On June 27th, iFLYTEK unveiled the Xunfei Spark Large Model V4.0 and its related applications in Beijing. The Xunfei Spark V4.0 has seen a comprehensive enhancement in its seven core capabilities, surpassing GPT-4 Turbo overall, and ranking first in eight international mainstream test sets, marking a comprehensive lead for Chinese large models.
The era of large model personalization has arrived! The Xunfei Spark APP/Desk has been upgraded with the release of "Personal Space," creating an AI assistant that understands you better; for personalized applications in professional fields, iFLYTEK has upgraded the Xunfei Xiaoyi APP, launching a personal digital health space, and creating a health assistant for everyone; the industry's first Spark Intelligent Grading Machine is introduced, with "AI Teaching Assistant" to help teachers reduce their workload and teach students according to their aptitude; the Xunfei AI Learning Machine has been upgraded with a 1-on-1 Q&A tutoring feature, creating an AI learning assistant for every child.
Facing the era of the Internet of Everything, the Spark voice large model has made further breakthroughs, releasing 74 language/dialect switch-free conversations, solving the problem of voice recognition in strong interference scenarios, and releasing extremely complex scene voice transcription technology. Through cloud-edge-end and integrated software and hardware solutions, it empowers the transformation of human-computer interaction in fields such as automotive, home appliances, and robots. In addition, for the last mile of enterprise "AI+" scenario value implementation, iFLYTEK officially released the Spark Enterprise Intelligence Platform and introduced typical intelligent entity cases such as Spark Business Opportunity Assistant and Spark Bidding Assistant, helping enterprises create value.
Advertisement
In the evaluation of eight international mainstream test sets, Xunfei Spark V4.0 has surpassed GPT-4 Turbo overall. In this year's real test of college entrance examination questions, Xunfei Spark's scores in Chinese, mathematics, and foreign languages all ranked first, and it was rated as "a large model that is better at solving problems"; in scientific research, Xunfei Spark has helped Professor Liu Haiyan's team from the University of Science and Technology of China to increase the protein design success rate from 0.1% to 20%, and the design time from six months to one day; empowering everyone, it has helped a 70-year-old person with no legal knowledge to successfully reclaim their pension debt, and helped a deaf person fulfill their literary dream... Xunfei Spark is becoming an AI assistant for everyone in China.
Since its full opening last September, the Xunfei Spark APP has accumulated 131 million downloads in the Android open market, ranking first among Chinese tool-type general large model Apps, and has emerged with a group of popular assistants loved by users, including writing, programming, work, and learning. This year's "618 promotion," smart hardware with Spark large model support has seen a year-on-year sales increase of over 70%, with an average monthly usage exceeding 40 million times, and more and more users are beginning to enjoy the benefits brought by large models.
Based on the country's first domestic ten-thousand card computing power cluster "Feixing No.1," the Xunfei Spark Large Model V4.0 was officially released. The seven core capabilities of Xunfei Spark V4.0 have been fully upgraded, fully benchmarked against GPT-4 Turbo, and have achieved overall surpassing in aspects such as text generation, language understanding, knowledge Q&A, logical reasoning, and mathematical abilities.
Xunfei Spark V4.0 has further upgraded its image recognition capabilities, and its application effects in scientific research, finance, healthcare, judiciary, and office scenarios have taken the lead over GPT-4. In addition, Spark's long text capabilities have also been newly upgraded, and it has introduced the industry's first source tracing function for long document knowledge Q&A illusions.
External authoritative test sets also reflect the leading position of Xunfei Spark V4.0. Among 12 large model mainstream test sets in China and abroad, Xunfei Spark ranked first in eight tests, surpassing international large models such as GPT-4 Turbo, and Chinese large models are comprehensively leading.
