"Only on the basis of a truly independent and controllable platform and a thriving ecosystem can there be a truly great future for AI in China." Today, as IFLYTEK advances on all fronts, unleashes productivity, and unleashes imagination, China's AI is further embracing the vast expanse of the universe.

Author/Tomato Sauce

Produced by/Xinzhai Business Review

As AI emerged as the biggest winner in the 2024 Nobel Prizes, it marked the first year of large model implementation, with a complete outbreak of AI implementation. Large models have facilitated a leap in AI performance, accelerating the industry's embrace of the vast universe of AGI. IDC predicts that global enterprises' investment in generative AI solutions will reach $143 billion by 2027.

Amidst the diverse implementations of AI, how can we break the "impossible triangle" of professionalism, generalization, and economics, and achieve breakthroughs in both technology and implementation? As the "first year of large model implementation" draws to a close, AI vendors are about to face their "settlement interface" and submit their answers.

Among them, IFLYTEK has showcased two sets of results in the large model arena and application field, achieving "nine firsts in international mainstream test sets" and "seven firsts in large model applications," becoming the industry's "heptagonal warrior."

On October 24, at the opening ceremony of the Seventh World Voice Expo and the 2024 IFLYTEK Global 1024 Developer Festival, IFLYTEK Chairman Liu Qingfeng unveiled IFLYTEK Spark 4.0 Turbo and showcased IFLYTEK Spark's "seven firsts" in the application field: first in winning bids from central and state-owned enterprises, first in the education and healthcare market, first in the smart vehicle market, first in large model developer ecosystems, first in the smart hardware market, first in empowering scientific research applications, and first in empowering industrial applications.

How did IFLYTEK achieve its seven firsts? It is thanks to IFLYTEK's keen strategic vision and clear implementation strategy since its launch last year. Long before the "war of the gods" in implementation had erupted, Liu Qingfeng stated that the key to competition in large models lies not in the timing of their release but in their early implementation into products, addressing users' essential needs, and achieving self-sustainability.

Success takes time, not haste. Looking back, we can see IFLYTEK's long-term strategic layout, with a strategic blueprint emerging for advancing into the era of AGI.

I. Foreseeing the future and building a solid foundation: Preparing for the dawn of AGI with a lengthy groundwork

IFLYTEK's "early start" stems from its clear understanding of the value and revolutionary interaction methods of AI applications. As early as 2019, Liu Qingfeng published a letter to all employees titled "Because We Foresee, We Are Determined," stating that over the next decade, 5G-driven Internet of Things will officially become the sixth wave in the IT industry, with voice becoming the most important human-machine interaction method. AI will profoundly change the world's production and lifestyle, empowering various industries and making human-machine coupling ubiquitous.

Based on this, IFLYTEK's implementation path focuses on both technical capabilities and implementation capabilities, achieving a "big platform plus heroism" approach. First, make the large model "stand tall and firm," i.e., achieve leadership in hardware-software integration from the base to the cloud, edge, and end devices, determining how far the large model can go in the era of AGI. Simultaneously, the large model will transform information acquisition, content production models, industrial competition landscapes, and scientific research paradigms, meaning that IFLYTEK must scientifically and reasonably implement the large model in essential scenarios.

This "dual-wheel drive" strategy enables IFLYTEK to continuously optimize its technology, enhance model performance, and ensure that while leading in technology, it can also meet the diverse needs of the market, feeding back into the base and driving positive momentum.

For example, IFLYTEK has further evolved its large model capabilities and refined its engineering efforts. In addition to the nine firsts in international mainstream test sets mentioned above for 4.0 Turbo, at the press conference, IFLYTEK upgraded its multi-modal capabilities, adding hyper-realistic and personalized abilities to its existing capabilities for far-field high-noise, full-duplex, and multilingual/dialect support.

This transforms multi-modal interaction from hyper-realistic voice to hyper-realistic digital humans. Users can make video calls to hyper-realistic digital humans and create digital avatars to converse with themselves, redefining the standards for multi-modal AIUI interaction in the era of intelligent connectivity.

Simultaneously, the domestic ultra-large-scale intelligent computing platform "Feixing 2" jointly developed by IFLYTEK, Huawei, and Hefei Big Data Asset Operation Co., Ltd., was officially launched. Additionally, IFLYTEK premiered ten hardcore products and innovative applications based on IFLYTEK Spark's capabilities, including the Spark Intelligent Office All-in-One and the VIAS robot for evaluating human-machine interaction effects in smart cockpits, preemptively positioning itself for the next wave of applications.

