What kind of core capabilities  product people need He Tuber to have when they cross over to AIGC? In this article, the editor will share some thoughts and opinions on crossing over to the AIGC field for Internet product people with ordinary backgrounds through personal experience.

This year's AI explosion is in sharp contrast to the sluggish Internet job market. Many Internet practitioners who have resigned or are currently working have turned their attention to the explosive field of AI 2.0, hoping to enter this field and maintain their continued stability. The path to career survival.

In this article, from the perspective of a PM with an ordinary background, I will share my experience on how to successfully cross-border and become an AIGC product practitioner. Before I start sharing, the editor wants to say that this sharing is to help "those PM students who currently have " ordinary backgrounds" ". If you graduate from a prestigious school, you will be a first-tier factory after graduation, and your journey has been smooth until now. This article is a reference for you It’s not of much value, because it’s not that you can’t get better, you want more. You hope to maintain your original workplace status after crossing over, and the salary can be higher. The editor’s sharing will not help your ambitions much. Don’t like it Don't spray, turn right when you go out, it's easy to leave without seeing you off. (Smiling face)

This article will answer the following two questions from the perspective of AIGC practitioners:

Do I need to know more about AIGC technology to switch to AIGC product manager?

Is it more profitable to work in a model factory than in a business factory to work on AI products?

1. Personal cross-border experience

Now, the editor will briefly talk about his experience of becoming a full-time professional (focusing on a real, simple, down-to-earth, smiling face). The editor has been a PM for seven or eight years and has worked in large, medium and small factories. At the beginning of 23, my last company had a wave of layoffs (force majeure), and the editor received a big gift package and became a member of the vast number of unemployed products. At the age of the editor, I have led a team and worked in execution. It is not just a part-time job, I just want to do something interesting. With the popularity of AI this year, I feel that this will be an interesting track, so I gritted my teeth and stamped my feet. Let's do this.

Then the next step is to find a job. Although the editor has been responsible for a content production efficiency improvement platform based on AIGC (technical means of AI 1.0) in a factory as early as 2018, since the editor does not have a background in AI 2.0, Large manufacturers directly skipped it and looked at small factories. At that time, the basic requirements for such positions were: being interested in AI, understanding and using tools such as ChatGPT, Stable Diffusion, and Midjourney.

1. Set a plan: lower expectations and get started quickly

The editor’s real situation at the time was: “Well, whether it’s GPT or AI painting, I’ve heard of it, but I’ve never touched it. But if I want to do this, what can I do?” (Maybe it’s like you who are preparing to change careers now. Same ), because in the spring, the AIGC direction has much more opportunities than the AGI direction. My first goal is to enter the AI ​​2.0 track (so I will choose whichever one has the lowest threshold, focusing on the idea of ​​stability ).

Stable-Diffusion and Midjourney were two research directions at that time, because the editor had previously made full-stack products and commercialized them. Of these two, I definitely choose Stable-Diffusion, because Stable-Diffusion is an engineering software, while Midjourney is a user product, with a lower-level engineering and more scalability for enterprises (there is another reason, Midjourney is too laborious to use) , you have to climb over the wall, and you have to spend money. The editor is poor~~~ Haha).

The editor read some cognitive courses related to Stable-Diffusion on Station B. I memorized the core logic, tried a few drawings, and then went for the interview.

I have forgotten how many companies I interviewed, but the final result is: except for a small AI entrepreneurial team that got an offer, the others were offers from non-AI directions (very helpless~).

This small entrepreneurial team had no direction. The boss had some spare money. When he saw the popularity of AI, he wanted to give it a try. At that time, he was making an SD launcher (something similar to the Qiuye launcher). The boss saw that I was more straightforward, and he also said directly: "You can come, but you can only sign a part-time agreement first. After all, we have no reliable direction. If you have any ideas, let us talk about it." I said: "Okay, this It’s not a problem, I just want to do AI.” Then I joined the job. (Did you think it was over here? No, this is the beginning)

2. Quick implementation: Determine the cut-off point, verify trial and error

I joined this team less than 2 weeks ago, and it was converted into a formal contract and served as the project leader. why? (You guessed it right, I found an AIGC application scenario with real needs that no Internet team has entered into in the market, and through a series of actions such as market research, industry analysis, opportunity insights, etc., I finally convinced people with reason. The boss gave up his project and gave me the team, and we will take care of this matter).

