From Tech to Application: iFLYTEK Ignites Finance

On August 30th, iFLYTEK's Xinghuo ultra-fast hyper-human interaction technology officially went live on the Xinghuo App, opening up the experience to the public for the first time.

This is the first major model product in China to benchmark against GPT-4o voice, offering a smooth and natural interaction experience that supports interruption and immediate response after interjections, as well as the ability to perceive human emotions.

Its performance is comparable to the GPT-4o voice dialogue function showcased by OpenAI earlier this year.

Looking back at May of this year, OpenAI's end-to-end real-time audio and video dialogue function, which was unveiled at a press conference, left a deep impression with its profound understanding of human emotions.

Advertisement

However, to this day, GPT-4o has not been officially launched to the market, while iFLYTEK has taken the lead in achieving product commercialization with its "Chinese version of GPT-4o," demonstrating the strong capabilities of "Xinghuo."

From continuous breakthroughs in general-purpose large models to in-depth development of voice large models, and then to precise empowerment of industry-specific large models, iFLYTEK has always been at the forefront of technological innovation, empowering various industries.

"In the era of general artificial intelligence, iFLYTEK aims to achieve a higher level of self-improvement and self-reliance in the global technology competition," said Shen Haibo, Vice President of iFLYTEK and General Manager of the Financial Technology Business Division.

Currently, iFLYTEK's core technology research and product development platform has gradually achieved full autonomy and controllability.

Computing power is the most important infrastructure for large models and the key to commercialization.

On October 24, 2023, iFLYTEK joined hands with Huawei to announce the official launch of the "FeiXing No.1," the first computing platform supporting the training of trillion-parameter large models.

On January 30, 2024, the hundred-billion parameter model Xinghuo V3.5, trained based on the "FeiXing No.1" computing platform, was officially released, with performance indicators at a leading level in China, verifying the reliability of "FeiXing No.1."

On June 27, the Xinghuo V4.0, based on "FeiXing No.1," was released, with comprehensive upgrades to its seven core capabilities, fully benchmarking against GPT-4Turbo, and once again achieving verification.

"The first requirement for financial institutions in applying technology is security," Shen Haibo said, citing the application of large models in the financial industry as an example.

He pointed out that in the face of the application goal of security and controllability, it is necessary to ensure both technical and service sides.

On one hand, relying on its profound accumulation in the field of artificial intelligence, iFLYTEK has constructed a complete technical system, allowing the industry's continuous growth to be rooted in the foundation of independent innovation and security, ensuring that Chinese enterprises maintain a technological edge in global competition.

On the other hand, in the process of serving financial institutions, iFLYTEK places great emphasis on and strictly controls five key aspects: data security, privacy protection, system security, risk control, and compliance review, building a comprehensive and multi-level security system to ensure that financial institutions can enjoy a high level of security and controllability when applying advanced technology, laying a solid foundation for the stable development of financial technology.

"Platform + Track" is iFLYTEK's market strategy for developing artificial intelligence business: relying on the first batch of "National New Generation Artificial Intelligence Open Innovation Platforms" - Xinghuo Open Platform, iFLYTEK continues to build an artificial intelligence industry ecosystem; at the same time, it deeply cultivates education, healthcare, smart cities, operators, automotive, finance, and other key tracks, building a sustainable "strategic base."

Currently, the Xinghuo large model created by iFLYTEK has become the foundation platform for its artificial intelligence industrialization, playing an increasingly important supporting and leading role in the company's various business tracks.

"Financial technology business mainly provides financial industry customers such as banks, insurance companies, and securities with a complete solution for digital management and operation," Shen Haibo introduced.

For a long time, iFLYTEK has been closely cooperating with financial institutions, accurately grasping their needs, and deepening the application of intelligent technology in the financial field, building an intelligent base and scenario application system for financial industry customers, providing financial services covering the entire process of customer service, marketing acquisition, risk control, and operation management, and helping financial institutions to achieve digital transformation and upgrading.

Since the release of the Xinghuo large model, iFLYTEK has actively carried out joint innovation with industry partners, accelerating the construction and deployment of large models in the financial industry, and is committed to exporting experience in the application and landing of large models, participating in the development of the China Academy of Information and Communications Technology's large model series standards, taking the lead in writing and releasing the first domestic banking industry large model standard, and promoting the healthy development of large models in the financial industry.

