Vertex Finance Seven is in charge of multimodal AI is expected to usher in the first year of applica

Mondo Three rural Updated on 2024-02-19

Guan Shipeng |Investment Advisors

a0380621040001|Practice number

OpenAI recently released the first Wensheng model SORA, and the generation technology has made breakthroughs, and multimodal AI is expected to usher in the first year of application.

The upstream and downstream of the artificial intelligence industry chain include the basic layer, the technology layer, and the application layer. The basic layer is the foundation of the artificial intelligence industry, which provides data computing power support for artificial intelligence, and can be divided into hardware facilities and data equipment, of which hardware facilities include AI chips and sensors.

The midstream technology layer is the core of the AI industry, including algorithms, general-purpose databases, and development platforms.

The downstream application layer is an extension of the artificial intelligence industry, and software and hardware products or solutions are formed for the needs of specific application scenarios, and the applications involve all walks of life, such as AI + security, AI + transportation, AI + medical, AI + manufacturing, etc.

Multimodality refers to collaborative inference of multiple heterogeneous modal data.

On December 6, 2023, Google released a new generation of large model Gemini, which is the first multimodal model released in the world, and the first model to surpass human experts in MMLU in terms of performance, and AI has entered the multimodal era.

Joaquin**'s latest research report said,openaiThe release of SORA means that after text generation and image generation, the generation technology has made a breakthrough, and this year will also become the first year of generation.

In terms of investment opportunities, on the one hand, at the application layer, we recommend focusing on creativity, design, education and other multimodal fields, especially those that generate strong correlation and take the lead in landing; On the other hand, the computing power requirements of the generation model are significantly higher than that of the text, which is reflected in the fact that the training materials increase the time dimension, greatly increase the amount of training data, and involve the process of high-dimensional data compression and decompression, which is usually more complex.

Technical: The multimodal AI concept index has continued to bottom out and rebound recently, and the shrinkage before the Spring Festival has risen by 685%, the market has a strong long mentality, and the market is expected to continue to rise.

Guan Shipeng (certificate number A0380621040001) Introduction:

He has served as a special guest of Zhejiang Economic Radio's "Fortune Evening Peak", "Finance at 8 o'clock in the morning" and "Venture Capital Heroes" columns for a long time, and has a keen perception of the mainstream direction of the market.

【Disclaimer】The content and views in this article are for reference only and do not constitute any investment advice. **There are risks, and you need to be cautious when entering the market.

Join the private club and get your daily picks with one click!

February** Dynamic Incentive Program

Related Pages