360 Zhou Hongyi: The development of AI is an inevitable trend, and we must fully master the core technology of large models

On May 31, 360 Smart Life officially launched the 360 ​​Intelligent Brain Vision large model and a variety of new AI hardware products, and announced that 360 Smart Life has officially entered the SMB market. After the meeting, Zhou Hongyi, founder of 360 Group, accepted interviews from the media on some hot topics related to large models in recent days.

Zhou Hongyi believes that the hallucination problem is currently the biggest bottleneck for large-scale models, but this problem is both its flaw and its feature. "There is an essential difference between large models and search. Search is simply copying knowledge. Large models try to understand all knowledge and "swallow" them, but this may cause details to be ignored."

He explained that the current large models can be used for some entertainment applications. For example, it can be used to write stories similar to "Monkey King vs. Ultraman". At this stage, adapting large-scale models to professional fields such as law, education, and healthcare is not feasible. Our goal is to transform the search engine into a knowledge base to solve the problem of knowledge ambiguity. When it comes to questions about knowledge accuracy, we can verify and revise answers through searches. ”

Zhou Hongyi believes that the visual large model released last night can be regarded as a large model vertical to the category. "To put it simply, the current large visual model is based on the large language model, from the initial understanding of language to the deeper interpretation of 'language images'."

At the same time, he also admitted that compared with other large models, the difference between 360 Intelligent Brain and other models is not that big. "360 Intelligent Brain has two characteristics. The first is the training data. 360 has screened out a lot of high-quality data to support the large model; the second is that after we change the search interface, we can achieve knowledge and timeliness through search. Enhancement, so that the most common hallucination problem of large models can be avoided.”

Regarding the real usage scenarios of the visual large model, Zhou Hongyi said that in the future, the visual large model can be combined with cameras to be applied to the fields of car navigation and security. In the field of security, he gave an example. For example, if a child stands on a high cabinet, the potential danger can be read through a large visual model and an alarm can be issued.

In the field of car navigation, through large visual models, potential dangers can be discovered during the car navigation process, and alarm processing can be performed while retaining the video, thereby reducing the probability of the driver being in danger. "For example, in a security incident that occurred on the streets of Shanghai two days ago, if the car behind had the support of a large visual model, it could identify the abnormality of the car in front. Then, it could automatically save the video and upload the alarm at the same time deal with."

At the end of the interview, Zhou Hongyi made some of his own opinions on some security issues related to the AI ​​era caused by "AI face-changing" some time ago. He emphasized that AI security issues should be taken seriously, and 360 has established an internal AI security team to focus on solving security issues in this field. He added: "AI professionals will need to conduct secondary verification in AI-generated works in the future, such as adding fingerprint or voiceprint recognition and other measures." ”

"AI is an industrial revolution, and its development is an inevitable trend. We cannot stop eating because it has some security problems. What 360 needs to do now is to minimize this security problem. It must not only solve network security , we must also solve data security and artificial intelligence security. At the same time, 360 must also maximize the research and development of large models. Because large models are the pinnacle of digitalization and an industrial revolution-level change, who does not master the core technology of large models? , who do not have practical application scenarios for large models, and thus are eliminated by the industry. Therefore, 360 must not only set up a dedicated research team to conduct research, but also find a better and safer solution through continuous attempts.” Zhou Hongyi said.

The above is the detailed content of 360 Zhou Hongyi: The development of AI is an inevitable trend, and we must fully master the core technology of large models. For more information, please follow other related articles on the PHP Chinese website!