2023 is coming to an end.AI Big ModelThe popularity remains unabated.

Just as Google announced that it would provide developers with a new version of the Gemini large model and promised to reduce the cost of use, Microsoft launched a new language model Phi-2 with 2.7 billion parameters. As the top giants are making frequent moves, mid-tier players are starting to form groups, such as Bio Geometry and Zhipu.AIStart to jointly build a large multimodal model of natural language and living language.

Although giants such as Baidu have already laid out large model technology as early as around 2019, 2023 can indeed be regarded as the "first year of large models". Almost all leading technology companies are deeply involved in research and development, and hot money continues to pour in, pushing the "Thousand Model War" to a new climax. However, in addition to the "arms race" of large models, more and more cold thoughts have emerged in the industry: There are more and more basic large models, but why are there so few that can be industrialized? In 2024, the productization, industrialization and commercialization of AI technology will be the top priority for the development of large models.

"Thousand Model War" climaxes

Industrialization becomes the number one problem

Judging from the scale of participating enterprises, the number of large models and the market size, China is already the world's second largest large model industry center after the United States.

As the "flag bearer" of domestic big models, Robin Li mentioned a set of data at the Xili Lake Forum last month: as of October this year, there were as many as 238 big models released in China, which was three times more than in June, and the number of text generation big models available for download on the Hugging Face platform was close to 30,000. In terms of proportion, the number of big models that have been launched/under development in the United States and China accounts for more than 80% of the world, far exceeding other countries or regions.

According to Sotu.com, the scale of China's large-scale model market in 2023 will be about 14.7 billion yuan, doubling year-on-year, and is expected to exceed 100 billion yuan in 2028. The huge market scale and the high attention paid by giants have tempted capital to continue to increase investment; AI is of great significance to improving production efficiency and economic quality, and is related to the country's core competitiveness to a certain extent, so it has also received high attention from relevant departments.It can be said that the big model’s rapid growth throughout the year is inseparable from policy support, the attention of giants and the enthusiasm of capital.

my country is at the forefront of the world in the orderly development of big model technology. Support measures from the central and local governments, such as the "Interim Measures for the Management of Generative Artificial Intelligence Services" jointly issued by the Cyberspace Administration of China and seven other ministries, and the "Several Measures for Promoting the Innovation and Development of General Artificial Intelligence in Beijing (2023-2025) (Draft for Comments)" issued by the Beijing Science and Technology Commission, have been successively released to clear obstacles for the development of big models, provide necessary resource allocation, and avoid disorderly development of technology.

In terms of capital, Baidu, Alibaba, Tencent, ByteDance, iFlytek, Meituan, JD.com, NetEase and other large companies are all deploying large model technology, and strong start-ups have become the sweet buns that VCs are vying for, and hot money is pouring in. According to a report from the China Institute for the Development of New Generation Artificial Intelligence, as of the end of October, there have been 38 large model investment and financing events in China, and there are more than 2,200 existing AI companies.

(Picture from Beike Finance)

On the technical level, domestic basic big models such as Wenxin Big Model, Ali Tongyi, iFlytek Spark, and Zhipu are ranked at the top in multiple lists and can compete with GPT to a certain extent.

The large model industry is thriving, but there are still some hidden concerns - such as the problem of industrialization implementation that plagues most practitioners.Any cutting-edge technology must be transformed into products or applications before it can be used by people and play its value. While AI big models are currently catching up with basic technologies, they also need to go deeper into the scenarios at the industrial end and play a role in the production and operation of enterprises or the life and study of users. In fact, the latter is exactly the advantage of China's AI industry: compared with playing chess, painting, and writing poetry, Chinese technology practitioners are more down-to-earth and good at applying technology to scenarios, making it used for products, applications or services - even if it is not so cool, it doesn't matter.

