Large model capabilities and large model applications

Currently, big model technology is developing in full swing, but everyone is still feeling their way forward regarding the application of big models; technical personnel do not know how to use big model technology, and product and business personnel do not know what big models can do.

But we must adhere to a concept - technology is just a tool, we just use tools to solve problems.

Since technology is a tool, we need to understand the tool from two aspects:

1. Tool Capabilities

2. Application of tools

In the previous article, we introduced the big model from technology to application to sorting out - Do you know the big model ecosystem? The content sorting out the big model from technology to application introduces the technology and application issues of the big model.

However, since large model technology is still developing rapidly, many people do not know what large models can do and how to do it; this includes the author himself.

Although the main application scenarios of big models now include AIGC, RAG and intelligent agents, and their application technology is also developing rapidly, in order to deepen our understanding of big models and make some technological innovations based on big models.

Therefore, today we will start from the essence of the problem to understand and use big models.

Any technology is essentially a tool, so there are two main aspects to understanding a tool: what the tool can do - the tool's capabilities, and what problems the tool can solve - the tool's usage/application.

Of course, many tools have multiple functions; for example, you use a hammer to hit a nail, but you can also use an axe to hit the nail. Of course, some tools have a single function and can only do one thing, such as a saw.

Why give the above example?

The reason is that many of us cannot distinguish between the capabilities and applications of technology, and even think that they are one and the same.

You don’t understand the capabilities of big models, let alone their applications; others just tell you what big models can do and how to do it.

Taking computer networks as an example, the network has only one function, which is information transmission; no matter how complex the technology to implement the network is, such as routing protocols, transmission protocols, etc.; their only purpose is to send information accurately to the destination.

Our product and technical personnel have built a variety of applications on this basis, such as websites, apps, and mini-programs; and various application scenarios such as entertainment, education, and life.

The same is true for large models, but the functions of large models are not as simple as computer networks; the same large model may have multiple functions at the same time, the most typical of which is the multimodal large model; in terms of data form, there are text, pictures and videos; in terms of expression form, there are generation functions and reasoning functions, etc.

Therefore, only when you understand the functions of big models, you will know in which business scenarios big models can be applied and what their potential advantages and problems are.

Of course, the capabilities of big models have been evolving. Today’s big models may be very different in functionality from those of two years ago. But these are all about how to make the big model a better technical tool, and to discover the shortcomings of the big model from specific application scenarios, so as to continuously optimize this technical tool.