Shulou(Shulou.com)11/24 Report--

In 2022, AIGC (generative AI) is a well-deserved Internet celebrity.

AI painting has been brushed on various major social platforms, ChatGPT has become popular at home and abroad, and digital people who rely on AI to generate voice, expressions and movements also show up frequently. In December 2022, Science magazine published the top ten breakthroughs in science in 2022, and sure enough, AIGC was selected.

Behind the popularity, the commercial potential of AIGC has yet to form a clearer path. At present, the most widely used and out-of-circle AIGC capability is AI painting (based on literary pictures), but we can see that a large number of ordinary C-end users are eager to play and leave, and the huge traffic rarely translates into a strong willingness to pay.

As for the B-side with professional needs such as illustration, auxiliary design and poster generation, the open source models of Google, OpenAl, Baidu, Tencent and other major companies are sufficient to provide support, and it is not clear how much commercial value this market can activate.

In addition, for the supportive procurement provided by the government, the G-end market is also an important part of accelerating the formation of a virtuous business cycle in the AI industry, and this kind of demand is mainly concentrated in smart cities, digitization of government affairs, big data platform, etc., in which scenarios AIGC can play a role remains to be explored.

The hot "false name" into a real money, to further enhance the commercial space, is the next top priority of AIGC. Let's take a look at how AIGC can make money.

AIGC economy, painting pilot AIGC needs to be commercialized, which is not difficult to make a judgment. However, how to commercialize it needs to be deduced step by step from the technical logic.

We believe that the commercialization of AIGC will first take place in AI painting, that is, in the field of literary graphics. At present, AIGC has given birth to text generation, code generation, image generation, speech synthesis, video generation, and even multimodal basic models and application scenarios.

The reason why AI painting will take the lead in exploring a path of commercialization stems from the three basic rules of AI commercialization:

First, AI technology is constantly evolving. Compared with other technologies, the commercialization of AI technology has a very typical difference: most AI systems will inevitably have errors or inefficiency after deployment, and need to go through iteration and continuous optimization to play a role. Therefore, after the AI system is applied on the ground, some errors can be tolerated. The key point is that the productivity gain and the speed of self-iterative evolution (that is, the ability to adapt to the running environment) should be able to offset the trouble caused by mistakes.

For now, only AI painting can greatly improve content productivity, while appropriate errors can be allowed.

The image generation models such as DALL ·E 2, Midjourney, NovelAI, Stable Diffusion, single-minded and intentional AI are very significant for the productivity improvement of art creation, design and other work. A painting that would take days or even weeks to complete can be done in a second through AI. And the models in this field are very curly and evolve very fast. So although there are some ironic problems in AI painting at first, such as painting people as dogs and girls who eat noodles but can't use chopsticks, these small troubles are nothing compared to the time and energy costs saved for professional designers.

In contrast, although the text generation is more advanced and ChatGPT shocked the world as soon as it appeared, it is difficult for NLP natural languages to achieve higher quality output and deeper vertical content. Code generation may have an impact on developer productivity in the short term, but this group is relatively small Voice, video, digital human and other AIGC application scenarios are worth looking forward to, but there is no subversive basic model like AlphaFold, and the application at the current stage is not mature enough. The short video generation system Make-A-Video released by Meta and Google's text-to-video tool Imagen Video have not set off a big response.

Therefore, in many AIGC application fields, AI painting is expected to be the first to enter the commercial track.

Second, AI technology is data-driven. People who know a little about artificial intelligence know the importance of data, and AIGC can not achieve excellent results without the training of a large number of high-quality data. This also makes AIGC products face technical, legal and ethical constraints when they enter the market.

At the technical level, it is necessary to solve the problems of data sources, tagging, privacy computing, training resources and so on, in which text and image data are easier to obtain and use. At the legal level, the compliance of data authorization must be solved in the commercialization of AIGC products, and the copyright fees for video and audio are relatively expensive. in contrast, AI painting can use open source image data sets to obtain licenses from professional painters or art websites, and the cost is more controllable. Ethically, AIGC is data-driven, so the final product may be contaminated by dirty data, or the original data may be biased or discriminated against. To solve these problems, generally speaking, more efforts need to be made in data preparation and data tagging. Image data annotation is now very mature and can be completed through crowdsourcing platforms.

