On November 28th and 29th, the 36Kr WISE2023 King of Business Conference was grandly held at the Beijing International Convention Center. With the theme of "The Sun Always Rises", this conference spans a main venue and six vertical fields. The main venue focused on six major chapters: "The Next 3650 Days", "In the Industrial Flood", "The Internet of Things in Attack", "AI and Business Increase", "Global Brands Look at China", and "Technology First and Sharing Innovation" The agenda invites business figures from all fields to engage in a two-day top-level business dialogue, asking questions now and giving answers to the future.
held a wonderful roundtable dialogue at the main venue of the conference on November 29 on the topic of "How to Continue to Advance AI Business Implementation". It is led by Chen Zibing, founding partner of Qianyue Capital, Fang Han, chairman and CEO of Kunlun Wanwei, Dr. Jue Ao, chief expert of Alibaba Group’s Turing Lab and head of Shenzhen Xiang Intelligent Technology Algorithm, Liang Jing, co-founder of Squirrel AI, Jinqiu Fund Executive Director Zang Tianyu, Founding Partner of Qingzhi Capital Zhang Yu and other major figures in the AI field jointly launched an in-depth discussion and sharing.
Roundtable dialogue scene
The following is the content of the roundtable dialogue, compiled and edited by 36Kr:
Chen Zibing: Hello everyone, I am Chen Zibing from Qianyue Capital. Today I am very happy to discuss and share with you some interesting topics related to commercialization and industrialization in the context of this new wave of AI technology. Let's first ask the guests to introduce themselves.
Fang Han: Hello everyone, I am Fang Han, the chairman of Kunlun Wanwei. Kunlun Wanwei is an Internet platform company going overseas. We will start to enter the track of large model research and development in 2020. In 2021, we have already released a large pre-trained Chinese model. In the same year, research on AI music generation began. The slogan we proposed in 2022 is All in AIGC. In December 2022, we released the earliest open source comprehensive training model in China. In April this year, we released our 100-billion-dollar model "Tiangong". In August, we released our AI search product. At the same time, we are also exploring AIGC products overseas such as music, comics, games, and social networking.
juao: Hello everyone! I am Wang Yan from Shenxiang Intelligence, a subsidiary of Yintai Commercial Group, and Jue Ao is my nickname. Our business positioning is to be a consumer solution provider driven by big data, including a brain and a pair of eyes. The brain refers to the business intelligence of retail entities, that is, embedding artificial intelligence in links such as people, goods, and sites to achieve cost reduction. Improve efficiency; Eyes refers to machine vision, that is, intelligent analysis of surveillance videos of offline business premises, so that surveillance cameras can not only see and see clearly, but also understand. With the "brain" and "eyes", the black box of business operations is opened.
Dr. Jue Ao, chief expert of Alibaba Group Turing Lab and head of Shenzhen Xiang Intelligent Technology Algorithm
Liang Jing: Hello everyone, my name is Liang Jing, from Squirrel Ai Intelligent Hardware Company. Squirrel Ai was founded in 2014. We We have been insisting on artificial intelligence adaptive technology from beginning to end, and have also added large model technology. The usage scenarios we generate are smart learning machine scenarios where students study independently at home or in study rooms. At the same time, on campus, we provide smart campus SaaS platform accounts and surrounding smart hardware ecosystem products. Our core is to always explore the essence of education. The essence of learning is actually to find the weak points of each student's knowledge and let students learn to understand, such as learning that knowledge point and some related abilities, rather than simply learning from On the surface, we can't solve problems, so from the beginning until now, we have attached great importance to and the core of research and development is that students must learn and understand the knowledge content from the root. In this case, once the foundation is laid, there will be a very good improvement and breakthrough in future learning and training of his abilities. Thanks!
Zang Tianyu: Hello everyone, I am Zang Tianyu from Jinqiu Fund. The investment of Jinqiu Fund mainly revolves around two main lines, one technical main line is AI, and the other business main line is globalization.Starting from the implementation of discriminative AI, AI and its industrial applications are actually the most focused direction of our team. And this year, we are also focusing on this wave of generative AI. I am very happy to follow you here today. Thank you for sharing.
Zhang Yu: Hello everyone! I am Zhang Yu from Qingzhi Capital. Qingzhi Capital was established with the support of Tsinghua University Intelligent Industry Research Institute (AIR) and focuses on investing in start-up companies in the AI field. Our slogan is "Focus on AI, Invest in the Future". We have thousands of square meters of physical incubators to incubate and invest in original AI projects from the source. If you want to have a good idea and prepare to start a business in the AI field, we provide free entrepreneurial space, incubation and investment opportunities in the Universe Center-Tsinghua Science and Technology Park. Thank you.
Chen Zibing: Thank you all guests for your introduction and sharing. Let us enter the first topic. We have seen that generative AI has been iterating and developing very quickly at the technical level this year. After a year, have you seen any of the leading implementations of generative AI? scenarios, or which industries do you think have development potential?
Fang Han: Before talking about this topic, a serious problem with generative large models is illusion, that is, the problems that arise are not necessarily correct, causing all walks of life - especially industries with strict fault tolerance requirements - to Acceptance of models has been slow. In the entertainment and social fields, accuracy requirements are not so high, and these fields will be implemented quickly. The one with the fastest progress in commercialization is Vincent Picture. Now AI has completed a lot of work for Taobao models and photographers. It used to cost 200 yuan a picture, but now it costs two cents a picture. This is a trend we have seen. On the C side, especially in content social networking and entertainment, I think AI will be implemented soon. Especially in some vertical industries, our intelligent assistants will gradually become popular.
Kunlun Wanwei Chairman and CEO Fang Han
Chen Zibing: Thank you Mr. Fang, what does Dr. Jue Ao think about this question?
Jue Ao: I strongly agree with . There is no industry without AI. If there is no industry scenario, there will be no AI. A few years ago, PPT, papers and patents were enough to attract investors, but now AI must have industry implementation scenarios in order to continue to develop. Even if it is listed, investors have to see if there are good implementation scenarios. . All industries deserve to be redone with large models. The
large model allows us to create content quickly and with high quality, benefiting industries including news, advertising, movies, and games, all of which have strong creative attributes. Our Intime Business is also using AIGC to create content. Intime has thousands of WeChat groups. We generate different marketing plans and beauty solutions for our customers every day to recommend to our customers, including trendy beauty recommendations. , clothing, etc.
