laitimes

Zhang Hongchao_Regarding the productization of GPT-4, everyone can feel the transformation contained in GPT

author:Xiao Tan Financial Business News

Now open the Internet and take a look, you will see the overwhelming "GPT-4 is so powerful", "too good", "the scientific and technological revolution is coming", "mankind is going to be unemployed"...

But how exactly did such a dramatic change happen? Because, from now on, we are still in the groping stage in many aspects. The answer seems a little vague. Only on ChatGPT has basic Q&A, ghostwriting and other functions, and Microsoft has begun to search, office is the current more clear large language mode landing plan. If there are only these points, it seems that they have not achieved the results brought about by the scientific and technological revolution.

Why is that?

The answer is that any technology hopes to eventually produce an effect, and both need to go through three stages of "technology-product-economic benefits, etc.". From the initial technology to the product to the market, this is the first step that every business needs to do. Among them, productization is a necessary link to carry on the upper and lower, many good technologies are due to the inability to achieve productization, and finally will only stop in specialized or niche fields, and there are many changes that do not need to determine the technical direction, just because the product is good, it can trigger changes. Products are one of the most important assets of an enterprise, and products are complex systems composed of multiple elements, so product development and production require a lot of resources and energy. Just as the iPhone incorporates many new technologies, however, it is a product innovation in itself.

In today's eyes, you can feel the transformative energy contained in GPT. However, how he is productized is little known, and the application solutions that can be seen are mainly implemented through APIs, and it is the world of geeks and developers.

Zhang Hongchao_Regarding the productization of GPT-4, everyone can feel the transformation contained in GPT

So in this round of AI boom, what kind of "killer" goods may be produced? What is a big language model? Let's study several key features of GPT-4 as clues to share with you the possibilities of large language model products.

Let's not say that the one that subverts the world is too empty, and don't just focus on the existing technical capabilities, then it is too real. From the current market point of view, artificial intelligence has begun to "horse race" expansion, but the real products that can make the industry have not yet appeared. Only in the mid-stage stage of productization is an opportunity for the current round of AI outbreak of the broad masses of the people.

Information Age 3.0, with "command" as the main line

The information age is indistinctly indisputable to two lines: access to information and personal entertainment. Access to information is becoming more diverse. Search, information flow, portals, live broadcasts and short videos, the formation of these Internet products is inseparable from the improvement of information acquisition efficiency and the enhancement of personal entertainment experience. GPT-4 stands for a large language model, which can produce massive content and even produce PPTs, websites and videos in a very short time, which are difficult tasks for AIGC in the traditional sense.

The most important product value derived from this ability is that the big language model has led to a complete change in the way humans interact with information.

At NVIDIA GTC 2023, Jensen Huang said: "Generative AI is a new type of computer, we can program in human language, and anyone can command the computer to solve a problem."

"Command" is a crucial term. When GPT-4 and larger-scale language models are used as the product base, the most likely way for users to interact with information products and entertainment products is commands.

In today's information 1.0, your needs are queryable, which means that the search mechanism and the product represented are search engines.

In today's information 2.0, your needs can be catered to, which requires a recommendation mechanism to achieve, and the goods represented are information flow and short videos.

When AI capabilities are further improved, your needs can be customized and met in a short period of time, and GPT-4 or even the new mechanism brought by 5, 6 and 7 is the command mechanism.

At this stage, the information is not formed, and then according to the user's needs to find, but the information itself is not available, when the user gives the command, re-execute AIGC. This is also a change in the way information is obtained. For example, I am a person in the era of search engines, and if I want to write a chip, I have to rely on keywords and key information to find and query various data reports, requiring users to have patience and the ability to find information. Now, artificial intelligence technology is developing rapidly, and many industries have begun to apply artificial intelligence technology. In the era of information flow, the AI recommendation mechanism will realize the automatic push of relevant content based on my own search and browsing of information about the chip, some of which are useful and some are useless, and need to be distinguished by users.

Zhang Hongchao_Regarding the productization of GPT-4, everyone can feel the transformation contained in GPT

When "command" is the main way to interact with information, I can directly instruct the AI to find all the recent information about the chip, and generate a dedicated report for my own use. I just need to type a command on it to complete the whole process - from the beginning to the end of the test, and then back to the test until the result is in. If there is a dissatisfaction, it can also be ordered to make corrections.

If it is such an imperative information collection, production, learning efficiency changes, and access to information means. It will bring humanity into a new era - the era of artificial intelligence. Then AI powered by similar technologies to generate video could dramatically change people's entertainment patterns. This is one of the reasons why many people are talking about "artificial intelligence" today. Today, we will feel that many short videos are patterned and fast-food, simply done by AI, and this worry may disappear in the future because the video is made by AI.

