After the "100 model war" in 2023, artificial intelligence technology will usher in the landing boom of thousands of industries in 2024.
In the first year of large-scale model application, artificial intelligence technology also reached the critical point of industrial application, and how to break through it has become the focus of attention of industry, academia and research community.
A few days ago, the round table forum with the theme of "(general) artificial intelligence for high-value industrial scenarios" organized by Xunyuan Science and Technology Co., Ltd. was held. A number of leaders of cutting-edge science and technology companies, well-known university experts and professors, representatives of transformation enterprises and front-line scientific research practitioners discussed how to break through the critical point of AI technology application in industry, how to achieve empowerment from technology to industry, and how to achieve AI's ecological future The value of investment and whether technology will replace human resources and other issues have their own views, and they share their own practice and thinking in their respective fields.
Before the arrival of AGI, what needs to be done to improve and perfect the technology? This is a real problem before us. Qiu Xipeng, professor of School of Computer Science and Technology of Fudan University, Li Dahai, co-founder of Mianbei Technology and Ding Ning, chief algorithm scientist of Xunyuan Technology, expressed their views on this issue.
Qiu Xipeng said that on the one hand, Scaling Law is a very Important rules can be significantly improved by expanding the model scale Performance. However, it is not enough to improve the scale of the model now, and many other technologies are involved, including the super alignment mentioned by OpenAI. On the other hand, for an AGI, it is necessary to be able to decompose complex problems into various simple problems to solve them. This requires careful decomposition and strengthening of the model capabilities to achieve AGI faster.
Li Dahai said, "Scaling Law means that the larger the model, the more language materials to be trained, the stronger the computational power to be trained, and the intelligent water of the model The more flat it is. The Scaling Law of the big model is more than that. We hope to advance our understanding of Scaling Law. For example, we have implemented a learning program called WSD Learning rate scheduler, using different learning methods at different stages of the training process Training with learning rate can achieve a better model with less computing power effect. Our ultimate goal is to train enough data to reach or receive Large model near AGI. "
Ding Ning believes that "sometimes what we need is not an AGI, but a model that behaves like an AGI. AGI is not a single technology that can be solved, but a systematic project. To achieve more professional AI, the first thing to mention is the model. At the model level, what we need to care about is not its efficiency, but its upper limit. Secondly, the exploration in decoding is a very good technology. Only by exploring can we achieve water beyond human beings Flat. " He said that the data is very It is important that the role of data is to give supervision signals to the model. Therefore, improving the density of supervision signals and finding suitable supervision agents are issues that need industry to focus on in their own scenarios.
If from a technical perspective, there is not only one way to AGI, and all sectors need to actively explore, then From the industrial side, how can we better realize the landing of AI value? What are your expectations for AI technology development? Visa Global Associate President, Greater China Product Department Pei Sheng, General Manager, Cheng Guoqiang, Technical Director of Yili Group's Digital Technology Center, and Chen Biyong, Founder of Nanxun Co., Ltd.
Pei Sheng exemplified the application of AI technology in real life, It is proposed that people have very high requirements for the upper limit of AI application, so the future development of the large model requires that the value it creates in the professional field must be accurate. In the training phase, it is necessary to choose a scene with high recognition, and someone should intervene.
Cheng Guoqiang, starting from the digital transformation of Yili Group, introduced how the Group implanted AI technology into the digital base of enterprise transformation, and deeply shared the application in specific businesses. When talking about the expectation of artificial intelligence, he said, "because we are real enterprises, we hope to enhance the internal awareness of the whole artificial intelligence, and also hope to deeply integrate technology and business in this process of improvement. In addition, the company attaches great importance to ecological construction, and we look forward to combining scientific and technological concepts and technologies with the actual scene of the company to truly help the company generate value. "
As the enterprise leader of "the last mile of AI commercial applications", Chen Biyong shared the key things that the company is doing. The first thing is to make the data more accurate through a lot of training, and the second thing is to achieve the final strategy through scenarios, thus helping brand operators improve their operational efficiency. He said, "Our approach at the AI level is to make intelligent predictions on sales, user repurchase, marketing response efficiency, etc., and optimize the content, materials, rights, channels, etc. of the brand to reach consumers through intelligent recommendations, so as to improve consumer repurchase. We also look forward to the joint use of generative AI marketing products. Only based on precise objects, precise time and precise rights and interests can we truly help the brand to achieve efficient consumer operation. "
No matter from the technical level or the industrial level, the development and future value of AI are expected. that On the capital side, How do investors and entrepreneurs understand the value of AI technology?
