Once you get the information, what next? When AI makes social crawlers zero-threshold, what is truly valuable is the ability to read people’s minds.
*▲ The machine is responsible for bringing the world’s voices to you, but you still have to rely on human judgment to understand which sentence and to whom it is spoken. *
After getting the information, what next?
Tonight’s AI gathering was attended by more than two hundred friends. The topic is social crawler and public opinion analysis, which sounds very technical, but He Jiaxun, the lecturer, did not rush to open the program code at the beginning, but first threw a question into the chat room: Why are you here today? Producing content, placing ads, doing customer service, or just out of curiosity?
I knew at that moment that this was not going to be an AI tools course that teaches you how to write prompts. It actually talks about a more fundamental thing: In this era when information is readily available, is it still a skill to capture the information?
After more than two hours of listening, Teacher Xiaogui demonstrated from the free Google News to his own AI middle platform that can automatically select topics and automatically write articles. The tools were thrown out one after another without hiding anything. But what I actually copied into my notebook was not one tool after another, but some seemingly basic questions.
Teacher Xiaogui spent a lot of space talking about brand public opinion. He said that the most embarrassing situation for marketing or public relations is when your brand is named by the media and your boss is discussed by netizens. As a result, you are the last person to know. So in the early years when there was no AI, he relied on free Google News: enter the brand name and boss name, and add double quotes outside the keywords, which means that the entire string of words must be completely hit before it is sent to you. This trick may seem like a down-to-earth way of making steel, but at least it keeps you from being kept in the dark.
With AI, there is almost no threshold for capturing data. He demonstrated DailyHot API, JINA Reader along the way, and captured hot lists and comments from Weibo, Zhihu, Baidu, Dcard, and PTT. But the point he emphasized was exactly the opposite: catching it back doesn’t mean anything. The point is, after catching it back, do you have any insights that can be used and realized?
This sentence resonates with me. To be honest, over the years I have seen too many people regard “I can capture the information” as an illusion of ability. When a crawler brings a thousand comments to you, do you really understand consumers? He proposed a very clear ladder: there are five levels of public opinion analysis. The first level is quantity, how many people are talking; the second level is positive or negative, whether the words are good or bad; and the next level is topic classification, comparison of competing products, and the most difficult early warning. He said that most people can only reach the first level, and there are not many people who can reach the second level.
In the second half he used an old story to tell the story more clearly. Two shoe sellers went to Africa to investigate. One reported that there was no business here because no one wore shoes; the other reported that there was gold everywhere because no one wore shoes. For the same piece of information, different behavioral insights lead to completely different conclusions. He put it very bluntly: Everyone is using this tool, the difference is who can turn thousands of messages into an insight that can be packaged and sold. Everyone can use the tools, but being able to identify consumers’ pain points, desires and hesitations from messages is something I can’t do because only the boss can make products.
I totally agree. When everyone knows how to capture information, and even the preliminary emotion classification AI has done it for you, the value of those of us who do content and research is forced back to the higher level: can you understand the human heart behind the information. This is the same as writing. Anyone can find the material. The watershed is whether you can develop a point of view. Technology depreciates, but judgment does not. This is also the core of what I talked about in <Five Things AI Can’t Do>: After AI makes execution cheaper, what is really scarce are those abilities that it cannot replace for you.
This also reminds me of the romance of a one-person company. It is easy for people to think that they should do everything by themselves and raise an AI army to build it from scratch. But what he reminded is the other side: the real leverage is not whether you can do it, but whether you spend time on things that only you can do? Buy it if you can and outsource the pain of maintenance. Sometimes it is the most rational for a one-person company. I also have a more complete review of this point in <One-person Company Management in the AI Era> Reading Experience.
Teach AI to learn from you, rather than making you more and more like AI
Tonight, Teacher Xiaogui walked through his AI platform completely, and I drew a flow chart on my notebook while listening. The system catches a hot spot, such as Shopee adjusting the free shipping threshold, and then uses RAG to find out his own views on pricing psychology, calculates a similarity score, and only starts writing when it is high enough; then Claude uses a skill that remembers He Jiaxun’s writing style to write a first draft, push it to his Telegram or LINE and ask him to review it, and only after that it is sent to Threads, social groups and websites. He will review the data seven days later to make the next topic selection more accurate. This set of ideas is almost the same thing as what I have been talking about Establishing a content production system: what really widens the gap is never a question or answer, but a whole production line that can run on its own.
The most interesting part, in my opinion, is that he taught the AI his own tricks. After the AI finished writing, he didn’t send it directly to him. Instead, he revised it himself. After the revision, he turned back to the AI and said: Please learn from what I just changed and write in my style next time. So each time the changes were made smaller and smaller. He also honestly admitted that the AI memory that everyone often talks about is essentially skill. This is exactly what I have been exploring in my own writing process. The first version produced by AI always has a plastic smell that is instantly recognizable. The real effort is not to get it right at once, but to establish a feedback loop: you change it, it learns, and it will be closer to you next time.
But he also left a very honest proviso: when you keep feeding AI with your own style, will AI become more and more like you over time, or will you unknowingly become more and more like AI? Tools save you effort, but your sense of language may also be slowly assimilated by the model you have cultivated. This is why I always believe that prompt words will expire, but you will not: When execution is made cheaper by AI, what is truly valuable is your questioning power and your unique voice.
The barrier he deliberately left behind
What impressed me the most was that he deliberately left a manual review checkpoint in the middle of the entire line. He could have let it run completely automatically, but he chose to let himself go in and take a look. He said it very truthfully: Unless the review mechanism can be very precise, he will not be able to jump through this level because he wants the quality of the content to be good.
This level actually answers the question about unmanned companies that I have been thinking about: How many things can one person and an AI army do? The answer is that automation does not remove yourself entirely, but hands over repetitive and mechanical links to machines, and then firmly leaves you in the only position that cannot be compromised: taste and judgment. The system can help you write eighty points, but the last twenty points, that is, whether it looks like you and whether it is worthy of your name, only you can sign.
If you also want to build such a content production line for yourself, hand over topic selection, gathering information, and writing the first draft to AI, leaving only the most critical judgment to yourself, this is what solo.tw <AI Content Production System Workshop> is doing with you; and if you even want to make the tool itself, <Vibe Coding Practical Workshop〉 will help you turn your ideas into online works on the same day.
**Like this idea of using AI on the knife edge? ** Welcome to subscribe to my Vista AI Inspiration Supply Station. I will share with you the daily selected AI tools, content, thoughts, and first-line observations as soon as possible to help you sharpen your judgment in the AI era.
To listen correctly, understand and act correctly
Tonight’s sharing, Teacher Xiaogui reminds everyone: listen correctly, understand and do correctly. Find the right voice, understand people’s hearts, talk to each other, and do the right thing.
In the end, this AI gathering that talked about social crawlers and public opinion analysis had nothing to do with crawlers. Instead, the connection between people is more important. What matters is that when machines can capture the sounds of the world for us, can we people still quietly understand one or two truly important words?
When I walked out of the venue after the show, I was not anxious, but rather at ease. Because this night once again confirmed: the person worth my time is always the person sitting in front of the screen and needing to make judgments.
If you want to go through the path of “grabbing data → insights → content → monetization” in a more systematic way, welcome to solo.tw to take a look at related courses: if you want to use one source of content for multiple purposes and automate production, see “AI Content Production System Workshop”; if you want to build your own data capture and public opinion gadgets, see “Vibe Coding” Practical Workshop〉; If you want to package your unique insights into a product that can be sold, see “Concept Realization Accompanying Running Camp〉.