Another one is smart shopping guide. We at Yintai Commercial use AIGC technology to recommend products and outfits that best meet their needs based on customers’ shopping history and preferences, as well as understanding of the product knowledge base, and provide interactive Q&A and after-sales services. etc.; In addition, everyone knows that shopping guides in shopping malls are very critical and directly determine sales. We digitize the long-term accumulated experience of excellent shopping guides and feed these data to large model training, so that excellent shopping guides can quickly Copy it and give it to every brand store to help brand stores in Yintai shopping malls improve their performance.
Chen Zibing: Mr. Liang Jing, what is your different perspective?
Liang Jing: My perspective on the education industry is about the application of AIGC in the entire education industry, such as personalized learning, auxiliary assistants, and automatic correction. For education technology companies, content generation is faster, broader, more accurate, and more Lively and interesting. For future potential, for broader outcomes like educational equity. At present, we see that if the large model product is actually only one-way interaction, when the user has a need, the user asks such questions and needs to the system, and the system gives the answer, which seems to be the feeling of our assistant teacher.
On November 28th and 29th, the 36Kr WISE2023 King of Business Conference was grandly held at the Beijing International Convention Center. With the theme of "The Sun Always Rises", this conference spans a main venue and six vertical fields. The main venue focused on six major chapters: "The Next 3650 Days", "In the Industrial Flood", "The Internet of Things in Attack", "AI and Business Increase", "Global Brands Look at China", and "Technology First and Sharing Innovation" The agenda invites business figures from all fields to engage in a two-day top-level business dialogue, asking questions now and giving answers to the future.
held a wonderful roundtable dialogue at the main venue of the conference on November 29 on the topic of "How to Continue to Advance AI Business Implementation". It is led by Chen Zibing, founding partner of Qianyue Capital, Fang Han, chairman and CEO of Kunlun Wanwei, Dr. Jue Ao, chief expert of Alibaba Group’s Turing Lab and head of Shenzhen Xiang Intelligent Technology Algorithm, Liang Jing, co-founder of Squirrel AI, Jinqiu Fund Executive Director Zang Tianyu, Founding Partner of Qingzhi Capital Zhang Yu and other major figures in the AI field jointly launched an in-depth discussion and sharing.
Roundtable dialogue scene
The following is the content of the roundtable dialogue, compiled and edited by 36Kr:
Chen Zibing: Hello everyone, I am Chen Zibing from Qianyue Capital. Today I am very happy to discuss and share with you some interesting topics related to commercialization and industrialization in the context of this new wave of AI technology. Let's first ask the guests to introduce themselves.
Fang Han: Hello everyone, I am Fang Han, the chairman of Kunlun Wanwei. Kunlun Wanwei is an Internet platform company going overseas. We will start to enter the track of large model research and development in 2020. In 2021, we have already released a large pre-trained Chinese model. In the same year, research on AI music generation began. The slogan we proposed in 2022 is All in AIGC. In December 2022, we released the earliest open source comprehensive training model in China. In April this year, we released our 100-billion-dollar model "Tiangong". In August, we released our AI search product. At the same time, we are also exploring AIGC products overseas such as music, comics, games, and social networking.
juao: Hello everyone! I am Wang Yan from Shenxiang Intelligence, a subsidiary of Yintai Commercial Group, and Jue Ao is my nickname. Our business positioning is to be a consumer solution provider driven by big data, including a brain and a pair of eyes. The brain refers to the business intelligence of retail entities, that is, embedding artificial intelligence in links such as people, goods, and sites to achieve cost reduction. Improve efficiency; Eyes refers to machine vision, that is, intelligent analysis of surveillance videos of offline business premises, so that surveillance cameras can not only see and see clearly, but also understand. With the "brain" and "eyes", the black box of business operations is opened.
Dr. Jue Ao, chief expert of Alibaba Group Turing Lab and head of Shenzhen Xiang Intelligent Technology Algorithm
Liang Jing: Hello everyone, my name is Liang Jing, from Squirrel Ai Intelligent Hardware Company. Squirrel Ai was founded in 2014. We We have been insisting on artificial intelligence adaptive technology from beginning to end, and have also added large model technology. The usage scenarios we generate are smart learning machine scenarios where students study independently at home or in study rooms. At the same time, on campus, we provide smart campus SaaS platform accounts and surrounding smart hardware ecosystem products. Our core is to always explore the essence of education. The essence of learning is actually to find the weak points of each student's knowledge and let students learn to understand, such as learning that knowledge point and some related abilities, rather than simply learning from On the surface, we can't solve problems, so from the beginning until now, we have attached great importance to and the core of research and development is that students must learn and understand the knowledge content from the root. In this case, once the foundation is laid, there will be a very good improvement and breakthrough in future learning and training of his abilities. Thanks!
Zang Tianyu: Hello everyone, I am Zang Tianyu from Jinqiu Fund. The investment of Jinqiu Fund mainly revolves around two main lines, one technical main line is AI, and the other business main line is globalization.Starting from the implementation of discriminative AI, AI and its industrial applications are actually the most focused direction of our team. And this year, we are also focusing on this wave of generative AI. I am very happy to follow you here today. Thank you for sharing.
Zhang Yu: Hello everyone! I am Zhang Yu from Qingzhi Capital. Qingzhi Capital was established with the support of Tsinghua University Intelligent Industry Research Institute (AIR) and focuses on investing in start-up companies in the AI field. Our slogan is "Focus on AI, Invest in the Future". We have thousands of square meters of physical incubators to incubate and invest in original AI projects from the source. If you want to have a good idea and prepare to start a business in the AI field, we provide free entrepreneurial space, incubation and investment opportunities in the Universe Center-Tsinghua Science and Technology Park. Thank you.
Chen Zibing: Thank you all guests for your introduction and sharing. Let us enter the first topic. We have seen that generative AI has been iterating and developing very quickly at the technical level this year. After a year, have you seen any of the leading implementations of generative AI? scenarios, or which industries do you think have development potential?
