"Is the big model hype or a real revolution?"
This sharp question raised by investment bank Goldman Sachs in a research report has unveiled the collective anxiety of the industry. Subversion, change, remodeling, the prefix to describe the transformation of the large model is so familiar. I still remember that in the first few rounds of technology boom, every once in a while, a concept would pop up to "slip" around, triggering a burst of hype and carnival, and then quietly extinguished. Will the large model become a chapter in this group of plays? This AIGC's "question of the times" seems real and urgent.
As a result, half a year after the wind of the general large model soared, the fate gears of the large model began to turn, and the clock pointed to the time zone where the vertical large model was applied. Whether it can be implemented and whether it has the ability to solve business needs has become the gold standard for detecting the quality of large models in addition to parameter competition.
From imagination to productivity
For a long time, the concept of AI has always been suspended in mid-air, often depicting an exciting rendering, but the specific how to "construct" and "build" have not been clearly answered.
After the emergence of large models, AI has found a new focus, and technology seems to be standing at the door of a new era again. The general large model represented by ChatGPT, Llama 2, and Wenxin Yiyan is easy to write poems and songs, chat and paint, showing a very high "IQ" and stunning the world.
After half a year of soaring, the wind direction changed.
After ChatGPT became the fastest product in the history of the Internet to exceed 100 million users, users began to show negative growth. This makes the "false fire theory" once again shroud the industry. Some people are worried that AI will once again be "cold", and it will only bloom like fireworks for a moment.
In fact, instead of getting colder, large models have become more pragmatic, and the imagination of technology is rapidly turning into productivity, and practitioners have reached a consensus: the architecture layer can only accommodate a limited number of players, and more participants must focus on the model layer and the application layer, turning good-looking and fun capabilities into top-notch and easy-to-use tools.
Therefore, in addition to the grand narrative of the large model changing the world and changing the industry, more people focus on the subtleties of the business in front of them, shift their attention from the general model to the path of vertical model landing in the industry, and are committed to using technological innovation to solve problems and improve industrial efficiency.
For example, at the press conference of Huawei's Pangu model, it clearly put forward the slogan of "don't write poetry, only do things", using the model to empower government affairs, mining, meteorology and other industries, and also announced that the Pangu model has been commercialized for the first time in the mining field. Jingdong Lingxi has also penetrated its tentacles into retail, logistics, health and other scenarios to solve real industrial problems; The office software WPS integrates large model capabilities in the document software to make creation easier; Recently, the leading enterprises in the consumer finance industry also said that they will release a large financial model based on their solid technology accumulation and financial industry data to help improve the quality and efficiency of their business.
The unanimous choice of this group of enterprises shows that the big model is not only a platform-level industrial revolution, but also a small opportunity for various segments. Application leaders are rolling up their sleeves and becoming doers, competing to enter the depths of the huge system of large models, and pulling the gears that can bring benefits to business scenarios.
Three horizontal and three vertical, rooted in the industrial scene
According to incomplete statistics, more than 106 large models have appeared in the country. They collectively "surfaced", giving the outside world the illusion that large models can be built quickly, but they are just changing the concept of existing technology to pack "old wine".
In fact, the big model didn't come out of nowhere. Before being detonated by ChatGPT, BAT was developing their own large models. The training notice of the "Institute of Equine University of Malaysia", an employee growth platform for immediate consumption, also shows that during the period of 2021-2022, the company's technical team has carried out internal technical exchanges on topics such as "language model application" and "machine learning model visualization" several times.
This shows that long before the advent of ChatGPT, these companies have been engaged in large-scale model research and development and application exploration.
What value can AI bring to the industry? How can large model technology go to the market? In addition to achieving technological leadership, we should pay more attention to two core issues: what efficiency can be improved on the supply side to form a broad market; How can the demand side improve quality and efficiency and reshape the experience?
For example, customer service, traditional manual customer service occupies huge human resources, and through the advantages of large models in natural language processing, service costs can be greatly reduced and efficiency can be improved; For example, risk management is the core of finance, and large models can predict and assess risks more accurately through data mining and machine learning, improving the accuracy of risk control technology.
It is understood that the artificial intelligence technology currently consumed immediately is mainly used in three major scenarios: first, financial intelligent dialogue, which realizes real-time human-machine collaboration, continuous learning, and credible security compliance; The second is the financial digital human, which realizes the digital human with temperature through the multimodal capability of large model + combined AI; The third is the AI core engine of financial services, which realizes an emotional human-computer interaction experience through the organic combination of the brain of the large model and psychology.