It is not difficult to see that compared to other vendors' "one-sided" implementation strategies, IFLYTEK's "all-out attack" is more comprehensive, unleashing its full force on a solid foundation, leveraging "a combination of general and specialized capabilities, end-cloud collaboration, and hardware-software integration" to drive growth.

Previously, during the wave of implementation, many vendors adopted a "team-based strategy" – for example, the "industry faction" focused on sectors such as education, finance, automobiles, and industry, diving headfirst into specific areas while abandoning base models or reducing the size of pre-training algorithm teams; the "software faction" positioned itself at the forefront of AI agents, concentrating efforts on developing platforms; the "technical faction" focused solely on base models, getting lost in a "technological flow" and neglecting emerging applications...

This led them to varying degrees of "localized fanaticism," resulting in a need to strengthen the implementation of specialized bases and address insufficient customization for diverse scenarios. IFLYTEK's answer is to embrace the market comprehensively, with a "master of all trades" attitude, deeply rooted and far-reaching, maximizing the potential of the AGI era.

II. A Combination of General and Specialized Capabilities, End-Cloud Collaboration, and Hardware-Software Integration: How to Unleash Full Force on a Solid Foundation

In today's context, the importance of a "combination of general and specialized capabilities" is self-evident. As Cao Feng, Director of the Artificial Intelligence Institute at the China Academy of Information and Communications Technology, stated, "Large models are at a critical juncture of expanding from general to specialized scenarios, and there are numerous high-value scenarios across all business scenarios of 'R&D, production, supply, sales, and service' that await deep integration with AI."

Based on years of deep cultivation across various industries, IFLYTEK, guided by "solving essential needs of the people," has focused on sectors such as education, healthcare, and justice, creating numerous industry-specific models and continuously refining them through problem-driven approaches. For instance, IFLYTEK introduced its latest product applications.

In the education sector, it unveiled a high school math intelligent tutoring system based on "problem chains," which can intelligently generate problem chains to assist teachers in inspiring students' thinking and gradually solving problems in a step-by-step manner. In healthcare, IFLYTEK released IFLYTEK Spark Medical Large Model 2.0, with significant upgrades and continued leadership in six core medical scenarios. In the justice sector, the Spark Legal Large Model empowers judicial scenarios such as court trial transcript preparation, with efficiency improvements ranging from 61.7% to 87.9% compared to the Spark General Large Model.

Beyond implementation, how to lighten AI, enable "end-cloud collaboration," and flexibly bridge the last mile of large model implementation is also an industry imperative. As CITIC Securities' research report notes, end-side AI represents the next stage of AI development, with the potential to unleash a wave of AI applications by empowering terminal hardware with large models.

Therefore, IFLYTEK empowers diverse scenarios, enabling large models to "exert force in the cloud and cultivate at the terminal," accelerating the delivery of AI benefits to users. In June this year, the IFLYTEK Spark app reached 140 million downloads on Android, planting the seeds for "killer apps."

In areas such as in-vehicle AI systems and smart homes, the introduction of end-side large models has significantly enhanced user experience. Previously, the Spark Large Model empowered intelligent interaction experiences for numerous vehicle models from automakers such as FAW, Chery, and GAC, receiving positive feedback from users. This time, IFLYTEK further released the end-side Spark Large Model for automobiles. Starting in the fourth quarter of this year, multiple vehicle models equipped with the end-side large model from Chery, GAC, Great Wall, and others will be launched for sale, further empowering users' quality of life.

On this basis, "hardware-software integration" forms a closed loop in IFLYTEK's implementation strategy, fully tapping into its potential. As cloud, network, edge, and end devices continue to converge, scenarios involving complex systems and computations are increasingly common, making hardware-software integration an inevitable trend.

On the consumer side, IFLYTEK seamlessly integrates its hardware ecosystem with industry capabilities. For example, at this launch, IFLYTEK will soon release its AI learning machine reading companion, embedded with Spark multi-modal AIUI capabilities, enabling children to "point and read" anywhere in a book, instantly transforming the text into a personalized digital human that jumps off the page, inspiring children's thoughts and questions, delivering cutting-edge technology to the front lines. Meanwhile, in 2022, IFLYTEK launched the Brain 2030 Plan, intensifying its efforts in the robotics sector...