Fang Han: Before talking about this topic, a serious problem with generative large models is illusion, that is, the problems that arise are not necessarily correct, causing all walks of life - especially industries with strict fault tolerance requirements - to Acceptance of models has been slow. In the entertainment and social fields, accuracy requirements are not so high, and these fields will be implemented quickly. The one with the fastest progress in commercialization is Vincent Picture. Now AI has completed a lot of work for Taobao models and photographers. It used to cost 200 yuan a picture, but now it costs two cents a picture. This is a trend we have seen. On the C side, especially in content social networking and entertainment, I think AI will be implemented soon. Especially in some vertical industries, our intelligent assistants will gradually become popular.
Kunlun Wanwei Chairman and CEO Fang Han
Chen Zibing: Thank you Mr. Fang, what does Dr. Jue Ao think about this question?
Jue Ao: I strongly agree with . There is no industry without AI. If there is no industry scenario, there will be no AI. A few years ago, PPT, papers and patents were enough to attract investors, but now AI must have industry implementation scenarios in order to continue to develop. Even if it is listed, investors have to see if there are good implementation scenarios. . All industries deserve to be redone with large models. The
large model allows us to create content quickly and with high quality, benefiting industries including news, advertising, movies, and games, all of which have strong creative attributes. Our Intime Business is also using AIGC to create content. Intime has thousands of WeChat groups. We generate different marketing plans and beauty solutions for our customers every day to recommend to our customers, including trendy beauty recommendations. , clothing, etc.
Another one is smart shopping guide. We at Yintai Commercial use AIGC technology to recommend products and outfits that best meet their needs based on customers’ shopping history and preferences, as well as understanding of the product knowledge base, and provide interactive Q&A and after-sales services. etc.; In addition, everyone knows that shopping guides in shopping malls are very critical and directly determine sales. We digitize the long-term accumulated experience of excellent shopping guides and feed these data to large model training, so that excellent shopping guides can quickly Copy it and give it to every brand store to help brand stores in Yintai shopping malls improve their performance.
Chen Zibing: Mr. Liang Jing, what is your different perspective?
Liang Jing: My perspective on the education industry is about the application of AIGC in the entire education industry, such as personalized learning, auxiliary assistants, and automatic correction. For education technology companies, content generation is faster, broader, more accurate, and more Lively and interesting. For future potential, for broader outcomes like educational equity. At present, we see that if the large model product is actually only one-way interaction, when the user has a need, the user asks such questions and needs to the system, and the system gives the answer, which seems to be the feeling of our assistant teacher.But in fact, learning is not just one-way, it should be two-way. The system must better understand the user's past learning situation and learning data, which requires the accumulation of a large amount of data and behavior. Two-way interaction and learning can truly provide personalized The current solution for Squirrel Ai is to combine AI adaptation with large model technology. In this way, we can provide two-way content, and we have already achieved a lot of advantages and results. In the future, we may feel that this path will go deeper, and We cannot only rely on AIGC technology, but also combine it with the essence of education and teaching to provide value technology to be more effective.
Squirrel Ai co-founder Liang Jing
Chen Zibing: From an investor's perspective, have you seen any of the leading commercialization areas of ?
Zang Tianyu: We believe that the most direct application of is in the Internet field, which is more focused on entertainment and social interaction. We have seen that many teams in China are doing entrepreneurship like Character.ai. In such a scenario, hallucination is not a problem, but a benefit. , including the recent face-changing short drama overseas, is also an area of entertainment direction. The second is applications that focus on efficiency and productivity, such as text and code generation scenarios, including processes that drive some enterprise automation, which can create clear value. Existing players in these fields combine new technologies to bring new experiences and implement them. Faster, there are many examples such as Microsoft Copilot, GitHub Copilot, Adobe firefly, etc. Thirdly, if we look at categories, we have seen many implementation cases in marketing, education, and law, generating products or marketing materials for companies, and many of them have made money.
Chen Zibing: Thank you Tianyu for sharing, and I would also like to ask Mr. Zhang Yu if he has any new views on this issue from our perspective.
Zhang Yu: I particularly agree with what the previous guests said. Application is very important. Our Qingzhi Capital has also invested in nearly 10 projects this year, including AIGC, large model, biomedicine, and robot-related artificial intelligence projects. From our perspective, we also have some opinions of our own:
comes first and pays more attention to industry models. The stronger the general ability of large models, the better, but after all, the field of large models only requires a small number of bases. There are thousands of industries. In-depth knowledge of the industry is very important and has more practical value. We have also invested in projects such as We will also take an in-depth look at the general model in medicine in the field of industry models.
Second, focus on industrial applications. Industrial applications are very important. Currently, AICG applications are common in media fields such as entertainment, live broadcasts, and text and graphics generation. On the other hand, the intelligent transformation and upgrading of traditional industries is also very important, such as manufacturing automation, industrial programming, etc., using AI for No-code/low-code program generation is equivalent to changing a paradigm standard. In the past, Microsoft spent billions of dollars every year developing Windows systems, and thousands of engineers collaborated on development. However, there were still many bugs, and version releases were often delayed. If programs are generated and run using large models without code in the future, the number of development programmers can be greatly reduced. If you want to meet deadlines, all you need to do is add more electricity and computing power, which was not possible in the past. Smart manufacturing is also a direction. We know that there are many lighthouse factories or black light factories. Together with our partners, we also cooperate with some large factories. In the future, in terms of agile manufacturing, we can quickly switch product types, and even produce completely personalized products according to individual wishes. Products of.
As an academic school, we also pay special attention to AI for Science. We know that the epidemic in the past few years has had a lot of impact on everyone. For thousands of years, the medical industry has always found diseases first and then medicines. So, can we use evolutionary theory and molecular dynamics to see how viruses develop and evolve? If we find the rules, can we do it first? Medicines for the next 5 to 10 years. This will be very interesting to explore. AI makes this possible!
Zhang Yu, founding partner of Qingzhi Capital
Chen Zibing: We have seen different new opportunities and touchpoints in the commercialization of generative AI. As we just talked about, our guests all come from different industries, so going deep into the industry Inside, how does AI integrate with these industries?First of all, I would like to ask Mr. Fang Han. Kunlun Wanwei is building its own large-scale model and has been deeply involved in AI for many years. Can you share with us the specific aspects of our layout or which industries we are more optimistic about?
