New Employees With an Annual Salary of 1 Trillion Won - Chapter 180

—————
This chapter was translated by Lunox Team. To support us and help keep this series going, visit our website: LunoxScans.com
—————

Chapter 180. Filtering (5)

Korea University, the representative university of Korea.

Especially Professor Park, who served as the head of the Computer Science department, had considerable connections.

Thanks to this, numerous overseas scholars and famous figures held invited lectures at Korea University.

Professor Hinton also readily accepted his visit to Korea through that connection.

“The core of artificial intelligence is data. It’s not simply about how much you collect, but like the human brain, it must be able to recognize patterns and learn on its own to be applied to various fields.”

I also sat quietly in one corner of the lecture hall, listening to his words.

And every time he spoke, each sentence penetrated deep into my heart.

I had been focused only on data collection until now.

Of course, I analyzed and utilized data through Omnis, but I hadn’t reached the concept of ‘self-learning.’

But if I could implement the picture that Professor Hinton was drawing?

Omnis would evolve hundreds of times more than now, and might even be able to pattern all the world’s data on its own.

“It may seem impossible now, but someday an era where machines think and judge on their own will definitely come.”

The short lecture came to an end.

Several students poured out questions, but most were SF-related questions.

Professor Hinton, seemingly accustomed to such situations, explained by drawing analogies to movies.

But I didn’t want to hear such science fiction-like Q&A.

So I threw a question I had been thinking about alone to the professor.

“To develop artificial intelligence, we need to collect and train vast amounts of data. But with current methods, the analysis speed is too slow, so no matter how much data we collect, won’t there eventually be a bottleneck?”

“That’s a good question. You’ve accurately identified the biggest problem in AI development. That’s right. With current CPU speeds, it’s difficult to properly implement artificial neural networks. But faster processors continue to be developed, so someday we’ll overcome that limitation.”

Professor Hinton was being honest about the current situation.

But I didn’t stop there and dug deeper.

“No matter how fast CPU processing speeds become, it will take at least 10 years, or as long as 20 years, to reach a level capable of implementing artificial neural networks. Then won’t the emergence of practical AI be postponed for decades?”

“That’s a good point. But scientists have always found answers. The limitations of artificial intelligence will be overcome before long. I’m betting my life on finding that answer too.”

With this question, Professor Hinton came down from the podium.

He shook hands with Professor Park and exchanged a few greetings, then pointed toward me and said something.

Soon Professor Park gestured toward me, and I approached and sat next to them.

“He’s not a Korea University student, but someone I’ve been teaching exclusively. He’s been collaborating with global companies and producing good results.”

“Nice to meet you. I’m Lee Jung-hoo. Please feel free to call me Mister Lee.”

“Nice to meet you. It’s been a while since I’ve met someone who showed such interest in artificial intelligence, so I wanted to have a conversation.”

A professor was a professor.

He was delighted like a child just by someone showing interest in the field he researched.

“I’ve been feeling the necessity of artificial intelligence acutely lately.”

“Artificial intelligence in movies often looks like a magic lamp. But even if AI is actually developed, it won’t be that omnipotent.”

“I don’t wish for a magic lamp. I just need something that can fill in human shortcomings. An entity that excludes emotions and variables, moving within perfect rules. That’s the artificial intelligence I desire.”

What I wanted was simple.

Perfect patterns, perfect rules.

And accuracy that wouldn’t make a single mistake even running 24 hours a day.

If only that were possible, I could create a much more perfect routine than now.

“Everyone has different expectations for artificial intelligence. But to reach that stage, we first need to increase the processing speed of artificial neural networks. If we can’t solve that part, it’s hard to move even one step forward.”

“There’s a way to solve that problem.”

“You seem to have a method in mind. I’d like to hear what kind of method it is. Shall we move to a quieter place to talk?”

Professor Hinton asked out of simple curiosity rather than great expectations.

But I declined his suggestion and pointed to the podium.

“I’d like to move after the next invited lecture ends. The solution I’ve thought of is deeply connected to the next lecture content.”

“Really? Then let’s listen to the next lecture together. We still have plenty of time.”

Professor Hinton wasn’t the end of the invited lectures.

Using Professor Park’s and my connections, we invited another speaker, and a young company CEO in casual clothes stepped onto the podium.

“Nice to meet you. I’m Jenson Huang, CEO of Nvidia.”

The other invited speaker was Jenson Huang.

