A clear signal: this year's Nobel Prizes in Physics, Chemistry, and Economics have all been awarded to scientists in the field.

Undoubtedly, technology, as an undeniable force, has risen and exerted its influence on the global scientific research stage. Liu Qingfeng, Chairman of iFLYTEK, stated, "It is bringing about a fundamental transformation in the paradigm of scientific research, becoming a new engine for technological development." Recently, iFLYTEK officially launched iFLYTEK Spark, which, based on real data and independent testing, has surpassed in code and mathematical capabilities. Meanwhile, the domestic ultra-large-scale intelligent computing platform "Flying Star II" was officially launched.

According to the report "China's Large Model Platform Market Share: The Year of Large Models—Initial Layout," the market size of China's large model platforms and related applications reached 13 billion yuan in 2023. Domestic large models have entered a period of rapid development. By the end of October 2023, a total of 238 large models have been launched in China.

In contrast to the booming development of technology, there has been a significant surge in commercial implementation. According to incomplete statistics, by the third quarter alone, the number of large model-related projects that have been publicly tendered and can be statistically tracked in the market has reached at least 100, far exceeding the total number of projects in the first half of the year. Among these, industries such as education, operators, government affairs, and energy rank prominently.

The government work report directly mentions the need to deepen research and application of big data, artificial intelligence, and to carry out the "AI Plus" initiative. With the introduction of the concept of "new quality productivity," as a crucial force in promoting high-quality development of the digital economy and constructing Digital China, central and state-owned enterprises have always been at the forefront of digital transformation, continuously deepening their layout of large models.

What is more noteworthy is that the adoption and application of new technologies by central and state-owned enterprises often drive industry trends and standard formation, contributing to the promotion of application innovation in vertical fields. In the third quarter, the project procurement parties were not only led by central and state-owned enterprises, but some local state-owned enterprises also joined in.

** 双向奔赴**

In the future, artificial intelligence will exist in all areas of society like water and electricity do today. As the vanguard of the national "Artificial Intelligence Plus" initiative, central and state-owned enterprises have an especially urgent need to achieve digital transformation. This urgency is both based on considerations for the long-term sustainable development of the enterprises and driven by the mission to deepen the implementation of national strategies.

As Sundar Pichai of Google once said, "Artificial intelligence is one of the most profound research directions in which humanity is engaged, even more profound than fire and electricity." With the support of large models, artificial intelligence is quietly revolutionizing the way energy is utilized, enhancing the efficiency of fire and electricity in specific production scenarios. The energy industry has seen a flurry of bidding completions, with many projects securing contracts worth over ten million yuan.

Among them, the recently released "Haineng" artificial intelligence model, jointly built by CNOOC and iFLYTEK, is a representative of the "integration of general and specialized knowledge." The "Haineng" AI model focuses on addressing pain points and difficulties in offshore oil and gas exploration and development, as well as daily office tasks. It simultaneously exerts efforts in computing power, data, large models, and scenario applications. With the support of the iFLYTEK Spark large model and a series of "assistant" applications, it serves as both an "industry expert" and a "dedicated assistant." Similarly, the Taishan large model jointly released by China Green Development and iFLYTEK is the first industry large model in the green industry. The Kunlun large model, jointly created by PetroChina, China Mobile, Huawei, and iFLYTEK, is the first large model in the energy industry to pass the record-filing process in China's energy and chemical industry. The iFLYTEK Spark large model provides core technical support for the development of "employee assistants" and "industry experts."

From the bidding projects in the third quarter of the year, it can be seen that enterprises across various industries are exploring the implementation of more segmented application scenarios.

Different large model vendors and enterprises explore various core scenarios, or conduct forward-looking technical research for certain scenarios, but each has its own focus. For example, Alibaba's strength lies in the cloud services provided by Alibaba Cloud, while iFlytek's strength is in the /end.

After years of deep cultivation in the field of artificial intelligence, iFLYTEK has accumulated strong technical expertise and application experience in multiple sectors including education, healthcare, industry, finance, and telecommunications. An industry insider analyzed, "Even without the technology of large models, iFLYTEK can still empower various industries with other artificial intelligence technologies."

With the acceleration of commercialization and implementation by leading enterprises, the benchmark radiation effect of iFLYTEK's Spark has become prominent, revealing broader growth potential from the leading effect of central and state-owned enterprises. This year, iFLYTEK successfully won 10 projects, with disclosed winning amounts totaling 1.1 billion yuan. Through data sorting from platforms such as the Chinese Government Procurement Network and the China Tendering and Bidding Public Service Platform, iFLYTEK's winning amount in a single month of October exceeded 1 billion yuan, ranking first in the industry. According to the third-quarter report, the company's sales receipts in the first three quarters reached 10 billion yuan, a year-on-year increase of 30%.

Data shows that as of the month, iFLYTEK has co-built multiple industry-wide models with leading enterprises, covering over applications scenarios.