Zhou Zhifeng, partner of Qiming Venture Capital, believes that "AI is the biggest technology wave in the next 10 years, and there is no one." He said that the company's China Invested Nearly 10 top AI companies, including Zhiyuan Technology and Face to Face Intelligence, are using technology to help customers reduce costs and increase efficiency. A breakthrough in business is valuable technology.
"From a certain point of view, today's AI start-ups have much higher requirements for entrepreneurs than before." Li Dahai mentioned that "large model start-ups, on the one hand, should have a deep grasp of the underlying technology of AI, and on the other hand, should have an accurate understanding of the scene and user value. These two major elements must have a relatively balanced understanding and landing."
Ding Ning believes that any scene can be intelligent, and the scene can be divided into high value and low value. Low value scenarios involve cognitive change. When talking about the high-value scenario, he took digital marketing as an example, "For example, when using the big model to do marketing, we tried 1 million times, and 990000 times failed, but we can generate 10000 blockbusters, which brings unparalleled value." At the same time, he also firmly sees the future of AI.
In the second half of the forum, everyone focused on industrial applications From a micro perspective, this paper discusses whether technology will make marketers become the first unemployed people in the value chain around the future application of AI capabilities in the marketing industry.
Pei Sheng said that the emergence of AI in industry applications may create some new products. At this time, the direction of marketing people may change, but for a long time, the actual application is similar to an implementation process. "The deep application of AI will create some new work in a period of time".
Cheng Guoqiang made it clear that he would "not replace". He believed that the biggest difference between people and machines was that people have emotions and understand business. "The big model is just the assistance of people. We hope to help marketing personnel with the tools of the big model. We are also building an agent with the knowledge base and the actual business logic. For employees, they should think about how to combine their business scenarios, let the process run in the scenarios, and achieve human-computer collaboration."
In Chen Biyong's opinion, the industry and brand should make use of technology to make all employees have the same ability Ping, so that more time and energy can be devoted to innovation Sex work. He said that a key pain point shared by the brand side is that the industry experience is difficult to precipitate and can be copied Poor sex. Many brands face different people to do planning, marketing, and service, which brings consumers a completely different sense of body, consumer activity, and repurchase. AI application can liberate the time of staff in brand enterprises and bring their industry experience to the same level Horizontal online, so that they can exert more creativity, link users in a warmer form, and truly return data to business, let people return to creation, and let services have temperature.
The role of technology in industrial development There is no doubt about its important role. In particular, the current consumer market has shifted from shelf e-commerce to content e-commerce, which has also enabled many large model companies to find value.
During the activities of the forum, Xunyuan Technology released the MODI Xiaoxian AI marketing brain (hereinafter referred to as MODI) after three years of polishing This is a core product based on the AI Native one-stop marketing automation base developed by the self research big model of Lianyuan, which fully reflects the differentiation ability of Lianyuan Technology in "integration of model and application, integration of general and professional".
The emergence of MODI is a milestone moment for both academia and industry. Thanks to the big model capability of Lianyungang Technology, MODI can help brand companies to continuously optimize after forming a closed loop from insight to innovation to marketing promotion, break the biggest pain point of no explosive products in the industry, and help customers sell explosive products on the basis of creating explosive products, which is also the goal and ambition of Lianyungang Technology.
When the big model is applied to the industry, it is bound to produce some products and enterprises with epoch-making significance and value.
As predicted by Zhou Zhifeng, in the next 6 to 12 months, we will compare with people Near industries, clusters may appear Sexual or explosive Sex companies, "a world-class super application" may appear this year.
In this regard, we are also full of expectations.
Disclaimer: There are risks in the market, so you should be careful! This article is for reference only, not for sale basis.