You may ask: "How did I do it? It's very difficult." In fact, it's not as difficult as you think. Because the essence of AIGC is to assist content generation through AI technology . The core is actually the business feature of " content " . The editor himself has been producing content for five or six years (his main business is in the field of content production and efficiency improvement, and his side business is content). So I still have a good understanding of the production process of various types of content. Please remember: AIGC is an optimization for content production, not an innovation.

What the editor has done during this period (in terms of AIGC):

Through various methods, we collect all kinds of advanced prompt word templates (20+ types) used in Wensheng diagrams in the industry, compare their advantages and disadvantages, and learn the advanced prompt word writing ideas of predecessors.

Based on the business characteristics, an AI prompt word template is designed that is easy to use in "long novel text → storyboard → continuous pictures with fixed characters". And test the token consumption ratio of different effect scores, and evaluate the prompt word template that is suitable for the business effect and the cost is acceptable to the user (the token consumption level during the test phase is 100 million).

Based on business scenarios, conduct one-on-one business exchanges with head, waist, and entry-level creators (hundreds of people) in the track to understand the understanding and pain points of AI tools at each stage.

Based on business stuck points, we studied Stable-Diffusion's engineering and plug-in codes, and through testing and comparison, launched the industry's first or only productized solutions such as fixed roles and multiple people on the same page to evaluate the degree of resolution of user pain points.

The project controlled by this editor was discontinued in September and October. Although our product has the best reputation in the industry and has a stable payment rate, after analysis, the tool is not the number one pain point in the industry and cannot be continuously monetized.

3. Review summary: sort out pitfalls and develop skills

So what did the editor gain from this not-so-long, not-so-successful entrepreneurial experience?

Where is the boundary of AIGC in actual business, and how to judge the degree of penetration into the business.

Through AIGC, we have completed the experience of how to ensure that the production pictures have content continuity based on the characteristics of long texts (novels, stories, film and television scripts, etc.) (all kinds of pitfalls and tears).

How to design valuable prompts in a commercial project, and how to set high-quality evaluation standards for prompts.

It can be said that until now, I have never seen a PM who is more professional than me in the vertical field of converting stories into pictures and videos. (I’m just bragging about myself, maybe the PM bosses in this track are too low-key and the editor can’t get in touch~~).

4. Determine goals: clarify advantages and go deep into the industry

With this entrepreneurial experience, I have a basic understanding of the current technology application capabilities of AIGC and AGI:

AIGC has great room for development in the direction of efficiency improvement for enterprises, and its technical reserves based on enterprise services have initially matured;

AGI is still in the technology update stage, and the technical reserves based on serving enterprises or serving individuals through enterprises are not yet mature.

Although like most product people, I am more optimistic about the development space of AGI, under the current national conditions, combined with my own advantages and disadvantages, facing a new round of career choices, I still choose the AIGC direction (let AGI’s bullets fly for a while) ~~~). In the interview for the AI ​​product position, I finally got several top offers from vertical fields, and I chose an offer from one that had the ability to self-research algorithms, the company’s revenue was stable, and its AIGC business could directly affect the company’s core value.

After talking about a lot of bits and pieces, the editor will now answer the two questions at the beginning from a product perspective and based on the above experience.

2. How much knowledge of AIGC technology do you need to switch to an AIGC product manager?

1. Who does it serve? technology or business

As a product, before thinking about this issue, it is best to think from the perspective of traditional product work ideas. Who does a PM serve in the company?

In technology-oriented companies, product positions mostly serve the technical side;

In business-oriented companies, product positions mostly serve the business side.

This kind of thinking is also common in the current AI2.0 track.

The AI ​​field is currently divided into four types of companies: model factories, major Internet companies with self-research capabilities for models, leading companies with model fine-tuning capabilities, and business companies that use third-party model services.

If you choose a model factory , then the person you serve is the algorithm team. You deal with algorithm models, data sets, training sets, etc. every day. Can you do it if you don’t understand?

If you choose a major Internet company with self-developed model capabilities , the objects you serve may be algorithms, or they may be business lines created based on model capabilities. Then you have to see clearly who you are serving, and determine where the technical knowledge you need lies.

