Chapter 149: Hyper-Personalized Community
Chilbong.
Including both before and after the regression, he is the designer I trust the most.
He is the one who successfully resolved numerous early issues with Quantum.
When he first joined as a new employee, he still had the air of a recent graduate.
Perhaps due to leading numerous large projects, he now shows glimpses of the demeanor he had when overseeing the Ahsung Comprehensive Research Institute before the regression.
“Is it… could it be that you drew this yourself, sir?”
Even Chilbong, who rarely seems surprised,
was taken aback when he saw the AI semiconductor concept I had sketched on the whiteboard.
“The structure is quite different from the semiconductors currently being developed at Ahsung, isn’t it?”
“Quite different is an understatement. I knew you were planning to enter the semiconductor business, but what kind of semiconductor are you trying to create?”
Chilbong asked directly, and I answered straightforwardly.
“An AI semiconductor with overwhelming performance, capable of toppling Ahsung’s semiconductors.”
As I stated this firmly,
Chilbong’s pupils dilated once more.
Knowing Ahsung’s technological prowess better than anyone, he grasped the weight of my words.
“But you said AI semiconductor, not just a regular one. What exactly is that?”
It was certainly unfamiliar territory for Chilbong.
Even in industry and academia, the concept of AI semiconductors began gaining attention only next year, in 2015.
“An AI semiconductor is, quite literally, a chip specialized for artificial intelligence computations. It’s a semiconductor with an optimized computational structure for learning data, predicting patterns, and performing inferences.”
I picked up a marker and briefly added a neural network structure and computational flow diagram.
“Conventional APs simply repeat predefined computations like executing apps or displaying screens based on user input, but AI semiconductors must make judgments and react according to situations. It’s not just simple calculations; the computational patterns themselves are variable.”
Chilbong stepped closer to the whiteboard, staring intently at the diagram I had drawn.
“So… is this structure intended for the AP in a smartphone?”
“That’s part of it. The AI AP will serve as the brain of Quantum. But there’s something even more important.”
I paused for a moment before continuing, looking at Chilbong.
“A dedicated AI semiconductor for interactive artificial intelligence. That’s my real goal.”
Chilbong’s expression grew more complex, and I couldn’t help but smile.
“I’m not entirely sure what a dedicated interactive AI semiconductor is, but I can tell it’s a far more intricate structure than an AP.”
I nodded.
“A dedicated interactive AI chip isn’t just about processing commands. It needs to interpret natural language, understand context, and modulate responses. It must accept not only text-based commands but also voice, gaze, and gestures as input.”
I turned on a small monitor next to the whiteboard and pulled up a pre-configured system block diagram.
Two chip designs appeared side by side on the screen.
[Quantum AI-AP]
[Project LUNA: Dialogue AI Core]
“The one on the left is the mobile AI AP for Quantum, and the one on the right is the dedicated interactive AI chip for the AI platform. As you can see, their purposes and optimized structures are entirely different.”
Chilbong took a deep breath.
“…So, you’re developing these simultaneously? In my opinion, even one of these would take years with a typical design team. Especially the one labeled Project LUNA—the dedicated memory bus structure for the language model alone is extraordinary.”
“That’s why I’m not dividing the design team.”
“…Excuse me?”
“I’ll handle it myself. I’ll take charge of the overall architecture design, and you’ll handle the computational structure and interface. I’ll provide separate documentation for internal voltage regulation, inter-cluster communication, and cache allocation structures. You can use that to develop the test flow.”
Chilbong seemed at a loss for words for a moment.
“But… even if we design it, without the equipment to manufacture it, it’s all just pie in the sky, isn’t it?”
Anticipating his question, I opened an additional blueprint.
It was a concept diagram of the testbed created by Team Leader Park Sang-bae.
“Equipment development began a month ago. I’ve instructed Team Leader Park to develop EUV, deposition equipment, etching machines, and CMP equipment for the 2nm process line. The plan is to complete the test line within three years and transition to mass production within five.”
Chilbong couldn’t help but exclaim.
“You’re serious. Everything is actually in motion!”
I nodded and opened a notepad to add some notes.
[Phase 1]
-
Smartphone AI AP development (Quantum AI-AP)
-
Computational optimization + power efficiency priority
-
On-device voice recognition and simple dialogue response processing
-
Specialized structure for Lunar + KTalk integration
[Phase 2]
-
High-performance interactive AI chip development (Project LUNA)
-
Natural language understanding + context comprehension + long-term memory
-
Capable of running ultra-compact LLM
-
Cloud + edge + server compatibility
[Phase 3]
-
AI chip integration structure testing and mutual interface design
-
Establish data sharing and feedback structure between AI AP and LUNA
-
Transfer data learned on the smartphone to server-based LUNA for continuous learning
※ Final goal: Enable high-performance interactive AI on smartphones and transition to physical AI
“Everything I’ve written here is about the semiconductor I want to create.”
To overwhelm Ahsung with superior technology, I must set the standard from semiconductors to interactive AI platforms.
Chilbong carefully examined the neural network processing unit structure I had drawn and nodded.
“As you said, I’ll take charge of the computational cluster configuration and interface design. But…”
“But?”