"In the context of digital finance, we have been thinking about what kind of 'large model engine' financial institutions need," Shen Haibo said frankly, from core technology breakthroughs to large-scale industrial applications is a high threshold process, the team must not only be able to break through from single points and further expand to the entire stack of technology on the business chain, but also be able to find a large enough scene, with an appropriate business model to form a mutual reinforcement between business and technology, otherwise, it can only become a single-point technology provider.

Visible and tangible application cases, scalable standard products, and application effects that can be proven by statistical data are the standards that iFLYTEK uses to measure the landing of large models.

Facing the financial industry, iFLYTEK has trained a large model exclusive to financial institutions, developed intelligent applications for financial scenarios, and established a product shelf system composed of "1+3+6", that is, with the Xinghuo large model as the core, based on the Xinghuo Zhisuan platform, Xinghuo Huoshi training platform, and Xinghuo intelligent body platform, to build product applications serving six major scenarios such as intelligent marketing, intelligent risk control, intelligent customer service, intelligent office, software development, and intelligent operation.

Shen Haibo introduced that in June this year, based on the Xinghuo V4.0 base, iFLYTEK released four intelligent products: Xinghuo Business Assistant, Xinghuo Due Diligence Assistant, Xinghuo Intelligent Customer Service 2.0, and Xinghuo Code Assistant.

Xinghuo Business Assistant can perform multi-dimensional customer group data analysis, generate customer group marketing suggestions, personalized business strategies, and provide real-time knowledge Q&A support for account managers; Xinghuo Due Diligence Assistant can automatically connect to multiple information sources and accurately extract key information, generate report drafts with one click through multi-dimensional analysis of credit risk, significantly improving the efficiency and quality of pre-loan due diligence document writing; Xinghuo Intelligent Customer Service 2.0 has upgraded technologies such as ultra-long text processing, knowledge tracing, and hyper-human synthesis, enhanced functions such as knowledge intelligence editing, intelligent body planning, and process arrangement, making user interaction simpler and business operations more efficient; Xinghuo Code Assistant has greatly improved the adoption rate of code recommendations through pre-training and SFT training, and integrates functions such as code review, automated testing, and programming suggestions, capable of automatically generating high-quality unit test code, effectively reducing developers' coding costs and thresholds.

"There are many companies in China that do large models.

Since 2023, we have been firm in one idea, to recognize our own advantages, based on years of industry cultivation, supported by the technology of the Xinghuo base, to create a benchmark application of large models in the financial industry, from 1 to N, and to pave a Xinghuo Avenue."

Not only in product development, Shen Haibo revealed that iFLYTEK provides a variety of service models for financial institutions: first, providing standardized product services; second, private deployment services based on business scenarios; third, enterprise-level large model joint construction services.

For financial institutions with specific needs and business processes, iFLYTEK can provide customized solutions, and these solutions are deployed and maintained on the financial institution's own infrastructure.

This means that iFLYTEK is not only selling products but also providing personalized services.

Private deployment ensures the security of data and the personalization of business, meeting the financial institutions' needs for data control and business specificity.

For example, iFLYTEK and Bank of Communications have jointly built a hundred-billion-level financial large model, creating end-to-end AI financial application scenarios in key financial business areas such as retail, inclusive finance, operations, and risk.

In addition, iFLYTEK has also jointly developed large models suitable for enterprise-level use with industry institutions.

Joint construction services mean that financial institutions are not only recipients of services but also participants in the creation process.

This model helps financial institutions obtain more precise and customized services, while also being able to use iFLYTEK's technological advantages to promote internal innovation and digital transformation.

For example, iFLYTEK has joined hands with PICC, based on the Xinghuo financial large model base, after private training and optimization, to create a unique enterprise large model for PICC, which has been applied and landed in hundreds of scenarios in various companies of the PICC Group.

"iFLYTEK not only wants to be a leader in technology but also a builder of the ecosystem, working with various forces to build an open, cooperative, and win-win financial AI ecosystem."

According to public information in the industry, iFLYTEK has signed strategic cooperation agreements with many banks and insurance institutions, including Agricultural Bank, Construction Bank, Bank of Communications, PICC Group, Shanghai Pudong Development Bank, China Minsheng Bank, China Everbright Bank, Shanghai Bank, and Changsha Bank.

"With the extensive application and continuous optimization of the Xinghuo large model in the financial industry, iFLYTEK will continue to expand the boundaries of technology and promote the intelligence and personalization of financial services to new heights."

As Shen Haibo said, from technological innovation to application landing, Xinghuo will surely set the prairie on fire.