On the spot, Liu Qingfeng demonstrated the effects of Xunfei Spark V4.0 in complex instructions, complex logical reasoning, spatial reasoning, high school mathematics, etc., and Spark's "intelligence quotient" has evolved again. Taking spatial reasoning as an example, "Bob is in the living room. He takes a cup and walks to the kitchen. He puts the ball into the cup, then takes the cup and walks to the bedroom. He turns the cup upside down and then walks to the garden. He puts the cup in the garden and then walks to the garage. Question: Where is the ball?" Xunfei Spark can infer the ball on the bedroom floor based on space and common sense, and these capabilities are significant for future embodied intelligence and home robots.The era of large model personalization has arrived! iFLYTEK Xinghuo launches the "Personal Space" for the first time, giving millions of users a one-click access to the "AI Smart Suite"
While large models bring convenience to work and life in China, there is a situation where the content generated by various companies is similar, generic, and not practical enough. How can we make large models more useful and create unique value in work and life? iFLYTEK has the answer - to create an AI assistant that understands you better.
How to create an AI assistant that understands you? Liu Qingfeng proposed that the AI assistant should be able to express itself based on user profiling, learn from usage history, and enhance learning based on personal information. When building a user's personal profile, the style can be chosen by oneself or dynamically improved based on conversations and usage history, thus forming a personalized expression style; the AI assistant, combined with personal information, can then generate personalized and targeted content.
Based on this, the iFLYTEK Xinghuo APP and desktop version have been completely upgraded and revamped, taking the lead in launching the "Personal Space". Users can upload their own work, study, life, health, and other types of materials to form a personal knowledge base for everyone, and then combine it with the character setting to allow the large model to generate more personalized content. In addition, iFLYTEK Xinghuo has launched 14 intelligent entities in the first batch, creating dedicated assistants for specific scenarios.
Liu Cong, Dean of the iFLYTEK Research Institute, demonstrated the "Personal Space" effect on site. When he uploaded his daughter's short essay and selected AI character tags that matched his daughter's style, Xinghuo generated a lively, cute, and more personalized article; when he uploaded the product poster of the iFLYTEK translation machine, user short videos, and related recordings, Xinghuo could also generate product training documents based on this multi-modal information, and multi-modal traceability of the generated information could also be performed. The large model has entered the personalized era, and the "usability" of the large model for work and study has soared!
In addition, the Xinghuo large model has also integrated the entire iFLYTEK C-end hardware and software product ecosystem, allowing millions of smart hardware users to have the "Xinghuo Suite" with one click. For example, files from iFLYTEK smart office books and smart recorders can be synchronized to the Xinghuo Personal Space with one click. Through data interconnection and operation linkage, synchronizing a meeting record from an office book to Xinghuo allows Xinghuo to perform official document writing, make PPTs, and generate to-do lists, bringing a more efficient office experience.
The Personal Digital Health Space is here! iFLYTEK Xiaoyi APP downloads exceed 12 million
Aiming at personalized applications in professional fields, iFLYTEK has upgraded the Xiaoyi APP and launched the Personal Digital Health Space, creating an AI health assistant for everyone and every family.
In the medical field, the iFLYTEK Xinghuo medical large model has been upgraded again, and its medical core capabilities have fully surpassed GPT-4 Turbo and GPT-4o. On this basis, the capabilities of the iFLYTEK Xiaoyi APP continue to be upgraded, covering 1,600 common diseases, 2,800 common drugs, and 6,000 common examinations and tests, meeting users' core health needs in the core scenarios before seeing a doctor, when taking medication, and after examination. Currently, the iFLYTEK Xiaoyi APP has a cumulative download volume of 12 million, with a user satisfaction rate of 98.8% and an active recommendation rate of 42%.
Liu Qingfeng introduced on site that the "Personal Digital Health Space" launched by the iFLYTEK Xiaoyi APP can build a personal digital health space based on individualized materials such as electronic medical records, examination reports, and physical examination reports. Before seeing a doctor, it can further analyze the causes of the disease, give personalized judgments on drug contraindications when taking medication, and provide data changes after examination by joint comparison. Through role switching, it can also understand the health status of other family members.The iFlytek Xiaoyi APP has currently passed multiple authoritative certifications for data security and privacy protection, further safeguarding the security of health data. Amid the current relative scarcity of medical resources, the emergence of the iFlytek Xiaoyi APP has effectively alleviated the urgent societal demand for medical services, providing a new model for personal and family health management.