By leveraging multidimensional strengths and launching comprehensive attacks, IFLYTEK is meeting the needs of the times and excelling in both market and scientific research, igniting a prairie fire.

III. The Needs of the Times and the Capabilities of AI: IFLYTEK's "Prairie Fire" in Strong Coupling with Markets and Scientific Research

Currently, on the "proving ground" of applications, IFLYTEK Spark is becoming the preferred choice of central and state-owned enterprises and receiving widespread market recognition. Public information shows that in the first three quarters of 2024, IFLYTEK successfully won bids for 38 projects, with a disclosed bid amount of RMB 216 million, maintaining a leading position in both the number of winning bids and bid amounts within the industry. As of October 2024, IFLYTEK has jointly established over 20 industry-specific large models with leading enterprises, covering more than 300 application scenarios.

Behind its emergence from fierce competition lies the philosophy of "meeting market needs with IFLYTEK capabilities." On the foundation level, IFLYTEK has constructed a comprehensive solution ranging from "computing power construction, data processing, model training" to "scenario implementation, security assurance, and refined operations." In terms of efficiency, IFLYTEK boasts a leading toolchain that significantly enhances the efficiency of "data processing and model training." In terms of practical capabilities, IFLYTEK has actual application cases covering over 300 industry scenarios, fostering a scalable effect of mutual reference and reuse. Additionally, it possesses an entirely domestically produced computing power platform.

Beyond technical prowess, IFLYTEK further excels in providing customized solutions that address the intricacies of implementation. For example, IFLYTEK's intelligent agent development platform for enterprises significantly lowers the threshold for businesses to interact and align with large model capabilities, enabling them to quickly build implementable intelligent agents through the provision of core capabilities.

Today, an increasing number of enterprises rely on IFLYTEK's intelligent agent platform to incubate their own AI assistants. For instance, the "National Energy Cup" competition hosted by the State Power Investment Corporation attracted 126 teams, ultimately incubating 54 scenario-specific intelligent agents.

In empowering scientific research, adhering to the mission of "AI for Science," IFLYTEK has also achieved a series of practical results. Liu Qingfeng explained that AI empowers scientific research through three tiers: enhancing basic work efficiency with research literature and code assistants, precisely modeling scientific tasks based on deep neural networks, and leveraging cognitive large models to learn domain knowledge and assist in designing scientific research experiment plans.

Based on this, IFLYTEK has flourished in multiple areas of large models and scientific research, such as collaborating with Professor Liu Haiyan's team at the University of Science and Technology of China to successfully design 48 novel proteins that do not exist in nature, and working with Dr. Li Xin's team at the Institute of Zoology, Chinese Academy of Sciences, to study single-cell gene expression...

IFLYTEK's AGI aspirations extend beyond these achievements. It aims to represent China on the global stage, building independently controllable heavyweights domestically and offering a second option to the world. To this end, IFLYTEK remains true to its original aspirations. This time, the first domestically produced 10,000-card computing power cluster platform, "Feixing 2," was launched, building upon last year's launch of "Feixing 1" on October 24.

According to Liu Qingfeng, over the past year, numerous "difficult and complicated issues" have been overcome, resolving more than 500 fundamental hardware and software issues and model adaptation problems. Based on this, "Feixing 2" will continuously adapt new models and algorithms and scale up intelligent computing clusters, continuing to lead the development of domestically produced computing power platforms and promoting ecosystem prosperity.

Faster alone, further together: Under the vision of ecosystem prosperity, IFLYTEK will also open up resources across scenarios, from technical capabilities to application implementation, achieving product success through the shortest path, sharing online and offline channels and resources, and accelerating developers' path to market success. Currently, according to IDC research reports and public market data, IFLYTEK leads in voice and semantic market share and has the largest developer base for large models, with 781,000 developers.

Looking ahead, IFLYTEK will take the lead in establishing an AI fund, using RMB 500 million in venture capital to drive developer entrepreneurship, accelerate the industrialization of cutting-edge technologies, and collaborate with local governments to provide industrial support for AI startups.

"Only on the basis of a truly independent and controllable platform and a thriving ecosystem can there be a truly great future for AI in China." Today, as IFLYTEK advances on all fronts, unleashes productivity, and unleashes imagination, China's AI is further embracing the vast expanse of the universe.