Complete solutions from computing power to operations

iFLYTEK can be considered one of the longest-standing companies in China dedicated to artificial intelligence.

The latest release of the iFlytek Spark Large Model, evaluated by authoritative units such as the Allen Institute for Artificial Intelligence and other domestic and international institutions, has surpassed the performance in several mainstream test sets covering different task types such as understanding and reasoning, comprehensive exams, mathematics and science, and coding. Among these, certain tests have exceeded -, and the training efficiency has been relatively improved by %.

And based on practical mathematical tasks, the test set construction - Xunfei Spark. In tasks such as calculation, finance, finance, and measurement, it surpasses - level across multiple dimensions, and has completed algorithm verification such as ultra-long thought chains, tree search, and self-reflection evaluation.

Of course, each major model company has its own strengths and weaknesses. The real-world demands from various industries are complex and diverse, requiring large model companies to have high adaptability. The iFLYTEK Spark Model has also implemented various "cloud, edge, and end" integrated full-scenario layout strategies for these vertical fields. For example, China Iron and Steel Research Institute, in collaboration with the iFLYTEK team, has built a Steel Research Intelligent Code Development Platform based on new large model technologies. The platform includes an intelligent coding assistant and supporting code quality and efficiency management tools. It is reported that, according to current metrics, the adoption rate of generated code by developers has reached %; since its deployment, the proportion of generated code in business code has reached %. Through the empowerment of the intelligent coding assistant, the R&D efficiency of employee algorithms and intelligent applications has increased by %.

Among numerous enterprise partners, why can iFLYTEK establish a large model that a company can truly use? When deploying large models in enterprises, the first step is to build computational power. iFLYTEK has partnered with Huawei to construct China's first domestically produced computational power cluster with ten thousand cards. We are familiar with various computational powers and can build the best computational power platforms. With computational power, the next step is to organize data. Can we organize data more efficiently and with higher quality, and can we then train models with this data? Can we establish models of various sizes to meet enterprise applications and implement them in tangible scenarios that can be seen and touched? At the same time, can enterprise large models ensure political and application security, and ultimately produce practical application results that can be proven through statistical data in actual operations? Here, I want to say, why are we the top bidder in terms of the number of bids, and why is our bidding ratio increasing? Because many enterprises can only reach the third step—training models, and the subsequent steps are far behind us. Even if they can achieve it, their capabilities in organizing data and training models are also far behind ours. Liu Qingfeng introduced that on the basis of continuously improving the foundational capabilities of general-purpose large models, iFLYTEK has constructed a complete set of solutions from "building computational power, organizing data, training models, implementing in scenarios, ensuring security, to precise operations."

To offer the world "a second choice"

The criteria for central SOEs and industry-leading companies in selecting large models go beyond technical capabilities; considerations of security and self-control come first. "Made in China" is the fundamental prerequisite for any technological innovation and self-control, greatly ensuring the security and independence of large model technology. As China's first large model trained on domestically produced computing power and open to the public, iFlytek's Spark undoubtedly gains a head start.

On a certain date, the domestic super-large-scale intelligent computing platform "Feixing No. 2" jointly created by iFLYTEK, Huawei, and Hefei Big Data Asset Operation Co., Ltd. was officially launched. On a specific date last year, the first domestic 10,000-card computing cluster platform "Feixing No. 1" went online, with the Spark Large Model continuously leading the development of domestic computing platforms. Liu Qingfeng introduced that over the past year, many 'difficult problems' have been overcome, and more than 100 basic software and hardware issues and model adaptation problems have been resolved.

"Flying Star II" will bring continuous adaptation of new models and algorithms, as well as another leap in the scale of intelligent computing clusters, continuously exploring uncharted territories and leading the development of domestic large-scale model foundations.

Liu Qingfeng has always said, "You can't build a building in someone else's yard," which is the fundamental reason why iFlytek has heavily invested in computing power. There is another "must-do" task: iFlytek has released the Spark multilingual large model. He said that the more a large model is used, the smarter it becomes, as it self-iterates, just like the birth of the internet, and will become the most basic underlying ability and tool for society. If all large model manufacturers in China only focus on Chinese, and there is no second choice globally, they will inevitably be confined within the Chinese market. The multilingual large model has already been applied in automobiles, helping Chinese cars go global.

Liu Qingfeng further pointed out that the key to the development of the artificial intelligence industry lies in five keywords: top-tier innovation, self-reliance, integration of general and specialized knowledge, cloud-edge collaboration, and software-hardware integration. He frankly stated, "Adhering to these five key points, our artificial intelligence will have a great future, truly liberating productivity, unleashing imagination, transforming industrial and competitive landscapes, bringing about changes in research paradigms, affecting various industries, and most importantly, fulfilling the people's aspirations for a better life! This is the direction of our future development."