Under the trend of big model technology, small and medium-sized enterprises are approaching the singularity of digital transformation

　　Focus: artificial intelligence, chips and other industries

　　Welcome all guest officials to pay attention and forward.

　　Foreword:

　　"Don't dare to turn, can't turn, can't turn" has always been the three mountains that are pressing on the digital transformation of small and medium-sized enterprises.

　　In essence, the big model will fundamentally reduce the cost and complexity of digital transformation of small and medium-sized enterprises.

　　Author | Fang Wensan

　　Image source | Network

　　Data is the "standard" of artificial intelligence.

　　Algorithm, computing power and data can be said to have become the three major factors to promote the development of artificial intelligence, among which data is particularly important.

　　In the process of training artificial intelligence model, if we want to make it smarter, a large number of diverse data are essential.

　　The application of small data and high-quality data has its premise, that is, it is necessary to fine-tune the model through small data on the basis of a large basic model (pre-training model), so that the model can serve specific application scenarios more accurately.

　　From this perspective, data will play a key role in the future when the basic model completes the downstream tasks, and the model is also a necessity.

　　The enthusiasm of the big model track is high

　　The advantage of large model is that it can learn more features and patterns, thus improving the understanding and processing ability of input data, and at the same time, it can predict and classify more accurately, produce more natural language generation results, or make more wise choices on complex decision-making problems.

　　At present, the industry is facing intelligent transformation. For enterprises and institutions, deep integration with artificial intelligence, adding industry-specific data and knowledge and accurately matching real application scenarios can greatly improve the efficiency and level of business processes and become an important force driving industrial transformation and upgrading.

　　The big model is diversified in the future, and the development trend is to do small things and specialize. The real opportunity lies in the enterprise market. Since the release of ChatGPT, leading enterprises of artificial intelligence have followed suit.

　　Baidu, Ali, Zhipu Huazhang, etc. have taken the lead in launching self-developed large-scale model products nationwide and opening up internal testing, which has certain industry migration and empowerment capabilities, and the enthusiasm for venture capital in large-scale model tracks is high.

　　AI big model empowers small and medium-sized enterprises to land in the industry

　　At present, many large models on the market already have a large amount of data, and even some open source models have been born, so the accumulation of data has spent a lot of money and time.

　　In this context, for many application-oriented enterprises, it is even more important to think about how to develop the subsequent industry model after labeling the model in the facing field.

　　At present, enterprises often face problems such as insufficient computing power resources, low data quality and insufficient scene opening when applying large models. There are two major challenges facing enterprises here.

　　The first is how to form an advantage in industry data, which tests the enterprise's understanding of the landing industry, such as how to mark the very professional processing data.

　　The second challenge is to invest in computing power, and model training needs to invest huge computing power resources.

　　Various industries are still exploring how to make full use of the existing large model foundation, give full play to the advantages of the scene and form the industrial ecology of general artificial intelligence.

　　Creating a better platform and environment supported by the government, bringing together upstream and downstream partners in the industrial chain, and promoting the cooperation of multi-parties such as technology, resources and scenarios will be conducive to promoting the research and development and application of large models.

　　The big model is not cheap enough, and there are costs besides the model.

　　According to the "Research Report of China Artificial Intelligence Large Model Map" issued by the Ministry of Science and Technology, China has released 79 large models with a scale of over 1 billion parameters.

　　There are so many models in a short time, but large enterprises need professional, customized and directly usable products, and the transformation is still challenging.

　　Obviously, not all enterprises can participate in the development of the basic big model, and its training cost is extremely high, and it is more about dealing with the big model from the user's point of view.

　　Although the price of API has been greatly reduced for enterprises adopting large models, for example, in March 2023, the price of ChatGPT API dropped by 90% compared with the end of 2022, and in June 2023, the price of input terminal dropped by 25% again.

　　Even so, this new calculation method is still more expensive than the traditional one.

　　According to Reuters, John Hennessy, chairman of Google's parent company Alphabet, said that the cost of interacting with large language models may be 10 times that of standard keyword search. Therefore, although the use cost of the large model has been declining, its price is still high.

　　In addition, the best effect may not be achieved only by connecting the large model to a specific field.

　　Therefore, before accessing API and putting it into commercial use, enterprises need to pay extra costs in many aspects. This includes the costs of data preparation and preprocessing, model training and tuning, deployment and operation and maintenance, model updating and iteration, and legal compliance.

　　All these factors make the overall cost of applying the big model still high. In the era of AI, these card points still exist.

　　However, the multi-domain generalization ability demonstrated by the AI model has the advantage of solving problems, and it also has to make SMEs accelerate their embrace of digital transformation.

　　The big model goes to the industry.

　　Large foreign models are often polished and matured in the laboratory first, but the reality outside Silicon Valley is complicated.

　　The domestic big models are often made by Internet companies and major industries.

　　At present, the application of China's big model in the four key areas of finance, media, cultural tourism and government affairs is relatively mature.

　　The financial sector is an information-intensive industry, and it is one of the best application scenarios of large model technology.

　　For example, the financial industry model "Pufa Baidu-Wenxin Model" jointly developed by Baidu and Shanghai Pudong Development Bank combines the industry data and knowledge accumulated in the Pufa scene.

　　Technical and business experts from both sides have designed targeted pre-training tasks such as financial report field discrimination and financial customer service question-and-answer matching, so that Wenxin Big Model can learn the knowledge of the financial industry and improve its efficiency in the application of typical tasks in Pudong Development Bank.

　　However, the high cost of new application development and the challenges brought by data compliance and security still lie in front of the financial industry and need to be further optimized.

　　Ending:

　　It should be noted that this capital and technology-intensive industry is usually difficult to combine with small and medium-sized enterprises with weak capital chain and lack of talents and technology.

　　However, large model manufacturers, digital solution providers and small and medium-sized enterprises will all benefit from this technological revolution, and the original market relationship and pattern will also change.

　　The content reference comes from: AI Nuggets: Can the "technical equality" brought by the big model benefit the digital transformation of SMEs? ; First finance and economics: comprehensively promote the application of large models and create a new pattern of industrial development-vigorously promote the landing and application of domestic large models in the industrial field; Multi-knowledge network: the commercialization of large-scale model application is slow, why is it slow? ; Economic Daily: The application of big model industry is accelerating.

　　The manuscripts and pictures published in WeChat official account are from the Internet and are only used for communication. If there is any infringement, please contact us and reply. We will handle the information within 24 hours after receiving it.

　　Please be sure to indicate:

　　"Name+Company+Cooperation Requirements"