Zhou Hongyi: The large model is fun and not easy to use, and it can be used as a poem, but not as a

Mondo Technology Updated on 2024-03-01

**: CBN.

In response to the previous chaos of AI course fees, Zhou Hongyi, chairman of 360 Group, has publicly stated that there is a great demand for AI science popularization, so he promised to launch free AI training courses to promote the popularization of AI science popularization. On the afternoon of February 29, Zhou Hongyi's first AI free class was broadcast, which lasted nearly 3 and a half hours, and its **live broadcast interface showed that there were more than 400,000 people**.

During the period, Zhou Hongyi talked about the direction of the large model in 2024, and a very important task is to "put wheels" on the large model and turn it from a "toy" into a tool; He also gave advice to AI entrepreneurs, saying that they should not touch the general model, and do not make some simple shell applications on top of the API (application programming interface) of the general model.

A small episode at the scene was that during the live broadcast demonstration of 360 AI search products, the product interface could not output content due to lag for many times, and the live broadcast was at an impasse for a while. Zhou Hongyi teased, "After making a large model, I understand more and more that all companies only dare to record the screen and demonstrate, but do not dare to actually demonstrate, because the results are unpredictable." At the end of the live broadcast, he explained that the demo was accidentally leaked**, and many users crashed the server the moment they entered.

In 2023, after talking about the large model for a year, Zhou Hongyi said that everyone may have misunderstandings and illusions, "The large model companies have boasted very well, they are all very fun, but they are not easy to use, you can compose poems but not do things, it is a toy, not a tool." He believes that in 2024, whether it is an entrepreneur or an Internet company, no longer show off the large model, "The large model is an engine, and it must be equipped with wheels, turned into a tractor, a sports car, and a car to use." ”

In 2024, many companies, led by 360, will come out with a series of tools, which are combined with scenario deepening, and I think there will be more and more such tools. Zhou Hongyi said.

Regarding the domestic large model, Zhou Hongyi mentioned that many people are very pessimistic because of SORA, but they should not be too pessimistic. "After ChatGPT came out, it took more than a year in China, although it has not caught up with GPT-4, but it has reached GPT-35 capabilities I think is more than enough. He said that in addition to China and the United States, the largest number of artificial intelligence companies in the world are in China.

I think Sora has given us an inspiration, OpenAI is now making super general large models, with more than one trillion parameters, and has very strong multimodal capabilities. So my suggestion is not to '100 model war', many small and medium-sized companies do not need to participate, this is indeed a long-distance race to compete for strength and financial resources. ”

Zhou Hongyi believes that the general general model is meaningless now, because it is open source, there is not much to compete for, he suggested that China should take the road of development of large models with its own characteristics, and he gave some suggestions in the live broadcast.

On the one hand, China can learn from the idea of open source, unite some giant companies, fully share and share technology, communicate with each other, give full play to the new first-class system, and work together to make 1-2 general super models, which may have a lot of application space in the future in the competition of great powers.

On the other hand, Zhou Hongyi believes that the large model should develop its strengths and avoid its weaknesses to find its own advantages, "the country is now talking about intelligent transformation and digital transformation, many first-class departments and enterprises have big data, how to further improve the big data into the large model, and use the large model to truly realize the transfer of the entire industrial Internet to the intelligent Internet and industrial intelligence, I think 2024 should be a focus of many start-ups, and it will bring many benefits to the verticalization and scenario-based transformation of the big **." ”

Zhou Hongyi said that there is no need for hundreds of billions and trillions of models, and only tens of billions of models are needed within the enterprise to solve a more important internal application scenario problem, he said, for example, "Just like we don't need to hire an Einstein in every company, we hire a college undergraduate, in terms of taxation, finance, and HR, special training for two years, because the blessing of internal knowledge can grow into a talent in a vertical field." "The most important thing in this field is scenario selection, enterprise knowledge accumulation and business integration.

On the basis of the above directions, Zhou Hongyi also put forward his own suggestions for entrepreneurs, "Don't touch the general model again, it makes no sense." ”

On the one hand, Zhou Hongyi believes that the direction can be found at the enterprise level, because the enterprise market is fragmented and is unlikely to be monopolized. In addition, it is to find the scene to c, "You must make the scene very heavy and vertical, and the shallow scene is worthless, because other large companies will fail if they add a function to the browser a little." ”

Zhou Hongyi said that SORA can only be used to generate materials, but if you really use it to make a documentary, there is a lot of engineering work in it, such as dubbing, editing, modification, etc., so the heavier you do in these aspects, the larger the model is just an engine for function, "There is no way to compete with it by adding a fan blade to the engine, so the scene should be heavy and vertical." ”

At the end of the speech, Zhou Hongyi said that there were problems with today's demonstration, but the team still worked hard and wanted to tell fans that the team had not relaxed on the large model. "This time we made an innovation, and it doesn't have to be successful, because it needs to be verified by users, but I think life lies in sports, and companies must keep tossing and turning, keep trying, and suddenly find a perfect marketing fit (the best fit between the product and the market), and then you will have the opportunity to stand out. ”

Zhou Hongyi admitted frankly that he is now working hard to learn and research artificial intelligence, and many people blame him, but he does not feel that he is a failure, "I still think that I have been working hard, then I can drive the company's business forward." ”

Related Pages