Three benchmark cases

See the road to industrialization of large models

There are many difficulties in the industrialization of big models. For example, the degree of digitization in different industries varies greatly. Enterprises of different sizes and fields have significant differences in their demand for AI applications and the costs they can afford. In addition to economic costs such as funds, there are also risks, time and marginal costs of using AI technology to transform businesses. Because of this, although many companies are currently paying attention to big model technology, there are very few companies that actually use big model technology to transform their businesses or even create AI native applications.

However, we can also see some benchmark cases of the combination of large-scale model technology and industry.

1. Du Xiaoman Xuanyuan large model: domesticThe firstOpen Source Financial Big Model

The data-driven financial industry is a highly digitalized industry. Digital infrastructure such as databases, storage, servers, automation, and information security are all first applied and popularized in the financial industry. In the process of popularizing AI technology, the financial industry has long been actively exploring the combination of AI with customer service, risk control, credit, marketing and other scenarios to reduce costs and increase efficiency while improving customer experience.

In 2023, big model technology will explode. In May, Du Xiaoman, a pioneering financial technology platform, took the lead in open-sourcing the domesticThe first"Xuanyuan", a large Chinese financial model with a scale of hundreds of billions; in September, "Xuanyuan 70B" will be open source and can be freely downloaded and used.As a large industry model born for financial scenarios, Xuanyuan is highly targeted in terms of intelligent capabilities, functional services and information security.

This targeting is reflected in many aspects: for example, the data set used by Xuanyuan for training contains a large amount of financial industry information such as institutional research reports, professional terms, market data, etc., which gives it a strong ability to understand and process financial information.

In terms of technical strength, Xuanyuan Model is not inferior. It has passed the CPA exam, banking/securities/insurance/fund/futures qualification, financial planner, economist and other financial fields.authorityIn the C-Eval large language model evaluation list jointly released by Tsinghua University, Shanghai Jiaotong University and the University of Edinburgh, and the CMMLU list jointly released by Microsoft Research Asia, MBZUAI and Shanghai Jiaotong University, Xuanyuan ranked first among all domestic open source models.FirstC-Eval and CMMLU are currentlyauthorityThe two major professional lists can be won at the same timeFirst, for a large industry model like XuanyuanabsoluteIt's considered a good result.

Du Xiaoman Xuanyuan model is being deeply applied in financial scenarios.

Internally, Xuanyuan Big Model has deeply empowered Du Xiaoman's marketing, customer service, risk control, office and R&D scenarios, and has achieved initial results. In terms of code assistant, the adoption rate of code generated with the assistance of Big Model can reach 42%, helping the company's overall R&D efficiency to increase by 20%; in the customer service field, Big Model has promoted service efficiency to increase by 25%. In the field of smart office, the current intention recognition accuracy of Big Model has reached 97%.

Du Xiaoman has always attached great importance to the export of financial technology capabilities. Du Xiaoman CTO Xu Dongliang revealed that when Xuanyuan was open sourced in May, hundreds of financial institutions applied for trial use. Judging from the feedback from corporate customers, Xuanyuan's large model has a good reputation for its professional capabilities. The length of the contextual dialogue in version 2.0 has been increased to 8K, and it can also give professional explanations for in-depth issues in the financial industry such as "non-interest income growth trend".

2. Alibaba Tongyi Qianwen model implements the "AI-driven" strategy in the e-commerce industry.

Alibaba will undergo many major changes in 2023, and "user first, AI-driven" will become its new strategic direction.When the Tongyi Qianwen big model was released on April 11, Zhang Yong, then chairman of Alibaba Group and CEO of Alibaba Cloud Intelligence Group, said, "All software is worth upgrading and transforming with the big model, and all Alibaba products will be connected to Tongyi Qianwen."

Alibaba has indeed done what it said. Its e-commerce business, which is Alibaba’s core business, has long been fully AI-based.Based on the Tongyi Qianwen model, Taotian Group has launched a series of AI tools for both B and C ends.