From the data level, AI painting is also easier to solve data bottlenecks, achieve data compliance, and lay a good foundation for subsequent commercialization.

Third, AI technology is based on the cloud. From training to reasoning, the computational complexity and computing power of the AIGC model vary greatly in different stages, which requires high infrastructure flexibility. Therefore, the generation model is often developed through cloud services. During deployment, the terminal demand also has some uncertainty, which may increase suddenly, and the computing demand will expand in a short period of time. It is also possible that the tide may ebb quickly, and users will soon lose interest after being addicted to it, so the cloud is the best pipeline for AIGC to provide services.

AIGC, as a cloud-based SaaS software capability, allows the demand side to introduce AIGC into the business by simply connecting to it when it is in use and bearing a certain amount of cloud or API service fees, without the need for self-training and development or self-built computer room, which is undoubtedly extremely cost-effective. For example, in today's AI painting software, users can type in a paragraph of text and generate alternative images in the cloud without the need for local GPU or high-performance chips to make the general public play.

The business model of cloud + AI pay-per-view will inevitably affect the prospect of AIGC products. Cloud manufacturers, for example, are naturally more willing to integrate image and video AIGC applications into the solution to increase business revenue. By comparison, text generation to recoup costs by accessing paid software on the cloud is a long way off. For example, GPT-3 costs as much as $12 million for training alone, but Davinci, the best and most expensive of its four commercial versions, costs only $0.060 per token (about four characters), and the cheapest version, Ada, costs as little as $0.0008.

Therefore, AI painting is easier to be concerned by cloud services and other industrial chains, combined with a wide range of industries, by driving the model API to pay, using cloud to complete the conversion of business value.

Judging from many angles, AIGC, especially AI painting, is expected to enter the stage of commercial application at a faster than expected speed. This is certainly good news for users, which means that better and cheaper AI painting products will be "rolled out" soon. But for AI companies, things may not be that simple.

ToC / ToB / ToG? Which road leads to Rome? Does finding a typical scene of AI painting mean finding a good business model? Big no special no.

At this stage, AI painting can play a significant role in improving productivity in three areas:

First, art generation, which can not only allow C-end users to generate paintings, but also generate clothing texture and other arts for game studios, creative institutions and so on.

Second, advertising creativity, also become the "Terminator of Party A", through automatic generation and design of creative sketches, reduce the cost of communication between designers and customers, quickly identify design needs, and avoid a large number of repetition or even rework.

The third is professional design, which combines AI painting with professional knowledge, such as 3D modeling, architectural design, medical treatment, industrial design, etc., so as to reduce the heavy cost of making effect drawings in these professional fields, first by AI to make rough sketches according to prompts, and then by professionals to complete the follow-up work.

Of course, there are applications such as meta-universe to generate digital communities, because they are still relatively minority, so I won't talk about them separately here.

In view of the above typical scenarios of promising large-scale applications, we will find that the commercialization of the three waves of power is different.

First, research institutions and their derivative companies.

The AIGC model requires pre-training on a large number of data sets and costs a lot of resources. One of the main creators of this basic model (foundation models) is scientific research institutions, such as non-profit research institutions such as OpenAI (GPT-3, ChatGPT, DALLE, etc.), or scientific research institutes such as the Institute of Automation of the Chinese Academy of Sciences (Zidong Taichu model).

Such organizations have few pressing concerns about commercialization, so they can focus on technological breakthroughs to create a strong basic model that, like cloud service providers, may provide services on a pay-as-you-go or on-demand basis.

For such organizations, the To C market has large traffic but limited payment capacity, which is more meaningful to help model iteration and optimization. The really feasible commercialization should be through the ToB market service industry, through the provision of API to achieve economies of scale, or by virtue of the neutrality of research organizations, to undertake certain government ToG projects, the application prospect of AI painting in digital intelligence projects, to undertake certain exploration tasks.

