回老家和朋友们聊了聊AI，发现跟我们不在一个世界, HSK5 reading

The author works in the AI industry. On a trip back to his hometown, he was eating barbecue with high-school classmates when someone asked: the news talks about large models and artificial intelligence every day—what is that, exactly? Can it help us get work done?

He immediately perked up. His own work is dealing with AI every day: testing models, building agents, writing workflows. In his information feed, eight out of ten posts are about AI. In WeChat Moments, people discuss new frameworks, new models, and new benchmark scores every day.

But the answers around the table surprised him. One person had used Doubao twice to ask for recipes. Another had asked DeepSeek to write a leave request; it came out badly, worse than writing it himself. A classmate who is a teacher sometimes uses it to look things up, but does not fully trust it, afraid it will "make things up." The rest had barely used AI at all. It was not that they could not afford it, or did not know how to download an app; they just felt it had nothing to do with them. One classmate said something the author still remembers: isn't AI just a chatbot? How is it different from the old intelligent customer-service bots?

He was shocked—not because they were wrong. Honestly, they were quite right. For most people, AI really is just a slightly smarter search box. But that feeling is a completely different world from the one he lives in at work. He suddenly realized he may be living inside a large information cocoon. It is not the kind created by algorithmic recommendations; it is the cocoon that comes with the in-group. Everyone around him talks about AI, so he assumed the whole world was talking about it. Outside that circle, many people only open Doubao once in a while to ask a question.

After he returned, he roughly divided AI users into six levels. The first level is the largest, perhaps more than 90%: they open an app, ask about recipes, the law, homework, or travel, then leave, without trusting the answers much. The second level tries extra features like making images or PowerPoints, but the quality is average; after a few tries they think "that's about it" and stop going deeper. The third level is where a real barrier appears: they save prompts, build workflows, and people around them already call them "AI experts." But these methods are hard to pass down, because first-level users think, "It's just asking a question—is that really necessary?" From this layer on, each step up means fewer people you can actually talk to.

The fourth level starts building agents, wanting AI to run on its own, but the results can only be shared with people like themselves. The fifth level is amateur enthusiasts: they use AI to write code and build knowledge bases, care a lot about value for money, and would rather spend two extra hours fixing things than pay another twenty yuan. The sixth level is the smallest, perhaps under 1%: they use the best tools, are willing to spend a few hundred yuan a month, and treat AI as a colleague. First-level users think AI is a chatbot; sixth-level users think AI is a tireless coworker. This is not a difference of degree. It is a difference of dimension.

He also found that for the first few levels, a small model that can run on their own computer is probably enough. Writing reports, organizing materials, and making simple plans do not require the strongest large models. Companies are similar: many firms mainly need a local model whose data stays inside and whose cost is under control. So while the industry chases parameters, reasoning ability, and longer context every day, the real beneficiaries are sixth-level users. For first-level users, going from an old model to a new one feels almost the same—they are still just asking for recipes. Technical progress is of course good, but the gains are very uneven.

When he went home, his friends' lives had not really changed because of AI: they still went to work and still took care of their kids. Nobody had lost a job for not using AI, and nobody had been promoted for using it well. For ordinary people, AI's impact is more emotional: news stories about "AI replacing another job" make them anxious, but when they actually try it, it does not feel that magical. Industry people are the opposite. They do use AI to work more efficiently, but they also have to keep learning; if they skip a week, they feel behind. The most ironic part is this: the people who do not use AI at all are the least anxious, while the people who understand AI best are the most anxious.

Finally, the author uses a 2005 network card as a metaphor. Back then, "going online" was something you had to do on purpose: install a network card, run a cable, go to an internet cafe. Later came Wi-Fi and phones, and the internet became like air. People no longer felt they were "using internet technology"; they were just living. AI is still in the stage where you have to open an app on purpose to use it. When it is truly mature, it should hide behind short-video recommendations, hospital exams, and company reports, becoming invisible infrastructure. At that point, his hometown friends will no longer say "AI is just a chatbot"—not because they learned how to use it, but because they are already using it without noticing. Until that day comes, his friends' judgment may be more accurate: AI really is still a chatbot. People in the industry just do not want to admit it.

In one sentence: people in the AI circle chase new models every day, but most ordinary people only use AI to ask questions; technical progress mainly helps a small group, while ordinary people mostly get anxiety. AI will be truly mature when it becomes like air, not when it becomes a stronger chat window.