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 131

Ding Pay

1,770 words 6 min read 2 views 1mo ago Source ​
Skip to the end of the chapter
Prev
Next

"Training doesn't require Open Things data," Han Luyi said.

Zhao Wenyuan put down his coffee. "Those two hundred or so..."

"This was specifically for you to check the quality. Since the quality meets the standard, we'll proceed according to this standard," Han Luyi said. "Since we're building a general-purpose dialogue model, it needs to understand what people from various industries are saying. The code-related data that Heavenly Craft has accumulated won't be very useful anymore; we need more diverse data."

"You have experience from your time at Google, how did they do it?" Han Luyi asked.

Zhao Wenyuan thought for a moment, "In the few years since Large Model came out, the process has been pretty much figured out. Large companies generally divide the data pipeline into three steps when building dialogue models: First, massive data collection, using web crawlers and authorized data sources, starting at the petabyte level, almost storing the entire Internet; then cleaning, deduplication, noise reduction, cutting off 80% or 90% of the garbage; finally, annotation, with a team of thousands of people annotating line by line according to the annotation specifications, the annotation cost alone can burn through tens of millions a year."

He glanced at Han Luyi: "We don't have the resources, but we don't need that much data for a prototype. There are readily available public datasets in open-source corpora, community Q&A, encyclopedias, and forums. Others have already done the collection and basic cleaning. What's left is fine cleaning and annotation. Five thousand high-quality annotations would cost over 100,000 yuan and take two or three weeks, even if we outsourced it, but the quality..."

Before Zhao Wenyuan could finish speaking, he said that the quality of the open-source corpus and the annotations outsourced was worlds apart from the annotations Han Luyi had just shown him. No matter how confident he was in his skills, he wouldn't dare say that something produced using this method could attract investment.

"That's what I was thinking too. Don't worry, give me the data, I guarantee it will be the same quality as before," Han Luyi assured him.

Zhao Wenyuan looked at him suspiciously. "Do you know what you're saying?"

Open Things's data labeling means you have data sources, user scenarios, and a deep understanding—what open-source corpora do you have?

Zhao Wenyuan tried to make his tone sound less offensive, "Anyone can get open-source data, and the annotations are all pretty much the same, so what's the difference?"

Good question, I have a solution.

Han Luyi could tell he didn't believe her, so she didn't press the matter. "Just leave it to me, you just need to set up the training pipeline first."

After their conversation, Zhao Wenyuan returned to the thirteenth floor to continue working.

Han Luyi leaned back in his chair, thinking about what to do next.

Zhao Wenyuan is right. Open-source corpora are available to everyone, and the annotations are generally similar, but that's for ordinary annotators. What is data annotation? Simply put, data annotation is about understanding. You're given an article, and the question is, "What does this person really want to do?" The annotator writes a standard answer. The current industry practice is an assembly line: the cleaning pipeline is responsible for deduplication, noise reduction, and format standardization. The annotator gets the clean text, extracts the core intent according to the specifications, and a 3,000-word blog post is ultimately reduced to a single sentence's intent tag.

But those details that the annotators overlooked were exactly what Han Luyi wanted most.

A user posted on a forum, asking how to tell their parents they didn't want to go back to their hometown to work. The post was over 200 words long, and included the sentence, "My dad just retired last month." The tag extracted by the commentator was "How to persuade parents to accept one's career choice."

But the phrase "just retired last month" was highlighted in the Sight.

The father has just retired, and the son is hesitant to return. This isn't really a matter of career choice; it's that the father's sudden free time and emotional needs have collided with the son's boundaries.

What this user really needs isn't persuasion skills, but how to reject someone who's just starting to feel lonely without hurting their feelings. The tag "How to persuade parents to accept one's career choice" doesn't even scratch the surface of the problem.

The more concise the corpus, the more difficult it is to use the Sight; the signals are all in those overlooked details, and the raw data is the gold mine.

What Han Luyi needs to do is use his Sight to extract the context that others perceive as noise but is actually relevant, and then add accurate intent annotations.

The same piece of data, others might label it as a single sentence, but he labels it as an entire demand map.

Turn stone into gold.

After talking with Zhao Wenyuan, Han Luyi went to Siheng Law Firm.

Gu Siyue was already waiting. The documents on the table were neatly arranged, and two cups of English black tea were placed on the table.

"You're here, have a seat."

Han Luyi sat down, picked up his black tea, took a sip, and found it to be alright.

"What's wrong? Is there something you can't say over the phone?"

Han Luyi explained his situation.