Fang Han: We will think about and judge this matter from the beginning of 2022. This AI wave should not be smaller than the Internet wave in 2000 and 2010. The relatively large companies that have grown up in this wave are, firstly, C-side companies, and secondly, they must be free models. The giant industries that have exploded must also be C-side companies. Since the C-side company's growth model is determined, it will most likely exist not in a subscription model, but in a free model. Third, we believe that it is most likely a UGC platform. The short video track gives us an inspiration. We believe that AIGC will make major breakthroughs in these directions.
All current AIGC technologies, whether they are Vincent pictures, chats, Vincent videos, Vincent 3D, all current technologies are materials. What is materials? What can be used after processing becomes content. Now most AIGC products are materials, not content. What real users can consume is content. At this time, there is a lot of room.
The threshold for large models is very high now. It requires Prompt Engineering capabilities, which makes it difficult for ordinary users to get started quickly. Therefore, we make sure to hide all technical details from the user. We technicians put all our engineering and algorithmic capabilities into it, ultimately hiding technical details and providing technical conveniences to users. We are basically working in this direction. We focus on AI search in China. We believe that AI search can compress the traditional search engine user experience from 5-10 minutes to 5-10 seconds.
Our overseas layout is mainly in AI music, such as full-chain solutions from automatic composition, arrangement, performance, singing, synthesis, etc., to AI games, AI comics, AI social networking and other fields.
This is basically our thinking.
Chen Zibing: Thank you Mr. Fang Han. Mr. Fang Han just mentioned to us many new opportunities to combine this wave of generative AI with some online products. We also know that Shenzhen Xiang Intelligence has been deeply involved in the offline retail industry for a long time, and offline supermarkets and department stores are an industry with very complex scenarios, a lot of data, and are changing all the time. So I would also like to discuss with Dr. Jue Ao, what new ideas will we have on the combination of offline business and this wave of new technologies?
Chen Zibing, founding partner of Qianyue Capital
Juao: From the very beginning, we have been taking implementation and large-scale application as the main driving force for , doing business intelligence and machine vision, targeting people and goods, and implementing it bit by bit. Plug-ins are built one by one, eventually forming our business intelligence operating system (MOS). In addition to the content generation and intelligent shopping guide mentioned above, we have also implemented the following three business intelligence plug-ins:
One is intelligent supply chain management and intelligent product selection. Intime Business can currently use AI technology to predict sales trends and popular products. and inventory needs, thereby optimizing product inventory and logistics distribution, which can help supermarkets and department stores reduce inventory costs and operating costs and improve profitability; AI can also conduct in-depth analysis and mining of operational data of supermarkets and supermarkets to discover consumer shopping habits Preferences and behavioral habits to select products and increase sales.
The second one is smart security and smart theft prevention. As shopping malls, supermarkets and department stores are crowded places, security is very important. Using AI technology, 24-hour intelligent value monitoring and timely warnings are achieved; in addition, in response to the problem of a significant increase in theft rates caused by the promotion of self-service cash registers in supermarkets, we provide an AI solution for full-site theft prevention.
The third one is smart experience. We use AI technology, including AR/VR technology, to realize virtual fitting, virtual makeup trying and other functions to improve consumers' shopping pleasure and satisfaction.
There are many challenges and difficulties in doing business intelligence for commercial entities. First, doing business intelligence requires four steps, namely standardization, online, and digitization. The last step is intelligence.Physical commerce is a very traditional industry. Many companies have not taken the first three steps well, and have not even completed standardization and onlineization, let alone the last step of intelligence. For example, when a large supermarket we served implemented a site-wide anti-theft solution, we found that its standardization and onlineization were very poor. We spent a whole month solving its networking problems, including replacing old ones. Old switches and routers etc.
Secondly, to do business intelligence, there must be corresponding organizational construction to ensure the implementation of digital intelligence. In particular, the No. 1 position must have vision and courage and be led by the No. 1 position. For example: when building an intelligent passenger flow system for a shopping mall, if the passenger flow at each entrance, exit, floor, and shop is accurate, it will not directly bring revenue to the revenue. However, the accuracy of the passenger flow is the most basic for the shopping mall, and it cannot be improved at this time. The passenger flow system depends on the decisiveness of position one.
We at Intime Business are at the forefront of the industry in terms of digital intelligence and are relatively mature. We have now formed "three growth curves." The first growth curve is to do a good job in the operation of the mall and double the square meter efficiency; the second growth curve is to use our capabilities to achieve online and offline integration. More than 60 stores have one tray, allowing service time and service space. can be extended and expanded, and the specific forms include Miaojie APP and cloud stores; the third growth curve is to export our polished "business intelligence" and "machine vision" products to help peers complete digital transformation and upgrading, and increase sales. scale.
Chen Zibing: Thanks to Dr. Jue Ao for sharing. Shenxiang Intelligence does have a lot of knowhow and experience in the entire retail industry. So we also want to look at another specific industry, which is the education industry. The education industry has always been a promising area for artificial intelligence. We also know that Squirrel AI Learning has been deeply involved in the integration of AI and education for many years and has experienced ups and downs in the industry. So I would also like to discuss with you, this wave of large model technology, compared with the previous AI technologies, will there be any new product integration points, or will there be some changes?
Liang Jing: Thank you host. If the other party knows Squirrel AI, they know that what we have been doing since 2014 is AI adaptive algorithm technology and engine. After combining the large model, we are actually different from other education technology companies in that we combine the AI adaptive algorithm with the large model. We know that on the market, whether domestic or foreign, for example, Khan Academy has a new product called Khanmigo, which is a collaboration with Open AI. It is a brand new large-scale educational product, as well as Duolingo. Launched new subscription app Duolingo Max after integrating with GPT-4. We believe that the real two-way personalized learning for thousands of people must be based on an adaptive engine platform plus large models. There are several core differences and values here. Let us take an example. If a child has been studying in this system for maybe three years, then he has generated a large amount of learning behavior data. This system has already I understand him very well. After adding large models to our system, everyone knows its advantages. This system with a large model and an adaptive engine is an AI virtual teacher. Based on his past learning content, basic knowledge, behavior, and personality characteristics, the information he will receive will be more accurate. If we compare it to a brand new, large-model product, then if the student starts learning, it will be equivalent to zero foundation. This large-model product requires a new understanding of the student's learning situation. So this is a relatively simple way to do a comparison. In addition, if you only retrieve some data information from the large model base, our Mr. Fang just mentioned some hallucinations or some inaccurate information. If it is used in the social entertainment industry, it is completely ok, but for education, At least what we have to instill is that you have to give people the correct answer, so it is not necessarily a correct answer. There can be multiple correct answers, but this answer is correct.However, if we do not filter, screen, or evaluate, but directly retrieve it from the large model, the problem-solving ideas may be completely incorrect. Therefore, in our entire integrated engine platform, we have Lots of agents. Different agents are agents, which have different functions. They perform different functions to evaluate and analyze some data retrieved from the large model, whether it is evaluation or push recommendations, etc., and then register it to our The core platform of the real engine.
