360 Zhou Hongyi: The development of AI is an inevitable trend, and it is necessary to fully grasp the core technology of large models

Tech 2023-06-01 14:41:24 Source: Network

Regarding the issue of the shortcomings of large models, Zhou Hongyi believes that the biggest weakness of large models at present is the illusion problem, but this is not only its disadvantage but also its characteristic. There is a fundamental difference between a big model and a search, where search simply copies knowledge

Regarding the issue of the shortcomings of large models, Zhou Hongyi believes that the biggest weakness of large models at present is the illusion problem, but this is not only its disadvantage but also its characteristic. There is a fundamental difference between a big model and a search, where search simply copies knowledge. A big model, on the other hand, attempts to understand knowledge and 'eat' it all, resulting in the loss of some details of the knowledge itself

He explained that at present, large models can be used for entertainment applications, such as allowing it to write stories like "Sun Wukong vs. Ultraman" without hesitation. But if big models are applied in vertical fields such as law, education, healthcare, etc., it is currently not feasible. An important tool for solving knowledge ambiguity is search engines, so we are now trying to turn search engines into a knowledge base. When it comes to knowledge accuracy issues, we can verify and correct their answers through search

Regarding the visual model just released last night, Zhou Hongyi stated that the visual model can be considered as a vertical model. Simply put, the current visual big model is based on the big language model, from the initial understanding of language only to the deeper interpretation of 'language+images' now

At the same time, he also admitted that compared to other large models, the difference between the 360 intelligent brain is not significant. 360 Intelligent Brain has two characteristics. The first is training data, which 360 has screened out a lot of high-quality data to support the big model. The second is that after we modify the search interface, we can enhance knowledge and timeliness through search, which can avoid the most common hallucinations of big models

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

In the field of car navigation, through visual large models, potential hazards can be identified during the car navigation process, and alarm processing can be performed while retaining the video, thereby reducing the probability of danger for drivers. "For example, if a security incident occurred on the streets of Shanghai two days ago and the rear car has the support of a large visual model, it can identify the abnormality of the front car. Then, it can automatically save this video and upload the alarm."

At the end of the interview, Zhou Hongyi provided his own insights on some security issues related to the AI era caused by the recent "AI facelift". He believes that the security issues of AI must be taken seriously, and currently 360 has also established an internal AI security team specifically to tackle them. He also added, "For example, AI practitioners will need to conduct secondary verification in AI generated works in the future, such as adding fingerprint or voiceprint recognition

AI is an industrial revolution, and its development is an inevitable trend. It cannot be abandoned just because it has some security issues. What 360 needs to do now is to minimize this security issue. We need to solve both network security, data security, and artificial intelligence security. At the same time, 360 also needs to achieve the ultimate development of big models. Because big models are the pinnacle of digitization and an industrial revolution level transformation, who is not in charge Whoever grasps the core technology of big models will not have the practical application scenarios of big models, thus being eliminated by the industry. So, 360 not only needs to establish a dedicated research team to research, but also needs to find a better and safer solution through continuous experimentation Zhou Hongyi said.

Disclaimer: The content of this article is sourced from the internet. The copyright of the text, images, and other materials belongs to the original author. The platform reprints the materials for the purpose of conveying more information. The content of the article is for reference and learning only, and should not be used for commercial purposes. If it infringes on your legitimate rights and interests, please contact us promptly and we will handle it as soon as possible! We respect copyright and are committed to protecting it. Thank you for sharing.(Email:[email protected])

Mobile advertising space rental

Tag: of is Zhou Hongyi The development AI an inevitable