news

Tencent Technology Author Su Yang

Editor Guo Xiaojing

"Everyone sent me WeChat messages to ask if my arm was broken. Luckily, it was nothing serious."

Zhou Hongyi responded to the experience of having his hand caught in the door of a GAC ​​car at the 12th ISC Internet Security Conference held today.

"People who work in security are very persistent. They think about security issues first when encountering anything, and they all want to conduct experiments themselves," Zhou Hongyi said. "In the future, when you conduct experiments on physical security issues, you should not use your hands to conduct experiments. Ham and cucumber may be more suitable."

Since he does not have a driver's license, Zhou Hongyi has been experiencing the various intelligent and safety capabilities of China's new cars from the perspective of a passenger in the past few months, and set up a "798 Small Car Show" downstairs of the company. This is also considered by the outside world to be a cross-border form to attract traffic for his personal IP and the Nezha Auto he invested in.

Regarding the Internet Security Conference, which has been held for 12 sessions, Zhou Hongyi said that when we founded it, we did not want this to be a closed-door meeting of 360, but wanted to make it a weather vane for the security industry, so this year the theme of the conference was upgraded to ISC.AI.

"AI will bring many security issues, but it can also become a means and technology to solve security problems," Zhou Hongyi said.

In the past two years, withChatGPTWith the emergence of security research, security research has become increasingly popular, and the number of professional papers published has increased rapidly.

Wu Shizhong, an academician of the Chinese Academy of Engineering, said, "Through keyword analysis of more than 2,500 security-related papers on the Archive official website, nearly 40% involved large model attacks and 30% involved large model security."

"In terms of the distribution of first authors of papers published in the past two years, China, the United States, Singapore, Germany and Australia are in the top five," said Wu Shizhong, "China and the United States are far ahead and form the first echelon."

But he also stressed that judging from the citations of papers, the United States still ranks first in scientific and technological influence. "The overall influence of our country's scientific and technological papers is still relatively weak, indicating that the quality needs to be improved."

Academician Wu Shizhong also revealed that in the past two years, includingOpenAISilicon Valley giants and startups including , Meta, and Antropic, as well as Chinese technology companies including Baidu and Huawei, are increasing their capital expenditures on security.

"(We) have built 200 data centers across the country," Zhou Hongyi said. 360 currently has 10,000 Petaflops of computing power in reserve, and the total number of security samples has reached 32 billion PB. "The annual maintenance cost exceeds 500 million yuan."

Zhou Hongyi said that Nvidia's market value becoming the world's first marks the true arrival of the era of artificial intelligence.

This round of artificial intelligence wave began in November 2022, with the birth of ChatGPT and the successive appearance of large language models based on the transformer architecture. The outside world vividly summarized it as the "Battle of a Hundred Models". Some trillion-parameter models began to appear. The industry's demand for computing power also pushed NVIDIA to the throne of the company with the highest market value in the world, but the industry's development direction and style are also constantly changing.

2023 is the year of models, and 2024 is the year of applications. Zhou Hongyi said, "Even Baidu's Director Li said that we should not focus on models anymore, but on scenarios and applications."

Zhou Hongyi said that security is the most appropriate and important application scenario for AI. In his opinion, the security industry has many pain points, including massive data analysis tasks, difficulty in rapid processing and tracing, and insufficient security manpower, all of which bring opportunities for the application of AI security.

"We have always said that cybersecurity is a confrontation between people," he said. "In the future, we must use AI to deal with the security issues of AI upgrades, and use magic to defeat magic."

Zhou Hongyi summarized the difficulties and adjustments in building a large security model into three points: interdisciplinary issues, the knowledge density of the model, and the deep integration with security business.

At the event, Zhou Hongyi also announced that the 360 ​​security big model would be free, saying that he wanted to bring the big model down from the altar so that every company could "afford to use it and use it well."