Zhitong Finance Editor, Guolian Securities released a research report saying that in the early morning of May 14, Beijing time, OpenAI released the next-generation flagship generation model GPT-4o and desktop app, and highlighted groundbreaking voice interaction capabilities. Since 2024, multi-modal capabilities have become the key focus of generative AI, and OpenAI, as a leading company, continues to lead the development of the industry at the technical and product levels. Furthermore, since OpenAI released Sora in February, video generation applications at home and abroad have been implemented at an accelerated pace. The release of GPT-4o is also expected to boost the implementation process of voice interactive AI applications, which are expected to take the lead in social networking, gaming, and education.

With the accelerated evolution of overseas AI model capabilities and the continuous catch-up of domestic model capabilities, the domestic AIC application implementation process has accelerated significantly since 2024, and it is expected that AI native “killer applications” will be incubated. It is recommended to focus on: 1) Kunlun Wanwei (300418.SZ) and Shengtian Network (300494.SZ), which are relatively fast implementing AIC applications; 2) Kaiying Network (002517.SZ) and Giant Network (002558.SZ), which are game targets with low valuation and excellent performance.

The main views of Guolian Securities are as follows:

A breakthrough was achieved in GPT-4O interaction capability, and the degree of “anthropomorphism” was further enhanced

Judging from the product effect, GPT-4o has achieved breakthrough progress in the field of real-time voice interaction, providing users with a more natural and accurate interaction experience: 1) Users can interrupt the model at any time without waiting until it ends to start talking, and the interaction is more in line with human interaction logic. 2) The real-time response capability has been greatly improved. The model has the ability to respond in real time, and there will be no embarrassing situation where users wait a long time for the model to respond.

3) The model has the ability to sense emotions, can generate voices with different emotional styles, and the interaction is more personable. Based on GPT-4o's strong interactive capabilities, rich application scenarios were presented at the press conference, including telling emotional stories in a rich voice, real-time video conversations, and real-time audio translation. From a technical perspective, GPT-4o uses a new technology where all inputs and outputs are processed by the same neural network to enable end-to-end training of text, vision, and audio.

1) Social networking: Currently, most AI+ social product forms focus on “user-AI agent” interaction. Users gain a sense of companionship and emotional value through the interaction process with personalized AI virtual people. Judging from product data, the leading overseas product, Character.AI, has reached the monthly activity level of 10 million, and “Hoshino” under Minimax in China is growing significantly. Judging from the implementation threshold, the task of the companion scenario is simple and the fault tolerance rate is high, so it is the fastest implementation AIC application scenario. Judging from user needs, AI agents “more like humans” are the core needs of AI social users. After the release of GPT-4o, it is expected that the user experience will be greatly improved in terms of multiple modes (from text interaction to voice interaction) and personification (more accurate identification of users' emotions and needs), thereby promoting AI social products to further break the circle and enhance commercialization capabilities.

2) Games: AI is being implemented rapidly in the game development process. Currently, the core focus is on gameplay innovation. Among them, AI+NPC has been launched in NetEase's “Against the Cold” and other products, but it is limited to text-based interaction, and the integration with core gameplay is also quite limited. As GPT-4o leads the transformation of interaction methods, in-game NPCs are expected to achieve real-time voice interaction with users, and the degree of personalization is expected to be further enhanced, greatly enhancing users' sense of immersion, thereby increasing activity and willingness to pay.

3) Education: Previously, overseas neighboring countries and the like had used generative AI in scenarios such as speaking practice, boosting a 57% year-on-year increase in paid users in 2023Q4. GPT-4o is expected to make “AI teachers” more personable after implementation, further improving teaching and training efficiency and user experience.

Risk warning: Technology development falls short of expectations, AI application implementation falls short of expectations, policy supervision risks.