I'm planning to start an independent model company, with the core team drawn from the Source Code division, primarily led by Zhao Wenyuan. The main question is: can both I and Zhao Wenyuan hold positions in both companies simultaneously?

Gu Siyue listened without saying anything, then turned around and pulled out the contract file from Source Code Technology from the cabinet.

She reviewed the Series A agreement for Source Code Technology, and she drafted Zhao Wenyuan's employment contract and Stock Options agreement. She knew the terms better than Han Luyi herself. She flipped through it for less than a minute and closed it.

"Let's talk about Zhao Wenyuan first. His situation is simple." Gu Siyue picked up a pen and drew a line on a sticky note. "For a non-compete clause to take effect, the Source Code has to file a lawsuit. If you were the CEO, would you sue him for working on a new project with you?"

"Then there's no problem?"

Gu Siyue glanced at him, wrote two words on the paper, and circled them.

"There's a potential problem: Stock Options."

Han Luyi listened quietly.

"Zhao Wenyuan has 3% of his unrealized Stock Options in the Source Code, and it will take a year to unlock the first batch."

Gu Siyue wrote a number on the paper: sixty million.

"Based on a 2 billion valuation for the Source Code, 3% would be this amount." She put down her pen. "He's only been there for less than two months and hasn't even received a single share. Once he leaves, it's all forfeited."

"What if you don't resign?" Han Luyi asked.

"If he doesn't resign, there's a legal issue of fiduciary duty; employees can't provide core technical services to related parties," Gu Siyue said. "But that's a weak clause. As long as the new company doesn't register, doesn't take up his time working at the Source Code, and doesn't use any of the Source Code's resources, no one can sue him." "Then there's no problem."

"No problem now." Gu Siyue emphasized the word "now," "The problem lies ahead. He Yunshen has a seat on the board of directors and has the right to review the company's operations and major personnel decisions. What if one day he discovers that your core technical lead is working on your off-balance-sheet projects?"

She held up her fingers and listed them off one by one. "First, I demand an explanation from you; second, I demand that Zhao Wenyuan terminate the agreement, or he will be fired and all unvested Stock Options will be forfeited; third, I'd also like to question whether you, as CEO, are actually working full-time."

She looked at Han Luyi.

"In short, he took a huge risk."

After saying that, Gu Siyue turned to a page of the sticky note.

"In your case, the CEO increased the number of shareholders, and the Series A investment agreement also included a full-time commitment clause."

"You simultaneously control another company, which constitutes a technical breach of contract. Strictly speaking, investors can demand that you rectify the situation within a specified period, and may even trigger a buyback clause."

Han Luyi was waiting for her "but".

Gu Siyue did indeed have a "but": "However, in practice, investors almost never take legal action for breach of the full-time contract clause. It's too damaging to relationships and seriously affects the company's normal operations."

She put down her pen.

"Legally, your situation is better than you think, but you need to be careful about one thing: the assets of the new company and the Source Code must be kept very separate." "You mean, this is the biggest legal risk?" Han Luyi asked.

Gu Siyue did not answer directly, "Do you know when Ding Pay was separated from the Dingsheng system?"

"2011."

"It's good that you know." Gu Siyue leaned back in her chair. "In 2011, Lü Yun transferred Ding Pay from Prosperity Group to a domestic company under his personal control, citing the central bank's requirement that third-party payment licenses must be held by domestic capital. Dingsheng's investors, as the major shareholders, were completely unaware of this beforehand, and afterwards they were furious."

"This matter was highly controversial at the time. Supporters said he was visionary, as Ding Pay would be Luyi without the license; opponents said he was greedy, using regulation as an excuse to pocket the company's most valuable assets. Wall Street was even more direct, with several investment banks downgrading Dingsheng's rating, citing the founder's lack of trustworthiness."

"In the end, it ended with a compensation agreement. Prosperity Group received 37.5% of the profits from Ding Pay's IPO, plus a cash compensation. On the surface, everyone was happy, but the rift in trust was never repaired. When Dingsheng later had its IPO, Ding Pay's valuation could only be calculated based on the agreed price, which was much lower than the market price. Moreover, ever since then, every time Dingsheng spins off a new business, investors always ask, 'Are you moving again this time?'"

Gu Siyue looked at Han Luyi.

"Lü Yun was able to push this through because he had absolute control at Dingsheng, and no one could stop him. Even so, he still paid a huge price in terms of trust. What about you?"

End of Chapter
Prev
NovelSummaries
Next

Explore the wiki

Han Luyi
Zhao Wenyuan
Gu Siyue

Comments

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