The Spark Intelligent Grading Machine reduces teachers' homework grading burden by 90%. Thanks to the upgrade of the base model and the further enhancement of image and text recognition effects for complex educational scenarios, iFlytek has released its first Spark Intelligent Grading Machine. It integrates intelligent grading, precise learning situation, and personalized learning. It supports free layout and homework of any paper size, and while supporting intelligent grading of multiple subjects and question types, it can also instantly generate multi-dimensional learning situation reports. It also provides materials for teachers' homework review and face-to-face tutoring. Liu Cong demonstrated the entire process of the Spark Intelligent Grading Machine grading homework on the spot, with 15 student homework assignments graded in half a minute. The grading simulates real handwriting, almost the same as when teachers usually grade homework.
With the Spark Intelligent Grading Machine, teachers have an AI assistant that reduces their workload and increases efficiency, tailoring teaching to individual students. Homework that used to take 90 minutes to grade can now be completed in just 5 minutes; manual analysis of learning situations, which used to take 60 minutes, can now be done by Spark in 1 minute; thanks to personalized homework, the student's error resolution rate has also increased from 50% to 73%.
In this year's middle and high school exams, iFlytek Spark was rated by the outside world as the "large model that is better at solving problems." This time, iFlytek Spark has further upgraded the AI 1-on-1 tutoring function of the iFlytek AI Learning Machine, which can not only provide multimodal heuristic explanations and free personalized answers but also engage in interactive inquiry-based learning and ultra-human-like guided companionship, giving children an additional "AI Tutor."
Data shows that compared to traditional problem-solving video learning, the AI tutoring method has increased children's learning completion rate to 90%, the error resolution rate to 93%, and children are more willing to think actively, with higher learning efficiency and enhanced self-confidence.
The Spark Speech Large Model has released 74 language dialects for "free conversation," solving the problem of speech recognition in strong interference scenarios. Recently, the "Key Technologies and Industrialization of Multilingual Intelligent Speech and their Applications" project, with iFlytek as the first completing unit, won the first prize of the National Science and Technology Progress Award. At the press conference, the award winner once again made a "royal flush," and the Spark Speech Large Model achieved a new breakthrough.
Liu Qingfeng believes that speech will become the main mode of human-computer interaction in the era of interconnected everything, and the most important scenarios for human-computer interaction are long-range, noise, multi-person speech, and multi-language. Therefore, the AIUI (Artificial Intelligence User Interface) in the era of interconnected everything must meet the standards of long-range high-noise, multi-language and multi-dialect, full-duplex, and multimodal. iFlytek also led the formulation of the full-duplex voice interaction ISO/IEC international standard, which was released in May 2023.
Facing the era of interconnected everything, the Spark Speech Large Model released this time supports 37 languages and 37 dialects for "free conversation" without switching. Among them, the recognition effect of the 37 languages leads OpenAI whisper-V3, and the recognition effect of the 37 dialects has been improved by an average of 30%. On the spot, iFlytek demonstrated the voice input effect of the iFlytek input method mixed with dialects and foreign languages, which can greatly improve the input efficiency.iFlytek has also released an integrated hardware and software iFlytek Simultaneous Interpretation system, which can support multi-scenario usage such as conference interpretation, meeting interpretation, exhibition hall interpretation, and tourism interpretation. The seats of the guests attending this conference are also equipped with iFlytek simultaneous interpretation listening devices, which can be worn to listen to multi-language AI simultaneous interpretation in real-time.
In response to the challenge of speech recognition in high interference scenarios, iFlytek has made a breakthrough in transcribing speech in extremely complex scenes with multiple people speaking at the same time. Even in a scenario where three people are speaking simultaneously, it can achieve a speech recognition accuracy rate of 86%. Three researchers from the iFlytek Research Institute tested on-site in a noisy environment, where it was difficult to hear clearly due to the overlapping speech. The multimodal capabilities of iFlytek's Spark not only separated the roles of the overlapping voices of three people but also transcribed in real-time what each person said, causing a sensational effect that led to continuous applause from the audience. In the future, multimodal voice recognition technology will be applied in iFlytek's smart office, smart screens, and other conference office products.