So in our opinion, grafting large model technology onto our current AI adaptive platform can be based on our algorithm accumulation over the past 9 years. Big data already has more than 10 billion learning behavior data, plus The technology, algorithms and content of the large model can provide students with a wider range of information and at the same time be more accurate, allowing children to know the most suitable content to learn at the moment without having to start from scratch. This is also the obvious small improvement after we grafted onto the large model. Results. We should organize an educational AI large model discussion forum in January and release our product technology at the same time.
Chen Zibing: Thanks to Mr. Liang Jing for sharing. I would like to ask Mr. Tianyu, from an investment perspective, after the big model comes out, does Jinqiu Fund have any new views on the industrial landscape of AI+ in different industries?
Zang Tianyu: When used discriminative AI technology in the past, technology and data were tightly coupled, and the threshold was relatively high. The application side needed to have the capabilities of algorithm architecture, engineering platform, and data pipeline. When investing from 2019 to 2021, Applications are mainly concentrated in wealthy industries or companies such as energy, finance, and advanced manufacturing. In this wave, because of the relatively powerful capabilities of the large base model, the application layer and the technical layer have decoupled and divided labor to a certain extent. Some companies provide basic base models, and some companies make fine-tuning and applications on this basis. Overall The threshold for the application of AI technology has been greatly lowered. We have also seen that more and more subdivided industries and scenarios are beginning to embrace AI. From the perspective of industry application, the territory has been greatly expanded. There are also some subdivided scenarios where the most advanced AI technology can be used, and many 2C AI native applications have been developed. Secondly, an independent base model layer was produced. As a very important position in the industrial chain, it also occupies a relatively large value. This is something we have not seen in the past.
Our focus is on the application layer, whether it is 2C or 2B, general or industry vertical. On the other hand, from the perspective of technological innovation, we will pay more attention to multi-modal models and embodied intelligence.
Zang Tianyu, Executive Director of Jinqiu Fund
Chen Zibing: Do you think that for an AI company, it is more important to run the business model and make money first?
Zang Tianyu: They are complementary to each other. If the business model can run smoothly, we will definitely make money.
Chen Zibing: just mentioned that Qingzhi Capital does both investment and early-stage incubation, and has accumulated a lot of experience in helping start-up companies from 0 to 1. From your perspective, what kind of support is needed for this wave of AI companies to start their own businesses, and what role can the government play in it?
Zhang Yu: The current business environment of is different from before. The differentiation is very obvious. In the past, there were usually more grassroots entrepreneurship, but now there are also grassroots entrepreneurship. But the result of the differentiation is that the number of startups that can obtain financing has changed compared with before. If there is less, each project that receives financing will get more money. Therefore, if entrepreneurs want to survive, they must grab the top position in the initial stage. Generally speaking, the current entrepreneurial environment is more demanding for entrepreneurs. Our Qingzhi Incubator focuses on the AI field and strives to provide better conditions for entrepreneurship in the AI field.
The government has a very strong guiding role in entrepreneurial companies. When starting a business in the AI field, especially in the field of large models, algorithms, computing power, data, and applications are all elements of development. The algorithm mainly relies on the company itself, but algorithm improvement is also related to data. Of course, many large models are now open source. You can use good open source models to quickly penetrate into practical applications to generate value.Of course, expert guidance is also important. Tsinghua Intelligent Industry Research Institute (AIR) has many top AI scientists. Many of the companies we invest and incubate have already communicated and cooperated with them, which is something other venture capital and incubators do not have. ; Computing power is currently relatively expensive. It would be a good solution if the government could build a large computing power center and provide it to enterprises for use. Our incubator also provides some free computing power to the startups in our incubator; data is more difficult, and we are still exploring how to protect data and use it reasonably. Our institute has experts who specialize in privacy computing and can use AI technology to remove data sensitivity and provide it to startups.
From a government level, supporting industry is a very important function of the government. The government has a lot of resources to support the industry. I think the most important thing is of course policies, application scenarios, data, and financial support. Our capital scale is pretty good at the moment. LPs are mainly VC and industrial capital. We don’t need government money for the time being. When we grow in scale in the future, we may get government support to some extent. Therefore, the government’s support for entrepreneurship and investment is the whole economy. An indispensable part of the system.
Chen Zibing: Thanks to all the guests for sharing. Finally, I hope to look forward to this industry with everyone. Everyone says that this year is the beginning of the golden decade of AGI. Standing at such a beginning, I hope you can make a simple prediction in one sentence: What are you most looking forward to in this industry in the next cycle of the next 3-5 years? Or going back to today’s theme of AI commercialization, how far do you think AI commercialization will go in the next five years?
Fang Han: What I look forward to most about is that I think technology is far ahead of products. In the AIGC or AGI market, the cost of technical personnel is currently the highest, and product students are not ready yet. I hope that after product students understand and understand the technical boundaries of large models, they can use their creativity to create a new disruptive business model. and product models. This is what Chinese product managers and Chinese entrepreneurs are better at. I firmly believe that the next generation of Byte and Alibaba will be founded and built by grassroots entrepreneurs. I also firmly believe that disruptive entrepreneurial models and product models are what we most want to see. .
Jue Ao: In the next 3-5 years, AI business implementation will definitely enter the stage of comprehensive popularization and in-depth application. Especially with the large-scale application of large models or general intelligence, it will definitely bring about profound changes in all walks of life, including Create new value and create new business models. As the guest just said, new giant unicorn AI companies will emerge in the next 3-5 years.
