The 23-year-old general-purpose large-scale model detonated the AI storm, and the 24-year-old vertical model reshaped thousands of industries丨Shanghai Artificial Intelligence Conference Golden Sentence Collection

Image credit: Generated by Unbounded AI tools

On July 7, 2023, the 2023 World Artificial Intelligence Conference entered the second day of the agenda.

On this day, dozens of forums, including large models, computing power, data, metaverse, governance, etc., came one after another, which was very lively.

If you participate in today's forum, you will definitely be impressed by a few words in your mind, that is, industry, vertical and application. Compared with the general language model of ChatGPT, what the entire industry is looking forward to is a vertical industry model that can truly empower thousands of industries.

A guest put it bluntly: 2023 is the year of the explosion of large-scale models, and 2024 must be the year of industrial applications. Large-scale models will retreat behind the scenes, and vertical industry models will come to the fore. These vertical industry models will promote AI and the industry to leap forward Development, this is a bit like the Cambrian explosion. We will wait and see.

The following content is shared by guests and organized by Babbitt.

General-purpose large models are not the only direction for model application, and models for vertical industries will become the tipping point of the value of large models. With the continuous iteration of technology, the industrial application of large models will also usher in an acceleration. —— Li Qiang, vice president of Tencent and president of government and enterprise business

In the future, the number of industrial large-scale models and vertical large-scale models will greatly exceed that of general-purpose large-scale models, and the implementation of large-scale models will be more reflected in vertical large-scale models such as industries and enterprises. The future challenge of large models lies in the "last mile", that is, to play value in enterprise business, which requires not only powerful and flexible basic software, an open and flexible white box model, but also practitioners who understand the business. However, the reduction in computing power cost will be faster than the increase in model size, and computing power will not be a gap. ——Shang Mingdong, co-founder of Jiuzhang Yunji DataCanvas

The general-purpose large model can solve 70%-80% of the problems in 100 scenarios, but it may not be able to 100% meet the needs of a certain scenario of the enterprise. If an enterprise conducts fine-tuning based on the industry's large model and its own data, it can construct a dedicated model to create a highly available intelligent service, and the model parameters are less than the general large model, the cost of training and reasoning is lower, and the model optimization is also more efficient. easy. —— Tang Daosheng, Senior Executive Vice President of Tencent Group, CEO of Cloud and Smart Industry Business Group

In the reality that computing power is in short supply, traditional computing architectures are losing competitiveness, and new computing models must be explored. According to preliminary calculations in the report, it is estimated that by 2025, the scale of my country's core computing power industry will not be less than 4.4 trillion yuan, and the scale of computing power related industries will reach 24 trillion yuan. ——Kang Yong, Chief Economist of KPMG

Facing the wave of AI, the financial field is actually very anxious. Practitioners often discuss how to apply large models in the financial field. However, due to the strict compliance requirements in the financial field, it is a great challenge to apply large models. Domestic companies want to overtake on curves, and the large model is a challenge. However, in the field of vertical large models, large language models that are suitable for the actual industry, because of domestic advantages in data and application scenarios, there are great opportunities for overtaking on curves. ——Haomai Fortune CIO Fu Xiaomin

There is a fierce competition in the "Hundred Models War" in the domestic large-scale model circuit. After trying these large-scale models, they will have a strong sense of impact and are easy to use. This will make people feel very optimistic about the domestic large-scale models . But the large model is not simply a technological competition, but also an ecological competition. It depends on how many companies and entities are willing to build applications based on your large model. This is what everyone is doing. Domestic large-scale models will have a prosperous ecology, and I believe they will perform well at the application level. ——Ma Qianli, co-founder of Unbounded AI

There are a lot of investments, which may be wiped out, 95%, maybe even higher, all gone after 2-3 years. But you have to vote because the opportunity cost is too high. In 1998, the Internet had great opportunities and had just begun, but the company had bubbles in areas such as valuation. AI will also go through this process. Artificial intelligence is not a bubble, some companies are bubbles. ——Zhang Yaqin, professor of Tsinghua University and academician of Chinese Academy of Engineering

There is a big difference in the distribution of talents in the AI field between China and the United States. In the United States, 19% are in Google, followed by Microsoft, META, etc., mainly because talents are concentrated in enterprises. And 12% of China's talents are in Tsinghua University, followed by Zhejiang University and Peking University. Among the top ten rankings, there is only one company, Alibaba, which ranks fifth, and the rest are universities. ——Zhou Zhifeng, Partner of Qiming Venture Partners

It is meaningless for everyone to make a large language model, and the large language model does not solve specific problems. Large language models can solve general knowledge problems, and there are unique problems in professional fields. On the B side, the large language model combines the data of specific application scenarios and various tools to truly complete industrialization and productization, such as upgrading tool software to make it more efficient, easy to use, and intelligent. This will be a structural Opportunity to upgrade. On the C side, the combination of large language models with robots and personal agents to help users make decisions will also have huge opportunities. In the next 3-5 years, I think there will be structured opportunities for the application of large language models. ——Wang Xiao, founder of Jiuhe Venture Capital