(Image source: Pexels)
The game of changing skins, professionalism, application companies and large models will all affect the development of the industry.
@商科星球original
Author丨Edited by Yuan Jing丨Ten Lixiang
ChatGPT exploded, making everyone envious.
The company headed by Meta launched the "encirclement and suppression" of ChatGPT. A few days ago, after Zuckerberg launched the LLaMA "Meta Artificial Intelligence Large Language Model" (Large Language Model Meta AI), the focus of this round of competition has been pushed to a climax.
There are many people behind ChatGPT: Microsoft, Google, Notion AI, etc.
In China, the project parties are eager to try and start a new journey. In a series of communications, some people hope to regain the glory of the mobile Internet era, some bet on the trend, some open their arms to AIGC, and some fall into extreme pessimism.
Although we try our best, no one can measure the development outline of AIGC with a ruler. At all moments when there is no answer, AIGC is undoubtedly in the middle of the competition.
This competition unfolds in a fog, in a fog. The only constant is: "everything changes" . Any slight changes in capital, labor, technology, policies and markets will cause fluctuations like butterfly wings. It is also in such a halftime game that secretly developing projects hope to learn more, while participants who have already revealed their cards may not necessarily laugh at the end.
01
three possible futures
One, AGI.
This can be the most joyful or saddest route. The so-called AGI can be called artificial general intelligence (Artificial General Intelligence). Like the human brain, AGI can also reason and make decisions and even solve problems due to uncertain factors. It will have the cognitive ability of common sense, the ability of work planning, the ability of self-learning and the ability to communicate in natural language.
The emergence of ChatGPT has shocked all walks of life, mainly because it initially reflects the signs of AGI. However, most people classify such a future in the mid-to-long term. At present, the adaptation of machines to the environment, reinforcement learning, and the interaction of intelligent agents are yet to be demonstrated.
People can scoff at ChatGPT's "serious nonsense", but deep-seated anxiety has arisen in their hearts-how far are we from its serious and correct doing? A practitioner who has surveyed more than 300 mainstream investment institutions said: "The large-scale model AGI will eat everything. Behind this ability is the pure scale effect, which is possible."
In this case, all application layers will be "sugar paper" except the underlying AGI provider. With the emergence of the highest level of monopoly, innovation opportunities will be greatly compressed, different projects will be "interfaced", and all capabilities will be assembled on the large model.
Second, the long tail of the API.
This is a logic based on the absence of an "ultimate monopoly". Different projects provide the bottom layer and API based on their own advantages, and different vertical industries obtain vertical applications based on expert training. Compared with general-purpose smart bodies, vertical smart bodies have higher benefits and lower costs.
Optimistically speaking, in the future, there may be hundreds of thousands of vertical intelligent bodies, clustered in their respective fields of expertise, and have the function of communicating with each other. At the level of problem solving, different APIs will cooperate with each other. "The relationship between vertical intelligent bodies is a bit like the current human society," said an investor.
Three, systemization.
This is a state that most people in China love to see, and its product form is closer to a "transparent layer". In this type of business, AIGC exists as an operating system, leaving enough room for APP to survive. In this way of thinking, AIGC can be regarded as an industrial upgrade of the original Internet products. As the underlying technology, AIGC is added to the iterative versions of various APPs, but it will not erode the market space of the original products. "If it is dominated by the Apple or Android ecosystem, then the operating system will still play the role of distribution," we learned in the closed-door discussion.
In some people's opinion, the operating system route based on the underlying large model will eventually be the "dish" of the big manufacturers. However, the good news may be that for the sake of stability, big manufacturers will transfer some of their benefits to developers, thereby establishing a new mobile Internet or PC ecosystem.
It is worth mentioning that the above three development directions have not yet reached a conclusion, and most practitioners we have contacted are more inclined to the second and third assumptions.
02
blind optimism
In the AIGC midfield battle, a very conspicuous problem is that new companies have been unable to find a landing scene for a long time. After communicating with star companies on the market, some investors have yet to find a fully convincing realization scenario. In this regard, some people put forward their own views: "I think that the answer may be found on alternative workflows."
Replacing and optimizing workflow has always been the only way for the long-term development of the digital economy. Simply put, this has been fully reflected in the current variety of cost-cutting and efficiency-enhancing products. In the field of investment, investors also apply this experience to the practice of judging whether a project is good or bad.