Liang Jing: From the perspective of the commercial value of AI, AI will eliminate 90% of white-collar jobs in the future. Planters and harvesters have already solved the basic work of farmers while improving work efficiency. From the perspective of human social behavior, in the future AI can reduce 70% of the time we spend looking at mobile phones or computers. We will have various voice and visual assistants to help us book cars, meals, tickets, taxis, and help us I write some report summary content. In this case, I think it will bring about another phenomenon. We can have more time to interact with others, go outdoors, and improve ourselves and our health. I hope AI will give What we bring does not solidify us, but liberates us, allowing us to have a higher quality of life and make us healthier.
Zang Tianyu: We feel that this year is the first year of large models. Everyone has seen changes in the paradigm. The capabilities of the base model are improving, and the context window is still lengthening. Where is the model boundary and how should AI native products be designed? How to build a data flywheel, I think, still requires a process of exploration. In the next 3-5 years, a mature AI product methodology will be born, and more applications with commercial value will be created, truly realizing hundreds of billions or even trillions of application market opportunities.
For Jinqiu Fund, we will pay great attention to two directions, one is multi-modal models, and the other is embodied intelligence. I think there will be very big changes in related underlying technologies in the next 3-5 years. Let us We are gradually seeing some convergence on the technical path, including the inference and application of large models on the device, which is a direction we are very concerned about.
Zhang Yu: This year for should be called the first year of generative AI. Of course, it is also the first year of large models. Especially in the second half of the year, we will start to see many AIGC entrepreneurial projects. I believe that the applications brought by AIGC will emerge in the next few years. From a business logic point of view, generative AI, including how large models can be closer to application scenarios, is very important. The application of large model generation is integrated with various industries, first of all with application scenarios, which is the main problem faced by large models in the near future. In the long run, large models will develop in two directions: First, autonomous intelligence. Through self-improvement and upgrading, large models will slowly grow up like children. This is still being improved, and we also see some prospects. Sexual research is developing in this direction; the second is the interaction and dialogue between models. Communication between people enables the rapid flow of knowledge among people, and the result is an improvement in the knowledge of all mankind. When models can communicate and dialogue independently, it is the rise of so-called autonomous agents, and then artificial intelligence will enter entered a new stage of development. Of course, this matter is a bit troublesome for human beings. Human beings may be threatened, but from the perspective of development, it may be the future. My summary is:
First, in the next 5-10 years, the digital economy with artificial intelligence as the core will become the mainstream economic form;
Second, AI can subvert all industries, and we fully believe that AI will subvert everything. Industry;
Third, the use of AI will become the basic competitiveness of enterprises;
Fourth, when AI develops to a certain extent, humans will face many deep-seated challenges. These challenges may be unknown to humans and are also worrying for everyone. Everyone needs to pay full attention to this issue.
Chen Zibing: Thank you all guests for sharing. I also look forward to seeing more product and commercialization innovations under this turbulent wave of new technologies, so that technology can exert greater social value and possibility. Thank you all. !
First of all, I would like to ask Mr. Fang Han. Kunlun Wanwei is building its own large-scale model and has been deeply involved in AI for many years. Can you share with us the specific aspects of our layout or which industries we are more optimistic about?Fang Han: We will think about and judge this matter from the beginning of 2022. This AI wave should not be smaller than the Internet wave in 2000 and 2010. The relatively large companies that have grown up in this wave are, firstly, C-side companies, and secondly, they must be free models. The giant industries that have exploded must also be C-side companies. Since the C-side company's growth model is determined, it will most likely exist not in a subscription model, but in a free model. Third, we believe that it is most likely a UGC platform. The short video track gives us an inspiration. We believe that AIGC will make major breakthroughs in these directions.
All current AIGC technologies, whether they are Vincent pictures, chats, Vincent videos, Vincent 3D, all current technologies are materials. What is materials? What can be used after processing becomes content. Now most AIGC products are materials, not content. What real users can consume is content. At this time, there is a lot of room.
The threshold for large models is very high now. It requires Prompt Engineering capabilities, which makes it difficult for ordinary users to get started quickly. Therefore, we make sure to hide all technical details from the user. We technicians put all our engineering and algorithmic capabilities into it, ultimately hiding technical details and providing technical conveniences to users. We are basically working in this direction. We focus on AI search in China. We believe that AI search can compress the traditional search engine user experience from 5-10 minutes to 5-10 seconds.
Our overseas layout is mainly in AI music, such as full-chain solutions from automatic composition, arrangement, performance, singing, synthesis, etc., to AI games, AI comics, AI social networking and other fields.
This is basically our thinking.
Chen Zibing: Thank you Mr. Fang Han. Mr. Fang Han just mentioned to us many new opportunities to combine this wave of generative AI with some online products. We also know that Shenzhen Xiang Intelligence has been deeply involved in the offline retail industry for a long time, and offline supermarkets and department stores are an industry with very complex scenarios, a lot of data, and are changing all the time. So I would also like to discuss with Dr. Jue Ao, what new ideas will we have on the combination of offline business and this wave of new technologies?
Chen Zibing, founding partner of Qianyue Capital
Juao: From the very beginning, we have been taking implementation and large-scale application as the main driving force for , doing business intelligence and machine vision, targeting people and goods, and implementing it bit by bit. Plug-ins are built one by one, eventually forming our business intelligence operating system (MOS). In addition to the content generation and intelligent shopping guide mentioned above, we have also implemented the following three business intelligence plug-ins:
One is intelligent supply chain management and intelligent product selection. Intime Business can currently use AI technology to predict sales trends and popular products. and inventory needs, thereby optimizing product inventory and logistics distribution, which can help supermarkets and department stores reduce inventory costs and operating costs and improve profitability; AI can also conduct in-depth analysis and mining of operational data of supermarkets and supermarkets to discover consumer shopping habits Preferences and behavioral habits to select products and increase sales.
The second one is smart security and smart theft prevention. As shopping malls, supermarkets and department stores are crowded places, security is very important. Using AI technology, 24-hour intelligent value monitoring and timely warnings are achieved; in addition, in response to the problem of a significant increase in theft rates caused by the promotion of self-service cash registers in supermarkets, we provide an AI solution for full-site theft prevention.
