news

author: an ran

editor: ye jinyan

produced by: deepnet·tencent news xiaoman studio

"we cannot blindly follow the super-large model of openai. if we want to launch a new industrial revolution in china and reshape each of our companies, we must pull it (openai) off the altar and then turn it from an atomic bomb into a tea egg. because for small and medium-sized enterprises, it is impossible for them to buy a lot of computing power and spend money on artificial intelligence." zhou hongyi, founder of 360, said at the 2024 yabuli china entrepreneurs forum.

in an interview with yu minhong, founder of new oriental and ceo of oriental selection, zhou hongyi made three judgments on china's artificial intelligence:

first, companies should not expect to use one ai to solve all problems, but should combine the company's internal business processes, find a professional scenario, use different large models to solve different professional problems, and then combine multiple large models to solve problems in the future. due to its lower cost, this methodology has now been widely accepted.

"for example, it is similar to a company recruiting employees. the company is not looking for a super versatile doctor, but hopes that each employee has specialties to form a team to work together," said zhou hongyi.

second, zhou hongyi does not think that generative ai is an operating system or a product. "a big model is not a product. no matter how powerful a big model is, it must be combined with a scenario to create a user experience. so i took the lead in using ai big models to transform search. internet giants will use ai to reshape existing products. for example, apple builds ai phones, and microsoft builds ai computers, both of which are empowering existing scenarios with big models."

third, the gap between china’s large-scale models and those of the united states has been shortened from two years to about half a year. some of the large-scale model functions announced by the united states half a year ago have basically been built by chinese internet companies half a year later.

zhou hongyi believes that china's computing power chips are still behind those of the united states, so he recently came up with an idea for an "avengers" organization to promote 16 well-known large models in china to cooperate according to the coe architecture and integrate them into all 360 browsers. users can enable different models for different problems, and they don't have to jump back and forth between different models. "it's like we found 16 individual champions, and their combined comprehensive capabilities should be able to surpass gpt-4." zhou hongyi said