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 129

The Real Use of a Cheat Ability

2,809 words 9 min read 2 views 3mo ago Source ​
Skip to the end of the chapter
Prev
Next

The morning flight took off at 10:20, and Han Luyi and Zhang Biao arrived at Beijing Daxing Airport around 8:00.

The waiting hall was quite crowded, and at a glance, half of them were using laptops—a typical scene of a weekday business trip. Han Luyi, holding a coffee, found an empty seat and sat down. Zhang Biao stood beside him, observing their surroundings.

A girl in a gray suit jacket sat diagonally in front of me, her laptop on her lap, an AI chatbot open in her browser, seemingly engaged in a battle with it. She typed a long message, waited a few seconds, stared at the screen for a while, her lips curled into a smirk, then opened a new chat and started typing again.

Han Luyi was positioned so that she could see the AI's reply on her screen, "Okay! Here is the outline of your Q4 performance report PPT [star emoji]", and then it quickly listed more than a dozen items, interspersed with asterisks, bolded characters, and dashes. Each line had a bolded title followed by an emoji, which looked well-organized, but it was all just generic templates with little information, such as "Review of this quarter's core performance indicators" and "Summary of team collaboration highlights".

The girl let out a long sigh, typing aloud as she muttered, "You're so stupid."

The AI responded quickly, "Sorry to make you think I'm stupid. I'm right here, not hiding or running away, I'll take this insult first."

The girl shook her head, rubbing her forehead. After a moment, she prepared to try again.

Han Lu looked at the girl's screen with curiosity.

This time, the prompts she gave were significantly more detailed than the previous one. "You are a manager with twenty years of experience in the Big Four Auditing business line. You need to prepare a Q4 quarterly performance report PPT for your partners. It should include revenue data, team productivity, and customer satisfaction. The wording should be professional, sophisticated, and concise. These are the raw data."

Then she dragged a few files from her desktop to the chat interface to upload them and pressed Enter.

Han Luyi laughed. Who has twenty years of work experience and is still a manager? This is a typical case of being fooled by online prompt templates. Then he opened the Sight.

A pale blue panel floated above the girl's head.

[Chen Jiujiu, "28 years old," is an Auditing manager at KPMG.]

[Mood: Tired, Irritable]

[Core Anxiety: Frustration with promotion, high work pressure, seeking a job change]

Let's continue reading.

[Requirement: Create a Q4 performance report PPT with a professional, sophisticated, and concise style]

[Underlying Needs (Expanded)]

Han Luyi, curious, inquired about the details.

[Reporting to: Auditing Practice Partner (not the client)]

[Core Request: Prove that the leading new business segment (data Auditing) has generated independent revenue and meets the criteria for being spun off from the existing business lines.]

[Key missing data: Lack of data dimensions required for partner decision-making, such as project independent personnel efficiency ratio, cross-personnel cost comparison with traditional Auditing services, and client repurchase rate]

[Current optimization suggestion: The prompt words have low information density, lack structured constraints, lack audience definition, and lack a core argument.]

Chen Jiujiu typed "performance report" in the prompt box, but what she really wanted to do was to create an independent proposal to convince her partner that the new business she was leading should not continue to be relegated to a supporting role under someone else's business line.

The PPT structures, data selections, and narrative logics of these two presentations are completely different.

But she herself may not have realized this difference, or even if she did, she didn't know how to translate this request into a prompt. Her needs and the prompts she wrote were worlds apart.

Han Luyi looked away and took a sip of coffee.

Interesting.

After boarding the plane, Han Luyi leaned against the window, closed her eyes, and continued to organize her thoughts.

Golden finger.

Since Han Luyi obtained the [Reality Code Sight] cheat, he has been using it every day and has upgraded it twice. With continuous use and summarization, his understanding of the Sight has become deeper and deeper.

First, let's discuss his theory on how the Sight works.

Information gathering was a function he understood from the beginning. By focusing his attention on people or objects, he could obtain relevant information; the accuracy of all such information could not be verified, so he simply assumed it was all correct.

However, there's one crucial point: this information must first be "known information." Known information doesn't mean information that Han Luyi himself knows, but rather information that has been definitively established—at least someone else knows it. For example, Han Luyi might not know about Chen Bowen's three bugs, and he himself might not even be aware of them, but the fact remains, and someone else must know.

Han Luyi tried to read quantum physics or astrophysics papers that he didn't understand. The Sight could provide mostly meta-information about the papers, such as author information, citation counts, and impact factors. However, the Sight couldn't provide answers regarding the factual parts, especially the theories that lacked experimental support.