Film commentary, star gossip, singing and dancing, AIGC will customize massive video content at any time in the future, internal movie information is captured by GPT, star gossip is collected by AI, the images of boys and girls in the dance are generated by AI, no longer need to find anything, no longer need to invite AI to suggest, nor do you need the participation of many short video creators, UP owners spent a lot of effort to make an edit, as long as a command, easy hours of entertainment time was beaten.

If the information age 1.0 is people looking for information, 2.0 is information looking for people, then, AIGC triggered the arrival of information 3.0, is that no one needs to look for anyone, AI itself can also be generated.

This productization opportunity has created a new product form called "command engine" after the search engine is the center and the flow of information. Aladdin's Lantern Club is a very creative game that allows us to feel a magical power before we know it, and this is what we want to achieve. I think "Aladdin's Lamp" would be a good name.

Game NPCs are on the lookout

On the night GPT-4 was released, I saw that there were many developers in the group talking about it, and the first piece of information about the use of GPT-4 came up, that is, there is a saying that it is good to make him an NPC in the game.

NPCs are allowed to use large language models as a base, come and enrich it with the player, multiple rounds of dialogue, high freedom, should make people think, but this friend did not say so.

A key innovative feature of GPT-4 is that it enables AI to recognize the picture and understand the details, connotations, etc. on the screen, for example, to understand what the humor of a funny meme is.

This ability, if used in-game, will give the game and its NPCs an unprecedented skill – meeting the player.

Since the advent of video games, players have only had to look at it and listen to the virtual world to be immersed. In this process, there is an interaction between the player and the real world – that is, physical interaction. Players should try to stay away from the reality around them in this process, and the sense of game is always one-way. Under such an "immersive" experience, many people ignore physical and psychological needs and focus only on sensory stimulation. Although AR games, somatosensory games have emerged, and to some extent, this situation has changed, so that the player has a beautiful posture, beautiful movements, force and play environment into the game, but the game itself does not know the players.

If GPT-4 is made a game NPC, or as a mechanic for the game itself, then perhaps we are not far from the game to understand the players and react accordingly. Among them, you may see a lot of people observing and thinking. When you observe the NPCs in the game, they may still be watching you, trying to understand your expression, posture, and interpret the confusion, excitement, or anger of breaking the handle after getting stuck.

Zhang Hongchao_Regarding the productization of GPT-4, everyone can feel the transformation contained in GPT

If this mechanic is placed in the hands of some game master, the final work will definitely be very shocking. In fact, AI generates levels, generates environments, completes automatic rendering and other technologies, has made great changes in game development, and has deeply combined GPT-4 with games, which is still a high probability in the future.

This integration has a high probability of two aspects, one is AR and VR games that focus on interaction, and the other is open world and multi-ending role-playing games that emphasize NPC free interaction.

In addition to competitions, the skill of "understanding" can also be applied to digital humans and even robots. This will open up new ways of thinking and more possibilities. It also opens up significant product opportunities.

"One mouthful"

Zhang Hongchao_Regarding the productization of GPT-4, everyone can feel the transformation contained in GPT

Talking about toC, let's talk about toB. This means to realize information sharing among various functional departments of the government, and to digitize and automate all business handling processes through the network, so that each government department and staff can know the matters they need to handle and related service processes at any time. Today, many people already know a word called "one-net-to-do". That is, to open up the boundaries of the information systems of various departments of the government affairs system, so as to avoid the way for clerks to travel back and forth to various departments, resulting in incomplete procedures, uncovered chapters, and unescaped roads.

Zhang Hongchao_Regarding the productization of GPT-4, everyone can feel the transformation contained in GPT

In many parts of the world, the all-in-one network office serves cities, government affairs and finance, finance and taxation as the center, digitalization-oriented, and the eastern coastal areas of the mainland and first-tier cities, "one-net-to-office" is also being promoted at a high speed.

However, there are actually pain points in the "one-net-to-office" model:

First, "through-office" is only the opening of data, but the logic of processing and the way of processing are still no different from before, users should not have, or should not, is the card is not a card, may spend a lot of time to repeat the inquiry, fill in the information, ask for opinions.

Second, the efficiency of "one-net-to-office" is highly dependent on the ability of operators. In the information age, the phenomenon of "information islands" is becoming more and more serious, and many systems are not interconnected, resulting in a large number of duplicate work. Affected by the limited energy of operators, insufficient professional ability and other factors, it is inevitable that they do not know anything about each business field, so once a complex occurs, there will be operators who can cope with it, but they do not know how to solve it, and they need to ask the leader layer by layer or communicate between departments, which will reduce efficiency.