“The physical AI you mentioned last… the artificial intelligence that speaks, thinks, and assists like a human. I want to handle the design of the chassis that will house it.”
Hmm.
His unexpected response made me chuckle.
Of course, I had no reason to refuse.
“Alright. You can handle the design of Luna’s body!”
I’ll take care of the head…
＊＊＊
Dojin Soft.
I lightly knocked on the door of Han Yu-ju’s office.
As I entered,
Han Yu-ju looked up in surprise, rising from her seat.
“CEO, what brings you here without notice?”
“I didn’t want to bother you if I called ahead. I heard you’ve been extremely busy lately. How’s the work going?”
“It’s tiring, but still fun.”
Her desk was cluttered with papers, suggesting she was too busy to even tidy up.
Still, her expression didn’t seem too bad.
I sat down and spread out a neatly organized blueprint and report on the table.
“This is the data center construction plan I discussed with CEO Park Cheol-jin. I’d appreciate it if you could review it and let me know if any changes are needed.”
Only then did Han Yu-ju realize the purpose of my visit and nodded.
“Executive Vice President Hong Chung-ki mentioned it. You secured the 20 trillion won funding for the data center construction yourself, right?”
“I said I’d take responsibility for securing the funds. So, Team Leader Han, just focus on constructing the data center we’re aiming for. CEO Park Cheol-jin has also agreed to help as much as possible.”
With so many internal projects underway,
Dojin Frontier had grown to the point where it could be called a conglomerate.
“You still plan to divide the data center construction into three phases, right?”
What was agreed upon in the last meeting was:
Phase 1: Three domestic sites
Phase 2: North America, Europe, and Japan
Phase 3: Other Asian and emerging markets
The plan was to construct a total of 15,000 PB of data center capacity within three years across these three phases.
“Since we’ve secured sufficient funding, there’s no need to divide the construction into three phases.”
During the meeting, it was because of insufficient construction funds.
With DJ Capital’s investment securing 20 billion USD, if necessary, we could consider proceeding with phases 1 to 3 simultaneously.
“However, we need to carefully consider whether a 15,000 PB scale is immediately necessary.”
“To be honest… even considering the monetization models currently being pursued, I’m not sure if a 15,000 PB scale is immediately necessary.”
Han Yu-ju chose her words carefully.
It was a reaction I had anticipated.
In my opinion, since we’re still in the early stages of monetization, 15,000 PB didn’t seem necessary yet.
“Then, what scale do you think is appropriate?”
“Considering the various services we’re currently pursuing, including the free cloud storage we plan to offer, it’s right to start with Phase 1, the domestic data center, immediately. Also, with Lunar’s global users growing explosively, we could expedite part of Phase 2, North America or Europe.”
Han Yu-ju clearly stated her opinion.
However, I had a slightly different perspective.
“By the way… have you reviewed the materials I sent?”
“Materials?”
Han Yu-ju looked puzzled.
“The AI semiconductor development documents.”
“Ah, the materials related to the interactive AI platform?”
“Yes.”
“I read them, but… is it really possible? An artificial intelligence that can reason and speak like a human?”
Han Yu-ju seemed skeptical about its feasibility.
I shook my head.
“It’s entirely possible. That’s why we need to realize it faster than anyone else.”
“…So, you’re planning to fully expand the data center for that?”
I nodded.
“Specifically, realizing interactive AI requires massive training data and computational resources. While the user data from Lunar and KTalk is substantial, it’s not enough.”
“…So, you’re saying we need additional data sources beyond KTalk and Lunar?”
“Exactly.”
I opened my laptop and showed Han Yu-ju a document I had briefly organized.
At the top of the document was a large title.
[Project Z: KTalk-based SNS Platform Construction Plan]
“…SNS?”
Han Yu-ju’s eyes widened.
“I’m planning to create an SNS platform based on KTalk and Lunar users. We need to collect text, image, video, and audio data, as well as secure a complex dataset including user interactions.”
“But the current SNS market is already dominated by global companies like Faderbook, Outstagram, and Twister. Is entry even possible?”
“That can be assessed with numbers.”
I flipped through the laptop screen to show the organized statistics.
“Looking at global SNS users last year, Faderbook had about 1.3 billion, Twister about 300 million, and Outstagram about 200 million. Our KTalk surpassed 300 million users last month, and Lunar’s OS market share exceeded 70%.”
“The gap isn’t as large as I thought.”
“Exactly. Especially since our users are mobile-centric and familiar with messaging and communities. Plus, we’re not creating a standalone SNS; we plan to integrate it as a default app within Lunar and KTalk. I think there’s enough potential.”
For KTalk users, there would be no need to sign up or log in; they could use it immediately through an update.
Han Yu-ju took a deep breath.
“The diffusion speed would indeed be tremendous. Existing SNS platforms rely on advertising or viral marketing to attract users, but we wouldn’t need that.”
“Exactly. Especially since Lunar can integrate SNS functionality at the system level, concepts like sharing, feeds, and following would feel much more natural.”
What I wanted wasn’t just a feed-based SNS, but a ‘hyper-personalized community’ where the entire user experience is interconnected.
“With that level of functionality and usage, the current data center scale definitely won’t last long.”
“That’s why I’m planning to proceed with the three-phase data center construction plan simultaneously.”