If the information seen is unreasonable or not in an optimal state, the Sight will also indicate "BUG" or "loophole". In addition, when activating the Sight face-to-face with someone, the Sight will also display the other person's emotions and hidden emotions. The accuracy of this is unknown, but in Han Luyi's experience, it is at least very useful in negotiations. What is the principle behind this ability? It is not as powerful as mind reading, nor does it read the other person's brainwaves. It is probably based on inferences from physiological information such as heartbeat and micro-expressions.

In this light, Sight's features seem somewhat like an enhanced version of a search engine that accesses real-world data.

The Sight has been upgraded twice so far. In addition to increasing the maximum energy and recovery speed, and greatly improving physical attributes, the first upgrade unlocked more information, and the second upgrade unlocked the [Path Compilation] ability.

[Path Compilation] is the ability to deduce the optimal solution based on existing information. When BugKiller was besieged by Dingsheng and CodeSafe, it was this ability that provided a solution to break the deadlock by cooperating with multiple cloud service providers.

However, the drawback of Path Compilation is that it has limited computational complexity, and the effort required increases proportionally with the complexity. When Zhao Wenyuan enabled it, Han Luyi tried to copy the best parameter combination of the model by having the Sight directly provide it, but only received a "insufficient energy" message. Other examples include financial products and weather.

When Han Luyi first acquired Sight, he thought it was because Sight discouraged speculative profit-making, but now it seems there are two reasons.

First, Sight cannot predict the future. Second, Sight have limitations when calculating complex systems. Weather and publicly traded financial products are typical examples of system models that Taleb, the author of "The Black Swan," calls complex models. Changes in one part of the model will cause changes in other parts, which in turn will cause even more changes, making it impossible to calculate all the changes in the entire model.

Taking the stock market as an example, if someone buys or sells a stock, it may have an extremely subtle impact on the stock price. This impact, in turn, will cause psychological changes in other market participants. Some people become greedy and their willingness to buy increases, while others become fearful and their willingness to buy decreases. These factors then influence their actions, which in turn affect the stock price. What's even more frightening is that market participants themselves are constantly affected by these external changes. Their own behavior, after being transmitted through external channels, will in turn affect their own behavior. This characteristic is what Soros calls reflexivity.

Soros made a billion dollars by shorting the British pound in 1992 using this theory, earning him the title of "the man who broke the Bank of England." However, the same Soros was directly confronted by the Hong Kong Monetary Authority when he shorted the Hong Kong dollar in 1998, losing a billion dollars and retreating in disgrace.

It's fair to say that no one can accurately predict this system, except for Laplace's demon.

From this perspective, the Sight is a search engine that connects to real-world data and a computer with extremely high Computing Power, but it's not Laplace's demon. So how does it usually use the Sight?

He uses the tool to check code bugs less and less, because he has fewer and fewer opportunities to write code.

To gather information and help him uncover Wang Zhiyuan's behind-the-scenes schemes: BVI structure, Flash fraud, and strategic acquisitions.

Read people, read Liang Yu, read He Yunshen, read Wang Zhiyuan.

They all work well.

But what does all this have to do with building a model?

Han Luyi thought of a few direct uses.

First: Use your Sight to understand your competitors. Find opportunities to meet with key personnel from companies like Dingsheng, Neus AI, or other model-making teams. Once you open your Sight, you'll know where their technical roadmap is, where they're stuck, and what their next steps are. Always know one step ahead of your competitors and always have an information advantage.

He thought about this approach for a while and found it useful, but not enough. Information advantage could help him make the right decisions, but it couldn't help him build a good model. Even if he knew Neus AI's technical roadmap, he didn't have forty researchers from OpenAI to execute it. Knowing someone else's answer and being able to write your own answer are two different things.

Second: Using Sight to recruit talent. The most valuable asset in the AI industry is talent, and recruiting anyone is a gamble. A person with a beautifully written resume might be a master of packaging, while someone who performs well in an interview might become complacent after three months. With Sight, he can tell a person's true abilities, loyalty, and whether they are there to work or to enhance their resume. This is a recruitment system that is abnormally precise.

This is indeed a competitive advantage, but it addresses the issue of "team capability," not "model capability." Even if they poach the ten best engineers in China, those ten people will still be using the same datasets and algorithms to train the models. Compared to other companies, their team might be slightly stronger, but there's no fundamental difference. Besides, what would they rely on to poach them?

Thirdly: Use your Sight for business negotiations. When negotiating with Computing Power providers, investors, and clients, always know where their bottom line is, and always get the best terms.

This is even more irrelevant. Winning the negotiation can save money, but saving money doesn't mean the model is better.

Han Luyi discovered a pattern: every application he came up with was to make Han Luyi himself stronger, not to make the model stronger. How many people could he meet in a day? Even if he used his Sight as a business weapon to the extreme, it would only make his personal decisions more accurate, but there was still a huge gap between "his personal decisions being more accurate" and "training a good model".

