patient's health portrait, generate a personalized rehabilitation plan and urge the patient to implement it according to the plan.
According to Wu Xiaoru, the data shows that after accessing the iFLYTEK post-diagnosis rehabilitation management platform, the management efficiency of doctors has increased by more than 10 times, the follow-up rate and consultation response rate of patients in the recovery process have reached 100%, and the satisfaction rate of discharged patients has reached more than 98%. At present, it has cooperated with Peking Union Medical College Hospital, Sichuan West China Hospital, Wuhan Tongji Hospital, Anhui Provincial Hospital, First Affiliated Hospital of Anhui Medical University, Zaozhuang Municipal Hospital and other hospitals in post-diagnosis rehabilitation management.
In the process of R&D and implementation of large models, openness and ecological construction are particularly critical. Previously, OpenAI has linked third-party applications to launch multiple plug-ins to enrich its application ecosystem.
At the conference, Liu Qingfeng, chairman of iFLYTEK, announced the opening of the Xinghuo cognitive big model development interface, fully opening the seven dimensional capabilities and 200+ assistant corresponding capabilities to developers, and supporting rapid integration of multi-terminal access, supporting private deployment, and building an ecosystem with the majority of developers.
He reiterated the goal of "this year's 1024 will benchmark ChatGPT: Chinese surpass, English equivalent", while emphasizing that it is not only to pay tribute to and catch up with OpenAI, but also to do independent innovation at the source. According to him, the National Key Laboratory of Cognitive Intelligence undertaken by iFLYTEK has been deployed in many fields such as brain-like intelligence, neural network large models, and game intelligence, exploring more potential paths and opportunities for cutting-edge cross-research.