Six adults, one classroom, and a very practical question
On a Saturday afternoon in Chengdu, six adults sit around a long table with laptops, phone chargers and paper coffee cups. The trainer writes a question on the whiteboard: What can you do after LLM training?
Nobody laughs, but nobody looks excited either. The question is too practical. These are not engineers. They are an office administrator, a salesperson, a small-shop owner, a middle-school teacher, a video creator and a retired community volunteer. They have paid between 0 and 2,000 RMB for short courses about using large language models, the technology behind ChatGPT and Chinese apps such as Doubao, Kimi and Tongyi Qianwen. They are not learning to build a model from scratch. They are learning what to do with one on a Tuesday morning.
The names below are pseudonyms, but the tasks are the kind that come up in almost every beginner class in China today. The short answer to the whiteboard question is not glamorous. It is a clearer email, a faster lesson plan, a better product description, a more prepared sales call. The longer answer is about how ordinary work changes when a general-purpose tool lands on a phone or laptop.
Why LLM training became a normal adult class in China
Short AI courses are now advertised in places that have little to do with tech: vocational schools, community centers, library workshops, even WeChat groups for neighborhood parents. Some are free. Some cost a few hundred RMB. Others are packaged as career upgrades for office workers worried about efficiency, or side-income hopefuls testing whether they can make money from content.
The pitch has shifted. A year or two ago, many courses spent hours on big concepts: what a model is, what parameters mean, why AI will change everything. Now the useful ones start with a task. Write a polite follow-up email. Turn a messy meeting note into five action points. Draft ten product titles for a shop. Make a lesson plan for a 45-minute class. Plan a three-day trip with a budget.
That change matters. For many Chinese adults, AI is not a research topic. It is a feature inside everyday digital life, next to WeChat, Douyin, Xiaohongshu and Taobao. People already learn new tools on their phones, often through short videos. LLM training is less like computer science and more like a driving lesson: you need someone to show you the controls, then you need hours on the road.
Case one: the office worker who got back 40 minutes
Chen Lili, 31, works in administration at a logistics company in Chengdu. Her day is full of small texts: internal notices, meeting reminders, supplier emails, Excel updates. Before the course, she often stared at a blank screen, rewriting the same polite phrases.
Now she uses an AI assistant to produce a first draft. She gives it the context, the audience and the tone: a short notice to warehouse staff, a careful email to a client who is already annoyed, a summary of a 40-minute meeting from her rough notes. She also asks it to explain spreadsheet formulas in plain Chinese. On a busy day, she estimates she saves about 40 minutes. Not a whole afternoon, but enough to leave the office before the worst of the evening traffic.
The checking has not gone away. Once, the AI invented a company policy clause that sounded completely reasonable. She caught it because she knew the real rule. Now she never pastes salary data, ID numbers or full client contracts into a public chatbot. She uses fake names and keeps sensitive details out. Her rule is simple: let AI draft, let a human decide.
Case two: the salesperson who rehearses difficult calls
Wang Tao, 29, sells building materials in Foshan. His customers are contractors, renovation managers and sometimes homeowners who are anxious about delays. A lot of his work happens on the phone. After a call, he used to scribble notes and forget half of them by the next morning.
He now records a short voice note to himself, transcribes it, and asks AI to turn it into a checklist: what the customer asked for, what price was mentioned, what needs follow-up, what cannot be promised. He also uses the tool to rehearse objections. He types a scenario: the client says the quote is too high and a competitor can deliver in two weeks. The AI plays the client. Wang Tao practices a calmer response.
Does that replace the relationship? No. The AI does not know that a contractor is under pressure from his own client, or that a delayed shipment is stuck at a warehouse. Wang Tao still calls. But he arrives at the call better prepared, and his follow-up messages are cleaner. The tool helps him sound less rushed, which in sales is already something.
Case three: the shop owner who writes faster, not automatically better
Zhou Min, 36, runs a small home-goods shop in Yiwu and sells through social platforms. She is good at choosing products and photographing them. She is less good at writing copy at 11 p.m. after a full day of packing orders.
She took a weekend course and learned to ask AI for product descriptions in different styles: practical for older customers, shorter for social feeds, a little warmer for holiday promotions. She also pastes batches of customer reviews into a chatbot and asks it to find repeated complaints. That is how she noticed that several buyers said the lids on a set of food containers leaked. She changed the packaging and added a clearer instruction card.
But she does not publish the first draft. AI copy often sounds too fancy for her shop. A simple spoon becomes a lifestyle revolution. She edits it back into her own voice, adds the exact size, the material, the delivery time. The useful part is speed. The dangerous part is sounding like everyone else.