Tools for the B-side include official customer service robots, intelligent image generation, and self-monitoring of marketing placements. During the Double 11 promotion this year, merchants used backend AI tools more than 1.5 billion times. For the C-side, Taobao Wenwen launched an AI smart assistant, and the number of people invited to try it exceeded 5 million within two months of its launch. B-side tools can improve merchant operating efficiency and reduce traffic costs, while C-side functions can significantly improve user experience, forming differentiated competitiveness when prices are skyrocketing in the e-commerce industry.

Alibaba has gone the fastest and furthest in combining big models with e-commerce scenarios. Jack Ma even mentioned the refreshing concept of "AI e-commerce" in his reply on Alibaba's intranet.

In order to further strengthen its large-scale model technology and deepen the integration of AI and business, Taotian Group was recently exposed to have secretly formed a new AI team and recruited AI talents with high salaries.top notchTalents should seize the time to train the exclusive big model "Turing" for the e-commerce industry. According to the information previously disclosed by Taotian Group, more AI tools will be released to merchants in the next year, including AI store opening, business consulting, smart weekly reports, etc. The service scope covers all aspects of merchants' daily operations. Driven by Alibaba, the combination of big models and e-commerce industry has just begun. It can be foreseen that in 2024, the leading e-commerce platforms will increase their investment in "big model e-commerce".

3. iFLYTEK Spark Big Model: a benchmark player in big model + education.

iFLYTEKFirstThe first label is voice intelligence, and the second label is a giant in smart education technology. Before the emergence of big model technology, iFLYTEK has been working on AI technology for many years, and a considerable part of its revenue comes from smart education services, such as oral assessment, educational hardware and other intelligent education services.

After the explosion of big model technology, the combination of Spark Big Model and the education industry has been even more vigorous. In May this year, the day after the release of iFlytek Spark Cognitive Big Model 1.0, the A-share education technology sector soared. In addition to iFlytek, XueDa Education, Action Education, and Guoxin Culture all followed the daily limit, showing a trend of "sparks setting the prairie on fire".

From 1.0 to 3.0, iFlytek Spark Big Model has always focused on conquering coding capabilities and multimodal capabilities, and based on technological breakthroughs, it has developed more functions and applications for schools, educational enterprises, teachers and students.For example, it provides student and teacher information management and school leaving application review functions for school management, teaching courseware production assistants tailored for teachers, and AI one-on-one heuristic dialogue functions for students. At the same time, iFLYTEK is also deeply applying large model technology to its educational hardware such as translation pens, voice recorders, learning machines, and office notebooks to strengthen product strength and consolidate its advantages in this category.

(Picture from iFlytek Spark official website)

The leading players in the three industries of finance, e-commerce and education can all gain new growth points through the transformation of big models. It can be seen that the industrialization of big models is not a pipe dream, but an inevitable trend.

Large Model Opening 2024:

Is there a secret to industrialization?

Du Xiaoman, Alibaba, and iFLYTEK have only made a good start. The industrialization level of big models still has a lot of room for improvement. In particular, industries with a long history and low level of digitization, such as agriculture, manufacturing, logistics and shipping, and energy, are in urgent need of embracing big model technology to improve production efficiency and achieve the leap from digitization to intelligence.In view of this, accelerating the productization, industrialization and commercialization of AI technology will be the top priority of the big model industry in 2024.Whoever can first run through the path of industrialization will be the winner in the "Thousand Model War". So, what enlightenment do the benchmark players bring to the industrialization of large models?

First, do not reinvent the wheel, select training parameters and design functional services in a targeted manner.

There are already many basic large models, what is lacking in the market is that they can compete with or even surpass GPT.top notchThe basic big model and the "industry big model" that can enable thousands of industries to lower costs, lower thresholds, and faster applications. To make a powerful industry big model, you need to "both understand AI technology and be an industry expert."

