A post about a Little Red Book.
[Tao Ran Residence | Today Scout APP]
The cover is a screenshot from a mobile phone, showing a light green interface with a rounded card in the center that reads "Today's To-Do". In the bottom left corner is a line of handwritten text. "Free, really free."
Text:
Sisters!! Dingsheng released a tool called [All Things Alive], which is better than Open Things. It uses AI to create apps for you!! I tried it for an afternoon and made a small daily check-in tool. It's very beautiful, and most importantly, it's free!!
A tutorial will be released tomorrow, providing step-by-step guidance. Even beginners can follow along. Follow me so you don't get lost!
All Things Alive #AIMakesApps #ZeroCode #BeginnerFriendly Hot Comments: "Finally, there's a free one! Open Things at 99 yuan." 1200 likes "A big company has made a move, can Open Things still survive?" 800 likes "Waiting for tutorials! Following." 800 likes "The membership system I made with Open Things is pretty good, but who wouldn't want to try a free one?" 700 likes "Didn't the blogger teach Open Things before? Haha, such a fast transition." 300 likes "Under Tao Ran Residence's latest update, the three Open Things tutorials posted in early October are quietly lying in the timeline, with views ranging from high to low. The earliest one, <Make a Homestay Reservation System in Ten Minutes with Open Things>, still has over 100,000 favorites. This one is still pinned."
Tao Ran excitedly refreshed Little Red Book again. As a top blogger who teaches people how to use Open Things, she was a key investment target of Dingsheng. She received 50,000 yuan in GG fees alone, and this wave of traffic helped her gain more followers. Even her husband came back to help her manage the homestay, allowing her to focus on her self-media.
Tao Ran's accommodation management system, created using Open Things, was running as usual, with new order notifications every day, but she stopped paying attention to Haicheng, Qiantan, and the Source Code Technology office.
When Zhao Wenyuan came over with his laptop, Han Luyi was looking at the weekly report sent by Shen Congyun. No, it was the third version of the synchronized template, which was half the length of the previous two versions. Shen Congyun was also iterating on it.
The results are in.
Zhao Wenyuan turned the computer around, revealing a table on the screen filled with dense numbers. Han Luyi glanced at one line at a time.
Generation accuracy: 78.4%.
The initial performance on the open-source platform was 68%, and after several rounds of training over seven days, it increased by ten percentage points. Of course, this also cost a considerable amount of money.
"The DP0 training pairs performed better than I expected," said Zhao Wenyuan. "The 'bug code → fix code' pairings accumulated by BugKiller naturally provide positive and negative examples for preference learning, resulting in a particularly significant improvement in the model's code quality judgment. On certain types of tasks, we outperformed Qianyuan's publicly available benchmark by two percentage points."
Han Luyi looked at the number. 78.4% meant that four out of every five generation attempts would be usable. This would reduce the communication between the Open Things and the model, decrease Tokn consumption, speed up generation, and significantly improve the user experience.
Zhao Wenyuan had other things to discuss, so he turned to the next page and opened a generated case.
Prompt: "Help me create a membership management system for a fruit store"
The generated code has a clear overall structure and complete functionality. However, on line 47, there is a database variable name:
taoran-vip db
Han Luyi stared at this line of code.
This variable name has no meaning in the context.
However, he knew Tao Ran, the guesthouse owner in Dali. She had created a similar membership management system on Open Things and even published tutorials. Jiang Yixin had shown him her Little Red Book posts.
"It's not just this one," Zhao Wenyuan said. "I ran a thousand test cases, and about 0.3% of the generated results contained information traceable to a specific user, such as variable names, comments, and business logic snippets. One case had a code comment that said 'Delivery Wednesday afternoon,' which is an actual business rule for a certain user."
Anonymization was performed during data cleaning, but it wasn't thorough enough. User business patterns seeped into the training set, and the model learned some side effects.
"The worst areas are variable names and comments, which even BugKil can't detect because, strictly speaking, these aren't bugs. But if we deploy it directly," Zhao Wenyuan closed the screen, "It could cause serious user privacy leaks."
"We don't collect data by default, so where does this data come from?" Han Luyi asked.
"It's user-initiated sharing, but our user profiles—I suspect many people don't even know what they've shared," Zhao Wenyuan said.
