So, where is your moat?
By Wang Ziwei @ Retail Wei Observation
Following Web3 and the metaverse, artificial intelligence, especially generative AI (Generative AI, or AIGC as it is called in China), has become a hot investment field. Many stocks that have incorporated AI concepts have experienced multiple "10 centimeter" or "20 centimeter" trading limit increases in the first half of this year, and many have doubled their market value.
What should generative AI and AIGC look like, and where is their value? This is a topic worth discussing. From the perspective of a primary market investment manager, this article analyzes the combination of AIGC and some industries, and attempts to answer questions about whether the business model is viable.
Disclaimer: This article represents a personal perspective and is inevitably biased and limited in information. Therefore, the views expressed are for reference and communication purposes only and should not be used as investment guidance.
[One] AIGC + Scripts/Copywriting#
Copywriting and scripts were one of the earliest entrepreneurial opportunities that emerged with the introduction of ChatGPT. The most classic example is Jasper, a foreign company that is now valued at 1.5 billion US dollars but is currently undergoing layoffs.
The reason is simple. Aren't you just using an API? You just made some fine-tuning for specific social media platforms, right?
Yes, that's the biggest problem. You have almost no technological moat. It could even be a side project for a few programmer friends or a small service built by programmers for their own company's marketing department.
This week, OpenAI announced the release of GPT-3.5 Turbo fine-tuning. OpenAI claims that the final customized model can match or even surpass the capabilities of GPT-4 in performing certain tasks. Moreover, OpenAI will release a more advanced GPT-4 this autumn. In other words, fine-tuning is becoming "simpler."
At the same time, when these projects are actually implemented, their essence is driven by operations: how quickly you can acquire customers, make them willing to pay, and retain them in the long term. Therefore, you will find that this field requires the monetization of entrepreneurs' "hidden assets." If you are a master of private domains, you don't need to worry about seed users for these projects.
In addition, most AIGC-powered companies that assist in scriptwriting and copywriting claim to help users reduce costs and increase efficiency through SaaS services. The problem is that while reducing costs may attract users and even lead to additional payments in the early stages, in the long run, these additional payments and commissions are almost non-existent.
As for increasing efficiency, it can indeed improve productivity. However, fundamentally, it may be helping operations and marketing personnel generate a large number of scripts and copywriting, spreading them on social media. It is a logic of hacking platform algorithms. Regardless of whether the platform will eventually allocate traffic (some platforms may directly limit such content), one thing is clear: you cannot get any additional income from it.
One more thing to mention is that compliance issues should be considered when implementing such projects. In other words, do not directly call OpenAI's API.
Therefore, from my personal perspective, AIGC's role in assisting copywriting can make money and even generate good cash flow. However, compliance is the most important issue (of course, you can avoid this issue by using the API of ChatGLM or Wenxin Yiyuan). As for whether it is valuable, that is a more complicated question.
[Two] AIGC + Design#
The emergence of MidJourney made us realize that AI can help us create beautiful images in such a "foolproof" way—just open a browser, no need for graphics cards or brushes. The only limitation is lack of imagination.
Compared to MidJourney, Stable Diffusion (usually referred to as SD) represents another extreme. You usually need to configure devices to make it generate images faster, but the results are also better. Compared to MidJourney, the complexity of operation is similar to the difference between Photoshop and Meitu Xiuxiu.
Therefore, the combination of AIGC and the design field has become a hot topic.
AIGC empowers various fields, from fashion design and jewelry design to home decoration and interior design, from real images to rendered images. It seems that designers are becoming redundant. And due to the clear "what you see is what you get" nature, AIGC + design can be described as a mess.
The underlying architecture of this field is essentially a fine-tuning of SD. Yes, if you say that you don't use OpenAI's API for copywriting and instead use Wenxin Yiyuan or ChatGLM, you can bypass OpenAI. However, it is almost impossible to bypass SD in the design aspect.
So, where are the barriers?
Among the high-quality projects currently seen, the most basic barrier is the prompt. Yes, you can learn everything from tutorials, but why can't you make it as beautiful as others?
That's the difference in prompts, including positive prompts, negative prompts, and prompts related to devices. You don't need just a few prompts, but dozens of them. So, in the short term, this can form a small barrier. However, in the long run, just like fabrics and colors in fashion, as long as you can produce it, I can replicate it in 2-3 weeks. Of course, I'm talking about one design. If it's dozens of designs, the workload is indeed not small.
The second barrier is technology. Indeed, some projects can create their own models, and there are various technical aspects such as accurate contour recognition that require research and development.
The third barrier is data. Where does your training data come from? Is it exclusive? For example, if you are in the construction industry, is your data from a top developer? If you are in jewelry design, is your data from a well-known jewelry brand? These are indeed barriers and can even help you go deeper. In other words, AIGC + design is likely to be an entry point.
AI design as an entry point is actually our fantasy about AI. That is, AI can help industries achieve 10 times or 100 times transformation. If not the entire industry, then we can find a node that can be disrupted by 10 times or 100 times.
Coincidentally, design is that entry point, which can improve industry efficiency in at least two ways:
First, it can generate a large number of related designs before production and then distribute them on social media to see real consumer feedback, achieving trial designs and generating interest.
Second, it can improve internal communication efficiency within companies. Taking the fashion industry as an example, there is often a communication gap between planners and designers. By quickly generating designs from the ideas in the designer's mind using AI and communicating with planners.
After entering the industry, can AIGC + design increase a company's revenue? It is almost impossible to measure, and the ceiling in this field is quite average. So, should you enter the supply chain after entering the industry? If so, how should you enter? Which part of the supply chain should you enter? These are questions that are worth considering.
These are also the questions that AIGC + design companies must answer.
[Three] Conclusion#
Overall, AIGC does bring the potential for cost reduction and efficiency improvement. However, from a business perspective, I believe that the following three business logics have not changed and still need to be considered in the AIGC boom:
First, reducing costs is difficult to generate long-term income. However, providing incremental value to companies that can be clearly distinguished and quantified can generate long-term income. This is essentially the logic of CPS, paying based on revenue increment.
Second, the ceiling is a problem that AIGC must consider. If the field you enter has an average ceiling, you must go deeper, whether it is the supply chain, more tools (such as Shopify), or others.
Third, project-based operations can help maintain the operation of your company in the early stages. However, in the long run, the capitalization value is very low.
"
"Retail Wei Observation" focuses on the latest strategies, tactics, and thoughts in the field of new retail and new consumption from a global perspective. The founder of the platform, Wang Ziwei, is an independent retail analyst.