The 100 model war is becoming increasingly fierce, but the doubts in the industry also increase.
With the entry of Tencent, Alibaba, Baidu and other giants, the competition in the field of basic large models is becoming increasingly fierce. But does the market really need so many basic big models?
Although the large model is concerned in the C end such as chat robot, it can bring stable business returns to enterprises. But this does not mean that enterprises can gain considerable profits from C-end users. In fact, with the expansion of user scale, related enterprises may face greater losses. The real business value is still concentrated on the B-end, providing services for retail, finance, manufacturing and other industries, and improving efficiency by reducing costs.
The demand of enterprises for basic large-scale models can be summarized into three categories:
Directly call the big model API to obtain relevant technical capabilities.
Carry out secondary development based on the large model to meet specific business needs.
Develop AI applications based on large models.
These requirements not only test the technical strength of the basic large model, but also test its ability to serve the enterprise.
The key is that, whether a startup or an Internet giant, a platform that can really quickly insight into customer needs and provide stable and efficient services will win.
The development of big models, the definition between technical strength and hype
Since the emergence of ChatGPT, the development of domestic large models has been changing. Before that, Only a few companies are committed to the research and development of basic large models. Because the large models have not been widely recognized by the market, they mainly focus on the deepening of technology and services. However, with the popularity of ChatGPT, the big model has suddenly become a new hot spot pursued by many investors and insiders.
In the context of such an active market, Some enterprises pursuing short-term interests choose to hype, emphasizing concepts while ignoring real technology upgrading. Some of them claim that they are only a few months behind GPT-4, but in fact they only carry out simple training on the basis of open source GPT-2. However, The progress of the large model is far from achievable in a few months; Debugging and optimization both take a lot of time. Technicians familiar with this process all understand the difficulties and uncertainties.
It is worth noting that, as early as before the fire of ChatGPT, companies that really believed in large model technology and deeply cultivated them have begun to layout. Compared with companies pursuing short-term interests, they recognized the potential value of the big model earlier. In addition, as the market has a deeper understanding of the complexity of the big model, companies that rely only on hype will be increasingly difficult to stand on, while enterprises that truly have technical strength can develop stably and withstand the test of competition.
Large model ecology, key to commercialization and application
In the process of large-scale model commercialization, only success in key application scenarios can initially establish a benign ecological cycle of self hemopoiesis.
Although the general large model is widely used, it lacks professionalism in the vertical field. Although the vertical large model is professional, its service scope is limited, resulting in an imbalance of costs and benefits.
To sum up, considering the advantages and disadvantages of the general model and the vertical model, the more appropriate idea for the current large-scale model in the process of commercialization is:, With the general large model as the base, it cooperates with retail, finance, manufacturing and other fields to realize the co construction of application scenarios and create value for daily work and production.
Due to the limitations of data, computing power and application scenarios, the number of open source large models that can truly run completely is limited. At the same time, the big model is similar to PC and mobile phone operating systems, and only one or two leading technologies may dominate the industry. It will be difficult to sustain the development of the large base model that fails to form a prosperous ecology.
From the perspective of the development history of PC and mobile phone operating systems, the first mover advantage is crucial. When Windows dominates the PC market and iOS and Android dominate the mobile phone field, there is almost no room for other operating systems to overturn 。 The big model field also shows this trend, in which the big model will start a prosperous ecology of AI applications. The data, skills and applications of individuals and enterprises can be quickly transformed into AI plug-ins, so as to improve the function of the big model and make it more practical and convenient
In the ecological competition of the large model platform, the common value and growth ability of the platform and its partners are crucial. Specific to office scenarios, such as generating presentations, writing articles, and rewriting styles, all have high requirements for the accuracy of the large model platform. Only the large model with superior technology can support these needs, and needs to be continuously optimized according to user feedback
Both giants and entrepreneurs need to accumulate and iterate from scratch. Therefore, the first mover advantage is crucial in the process of ecological construction, which also explains why early start-ups can compete with large companies with rich resources.
Who will adapt to the Chinese market in the 100 model war?
In the current lively "100 model war", the focus of competition has been very clear: The technology and service capabilities of the large model platform, as well as the strength of its ecological construction, are the key to determine the outcome of the competition.
The construction of these capabilities takes time to accumulate, but first mover advantage does not simply mean first. In addition to the time difference, it is also necessary to have accurate insight into the market demand and implement the strategy quickly and firmly. Unclear strategies or detours may lead to the loss of time advantages.
As the big model platform gradually shifts its focus to ecological construction, strategic concentration and execution in ecological competition become particularly critical. Who can more efficiently accumulate resources in different fields and scenarios will have a more prominent competitive advantage.
In the complex Chinese market, due to the diversity of enterprise regions, scales and needs, it is difficult to find standardized solutions. Large model companies may be lost in various demands, and it is difficult to form a super platform from core technology to project outsourcing companies.
Different from OpenAI, the domestic big model platform needs to pay more attention to the details of commercial landing, because these details determine the relationship with external partners and the landing speed of the big model. Just releasing a large model is not as difficult as the market expected, but only with high-quality data application scenarios can we achieve continuous optimization and establish a competitive advantage. In this context, having high-quality data scenarios and working with more partners has become the key to forming competitive barriers.
In view of the diversified requirements of text generation, intelligent customer service, and data tagging, as well as the characteristics of enterprises of all sizes, Enterprise Big Data has launched its intelligent customer service system solutions to meet the specific needs of enterprises. Different from the rigid and single traditional customer service system, AIW's AI customer service agent has the ability of deep learning and problem decomposition, and can start from the actual needs of customers, keenly catch and accurately respond to their multi-dimensional problems, For example, "Long life product recommendation". This intelligent customer service can not only provide faster and more accurate answers than manual work in most cases, but also 24/7 online , has always maintained a level of intelligence equal to or even higher than that of human beings. With the support of multidimensional data, it It can achieve self iteration, accurately meet the needs and expectations of customers, Redefine the role and function of customer service in business.
Although many people think that the giant companies with abundant funds have more advantages, the real competition is to dig into details, whether it is start-ups or large companies. The capital market has made a choice as to who is the most suitable base model for China.