Large models are driving the transformation of human-computer interaction, and all applications in the field of voice are worth being restructured. With the support of large models, the Spark intelligent cockpit has been upgraded, not only with "free interaction" in multiple languages and dialects but also with super-human interaction with multiple emotions and modalities, making the human-vehicle interaction more warm. Currently, iFlytek's voice interaction products rank first in the Chinese market and are widely exported to various parts of the world. The Spark large model has endowed many models of car companies such as FAW, Chery, GAC, JAC, and Great Wall with highly intelligent interactive experiences.
In order to better implement large models, iFlytek has also created a cloud-edge-end integrated and hardware-software integrated solution to empower more industry scenarios such as home appliances, operators, and robots. In response to the needs of embodied intelligence and humanoid robot companies, iFlytek officially released the Robot Super Brain Platform 2.0, the first in the industry to support multimodal interaction. Currently, more than 400 robot companies have adopted the iFlytek Robot Super Brain Platform.
The Spark Enterprise Intelligent Body Platform was officially released, creating a dedicated AI assistant for each position.
Since its release on May 6 last year, the iFlytek Spark large model has become the first choice for leading companies in various fields such as the State Energy Group, China National Petroleum Corporation, China Mobile, China People's Insurance, Pacific Insurance, Bank of Communications, Chery Automobile, FAW Group, Volkswagen Group, JAC Group, Haier Group, and Midea Group.
Spark has achieved application effects in multiple typical scenarios such as code, compliance review, customer service, bidding evaluation, and intelligent interaction. Taking Bank of Communications as an example, the product iFlyCode based on the Spark large model covers more than 6,000 R&D personnel, with a code adoption rate of 38%, significantly improving work efficiency.
How to better solve the last mile problem of enterprise large model application? Liu Qingfeng said that enterprises must first scientifically understand the boundaries of large model capabilities, choose the right solution according to the difficulty of the task, and create a dedicated large model for the enterprise with less computing power and higher efficiency. With the release of Spark V4.0, he believes that the time to create a dedicated assistant for each position with the intelligent body platform has come.
The Spark Enterprise Intelligent Body Platform was officially released on-site. Focusing on the three key capabilities of building an intelligent body, the current enterprise intelligent body platform has covered more than 400 AI atomic capabilities, integrated more than 90 external information sources, and connected more than 100 internal IT systems, which can be quickly built by enterprises in conjunction with business scenarios to create intelligent body applications that can be implemented. The platform also launched 32 enterprise intelligent bodies in production, scientific innovation, office, and management domains for enterprises to use as plug and play.
Based on the enterprise intelligent body platform, iFlytek has created typical application cases such as the Spark Business Opportunity Assistant and the Spark Bidding Evaluation Assistant, setting an example for enterprise applications.In the intelligent agent iFlyCode, it integrates six major scenario agents including a code generation assistant, architecture design assistant, code Q&A assistant, testing assistant, database optimization assistant, and code review assistant, which has increased the adoption rate from 30% to 52%, significantly enhancing the practicality of corporate intelligent agents.
Spark Business Opportunity Assistant can achieve comprehensive awareness of business opportunities, improve the quality and efficiency of customer visits, and intelligently analyze sales management, helping to enhance the effectiveness of frontline sales and business opportunity management. Spark Bidding Assistant, through functions such as pre-bidding sourcing, intelligent bidding evaluation, and bid approval review, achieves a 98% consistency rate between human and machine in intelligent bidding results, and a bid anomaly detection rate of over 80%. This significantly improves the efficiency of corporate bidding while reducing procurement costs.
Spark Developer Ecosystem Accelerates Growth: Over 1 million developers added in 5 months, with a total number of developers exceeding 7 million.
While the iFlytek Spark large model empowers industries, it is also promoting the vigorous development of the developer ecosystem. Since the release of iFlytek Spark V3.5 on January 30th of this year, in just 5 months, the Spark developer ecosystem has accelerated its growth, with the number of developers increasing from 5.98 million to 7.02 million, adding over 1.04 million; the number of overseas developers exceeds 400,000; and the number of large model developers reaches 570,000. More and more developers are joining the Spark ecosystem, unleashing the application value of more demand-driven scenarios.
Liu Qingfeng stated that only a self-controlled and prosperous ecosystem can ensure a bright future for China's general artificial intelligence. It is necessary to understand the comprehensive gap between China and the United States in large models scientifically and rationally, and also to have confidence in catching up quickly.