"We definitely need to address the privacy issue. What are your plans?" Han Luyi asked.
"Add One data anonymization to the data pipeline, replacing all identifiable information with generic placeholders before feeding it into the training. However, there are side effects: accuracy may decrease, costs may increase by 10% to 20%, and most importantly, the current results need to be retrained."
How long will it take? How much will it cost?
"The data pipeline upgrade and rerun will take a week, and the second round of training will be postponed by a week. The Hardware cost will be between two and three million."
Han Luyi didn't hesitate. "Then let's do it this way."
Zhao Wenyuan nodded, stood up to pack up his computer, and suddenly said something.
"Qianyuan definitely didn't do it this way. The code that CodeSafe obtained was basically all embedded into the model."
Zhao Wenyuan walked out and closed the door.
Han Luyi took out his phone, the variable name still stuck in his mind.
taoran-vip-db
He opened the Little Red Book, searched for [Tao Ran Residence], and the first post that popped up was about how to use All Things Alive.
In the conference room, Su Niannian pointed to the projection screen.
"Daily Active Users dropped from 51,000 to over 43,000, and new registrations decreased from 600-700 per day to 300-400."
No one spoke in the meeting room. Outside the Qiantan office building, the view of Haicheng was overcast and dimly lit.
"The good news is that the paid conversion rate hasn't changed; it's targeting the most superficial users—those who registered, tried it out, but didn't develop a habit."
Jiang Yixin raised her hand. "I downloaded and tried that 'All Things Alive. It works fine for simple things, but it fails for slightly more complex ones. When I generate something with a database, the page layout gets messed up after just one refresh."
"Users won't test complex scenarios first," Su Niannian said. "For the first use, let's do the simplest thing: check-in, expense tracking..."
"To be done, I'll rate the tools."
"So this time, it's not that we've gotten worse, it's that All Things Alive has stolen our 'first attempt'."
She turned off the screen mirroring.
"But what truly retains users is product strength, and we focus on developing our own strengths."
After the product launch, Su Niannian began organizing the data from the first week of the startup template section's launch.
As she looked at the backend data, a slight smile finally appeared on her lips.
Among the ten templates, [Milk Tea Shop Membership System], [Appointment Scheduling], and [Small Commodity Inventory Management] ranked in the top three. Template users had Seven-Day Retention 15 percentage points higher than regular users. Spontaneous sharing of "Using Open Things templates to complete XXX in one hour" has started appearing again on Little Red Book social networking site).
These figures are decent, but there are even more impressive ones.
She asked Li Ting to compile real user cases of people who made money on Open Things, anonymize them, hide personal information, retain industry, scenario, and income data, contact the users to obtain authorization, and finally make them into the "Open Things Entrepreneur Stories" series, which was placed on the product homepage, pinned at the top of the community, and under various official accounts.
It's a GG (Goblin) game, but it uses real people, real events, and real data.
The first episode told three stories: a fruit shop owner saved 20,000 yuan in outsourcing fees by using Open Things to create a membership system; an Open Things to help small businesses create scheduling tools; and a college student earned his first pot of gold by using Open Things to create a class schedule management program and selling it to his classmates.
Within two days of its release, the series of pages had already garnered over a million views. The comments section was completely different from the K0L review on All Things Alive.
Over there, it's all "So cool!" "So amazing!" "So beautiful!", while over here it's all "How did you make it?" "Where's this template?" "I want to try it too!".
Jiang Yixin sent over the list of the top 20 commercial application scenarios she had selected. The twenty scenarios were sorted by retention rate, and each one included a typical user profile and usage frequency. Su Niannian scanned through them and circled eight to move on to the next batch of template candidates.
She opened Lark and Han Luyi a message: "The first week's data for the template section is ready. Take a look when you have time."
Before anyone could reply, she was already writing a batch of template PRDs.
The Yuanlimu Nlili Cafe is located downstairs in an office building, and it's not crowded on weekday afternoons.
When Han Luyi, Gu Siyue was already sitting by the window, with two documents spread out in front of her. She was wearing a dark gray suit today, her hair was tied up, and she had changed her earrings, replacing the pearls from last time with a very small diamond.
