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 461

Westworld

1,417 words 5 min read 1 view 25d ago Source ​
Skip to the end of the chapter
Prev
Next

Around 7 a. m., Zhang Wenduo brought a cot to the office.

The original eviction date for the student apartments was July 15th, Zhang Wenduo's new rented apartment also started on July 15th.

However, plans can't keep up with changes. Suddenly, the compound issued a notice saying that the building where he lived would be renovated before the summer vacation, and everyone had to move out before July 10th.

Students currently enrolled will be arranged to stay in another dormitory building for the summer vacation. However, for students Zhang Wenduo who have already graduated, living on campus is a free benefit that no one will question. Given this unexpected incident, naturally no one will arrange a replacement dormitory for him.

Suddenly, Zhang Wenduo had nowhere to live.

There were a few classmates around him in the same situation, who were planning to stay at his junior's dormitory for a while and sleep on the floor.

Zhang Wenduo thought to himself, "Since I'm sleeping on the floor anyway, what difference does it make where I sleep?" So he simply borrowed a cot and brought it directly to his office. This way, he could stay a little later when doing experiments and save time commuting.

Anyway, he's usually the first to arrive at the office and the last to leave. Even if he slept in the office for a week, it wouldn't affect anyone.

Showering is also easy to take care of; there's a gym in the building with showers, and you can wash up in the office restrooms.

As he tucked a cot under his workstation, Zhang Wenduo sighed, "If only I had my own private office like President Zhao, how convenient it would be to make up a cot!"

Suddenly he heard a voice behind him call out, "Dr. Zhang?" Zhang Wenduo paused.

Who's here so early? Is it President Zhao again?

When Zhang Wenduo turned around, he saw that it was Han Luyi, and there were two other people beside him. Zhang Wenduo remembered that they were Boss Han's secretaries.

"Oh, Boss Han, when did you come to Beijing?"

"I've been here for two days, but I haven't seen you yet. I heard some noise over here just now, so I came to check." Han Luyi pointed to the folding cot at Zhang Wenduo's feet and asked, "What's this?"

A slightly embarrassed look appeared on Zhang Wenduo's face.

He had mentioned to Zhao Wenyuan that his dormitory lease was about to expire. Although he hadn't asked for permission, he was now familiar with Zhao Wenyuan and didn't think that staying in the office for a couple of days was a big deal.

I never expected the big boss to see it.

Han Luyi was also puzzled at this moment.

How does Old Zhao usually manage things? Has our Origin culture become so chaotic? There's no need for him to sleep on the floor in the office. If this gets posted online, won't I end up with the same reputation as Elon Musk?

"My new apartment isn't ready yet, so I'll be staying in the office for a couple of days," Zhang Wenduo explained.

Han Luyi didn't comment, but changed the subject, saying, "I heard from Wenyuan that you've been researching something new recently. He didn't tell me in detail, but now that you have some time, can you tell me about it?" At this, Zhang Wenduo perked up.

"President Zhao tell you about the Social World Model?" He turned on his computer, typed a few words on the keyboard, and a lot of data appeared on the screens in front of him.

"Our idea is to use Glutinous Rice Ball's intention understanding to predict the behavior of people after they interact and influence each other. We've already achieved initial success with a scale of several hundred individuals. Here are the results," Zhang Wenduo said, pointing to several sets of data on the screen. "Boss Han, look, this is a virtual, closed town we created, set in the United States West at the end of the 18th century. Every character in it is an Agent. We let them live here for 10,000 rounds, each round representing one hour. During this time, they can move, interact, and influence each other. At the same time, there is a trained Social World Model to predict the final outcome."

Han Luyi look and said that although it was called a small town simulation, it was actually all text and some metadata, such as "The mayor went to the bakery to buy bread and the owner gave him a discount" or "The cowboy went to the pub to order drinks and got into a conflict with the bartender". The metadata was environmental information such as time and weather.

Zhang Wenduo excitedly displayed the interim results. "Boss Han, look, after several rounds of training, our Social World Model can now predict the townspeople's actions with relatively high success rates, achieving an accuracy rate of 82%."

Although Zhang Wenduo's concept was somewhat interesting, reminiscent of the American TV series "Westworld," the data presented was actually dry and uninteresting. It could only serve as part of an experiment and was even too uninteresting to read as a novel.

After reviewing it, Han Luyi asked doubtfully, "Where did this data come from? Without real-world data, does anything trained on it have any practical meaning? These Agent are just following Glutinous Rice Ball instructions—wait, are you using the Tangyuan Model? Or have you also integrated other models?"

Zhang Wenduo turned to look at Han Luyi, somewhat surprised by the boss's keen insight. How could he have spotted the biggest problem in the entire project with just one glance? And he hadn't even said anything yet.

"You're right on both points. What we lack most is data. Initially, we considered using real data, but the amount of data that can be collected in reality is extremely limited, and attribution is very difficult. It's hard to establish a correlation between what someone did and why they did it, let alone causation. So we came up with this 'clumsy' method: using a cluster of Agent to generate simulated data, and then using this simulated data to train Social World Model. This method can at least verify whether our direction is correct."

"The Agent also used various commercially available models, mainly to provide variables for this experiment. Otherwise, if we used the same model for everything, we were worried that it would cause the model to collapse."

Han Luyi shook his head after hearing this. Using data generated by the model to train the model is like trying to climb a spiral with one foot on the other; it's hard to say how effective the model trained in this way will be in real-world scenarios.

The model collapse mentioned by Zhang Wenduo is a typical problem. If the training data for the next generation of models all use model-generated data, the implicit biases in this data will be amplified, creating an echo chamber effect, or what could be called an information cocoon for the model. As a result, some capabilities of the next generation of models may deviate from reality, which is what is known as model collapse. However, data is indeed unpredictable; the real world is just that complex. Unless there is Sight to achieve accurate attribution, how can one obtain the causal relationship for a person's behavior?

However, Han Luyi does not intend to use the Sight to upload the data this time.

Leaving aside the question of what this so-called Social World Model can actually be used for and what its training will be useful for, even if we obtain the data using the Sight, it doesn't necessarily mean we can complete the training.

There is another very alarming issue here: how to explain the source of the data.

The previous Annotated Data, after all, had the original corpus available. Even if some people had doubts about the level of detail in the annotations, the information Han Luyi obtained was all inferred from the corpus and could be clearly explained.

However, when people attribute their behavior to other factors, the real reasons are often not in the information provided. The Sight can be seen and written down makes the data itself problematic.

So Han Luyi didn't bring up the data source issue, but instead asked, "So, why did you choose a Western town setting? Is it because there's more raw data of that type? It can't be because you like watching 'Westworld,' can it?"

End of Chapter
Prev
NovelSummaries
Next

Explore the wiki

Han Luyi
Zhao Wenyuan

Comments

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