Taking the Zidong Taichu model developed by the Institute of Automation of the Chinese Academy of Sciences as an example, it has the ability of multimodal generation such as "sound generation by sound" and "sound generation by sound". Such as intelligent cockpit, industrial design, literature and travel, sign language services and other fields.

Second, large-scale science and technology enterprises. Technology giants are actively involved in the research and development of large models. The main purpose is to see that large models, as the basic model, will be the new generation infrastructure of the AIGC economy. Large technology enterprises tend to have an advantage in products because they have a lot of data, and their salary and working environment are more likely to attract elite technical talents. Therefore, the general class generation model is more and more concentrated to the head enterprises. Google, Meta, Baidu, Tencent, Huawei and other companies are actively involved.

The successful occupation of large technology enterprises in the AIGC field can attract a large number of AI developers and ISV service providers to gather in their own ecology and build an active business atmosphere. So, after the ecology has been set up, where to collect the money?

At present, AIGC's business model is very consistent with the development logic of AI to B model, and it can even be said to be an inevitable choice. First of all, basic products + project system.

The To B market has various levels, among which some projects with reasonable revenue are mainly smart cities, transportation transformation and other projects, among which large-scale science and technology enterprises have innate technological advantages, brand advantages and executive capabilities, which act as traction to provide integrated and customized AIGC capabilities for large-scale projects, so as to achieve R & D recovery.

The other is basic products + cloud services. Provide basic modeling capabilities through API, embed their own AI capabilities into application scenarios of various industries through a large number of downstream enterprises, unlock more industrial value of AIGC, and also drive the growth of cloud services, algorithms and technology solutions of technology enterprises.

The commercialization challenge of large technology companies comes from the strict supervision that usually attracts managers, as well as moral review and ethical supervision from the general public.

For example, if the AI painting software of large enterprises illegally uses artists' paintings for training, it will inevitably cause a storm of public opinion; in some areas, Google and Meta have been issued huge fines for the bad use of data, and the supervision on the development and deployment of AI by large technology companies is also being strengthened.

Third, small and medium-sized enterprises and start-ups. Not all enterprises need to train and develop AIGC models themselves, and it is impossible for a technology giant to take all the algorithmic models. As the above two types of organizations open up the basic models and resources, the deployment cost of AI painting is gradually reduced, and a large number of small and medium-sized enterprises and start-up teams can explore new business models, products or services on the basis of the general model. Form an AI software ecology of a single platform / model + a large number of enterprises + countless developers.

For this kind of enterprises, due to the limited time and resources, through the call of API re-innovation, quickly build customized products and services, quickly respond to market demand, and obtain revenue. For example, after the popularity of AI painting, a large number of AI painting Mini Program and tools developed by individual developers or startups have been launched.

Such companies tend to produce star apps, such as the recent Italian AI painting Mini Program, which added 657000 users a day on November 11, a sign of its popularity. However, the core challenge of ToC application is to use a single scenario. Once the user interest fades, the customer acquisition and operating costs will increase suddenly, and the product must re-explore the way of growth. The exit channel of the capital market, that is, to complete the exit through listing / acquisition / multiple rounds of financing, has become very difficult today.

(statistics from the casual AI painting platform) another market that may be the first to become hot is enterprise services, which are combined with vertical industries to form ToB vertical solutions with a high degree of standardization and mature cost and feedback models based on basic models. Judging from the market response over the past year, the vertical application of AI painting will be the first to become popular in creative design, e-commerce, industrial design, architecture, urban transformation and other industries, mainly in the automatic generation of tedious art tasks and commercial realization in the form of software revenue, service fees, subscription fees, and so on.

What is obvious is that when these three types of enterprises: scientific research institutions, large technology enterprises, small and medium-sized and start-up teams, can find their own niche market and form economies of scale in the B-end industry scenario, then it means that the AIGC commercialization cycle is really open.

Can 2023Gy AIGC start to make money? In 2022, AIGC models came out one after another, which not only achieved a high degree of activity, but also gave birth to a new market map. So, can AIGC models start making money in 2023?

