deuceswildpokeronline| Zhang Xuewei of the Invesco Great Wall Science and Technology Corps: GPT-4o is a huge improvement in the multimodal AI model, which is good for the computing power demand on the reasoning side

This week, the release of the new multi-modal large model GPT-4o on OpenAI has once again triggered market concerns about AIdeuceswildpokeronlineconcern. Regarding the highlights of GPT-4o, its significance to the development of the AI industry and the investment opportunities it brings, Zhang Xuewei, a member of the Invesco Great Wall Science and Technology Corps, made a detailed analysis in the recent live broadcast of "Technology New Insights".

Zhang Xuewei believes that GPT-4o is a huge improvement in the multimodal AI model. Compared with previous versions, GPT-4o is 5 times faster and has a response speed of only 320 milliseconds when the API price is 50% cheaper. The interactive experience is more like a real person, like a "personal intelligent voice assistant", which will greatly enrich downstream applications of the AI model. Its powerful functions create an intelligent voice assistant with high integration of "text/audio/image", human emotions, and low-latency real-time interaction. It is free and open, which helps to promote the improvement of AI penetration and the mature upgrade of the application side.

In terms of investment opportunities, Zhang Xuewei believes that it is mainly beneficial to inferential computing power and AI+ related hardware. This ultra-low latency of GPT-4o will put forward higher requirements on reasoning power. The overseas Nvidia supply chain continues to benefit. Reasoning will relay training in the future, driving the scale of computing power to continue to increase. The continuous improvement of training computing power will promote the intelligence of large models, and the explosion of applications will drive the total demand for reasoning computing power.

In terms of domestic computing power, the progress of overseas large models will also promote domestic ones to accelerate catch-up. Affected by U.S. sanctions, the purchase of high-end overseas chips is hindered. Therefore, we are currently accelerating the layout of AI large models based on domestic computing power in China, and related computing power chip companies It is a good development opportunity; in addition, the development of global AI will also promote application maturity, and the demand for domestic computing power on the inference side will also increase, with more opportunities.

In pan-entertainment scenarios, AI provides a better virtual and real interactive experience, etc., and AI+ represented by MR becomes possible. Including AI+ mobile phones, PCs, etc. For example, Apple is expected to reach a cooperation with openai to support ChatGPT in the iPhone, which is expected to promote the maturity of end-side AI, accelerate the frequency of replacement of consumer terminals, and be optimistic about opportunities in related industry chains.

Looking forward, Zhang Xuewei believes that the large model is developing towards multi-modal. From previous text and voice to current pictures, including future videos, new applications will be born in different media, which will improve the generalization ability of the model. In the future, for example, advertising recommendations and user portraits will be closer to the large model. At present, our domestic large models are in a period of rapid development, continuing to develop in the direction of platform and multimodal, and actively exploring the path of commercialization. Among them, the more excellent ones are Moon's Dark Side, Baidu, Ali, etc. We have seen that the improvement of the next generation of models will be in the ability to reason and analyze, as well as the ability of multimodels. Currently, overseas leaders rely on long form data to improve, and domestic companies still have great market potential in data training.