The third one is smart experience. We use AI technology, including AR/VR technology, to realize virtual fitting, virtual makeup trying and other functions to improve consumers' shopping pleasure and satisfaction.
There are many challenges and difficulties in doing business intelligence for commercial entities. First, doing business intelligence requires four steps, namely standardization, online, and digitization. The last step is intelligence.Physical commerce is a very traditional industry. Many companies have not taken the first three steps well, and have not even completed standardization and onlineization, let alone the last step of intelligence. For example, when a large supermarket we served implemented a site-wide anti-theft solution, we found that its standardization and onlineization were very poor. We spent a whole month solving its networking problems, including replacing old ones. Old switches and routers etc.
Secondly, to do business intelligence, there must be corresponding organizational construction to ensure the implementation of digital intelligence. In particular, the No. 1 position must have vision and courage and be led by the No. 1 position. For example: when building an intelligent passenger flow system for a shopping mall, if the passenger flow at each entrance, exit, floor, and shop is accurate, it will not directly bring revenue to the revenue. However, the accuracy of the passenger flow is the most basic for the shopping mall, and it cannot be improved at this time. The passenger flow system depends on the decisiveness of position one.
We at Intime Business are at the forefront of the industry in terms of digital intelligence and are relatively mature. We have now formed "three growth curves." The first growth curve is to do a good job in the operation of the mall and double the square meter efficiency; the second growth curve is to use our capabilities to achieve online and offline integration. More than 60 stores have one tray, allowing service time and service space. can be extended and expanded, and the specific forms include Miaojie APP and cloud stores; the third growth curve is to export our polished "business intelligence" and "machine vision" products to help peers complete digital transformation and upgrading, and increase sales. scale.
Chen Zibing: Thanks to Dr. Jue Ao for sharing. Shenxiang Intelligence does have a lot of knowhow and experience in the entire retail industry. So we also want to look at another specific industry, which is the education industry. The education industry has always been a promising area for artificial intelligence. We also know that Squirrel AI Learning has been deeply involved in the integration of AI and education for many years and has experienced ups and downs in the industry. So I would also like to discuss with you, this wave of large model technology, compared with the previous AI technologies, will there be any new product integration points, or will there be some changes?
Liang Jing: Thank you host. If the other party knows Squirrel AI, they know that what we have been doing since 2014 is AI adaptive algorithm technology and engine. After combining the large model, we are actually different from other education technology companies in that we combine the AI adaptive algorithm with the large model. We know that on the market, whether domestic or foreign, for example, Khan Academy has a new product called Khanmigo, which is a collaboration with Open AI. It is a brand new large-scale educational product, as well as Duolingo. Launched new subscription app Duolingo Max after integrating with GPT-4. We believe that the real two-way personalized learning for thousands of people must be based on an adaptive engine platform plus large models. There are several core differences and values here. Let us take an example. If a child has been studying in this system for maybe three years, then he has generated a large amount of learning behavior data. This system has already I understand him very well. After adding large models to our system, everyone knows its advantages. This system with a large model and an adaptive engine is an AI virtual teacher. Based on his past learning content, basic knowledge, behavior, and personality characteristics, the information he will receive will be more accurate. If we compare it to a brand new, large-model product, then if the student starts learning, it will be equivalent to zero foundation. This large-model product requires a new understanding of the student's learning situation. So this is a relatively simple way to do a comparison. In addition, if you only retrieve some data information from the large model base, our Mr. Fang just mentioned some hallucinations or some inaccurate information. If it is used in the social entertainment industry, it is completely ok, but for education, At least what we have to instill is that you have to give people the correct answer, so it is not necessarily a correct answer. There can be multiple correct answers, but this answer is correct.However, if we do not filter, screen, or evaluate, but directly retrieve it from the large model, the problem-solving ideas may be completely incorrect. Therefore, in our entire integrated engine platform, we have Lots of agents. Different agents are agents, which have different functions. They perform different functions to evaluate and analyze some data retrieved from the large model, whether it is evaluation or push recommendations, etc., and then register it to our The core platform of the real engine.
So in our opinion, grafting large model technology onto our current AI adaptive platform can be based on our algorithm accumulation over the past 9 years. Big data already has more than 10 billion learning behavior data, plus The technology, algorithms and content of the large model can provide students with a wider range of information and at the same time be more accurate, allowing children to know the most suitable content to learn at the moment without having to start from scratch. This is also the obvious small improvement after we grafted onto the large model. Results. We should organize an educational AI large model discussion forum in January and release our product technology at the same time.
Chen Zibing: Thanks to Mr. Liang Jing for sharing. I would like to ask Mr. Tianyu, from an investment perspective, after the big model comes out, does Jinqiu Fund have any new views on the industrial landscape of AI+ in different industries?
Zang Tianyu: When used discriminative AI technology in the past, technology and data were tightly coupled, and the threshold was relatively high. The application side needed to have the capabilities of algorithm architecture, engineering platform, and data pipeline. When investing from 2019 to 2021, Applications are mainly concentrated in wealthy industries or companies such as energy, finance, and advanced manufacturing. In this wave, because of the relatively powerful capabilities of the large base model, the application layer and the technical layer have decoupled and divided labor to a certain extent. Some companies provide basic base models, and some companies make fine-tuning and applications on this basis. Overall The threshold for the application of AI technology has been greatly lowered. We have also seen that more and more subdivided industries and scenarios are beginning to embrace AI. From the perspective of industry application, the territory has been greatly expanded. There are also some subdivided scenarios where the most advanced AI technology can be used, and many 2C AI native applications have been developed. Secondly, an independent base model layer was produced. As a very important position in the industrial chain, it also occupies a relatively large value. This is something we have not seen in the past.
Our focus is on the application layer, whether it is 2C or 2B, general or industry vertical. On the other hand, from the perspective of technological innovation, we will pay more attention to multi-modal models and embodied intelligence.
Zang Tianyu, Executive Director of Jinqiu Fund
Chen Zibing: Do you think that for an AI company, it is more important to run the business model and make money first?
Zang Tianyu: They are complementary to each other. If the business model can run smoothly, we will definitely make money.
