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 446

The more, the better

1,938 words 6 min read 1 view 29d ago Source ​
Skip to the end of the chapter
Prev
Next

Haicheng, Origin Group headquarters, in the Chairman's Office.

Han Luyi sat in front of the computer, the video conference was connected, and Zhao Wenyuan's face appeared on the screen.

"Wenyuan, how have you been lately?" Han Luyi smile.

"Boss Han, everything is fine," Zhao Wenyuan replied.

On Friday afternoons, the two have a regular one-on-one meeting. There's no fixed topic. If there's any major work progress to report, they'll discuss work; otherwise, they'll talk about their strategic judgments, or even just chat about life.

This was an arrangement made by Han Luyi himself. He holds such meetings with each senior executive, and the frequency varies depending on their rank.

However, he attached great importance to the meeting with Zhao Wenyuan, mainly for two reasons.

Firstly, it's due to the importance of the Beijing Research Institute. Currently, all R&D work related to Large Model, the Interconnect technology team, and pilot projects in the Medical and power industries are all under the Beijing Research Institute, which can be considered the core of Large Model R&D.

The second reason is due to the unique characteristics of remote management.

Although it doesn't require overcoming time differences like multinational corporations, the feeling of working in two separate locations is completely different from working in one office.

When you're in the same office, you see each other all the time, and you often chat about small things when you bump into each other in the hallway.

In case of emergency, there's often no need for advance appointments; you can come anytime. You can even have lunch together, which can strengthen your relationship and allow you to talk about work.

However, working in different locations eliminates these conveniences. Each meeting has a fixed agenda, ends once the discussion is over, and relationships between people inevitably become distant.

Therefore, this meeting, which has a fixed time but no fixed topic, is particularly important.

"What's on the agenda today?" Han Luyi asked.

Zhao Wenyuan nodded. "Indeed, there are some things to report. We recently discussed the next stage of Glutinous Rice Ball's training plan. The current idea is to invest more Computing Power and increase the number of parameters based on the current capabilities. This is mainly based on our consideration of some shortcomings in Glutinous Rice Ball's current capabilities."

"Glutinous Rice Ball's main strengths at this stage lie in intent understanding and Long-Range Tasks planning, but it is not actually leading in logical reasoning and solving complex problems; it can only be said that it is not lagging behind. This is because the improvement brought by Annotated Data is mainly concentrated in intent understanding. To improve mathematical ability and problem-solving ability, the most important thing is to increase the number of parameters; otherwise, it will struggle to compete with top international models."

This raises the question of the iteration direction for Glutinous Rice Ball 2.0: should it compete with overseas models in terms of parameter size and join this arms race? And if so, where will the Computing Power come from?

Currently, most of the Computing Power on the Origin side is allocated to inference, leaving only a small portion for training.

This small portion was sufficient for post-training the Glutinous Rice Ball 1.0 dataset, but if we want to create a Glutinous Rice Ball 2.0 with larger parameters...

The existing infrastructure is definitely far from sufficient; where will the insufficient Computing Power come from?

A portion of the L100 in the Inference Cluster could be moved over for training, but this would inevitably affect the supply capacity of all product lines that rely on the Tangyuan Model —including Open Things, Riding the Wind, and Tangyuan Chat, as well as the publicly available Glutinous Rice Ball API.

Even without considering the growth of new users, the experience of existing users cannot be compromised; otherwise, the reputation will collapse in no time, and no amount of modeling capabilities can save it.

How should we weigh these factors?

Han Luyi asked, "How many L100s do you think Training 2.0 will need?"

Zhao Wenyuan had clearly already calculated it, and blurted out, "Ten thousand sheets."

After saying that, he quickly added, worried he hadn't said enough, "At least ten thousand, but the more the better."

Han Luyi this. There are only a little over 18,000 L100s in total, and this is more than half of them. Although other Computing Card clusters are gradually going online to take on some inference tasks, there is still a long way to go before they can replace these 10,000 L100s.

"Okay, I agree with the direction of improving the number of parameters and mathematical capabilities. You can start by planning the training scheme. As for the Computing Power, I'll figure it out," Han Luyi said.

When Han Luyi said this, he was actually having a headache. He didn't have enough Computing Power and didn't really have a solution for the time being.

Even with cheat codes, solving the problem is still so difficult.

Zhao Wenyuan, on the other end of the internet, was delighted to hear this. As expected of the all-powerful Boss Han, there was no question that could stump him.

At this moment, Han Luyi asked, "By the way, Wenyuan, I also have a few questions for you. The current capabilities of Glutinous Rice Ball 1.0 are actually a bit excessive in many scenarios. Shouldn't we consider training some smaller models, whether for internal use or for external sales, to provide more options?"

After listening, Zhao Wenyuan nodded and said, "You mean the Flash version, right? Releasing a lightweight version to increase sales volume after the flagship model is stable is indeed a common strategy in the industry. However, regarding Computing Power, Zhao Wenyuan didn't finish his sentence. Generally speaking, after the lightweight version is launched, it will attract more users who are more sensitive to cost-effectiveness due to its lower price. Although the unit Computing Power consumption is less, the total Computing Power required may increase again."