Case four: the teacher who uses AI, then checks it twice
Liu Yan, 41, teaches English at a middle school in Hefei. Her evenings are a pile of lesson plans, exercise sheets and parent messages. She uses AI to generate a first outline for a reading lesson, to create extra practice questions, and to explain a grammar point in simpler Chinese for students who are falling behind.
She also uses it for oral practice. Students record short answers to AI-generated questions, then bring the recordings to class for discussion. The point is not to let a chatbot grade them. The point is to lower the fear of speaking before class.
Her concerns are concrete. AI sometimes writes awkward English or gives a grammar explanation that is technically true but confusing for a 13-year-old. She checks it against the textbook. She worries that students will use AI to write final essays, so her rule is clear: AI can help you practice, brainstorm and check vocabulary, but the final work must show your own thinking. She does not let it grade essays alone. The teacher still has to know the student.
Case five: the video creator under pressure to sound different
He Wei, 27, works at a media company in Zhengzhou and runs a weekend food account. He uses AI to brainstorm titles, outline 60-second scripts, write captions and turn a long interview into short clips. It helps him get past the blank page.
It also makes him nervous. When everyone in a training group uses the same tool, the first drafts start to sound the same. The same opening line. The same three adjectives. The same tidy ending. He now spends more time on details that cannot be generated: the exact price of a bowl of noodles, the owner’s story, the sound of a wok in the background, the way a regular customer orders without looking at the menu.
For him, the opportunity is real. A solo creator can now produce more drafts, test more formats, cover more small shops. The pressure is also real. AI raises the floor, but it does not decide what is worth filming. That still comes from walking around, talking to people and noticing what is odd or funny or true.
Case six: the retiree who plans trips and asks better questions
Zhang Shuyi, 58, is a retired accountant in Shanghai and a volunteer at her neighborhood community center. She joined a free AI class because her daughter kept saying she should try it. Her first question was not about work. It was about a trip to Yunnan.
She used AI to compare train routes, estimate altitude issues for an older traveler, and build a loose budget. Then she used it for community work: drafting a notice for a flea market, making a volunteer schedule, turning a long announcement into a short message that older residents could read on their phones.
Health information is where she is most careful. She may ask AI to explain a lab report term in plain language, but she takes the answer to a doctor. She checks official hospital accounts and does not trust a chatbot with symptoms alone. The benefit is not medical advice. It is confidence to ask a better question at the next appointment.
What they have in common: a tool, not a replacement
Six different jobs, one pattern. The course gives people vocabulary and a first small win. The real value comes later, when they attach the tool to one repeated task: the weekly report, the customer call, the product listing, the lesson plan, the video script, the trip budget.
They learn simple prompt habits. Give context. Name the audience. Say what format you want. Add an example. Then check the result. None of them talk about AI as a magic answer. They talk about time, drafts and mistakes.
For readers outside China, the closest comparison may be the spreadsheet or the smartphone. Both are general tools. Some people use them deeply; others barely touch the advanced features. The difference is rarely intelligence. It is whether someone has a concrete task and a safe place to practice.
The bill, the mistakes, and the parts AI still cannot do
The costs are not zero. Free versions have limits. Paid chatbots may cost 20 to 100 RMB a month, depending on the app and plan. Courses range from free community sessions to several thousand RMB. The bigger cost is time: learning what the tool is good at, and unlearning the idea that a long prompt is always better.
Mistakes are common. AI invents facts. It writes confident nonsense. It can leak private information if you paste it in. Small businesses need to check platform rules on AI-generated images and text. Teachers need school policies. Salespeople need company data rules. Copyright is still messy, especially for images and long text.
So the limits matter. Do not put ID numbers, client contracts or student records into a public chatbot. Do not trust a legal, medical or financial answer without a human check. Do not assume a polished paragraph is a correct one. The most expensive part is not the subscription. It is judgment.
A small shift you can see from the street
None of these six people describe a sci-fi transformation. Chen leaves the office a little earlier. Wang makes a more careful call. Zhou notices a leaking lid in her reviews. Liu has a sharper lesson plan and a clearer rule about student essays. He Wei spends less time on first drafts and more time finding a good story. Zhang asks her doctor a better question.
That is what LLM training looks like from the inside: not a revolution in every room, but a series of small adjustments in how people write, plan, sell, teach and prepare. The question on the whiteboard sounds simple. The honest answer is simple too. You can do your work with a little less blank-page time, as long as you keep your eyes open.
Start the discussion at forum.chinacomes.com