Chen Zibing: just mentioned that Qingzhi Capital does both investment and early-stage incubation, and has accumulated a lot of experience in helping start-up companies from 0 to 1. From your perspective, what kind of support is needed for this wave of AI companies to start their own businesses, and what role can the government play in it?
Zhang Yu: The current business environment of is different from before. The differentiation is very obvious. In the past, there were usually more grassroots entrepreneurship, but now there are also grassroots entrepreneurship. But the result of the differentiation is that the number of startups that can obtain financing has changed compared with before. If there is less, each project that receives financing will get more money. Therefore, if entrepreneurs want to survive, they must grab the top position in the initial stage. Generally speaking, the current entrepreneurial environment is more demanding for entrepreneurs. Our Qingzhi Incubator focuses on the AI field and strives to provide better conditions for entrepreneurship in the AI field.
The government has a very strong guiding role in entrepreneurial companies. When starting a business in the AI field, especially in the field of large models, algorithms, computing power, data, and applications are all elements of development. The algorithm mainly relies on the company itself, but algorithm improvement is also related to data. Of course, many large models are now open source. You can use good open source models to quickly penetrate into practical applications to generate value.Of course, expert guidance is also important. Tsinghua Intelligent Industry Research Institute (AIR) has many top AI scientists. Many of the companies we invest and incubate have already communicated and cooperated with them, which is something other venture capital and incubators do not have. ; Computing power is currently relatively expensive. It would be a good solution if the government could build a large computing power center and provide it to enterprises for use. Our incubator also provides some free computing power to the startups in our incubator; data is more difficult, and we are still exploring how to protect data and use it reasonably. Our institute has experts who specialize in privacy computing and can use AI technology to remove data sensitivity and provide it to startups.
From a government level, supporting industry is a very important function of the government. The government has a lot of resources to support the industry. I think the most important thing is of course policies, application scenarios, data, and financial support. Our capital scale is pretty good at the moment. LPs are mainly VC and industrial capital. We don’t need government money for the time being. When we grow in scale in the future, we may get government support to some extent. Therefore, the government’s support for entrepreneurship and investment is the whole economy. An indispensable part of the system.
Chen Zibing: Thanks to all the guests for sharing. Finally, I hope to look forward to this industry with everyone. Everyone says that this year is the beginning of the golden decade of AGI. Standing at such a beginning, I hope you can make a simple prediction in one sentence: What are you most looking forward to in this industry in the next cycle of the next 3-5 years? Or going back to today’s theme of AI commercialization, how far do you think AI commercialization will go in the next five years?
Fang Han: What I look forward to most about is that I think technology is far ahead of products. In the AIGC or AGI market, the cost of technical personnel is currently the highest, and product students are not ready yet. I hope that after product students understand and understand the technical boundaries of large models, they can use their creativity to create a new disruptive business model. and product models. This is what Chinese product managers and Chinese entrepreneurs are better at. I firmly believe that the next generation of Byte and Alibaba will be founded and built by grassroots entrepreneurs. I also firmly believe that disruptive entrepreneurial models and product models are what we most want to see. .
Jue Ao: In the next 3-5 years, AI business implementation will definitely enter the stage of comprehensive popularization and in-depth application. Especially with the large-scale application of large models or general intelligence, it will definitely bring about profound changes in all walks of life, including Create new value and create new business models. As the guest just said, new giant unicorn AI companies will emerge in the next 3-5 years.
Liang Jing: From the perspective of the commercial value of AI, AI will eliminate 90% of white-collar jobs in the future. Planters and harvesters have already solved the basic work of farmers while improving work efficiency. From the perspective of human social behavior, in the future AI can reduce 70% of the time we spend looking at mobile phones or computers. We will have various voice and visual assistants to help us book cars, meals, tickets, taxis, and help us I write some report summary content. In this case, I think it will bring about another phenomenon. We can have more time to interact with others, go outdoors, and improve ourselves and our health. I hope AI will give What we bring does not solidify us, but liberates us, allowing us to have a higher quality of life and make us healthier.
Zang Tianyu: We feel that this year is the first year of large models. Everyone has seen changes in the paradigm. The capabilities of the base model are improving, and the context window is still lengthening. Where is the model boundary and how should AI native products be designed? How to build a data flywheel, I think, still requires a process of exploration. In the next 3-5 years, a mature AI product methodology will be born, and more applications with commercial value will be created, truly realizing hundreds of billions or even trillions of application market opportunities.
For Jinqiu Fund, we will pay great attention to two directions, one is multi-modal models, and the other is embodied intelligence. I think there will be very big changes in related underlying technologies in the next 3-5 years. Let us We are gradually seeing some convergence on the technical path, including the inference and application of large models on the device, which is a direction we are very concerned about.
Zhang Yu: This year for should be called the first year of generative AI. Of course, it is also the first year of large models. Especially in the second half of the year, we will start to see many AIGC entrepreneurial projects. I believe that the applications brought by AIGC will emerge in the next few years. From a business logic point of view, generative AI, including how large models can be closer to application scenarios, is very important. The application of large model generation is integrated with various industries, first of all with application scenarios, which is the main problem faced by large models in the near future. In the long run, large models will develop in two directions: First, autonomous intelligence. Through self-improvement and upgrading, large models will slowly grow up like children. This is still being improved, and we also see some prospects. Sexual research is developing in this direction; the second is the interaction and dialogue between models. Communication between people enables the rapid flow of knowledge among people, and the result is an improvement in the knowledge of all mankind. When models can communicate and dialogue independently, it is the rise of so-called autonomous agents, and then artificial intelligence will enter entered a new stage of development. Of course, this matter is a bit troublesome for human beings. Human beings may be threatened, but from the perspective of development, it may be the future. My summary is:
First, in the next 5-10 years, the digital economy with artificial intelligence as the core will become the mainstream economic form;
Second, AI can subvert all industries, and we fully believe that AI will subvert everything. Industry;
Third, the use of AI will become the basic competitiveness of enterprises;
Fourth, when AI develops to a certain extent, humans will face many deep-seated challenges. These challenges may be unknown to humans and are also worrying for everyone. Everyone needs to pay full attention to this issue.
Chen Zibing: Thank you all guests for sharing. I also look forward to seeing more product and commercialization innovations under this turbulent wave of new technologies, so that technology can exert greater social value and possibility. Thank you all. !