Moreover, the current Computing Power of Origin is already stretched thin.

"Computing Power for now, let's focus on building the Lightweight Model first," Han Luyi said.

Zhao Wenyuan nodded and continued, "By the way, Boss Han, I have two more things to report to you. One is about Zhang Wenduo's recent research. He's researching using models to extrapolate group relationships, which I find quite interesting. You can take a look when you have time. The other is a company that Jiang Jiusi introduced."

"Jiang Jiusi? The one who works on Medical AI?" Han Luyi asked. "I remember you gave a report on their project last time; it seemed to be going quite well. What did you talk about?"

"There's a Brain-Computer Interface startup called Mengnao Technology. They'd like to ask if you're considering investing. I just did some research; the founders and core technical team are all from top universities and research institutions, so their research backgrounds seem quite reliable. The company is based in Beijing, but if you're interested, their founders are willing to come to Haicheng to discuss it with you," Zhao Wenyuan said.

Brain-Computer Interface?

Han Luyi glanced at the screen; the browser on the desktop opened automatically, and the Tangyuan Chat interface popped up. A line of text automatically appeared in the dialog box. "What kind of company is Mengnao Technology?" The text input speed was extremely fast, far exceeding that of typing. Then Tangyuan Chat began searching and quickly provided the answer.

Han Luyi looked at the document, he thought to himself, "The Brain-Computer Interface really came to the right person to invest in me."

Putting aside everything else, he can offer a lot of help just in terms of user experience and product direction.

He dares to say that no one in the world has a better understanding of the experience of using Brain-Computer Interface than he does.

Han Luyi has recently become more and more accustomed to using the upgraded Sight to operate electronic devices. He can connect to any signal that accepts Bluetooth, UHF, or WiFi, so he rarely uses traditional input devices like mice and keyboards anymore. He only pretends to use them when others are around. It feels like someone who is used to driving sports cars suddenly going back to driving a tractor, which is extremely uncomfortable.

If Brain-Computer Interface can truly reach this level of Sight, there will definitely be no shortage of market demand. Han Luyi is very interested in this investment.

But the question remains: how many technical hurdles will need to be overcome? Is Mengnao Technology a suitable partner?

Pulling himself out of his thoughts, Han Luyi asked, "Have you seen it? What did you think of it?"

Zhao Wenyuan's expression showed a hint of embarrassment.

After doing his research, he realized that his preconceived notions about Jiang Jiusi were completely wrong.

The founder's surname is Meng, not Meng (孟), which is indeed rare, and that's understandable. But she's a woman, not a young talent, but a remarkable woman.

Zhao Wenyuan cleared his throat and said, "Jiang Jiusi showed me a video, and I've done some research on it as well. They've developed a helmet-style, Non-Invasive brain-computer interface that can move Robotic Arm easily. This technology is quite cutting-edge. While Non-Invasive is safer than Invasive, it also presents much greater technological challenges. I think it's worth discussing. As for investment, that's up to you."

Han Luyi nodded and said, "Okay, send me the information, I'll take a look too."

After the video ended, Han Luyi searched online for more information about Brain-Computer Interface.

Surprisingly, just now, Neuralink, a company invested in by Elon Musk, released a video, a pre-recorded presentation hosted by Elon Musk himself.

Han Luyi quickly watched the video and gained a new understanding of the progress in Brain-Computer Interface.

Neuralink uses Invasive technology, which involves implanting the device directly into the cerebral cortex. At the press conference, they stated that seven people had already connected to their device, with usage time reaching 100 hours per week.

Moreover, these users are already able to perform very complex operations through Brain-Computer Interface, such as typing, playing games, and operating external devices. Their interaction speed with the devices has even broken the world record that humans can achieve with a mouse.

However, this technology is not without its advantages.

Neuralink's seven participants were all patients with high-level paraplegia and Amyotrophic Lateral Sclerosis. One of the most important reasons for this was that such Invasive procedure was too risky, potentially causing permanent brain damage or even death. These terminally ill patients were willing to take such risks, given that their current quality of life was already poor enough.

However, for healthy individuals, the risk of implanting electronic devices through the skull and into the cerebral cortex simply to save their hands seems too high.

In other words, if Mengnao Technology's technology can truly achieve the same level of integration as Neuralink in Non-Invasive manner...

Similar results are not necessary, or even not. As long as it achieves a smooth interaction with the computer, it is already a great success.

Han Luyi glanced at the mouse and keyboard on the table in front of him; they were now basically just decorations.

The interaction method of the Sight is indeed easy to use, but if you need to have a hole cut in your head to use it...

Han Luyi gently touched the top of his thick head of hair and couldn't help but shiver.

It's probably better to avoid brain surgery then.

After watching the online video and reading the materials sent by Zhao Wenyuan, Han Luyi sent Zhao Wenyuan a message.

"Please contact Mengnao Technology for me. I'm going to Beijing next week to visit their company in person."

End of Chapter
Prev
NovelSummaries
Next

Explore the wiki

Han Luyi
Zhao Wenyuan

Comments

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