"Sign a supplementary agreement." She pushed a page towards me. "The registration of Siheng Law Firm is complete. The legal counsel contract has changed from my personal position to the law firm. The terms remain the same. Please take a look and sign."
Han Luyi through it and signed it.
Gu Siyue put the documents into her briefcase, showing no intention of leaving.
Have you seen "All Things Alive"?
"I saw it."
She pushed another document on the table in front of Han Luyi. It was the All Things Alive User Agreement, one clause of which was underlined with a highlighter.
"Intellectual property rights and usage authorization for user-generated content." Pointing to the underlined section, she continued, "Dingsheng Technology has the right to use all content generated by users on the All Things Alive platform for 'product improvement and technology optimization,' including but not limited to Model Training and algorithm iteration."
Han Luyi glanced at it.
"This statement is much broader than Open Things's. Our agreement is 'anonymized statistical analysis,' while theirs is 'product improvement and technology optimization,' the latter of which can cover almost any data usage."
She handed over the documents.
"I might need it later."
Han Luyi took it.
At this moment, Gu Siyue's gaze passed over his shoulder and paused for a moment.
At the next table, Zhang Biao sat with a glass of ice water in front of him. He sat upright, his hands on the table, looking straight ahead. People around him were chatting, making phone calls, and typing on keyboards, while he seemed like a piece of granite that had been mistakenly thrown into the coffee shop.
Gu Siyue glanced at him, then at Han Luyi.
"--This is?"
"It's a long story. This is Biao, he's my bodyguard now."
Gu Siyue didn't press the matter. She put away her briefcase, stood up, and said, "Your security budget can be managed by a law firm; there's room for tax optimization."
Han Luyi said, "It wasn't me who paid for it; it was the investors."
"The key person clause, right? He Yunshen seems to care about you quite a bit," Gu Siyue said. "Let's meet in the office next time; your colleague is too conspicuous sitting in the coffee shop."
The 17th floor of Dingsheng Tower.
A message popped up in the Lark group chat.
Wang Zhiyuan: "@Chen Bowen, you've worked hard. The first week's data for All Things Alive is good. DU has stabilized, and registrations are still growing. Prepare the data for the chairman's special report next Monday."
Chen Bowen looked at the message.
President Wang's public praise.
Last time in his office, Wang Zhiyuan said, "Don't let me down."
Chen Bowen quickly typed and sent a message: "Received, I'm ready!"
I will give a special report to the chairman next Monday.
I'm reminded of the list of red and orange bug tickets before the launch.
He didn't dare think any further.
Chen Bowen worked on the PPT until late at night, leaving only one person in the office.
He opened the internal user feedback group for All Things Alive and started scrolling through the messages from day one.
Day 1: Screenshots flooded social media, with comments like "So cool!", "Even better than I imagined!", and "Dingsheng finally did something decent." These were all simple scenarios: diary entries, check-ins, and expense tracking.
Day 2: Still taking screenshots, the novelty hadn't worn off, mainly simple scenarios. Some people started trying a slightly more complex one, "creating a timetable management system," and it was generated and usable.
Day 3: Caught between positive feedback, different voices began to emerge.
"The check-in button isn't working; I've refreshed it three times."
The generated page is misaligned on mobile devices, but works correctly on PCs.
Why are my weekend classes gone from my schedule?
Day 4: Even more negative news.
"I was keeping an account, and after adding ten entries, the total amount was calculated incorrectly."
The generated mini-program cannot be opened when shared with friends.
"The submit button only works after I tap it three times; I thought it was a problem with my phone."
Negative reviews haven't exploded yet, but a few more are appearing every day, mixed in with "cool" and "awesome," like seawater slowly seeping into the bottom of a ship.
The product team, backend, and frontend developers are all working overtime to fix it. But Chen Bowen knows the core issue isn't the few minor bugs reported by users. Currently, those using All Things Alive are mostly novice users; the programmers have a natural aversion to no-code and low-code products. Novice users lack confidence and sometimes, when they encounter a bug, they wonder, "Am I using it wrong?" But let's wait a few more days—
He placed his phone face down on the table.
With the screen facing down, unread messages in the feedback group are blocked by the desktop.
There was a three-second silence.
He turned the phone over again.
No new messages.
He breathed a sigh of relief.