Sight can help him win a negotiation, but it can't win a technological race.

When you can't figure something out, try thinking about it from a different perspective.

What can't Sight do?

Han Luyi stared at the back of the front seat, piecing together the information.

Sight does not create information, invent knowledge, predict the future, or penetrate causal chains.

It only does one thing.

It presents existing information in a structured way that is not directly accessible to human senses.

The vulnerability in the code was there all along; the Sight allowed him to see it.

Humans' hidden emotions are always there: micro-expressions, tone of voice, breathing rate, word choice habits. These signals are constantly being sent out, and the human brain can't process them all. Sight extracts them and translates them into a panel that it can understand.

The fabrication surrounding the "All Things Alive" press conference was always there, and Wang Zhiyuan's BVI architecture was always there.

Sight is an information processor, not an information creator.

Han Luyi opened his eyes and glanced out the window.

The plane was inside the clouds; everything outside was a vast expanse of white, and nothing could be seen clearly.

He has an ability that others don't: he can see the truth.

But this ability is tied to him alone; he can't tell others about it, nor can it be scaled up.

His trump card was his golden touch, but he couldn't figure out how to turn it into the key to solving this problem.

Then he thought of BugKiller.

Why was BugKiller successful?

On the surface, it's because the product is excellent, the engine is powerful, and it was the first to enter the ambient programming market—these are all important, of course. But what is the most fundamental reason?

He used his Sight to see the bug patterns that others couldn't.

Other companies are also developing code inspection tools, with algorithm teams, millions of codebases to work with, and decades of inspection rules in the open-source community. However, their inspection rules are based on summaries of past experience, working backward from "past mistakes" to "potential mistakes."

They're guessing.

Han Luyi is watching.

He saw things that others couldn't guess, then extracted the patterns he saw, wrote them into detection rules, and let BugKiller replace his eyes.

The essence of BugKiller

It's like copying the ability of Sight to "see vulnerabilities" to a piece of software.

Han Luyi sat up straight.

Can this logic be applied to Model Training?

For a generic model to become an entry point, the real challenge lies in understanding what users actually want.

Large Model get larger and benchmark scores rise, improvements in user experience become slower and slower. Why?

Because the model doesn't understand people.

"Prepare a Q4 earnings report PowerPoint presentation."

Ten people saying this may have ten completely different real needs behind it.

Chen Jiujiu wanted an independent proposal. Some people might want to use it to claim credit, while others might have no specific requirements and just want to leave work early. How can this problem be solved? How do ordinary companies handle this? They hire a labeling team to guess. The manager writes labeling guidelines, the labelers follow the guidelines, and at most they guess correctly 60-70% of the time; the rest is just noise. The model learns only the average level, forever like an intern trying to guess your thoughts. What can Han Luyi do?

He can see what each person truly wants when they say those words.

The same logic applies.

The Sight sees the corpus, can see the underlying intent, extract the bias patterns, and transform it into high-quality Annotated Data, allowing the model to replace the Sight. BugKiller replicates the Sight's ability to "see bugs".

The new model aims to replicate the ability of the Sight to "see intent".

The real use of the cheat code isn't to directly use it to do things.

It's about turning it into a product.

Han Luyi leaned back in his chair.

The path is very clear.

After returning to Haicheng, I used the Sight to observe real user scenarios, recorded the discrepancy between "surface instructions" and "true intentions," accumulated enough use cases, found the patterns of intent deviations, and then used this data to run a fine-tuned version on an open-source platform to see the results.

If the fine-tuned model shows a significant improvement in intent understanding...

This was the bargaining chip he used to negotiate with Dingsheng.

The intent behind what the Sight sees is data quality that others cannot obtain.

That's his stronghold!

This is also the next breakthrough in the field of Large Model!!

The clouds outside the window were sinking, and the ground in the distance began to reveal its outline. The buildings and roads changed from blurry patches of color to clear lines, like a process of focusing.

The plane was taxiing after landing when an announcement came on.

"Please keep your seatbelt in place and remain in your seat until the plane comes to a complete stop and the seatbelt indicator light goes out..." Han Luyi took out his phone and sent a message to Zhao Wenyuan: "Wenyuan, are you in the office? I'll be there in an hour, let's go over some details." Zhang Biao turned to him and asked, "Boss Han, shall we have lunch when we arrive?"

"No, I'm going back to the company."

BugKiller took two months from idea to release.

This time it needs to be even faster.

End of Chapter
Prev
NovelSummaries
Next

Explore the wiki

Han Luyi
Zhao Wenyuan

Comments

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