Intelligence Finance app learned that GTJA has released a research report, stating that the open-source capabilities of large-scale models are improving rapidly and the trend towards lightweight is significant. It continues to be bullish on the promoting role of AI technology in the content industry. With the rapid progress of AI large-scale model open-source and lightweight, and modalities such as video and audio, some application scenarios are expected to change.

GTJA's main opinions include:

Continuing to be bullish on the promoting role of AI technology in the content industry. With the rapid progress of AI large-scale model open-source and lightweight, and modalities such as video and audio, some application scenarios are expected to change. You can layout along the following train of thought: 1) Transformation of applications such as games, recommend G-bits Network Technology, Kingnet Network, Perfect World, Meitu, and symbols to benefit from Tencent, Netease, Kuaishou, and Giant Network Group; 2) Education track, symbols to benefit from Southern Publishing and Media, Anhui Xinhua Media, and Astro-Century Education & Technology; 3) Emotional companionship and socialization, symbols to benefit from Kunlun Tech and Hubei Century Network Technology Inc.

Large-scale model side: open-source capabilities are improving rapidly, and the trend towards lightweight is significant. Since 2024, the development of large-scale models has shown three major trends: 1) the development of open-source models, with capabilities rapidly approaching those of closed-source products; 2) "lightweight," with model "cost-effectiveness" rapidly improving; 3) development of edge-side models, AI hardware has begun to lay out. All of these indicate that the development of AI large-scale models is moving towards practical and feasible directions.

AI-generated videos: Capability realization is expected to accelerate. Since the release of the OpenAI Sora demo video in February 2024, the industry standards in the AI video field have been significantly improved. After four months of accumulation, many domestic and foreign teams have produced the first answer sheets of "Sora-like" products in June and July: in China, there are G-bits Network Technology with multiple iterations, facing the global market and rapidly commercializing Kuaishou Kelin, a unicorn enterprise derived from text large-scale models; overseas, there are Runway Gen3 Alpha that consistently maintains high quality output and Luma that has relevant experience in 3D generation. The output of these products has significantly improved compared to the "pre-Sora era" products, indicating that the expected results in the field of AI video may be accelerating realized.

Voice function: may help promote the iteration of AI products. The advanced speech function of GTP-4o has been tested on a small scale since late July. This function enables AI to obtain richer information such as emotions and intonation from user speech, and can also reflect richer expressions such as intonation when responding. The interaction delay is low and can be interrupted, and the experience of AI voice interaction is expected to be significantly improved. The development of AI technology in this field is expected to have improved effects on application scenarios such as education and emotional companionship, and will help to enhance the experience of human-computer interaction.

Risk warning: Slow progress in the promotion of AI applications, AI-related commercial realization falls short of expectations, and risk of regulatory oversight on generated content.