How are alternative workflows manifested in the AIGC industry? You can take Jasper.ai as an example to find out.
Looking at the relationship between Jasper.ai and GPT-3, we can see that the two are in a complementary relationship. The former did not produce content completely according to the results provided by OpenAI, and in the marketing copywriting industry that Jasper.ai specializes in, it also optimized the results provided by GPT. In this way, there are three criteria for judging AIGC industry projects: one is to replace the workflow; the other is to combine with the large model; the third is to have user needs.
But at present, domestic Jasper.ai is not easy to find.
"I think many domestic projects don't understand that you don't need to say how good the technology is before applying it, but before applying it, you must base it on user needs," an early investor complained. In his opinion, many companies take things for granted. “For example, I think the authenticity of user needs in specific scenarios such as AIGC+ script killing and AIGC+ games is often questionable.”
"Now it's a bit like the time when cloud native first came out. At that time, people thought that cloud native could replace everyone and subvert all tools." Later, the cloud native industry moved towards empowerment. In the logic of subversion, there was no There are too many bright spots. In the eyes of this investor, the current sentiment in the AIGC industry is optimistic but full of blindness.
In fact, in order for AIGC to quickly land, there is another variable—that is, the development of the local open source community. However, objectively speaking, compared with foreign countries, the development of the domestic open source industry is still not as good as expected, so that some practitioners describe this industry as "not so healthy all the time . "
To sum up: the open source community can speed up the production efficiency of AIGC products, and the ecological niche of start-up companies is to adjust the trained model. On the user side, the benefit-to-output ratio that AIGC can provide may become a key link after implementation, and this is also the core essence of digitization itself.
03
Wen Shengwen and Wen Shengtu
In the AIGC industry, Wenshengwen and Wenshengtu are the two major technical directions. We have written about this in our series on Vincentian diagrams . In a series of dialogues with industry practitioners, we found a new path for the future development of the Vincent graph industry.
Otaku scene.
"The growth we've seen in the otaku scene is terrifying," said an investor. In terms of underlying technology, the investor sees that the speed of image generation is rapidly increasing. "This unlocks new possibilities, and the current production volume of pictures per day is already a thousand times higher than before, and it has already been shown in the subdivision scene."
(Interestingly, the Vincent diagram has accelerated the further iteration of the rendering industry - in order to have more granular and higher-resolution images, the parameters of AIGC products will become larger and larger, but objectively this will affect the rendering more computing power and memory requirements).
What excites practitioners the most is that, logically, there is a positive correlation between image generation speed and customization. This means that the era of "super customization" in Vincent's industry is coming. Some insiders said that when the content volume is enlarged to 100,000 times, personal aesthetic needs for pictures will be further satisfied, and this kind of picture needs will be quickly filled into design, architecture, fashion or alternative model scenes among.
Compared with Wen Shengwen, it is easier to find the usage scenarios of the track in the picture. In Wenshengwen, its use scenarios are still locked in script collaboration, story creation, etc. Some investors believe that Wenshengwen's scenarios are "too small and too fragmented . "
Moreover, the Wenshengwen industry in the future will move towards specialization, and various Wenshengwen products need to prove their professionalism in their respective fields, which in turn requires enterprises to invest more research and development resources in it.
"In addition, Wen Shengwen's usage scenarios and willingness to pay when people call GPT-3 are open to discussion." According to an investor, domestic small companies tend to call APIs, but this "skin-changing" ChatGPT The product did not show very good profitability. In the case of not owning the underlying product, the logic of the "skin-changing" product is more like an authorized agent. In a general sense, the commercial value of an agent is far less than that of an original company.
Ending: So far. We seem to have found the development logic of the AIGC enterprise: under the premise of seizing the traffic entrance, seek a balance in application scenarios, data training, and data labeling, and form a growth flywheel with data as the core. As data labeling becomes more detailed and iterations become faster and faster, the retention rate of customers will further increase, and by then, AIGC products will become possible to transform the original industry.
However, at a time when people are paying unprecedented attention to ChatGPT, the refinement and precision weaknesses of AIGC products are gradually being exposed. What is certain is that under this trend, people's interest in artificial intelligence products has gradually shifted from early adopters to their needs for language fluency, depth of thinking, logical ability, and details (resolution).
People need better AIGC products, and technology is developing rapidly, but can they enjoy this wave of dividends? It's hard to say now.
—END—