Skip to main content
  • Home
  • Library
  • Read
    • Popular
    • Charts
    • Browse
    • Categories
    • Wikis
    • Glossary
  • Request
  • Activity
  • Leaderboard
  • Help
Our Apps
  • DaoSearch
  • Narratorr
Social
  • Discord
  • Patreon
The AI Wave Tycoon's Rise
Chapter 130

Consider It a Loan from Me

1,759 words 6 min read 1 view 1mo ago Source ​
Skip to the end of the chapter
Prev
Next

Zhao Wenyuan had been waiting for a long time.

When Han Luyi pushed open the office door, Zhao Wenyuan was already sitting on the sofa, with two cups of Luckin Coconut Latte on the coffee table in front of him. As expected of you, the spokesperson for Coconut Latte.

"What did Boss He say?" Zhao Wenyuan asked.

Han Luyi didn't answer. He put down his backpack, took out his laptop, turned on the screen, and turned it in front of Zhao Wenyuan.

These are the data he manually marked on the car when he returned.

"Let me look at something first."

Zhao Wenyuan glanced at Han Luyi.

Why does this person look so excited?

The screen displayed a spreadsheet software with four columns. The data source was anonymized user records exported from Open Things. Zhao Wenyuan immediately recognized the first three columns: user input, AI-generated results, and actual user behavior. These columns were already recorded in Open Things's data.

The fourth column is newly added.

Column Name: True Intent.

Zhao Wenyuan's gaze lingered on the first line.

User input: Help me create a customer management system.

AI-generated result: A standard CRUD customer management page, complete with four functional modules: list, add, edit, and delete. Actual user behavior: The add, delete, and modify functions were removed, leaving only the remarks field. The remarks field was then manually expanded into a customer follow-up record page with a timeline. This process was modified three times and regenerated twice.

How should standard annotations be written? Zhao Wenyuan asked himself.

"Misunderstanding of requirements, user dissatisfaction with generated results, partial adoption." That's how he would write it.

The fourth column reads: The user is in sales. The company already has a CRM system, but the remarks section is too rudimentary. What she needs is not a customer management system, but a customer follow-up log tool to supplement the existing CRM.

So detailed? Zhao Wenyuan slid his finger across the touchpad and scrolled down.

Article 2.

User input: Create a work schedule.

AI-generated: Standard shift calendar with drag-and-drop overtime templates.

User behavior: The entire scheduling UI was deleted, only the data export function was kept, and the system was manually modified to highlight statutory holidays and automatically accumulate overtime hours. This was done five times.

The fourth column shows a user who is an HR professional doing year-end closing. She doesn't need a scheduling tool, but rather an overtime pay compliance calculator, requiring actual attendance data to be cross-referenced with the definition of statutory holidays to calculate the overtime multiplier.

Zhao Wenyuan paused for a moment.

He reviewed the third column again; five modifications and the complete deletion of the UI were behavioral data recorded in black and white in the system logs. The annotations in the fourth column explained the "why" behind these behaviors.

He casually flipped through a few more.

A user entered "create a meeting minutes template," but actually deleted the generated template down to just a table frame, and then manually added two columns, "Pending Follow-up Person" and "Next Check Date." The fourth column is labeled: The user didn't want to create meeting minutes, but rather a project progress tracking dashboard, because the company doesn't have project management tools, and she's using meeting minutes as a substitute.

Zhao Wenyuan chose this option for verification. The user deleted the template, leaving only the table, and the behavioral data matched. Follow-up personnel and inspection dates were manually added, and the operation logs showed this. The conclusion was: Use meeting minutes instead of project management tools.

He thought about it and felt it made sense. In fact, it wasn't just that it made sense; if it really was project management, the user's next requirement would most likely be a Gantt chart or Kanban view, rather than a better meeting minutes template.

This level of annotation precision is no longer just "accurate".

It means the annotator understood the user's work scenario.

Zhao Wenyuan looked up at Han Luyi.

"Is this a further processing mark?" Zhao Wenyuan asked. "Used to further improve Heavenly Craft?"

"It's not meant for Heavenly Craft; it's merely an example."

Zhao Wenyuan frowned. "If not Heavenly Craft, then what are we doing? Heavenly Craft is a code- Special Project model, and the annotations of these Open Things data fit Heavenly Craft very well." "What if public datasets of user behavior could also have annotations of this precision?" Han Luyi said.

Zhao Wenyuan shook his head directly.

"Impossible. Your accuracy is due to the closed-loop behavior of the Open Things: how the user modified it, how many times it was regenerated, and what was ultimately retained. These are all objective signals. Public datasets don't have these; relying solely on manual annotation is pure guesswork."

Han Luyi did not argue.

He can't do it, but Sight can.

He changed the subject. "How much data do we need to make a prototype?"

Zhao Wenyuan was stunned for a moment before realizing what Han Luyi was talking about. It wasn't about improving Heavenly Craft, but about fine-tuning the intent understanding direction on an open-source, general-purpose platform.

"Through fine-tuning, can we verify changes in intent comprehension?" Zhao Wenyuan thought for a moment, "Five thousand to eight thousand samples of this quality would be enough to show a significant improvement."

He paused for a moment, then added, "But it's pointless to build it. No matter how well you fine-tune a small model, its performance is nowhere near that of GPT-4."