Today, large models continue to emerge and iterate constantly, AI infrastructure is becoming more and more perfect, and the enthusiasm of technology enterprises and developers is also very abundant, but there is still a certain information gap with the broad industrial world. If the gap between the two is not shortened, AIGC commercialization will not come. Only when the number of users of AI painting model and the depth of application scenarios reach a certain scale, will it mean that the To B long-tail market of enterprise services has been pried open completely.

As you can see from the previous article, institutions and technology enterprises that build a new generation of infrastructure through basic models and API are the foundation of the AIGC industry, so in the next 2023 years, such institutions and enterprises need to undertake the task of accelerating the commercial maturity of AIGC.

If all goes well, we will see in the coming year:

1.AIGC products are instrumented. At present, the application threshold of some large models of AI painting is still on the high side, and there are still many challenges for large-scale applications. There is also a "running AI painting" service on the second-hand trading platform, which can help customers use overseas AI painting software to generate works, or optimize keywords to generate more accurate and reasonable works. In the future, large models such as AI painting will encapsulate capabilities to be more perfect, simple and easy to use, and communicate horizontally and vertically with vertical industry knowledge and diversified computing resources, so as to meet the application needs of various types of developers and enterprises, and complete the transfer of AIGC capabilities at the lowest cost.

two。 Autonomy of large model technology. The combination of AIGC application and digital intelligence is still very novel and imaginative at this stage. For example, based on AI model to generate urban traffic design, urban green space planning and so on. Once we enter the stage of industrial scale application, we need to face a problem. The basic model is the support of all AIGC applications, while some large overseas models such as OpenAI series do not support access to the mainland. When everyone is shocked by GPT3.5 and ChatGPT, we can not ignore the smell of "choking" on the software.

In 2023, it becomes more and more urgent for AIGC to integrate with industrial intelligence and to solve the problems of security, controllability and leadership of the underlying model.

3. The industrial chain tends to be perfect and smooth. In 2022, we will see all kinds of AIGC models catch up with each other and compete with each other. In order to change AIGC from minority demand to public demand, and to further enhance the commercial space of AIGC, it is not the response of a certain model, but the industrial chain roles such as developers, ISV service providers, cloud vendors, Internet companies and traditional enterprises that can align with AIGC and know how to make good use of AIGC. How to find / sell the AIGC products you need.

At present, the degree of industrial concentration in the field of AIGC is still relatively low, and the application scenario is relatively simple, which requires not only enterprises with basic models to educate the market and build typical cases, but also a large number of agents and cloud service providers to promote the matching of supply and demand. Developers fully release their brains and creativity and explore the actual scene of AIGC. All these need a perfect and smooth industrial ecosystem.

4. Industry standards and ethics basically form a consensus. The agreed technical standards of the industry are the important promoters of the commercialization of AI. In particular, AIGC, which involves personal creation, may lack transparency and interpretability in the process of model training and development. Additional efforts must be taken to cultivate public confidence and avoid mistrust of AI technology due to data abuse and copyright problems.

In this regard, the developers of the basic model have more ability and responsibility to promote the establishment of industry technical standards and market norms, on the one hand, they can reduce the long-term risk of subsequent commercialization and avoid the possible cost of modification; second, developers / agents / users' trust in AIGC products can be established at an early stage to ensure that products comply with ethical norms and laws and regulations. Third, technical standards also contribute to the establishment of competitiveness and establish boundaries for follow-up market activities.

In 2022, AI painting has aroused the resistance and concern of a large number of individual artists, and there is no clear consensus and definition of copyright, which is expected to bring about changes through industry standards, norms and consensus in 2023.

The commercial prosperity of AIGC is essentially the construction of an AI landing channel from the laboratory to the industrial zone. Only by building the above-mentioned cornerstones one by one and completing the preliminary preparations for commercialization, can we really usher in the "singularity" of large-scale explosion.

This article comes from the official account of Wechat: brain polar body (ID:unity007), author: Tibetan fox

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.