Du Xiaoman is a good example. On one hand, it has the AI technology foundation backed by Baidu, and on the other hand, it has the industry knowledge, capabilities, scenarios, ecology and other resources accumulated over many years of deep involvement in the financial technology industry.

It is reported that although Xuanyuan is trained based on the Bloom large model with 176 billion parameters, it is also inseparable from the hundreds of billions of tokens Chinese pre-training data sets accumulated by Du Xiaoman over the years, including basic knowledge and huge parameters of industries such as banking, insurance, and funds. Because of the latter, the Xuanyuan large model has a financial information processing capability that far exceeds similar competitors and general large models, and can also provide targeted functional services for the pain point scenarios of the financial industry.

Second, "customize" large-model functional services in-depth to meet industry needs, rather than developing them in isolation.

Technology companies are prone to the problem of "looking for nails with a hammer". If it cannot meet real needs, no matter how powerful the technology is, it may just be self-satisfaction.

Why can Du Xiaoman, Alibaba and iFlytek be the first to reap the benefits of large-scale model industrialization? Because Alibaba itself is the leader of the e-commerce industry, Du Xiaoman has been deeply involved in the construction of the domestic technology and financial industry since its establishment, and iFlytek has also been deeply involved in the smart education industry for more than ten years. Their understanding of the corresponding industry is beyond the reach of ordinary companies. By understanding the operating logic and deep-seated problems of the industry, we can gain insight into the real pain points of enterprises and practitioners and provide effective solutions.

Taking Du Xiaoman as an example, based on the four basic capabilities of understanding, generation, logic and memory, Xuanyuan Big Model integrates the usage habits of the financial industry, optimizes needs, and provides a series of targeted functions. For example, for personal credit management services, Xuanyuan Big Model provides bank customers with customer history information management and user multi-level demand analysis functions, and provides users with natural language interactive question and answer services for professional questions, fully improving the processing efficiency of both parties. Du Xiaoman has insight into many needs when serving financial institutions and its own customers, so that it can make truly usable, useful and easy-to-use financial big model products.

Third, when everyone adds fuel, the flames rise high. The big model is not a one-man show and must benefit industry participants.

Small and medium-sized enterprises are the main force of the industrial chain. However, due to the limitation of financial strength and talent resources, it is often difficult for them toFirstIt is time to apply new technologies, especially new technologies with high barriers to entry. Compared with deep learning, large models require huge computing power, huge amounts of data, and huge amounts of algorithms. The barriers to entry are much higher, and for many companies, it is somewhat out of reach. This is an opportunity for the leading players. If they adhere to the inclusive and open route, they can not only make the large model technology have a landing point for "industrialization", but also obtain corresponding value in the industrialization of large models.

At the Big Model Technology and Application Forum jointly organized by Du Xiaoman and Peking University Guanghua School of Management, Du Xiaoman CTO Xu Dongliang expressed a similar view. He believes that big models are an opportunity for small and medium-sized financial institutions to break out because they can accelerate the digital and intelligent upgrading process through application innovation and then cross the digital divide.

It is not difficult to find that "openness" has become a large model for the successful implementation of the industry.maximumCommon denominator. Xuanyuan of Du Xiaoman, Tongyi Qianwen of Alibaba, and Spark of iFlytek all take the open source route. As Xu Dongliang said, opening up the big model capabilities to financial institutions can not only accelerate the popularization of technology, but also lower the threshold for use. It is an inevitable choice to achieve technology inclusion.

Unlike the short-lived popularity of emerging technologies such as blockchain, the popularity of big models will not drop suddenly. On the one hand, big model technology will penetrate into more industries in 2024. On the C-side, big model-driven explosive phenomenal applications will definitely appear, and on the B-side, there will be more and more cases of big model industrialization. On the other hand, big model technology is essentially a continuation of deep learning technology. AI technology has been developed for more than 10 years and will be the basic technology of the technology industry for decades to come. Big models are the first step in the AI wave.maximumThe AI wave will continue to surge.