Han Luyi said, "I'll take it to see the investors."

Zhao Wenyuan understood.

A prototype is not a product; it's a proof of concept, demonstrating feasibility. You can say something like, "Look, I can achieve this effect with just a few thousand data points on a 7B model. What if we had high- Computing Power and big data?"

"Can you deliver it in two to three weeks?" Zhao Wenyuan asked.

"Can."

Zhao Wenyuan scanned the table again; there were 287 entries, and the quality of each annotation exceeded the level of the top annotation teams he had ever seen.

If we could use this kind of data for training... Zhao Wenyuan gave a self-deprecating laugh, "Don't be delusional."

Han Luyi business trip? That looks like he was working dozens of hours straight.

Zhao Wenyuan didn't ask him how he did it. The most reasonable explanation is that Open Things's behavioral data closed loop has user operation logs at hand. With someone who understands the product well enough and spends enough time, it is indeed possible to achieve high-precision annotation.

Moreover, Han Luyi was the designer of Open Things.

The only problem is that the data obtained this way is too expensive and not worthwhile.

Just then, there was a knock on the door.

Su Niannian pushed open the door and came in, carrying a convenience store bag in her hand.

"Biao said you didn't have lunch."

She placed the bag on the coffee table. It contained a sandwich, a rice ball, and a cup of coffee.

Zhao Wenyuan visibly stiffened.

He glanced at Han Luyi subconsciously, swallowing back the words he was about to say. They were discussing the new company's core data strategy. Su Niannian was the co-founder of the Source Code, and if she didn't know Han Luyi's plan...

"Nian Nian knows." Han Luyi's hands didn't stop unpacking the sandwich.

Zhao Wenyuan: "... What do you know?"

Han Luyi glanced at him, but didn't expose his clumsy attempt at concealment. "I know."

Su Niannian sat down on the chair next to her, glanced at the form on the screen, but didn't look at it in detail.

Where were we in our conversation?

Zhao Wenyuan finally breathed a sigh of relief.

"Let's talk about data," Han Luyi mumbled, biting into a sandwich. "We need 8,000 highly labeled data points to make the prototype model, and we need to get that done within two weeks." Zhao Wenyuan thought he meant the data would be ready within two weeks, but Han Luyi meant the model would be ready within two weeks. But no one realized the difference.

Su Niannian nodded, didn't ask for details, and changed the subject.

How's it going with Boss He?

"The deal fell through; he's not optimistic."

Su Niannian didn't seem surprised.

After a brief silence, Zhao Wenyuan glanced at Han Luyi and then at Su Niannian; the atmosphere was somewhat heavy.

Failing to secure funding is the worst news imaginable for any startup; for a company that hasn't even been established yet? It's a devastating blow.

"I still have sixteen million here," Su Niannian said. "I'll lend it to you first, no shares required."

Zhao Wenyuan was stunned.

Han Luyi eating his sandwich.

"I haven't touched the money from selling the old shares last time," Su Niannian said calmly, her tone revealing no emotion. "If you're starting a new company, what you need most is seed capital. Consider this money a personal loan to the new company; I don't want any shares."

The initial investment was 20 million, and the tax rate was 20% of the capital gains. After tax, the net profit was a little over 16 million.

Han Luyi looked at her.

Su Niannian said with a smile, "Don't give me too low an interest rate. I'm counting on this money for my retirement."

Zhao Wenyuan sat quietly to the side. He didn't know much about the specific details of the cooperation between Su Niannian and Han Luyi, but this figure was probably all the wealth Su Niannian could offer.

She doesn't want shares, just an IOU. If the company succeeds, she'll earn a few percent interest; if it fails... what assets could an unregistered company possibly have to pay off its debts?

"Don't worry, I have some money too," Han Luyi said. "Besides, I have ways to get financing."

What method?

Han Luyi didn't give a direct answer; he hadn't figured out how to bring up the matter of asking Dingsheng for Computing Power —neither Su Niannian nor Zhao Wenyuan had had a pleasant experience dealing with Dingsheng.

Seeing that he didn't say anything, Su Niannian didn't press him for details.

She understood Han Luyi's rhythm; if she had thought it through, she would speak; if she hadn't, it meant she was still thinking.

"I won't be polite when it really comes down to it," Han Luyi said.

Su Niannian stood up.

"By the way, Neus has arranged a video conference with Ryan next Wednesday at eight in the morning. Don't be late." Han Luyi nodded. He had to attend this meeting to find out where Neus was.

"I've sent you the agenda Lark," Su Niannian said before leaving the room.

Zhao Wenyuan watched her leave, then turned his head.

"You and President Su..."

Han Luyi looked up. "Hmm?"

Zhao Wenyuan did not continue.

Han Luyi explained, "We're old classmates."

Zhao Wenyuan glanced at him but didn't expose his clumsy attempt at concealment.

Han Luyi continued eating her sandwich, wiped her hands after finishing, and pulled the screen back to the spreadsheet page.

If the quality is good enough, what else is there to say?

Brother Sight, it's up to you.

End of Chapter
Prev
NovelSummaries
Next

Explore the wiki

Han Luyi
Su Niannian
Zhao Wenyuan

Comments

No comments yet.
AboutStatsAPIChangelogStatusContactPrivacyTermsDMCA
© 2026 RandomTranslator·v3.26·RSS