If you choose a leading company in the vertical track with model fine-tuning capabilities , and your service targets are basically the business side, then do you need to have in-depth knowledge of model algorithms? I can responsibly tell you that it is not necessary, because the work of the algorithm team is a black box for the business side. If you can understand a few model-related terms spoken by the algorithm, it is enough for you to work.

If you choose a business company that uses third-party model services , the object you serve is still the business side. In this case, you also know that it is not important whether you understand the algorithm model knowledge. Then the question is, why are there still so many businesses? The company itself does not have the ability to self-research models, but AIGC positions require in-depth understanding of AIGC and preferably project experience?

2. Job boundaries? Project establishment or execution

Because the vast majority of these companies belong to the third or fourth situation, their AI business is either not in the project exploration period, or has completed the project and needs to be implemented. And they themselves do not have the ability to practice AIGC projects. They hope that you at least know how to use it and have some experience in using it.

To sum up, AIGC product managers do not necessarily need to understand the model characteristics and algorithm logic very well, but they must understand the business. Understanding the business here refers to the real pain points of the business corresponding to the current position. If we want to use AI technology, how to empower this business, where is the threshold, and what is the ROI. If you can explain these clearly during the interview, then as long as it is not a technical position, you will be more than half successful. The remaining probabilities are all about your advantages over other candidates.

3. Is it more profitable to work in a model factory than in a business factory to work on AI products?

This question, I think most people may think that this should not be a question, but a statement. The editor's point of view is slightly different. Looking at the current employment scene alone, the salary package in model factories is indeed higher than that in other types of companies (after all, scarcity is more valuable~~~).

1. Long board theory or barrel theory

But when you serve technology, it means that you will be far away from the business (compared to serving the business). If your original advantage comes from business (for example, you are in content, commercialization, growth, social networking, etc.), then you have to think about whether you want to give up your original advantage to do something that you are not good at and pay a fee for it. Longer timeframe? Where is the value for money? Of course, you might say that in the era of explosive AI, you want to make up for the shortcomings of algorithmic cognition. Make your overall strength stronger.

But what the editor wants to say is: in the current competitive environment, the longboard theory is more practical than the barrel theory.

So the editor’s argument is:

Each has its own merits, there is no absolute good or bad.

But as a product person, the core competitiveness is the ability to solve problems for customers . What is your background and what you are good at will determine whether it is better for you to choose a model factory or a business factory.

You may sneer at the above sentence (it sounds like a false and empty conclusion). Let me give you an example.

2. Case sharing

A friend of the editor, Xiaozhi, has many years of experience in content and community products. He is currently responsible for content and community business for a long-established leading company in a niche field. After AI 2.0 became popular this year, the company gave him a task: to empower the company's content business through AI capabilities and increase the platform's daily user activity and retention.

After several in-depth exchanges between the editor and Xiaozhi, Xiaozhi, as the owner of the AI ​​project, is concerned about:

P0: By predicting the upper limit of capabilities that AIGC can achieve in the next year, what points in content can improve user consumption retention (project height).

P1: Based on the characteristics of application cases in other fields that AIGC has passed at the current stage, whether it is possible to learn from and make a certain degree of content empowerment MVP attempts (project cut-off point).

After some business suggestions and specific implementation strategies provided by the editor, Xiaozhi has passed the stuck points of the above two issues (the project establishment story has been established, and the MVP has been verified), and is now getting started~~~

At this time, the editor would like to ask all readers and friends: Is Xiaozhi working on the AIGC project, and is he an AIGC product person? Will this project experience add points to Xiaozhi?

4. Editor’s suggestion

Having written this, you may have understood what is the threshold for switching to the AI ​​2.0 field as a product? But I still don’t know how to change careers. Then the editor gives you some personal suggestions:

Choose a specific track based on your existing experience and the characteristics of AIGC or AGI. Choose opportunities in this track. This point is very important.

Analyze which of the above types of companies exist in this track, and which type of companies can your model’s cognitive ability hold.

If you don’t have relevant practical experience in AI, make one yourself (please note that you are not making it up, you are doing it yourself). Start a small AI-based project or small business by yourself.

If you think it is difficult to build one, then you may have two problems. One is that you have too high expectations for the results of the project to be built, and the other is that you have seen too few cases of "integration of technology and business" and have little.