Integrating large language models into the above aspects inevitably solves some of these problems. In the age of artificial intelligence and the internet, the way we work is changing profoundly. Clerks can directly make suggestions to AI, understand the internal logic and methods of complicated departments and businesses, so as to complete the needs of doing things at one time as much as possible; Employees can use artificial intelligence to learn knowledge and accumulate experience, and through the analysis of links or elements in their own workflow, obtain relevant information and use it, and improve work efficiency. Moreover, operators can also use AI to obtain cross-departmental operation methods, operation examples, etc., so that more complex situations can be handled efficiently. Of course, all this requires a premise - artificial intelligence technology must have sufficient accuracy and precision. The key is that similar capabilities can use AI more to replace human labor.

This kind of service platform that relies on the knowledge capture and logical analysis capabilities of large language models, "one-net-to-office" can be upgraded to "one-mouth-to-do", and the clerk clarifies the demands and solves the problem thoroughly. In the era of "Internet + government affairs", the thinking mode of the big model has been widely accepted and applied to various fields, and it will become one of the new directions for the construction of smart government in the future. Of course, this title is purely a joke, but with urban services as the center, it is necessary to realize the transformation from "network" to "people", which is undoubtedly the core opportunity for the development of toB products in the era of large models.

Among the similar capabilities, the largest demand for landing is government affairs and finance, which involve a wide range of matters, are more professional, and closely related to the needs of the people. GPT stands for Big Language Model that pushes the boundaries of public service and enables AI to connect expertise with user needs.

The integration of functions such as advice, data aggregation, matter processing, and document certification and other new AI service platforms will become the central opportunity for the next development of AI technology in the toB market.

Ultra-no-code development

It has set off a climax of entrepreneurship on the mobile Internet, and people often joke, "I have ideas, opportunities, and methodologies, and I am one programmer away from getting financing."

This is a joke, but we may find that even the programmer is no less.

One of the amazing things about ChatGPT is that the code is automatically generated and modified. Isn't this the artificial intelligence used in traditional software development? In fact, many programmer friends said that similar to the AI model, its function is also mainly auxiliary work, which is completely unable to replace people who understand the business, understand product logic and implementation methods, programmers.

Zhang Hongchao_Regarding the productization of GPT-4, everyone can feel the transformation contained in GPT

But complex development does not necessarily succeed, simply impossible? If you're a newbie, you can try this app — it offers a lot of features, including modeling, layout design, and animation. GPT-4 demonstrated an innovation during the rollout by making the model automatically generate sites as required, which, although the results produced were crude, but met the basic requirements for development.

As far as the cloud computing and software industry is concerned, low-code development or even no-code development has been a popular concept for the past two years. The so-called "zero programming" is to entrust all the complex and repetitive work in software development to a single software developer. The core logic is to use the code base and make a quick copy of the existing development examples, but GPT-4 is significantly more advanced in logic, and he does not even have the ability of users to understand the architecture, and directly generates development programs on demand.

This function may not be possible to produce complex software applications in the short term, but for many people who do not have programming and development capabilities, it is reasonable to use AI to obtain a mini program, an e-commerce platform, and a simple APP.

Using some templates with GPT-4 can form a "super code-free" product form. By using this template during the development phase of a software project, communication between developers and customers can be made quick and easy. Users only need to send out demand and pay, they can get their own software applications, and the labor cost of the whole process of development is close to 0.

It has brought people such AI-led, template-plus-fine-tuned software development, and maybe it will become a new outlet. That's the beauty of AI, because it makes humans smarter and more human. In fact, think about it carefully, today you can pull a group chat chat, in the early days of the Internet chat room, this situation is completely unimaginable, and then in the future there are several APP, a large number of sites, a group of digital people may not be difficult.

From the rise of pre-trained large models, to today's ChatGPT, GPT-4, among which the driving effect of intelligent emergence effect is the most eye-catching, as information common sense, some content has changed. Just like the human brain in the learning process, will go through a complex thinking process of "perception-judgment-execution", each cognitive stage has its own specific standards and rules, only in this way can the best learning effect be obtained. In other words, GPT-4 is a condition, not a result.

Only by taking advantage of these new conditions, acquiring new methods, and creating new products can we achieve the best results in the AI era.

Some people will worry about whether they have subverted their work and life under such a huge change?

In fact, this anxiety exists in any situation, but it never really becomes a disaster.

In 1999, Pu Shu owned a song called "New Day", which has the lyrics that read, "Come on the Pentium computers, let them think for me."

Today, more than twenty years later, human beings are still wondering where the Pentium computer is going?