Jiang Ning, CTO of Immediate Consumption, said: "Large models have a wide range of application prospects in the financial field and can build a personalized service experience for users. In the context of the construction of digital China, the large model will effectively improve the efficiency of the value chain such as marketing and operation in the financial field, further expand the innovative application effect of data decision-making in the field of risk control, and help the digital transformation of the financial industry to make a substantial leap. ”
Consume CTO Jiang Ning now
From the perspective of immediate consumption, the large model is an "application question", and the answer should be based on use, penetrating into the scene, and forging the ability of self-hematopoiesis and commercialization.
Generative AI, as the name suggests, outputs the result of a large model "emerging", and there is a possibility of "nonsense". This uncertainty can be fatal when it comes to industries that require a high degree of precision. This is especially true for financial institutions with strict compliance requirements.
Moreover, the financial industry has strict requirements in terms of data security, privacy protection, consumer rights protection, etc., and the financial model must be both smart and reliable to ensure the accuracy of the facts and the controllability of the logic.
Based on these characteristics, Jiang Ning said that in order to achieve the lead in the implementation of the large model, the "three horizontal and three vertical" strategy has been formulated for immediate consumption, forming a unique methodology.
In the "three horizontal" capability dimensions, including continuous learning, model control, and combined AI, it is necessary to achieve industry-leading leadership; In the "three vertical" scenarios, including real-time human-machine decision-making, multi-modal large models, and data intelligence, it is necessary to do in-depth and thorough work to achieve smarter, more stable, safer and controllable.
In the AI coordinate system, the general large model is like a horizontal, with more than breadth and insufficient deep cultivation. Combine the powerful transfer learning and generalization capabilities of generative models with the usability and professionalism of vertical discriminative models, embed and optimize the processes of the financial industry, find accurate value points that reshape user experience, and accumulate potential energy.
From data to knowledge, from single point to matrix
At present, scientific research institutions around the world have still not been able to understand the internal operation mechanism of large models, but a consensus has been formed on how to do a good job in large models: participants must have a deep accumulation of technology pools and high-quality industry data, otherwise, no amount of concepts and outlets will be of any avail.
The financial industry, especially consumer finance, has always been extremely sensitive to technology and oriented to industrial applications. Because in massive business processing scenarios, a small step of technological improvement may bring a huge improvement to the operational efficiency of the industry and reshape the customer experience.
In the past 8 years of digitalization, Instant Consumption has been committed to mining the value of data, realizing data assetization, and accumulating high-quality native data, including 3.2 billion real user dialogues and trillion-level tokens in the financial field, forming a multi-modal data asset.
And data doesn't equal capability. Lu Quan, president of the Institute of Consumer Artificial Intelligence, believes that "the data that can be shared in the financial industry is very limited, and the large model should be able to precipitate the data into knowledge and reuse it in other users in the same industry to achieve a win-win situation." The key to this precipitation and transformation is the technological "insight" rooted in the industry.
As a leading technology-driven consumer financial institution in China, in the long-term climbing movement, the company has a systematic understanding of the financial industry, has a lower level of thinking about technology, and has established a moat on the three pillars of computing power, algorithm and data, which can make data the fuel to drive the large model to do deep and thorough.
In response to the national strategy of expanding domestic demand, consumer finance with small, decentralized and inclusive consumption as the core must use technology to promote the reduction of operating costs and open up the road from "universal" to "beneficial". Up to now, the company has independently developed more than 1,000 systems, set up a R&D team of more than 2,000 people, and submitted more than 1,300 patent applications. These technologies have provided automated marketing, risk control and other services for more than 100 million users.
Nowadays, they also support the growth and application of large-scale consumer model technology.
To promote the large-scale model, which is regarded as an industrial revolution-level change, it is necessary not only for the leading enterprises to play a leading role, but also for the upstream and downstream to form a joint force. As far as the financial industry is concerned, a number of hard-core enterprises such as Immediate Consumption have leveraged business-focused large-scale model solutions to leverage the transformation of financial ecological partners and form a matrix effect.
At present, the company focuses on the strategy of "self-operated + open platform + financial cloud", and continues to export its own technology solutions to its peers. It is foreseeable that the large model will also become a part of this strategy, using a secure, credible, and compliant large model to solve more user needs, and also allowing ecological partners to enjoy the dividends of technology inclusion.
In this large-scale model trend of embracing, betting and catching up, immediate consumption is a time-moving person and a wave-maker, which will continue to penetrate into the financial service scene and pave the "thick snow" on the "long slope" with more real scientific and technological strength.