The reaction wasn’t enthusiastic, perhaps because not many students knew him.

But he enthusiastically lectured about the new future that Nvidia envisioned.

“Many people think of graphics processors as just ‘chips for games.’ But that’s only half the truth. The applications of GPUs are limitless!”

Professor Hinton looked at me.

His eyes contained the question of whether CPU problems could be improved with GPUs.

Instead of answering, I looked at Jenson Huang, who continued speaking.

“GPUs aren’t just auxiliary devices used only for graphics work. Someday they’ll become a new architecture capable of processing even CPU calculations in parallel. And through GPUs, I believe personal computers will soon be able to perform on par with supercomputers.”

It was still close to fantasy.

GPUs hadn’t been around for very long.

They hadn’t yet shown proper computational performance.

Moreover, Nvidia’s situation wasn’t all that good.

While they were evaluated as having faster processing speeds than competitors.

They had many weaknesses in other areas like heat generation, power efficiency, and driver stability.

In fact, market share was also declining with continued poor sales.

Nevertheless, Jenson Huang was confident.

He raised his voice with all his passion to convey even a little more of the vision he was drawing.

“Over the past 20 years, computer processing speeds have become more than a thousand times faster. Thanks to this, human life has changed. But now visible innovation has stopped. Nvidia’s GPU will change that stagnant flow!”

The lecture ended, but the applause wasn’t loud.

Instead of the students who didn’t move their hands, only I clapped enthusiastically.

Seeing this, Jenson Huang smiled and approached us, expressing his gratitude to Professor Park.

“Thank you for proposing the invited lecture. It was a good opportunity to convey Nvidia’s vision to Korean students.”

“Haha, I also wanted to hear CEO Jenson Huang’s speech. Actually, the person next to me asked me several times to invite you.”

Professor Park naturally introduced me.

Jenson Huang looked at me, then widened his eyes as if remembering something.

“Are you perhaps affiliated with Tiger Fund? I think I passed by you once at Daehyeon Semiconductor.”

“Let me formally introduce myself. I’m Lee Jung-hoo, in charge of Tiger Fund Korea Branch.”

“I knew it! I’d only heard rumors, but now I get to meet you directly. Thanks to you, we’re maintaining a good relationship with Daehyeon Semiconductor.”

Jenson Huang warmly offered a handshake.

Daehyeon Semiconductor, which stably supplied semiconductors of desired performance at low prices, was a precious partner to him.

And Tiger Fund, which owned that Daehyeon Semiconductor, was an indispensable core partner for Nvidia.

“If you don’t have any other schedule, could we talk together for a moment?”

“Actually, I do have one more schedule. I’m supposed to hold a GPU presentation at Yongsan Electronics Market. I have about an hour free, so if it’s okay, I’d like to join the conversation during that time.”

Jenson Huang was a hands-on CEO.

He was personally heading to Yongsan Electronics Market to plan GPU promotion and coordinate schedules.

He was that desperate, and I could feel his determination to somehow make the vision he was drawing into reality.

“It will be a better time than the schedule at Yongsan Electronics Market. And I’ll make sure not to keep you too late.”

“In that case, I’ll stay for a while. Thank you for the invitation.”

We moved locations.

When we moved to the lab room that Professor Park had prepared in advance, there was a large table in the center.

“Well then, have your interesting conversations. I have a next class, so I’ll excuse myself first.”

“I’ll come see you when the conversation ends.”

“Haha, it’s regrettable that I can’t join such a conversation, but a professor can’t skip lectures. Well then, see you again soon.”

Professor Park left us the space.

I began to bring up the story I had wanted to tell Professor Hinton and Jenson Huang in earnest.

“First, the reason I brought you two together is to discuss how we can increase the speed of artificial intelligence development.”

“Haha, so it turns out it wasn’t Professor Park but Mister Lee who called us here.”

“I thought you might feel burdened if I invited you in Tiger Fund’s name, so I used a little trick.”

We started the conversation playfully.

But the content of the conversation flowed seriously from the beginning.

“So how do you plan to increase the speed of AI development?”

“Ultimately, the core issue is computational speed. Tiger Fund has a platform called ‘Rollbook’ under its umbrella, which recently started a video service. The problem was that rendering alone took dozens of hours, making service expansion difficult.”

“Did you happen to solve that problem with GPUs?”

Jenson Huang showed great interest.

Instead of answering, I played a prepared video.

“As you can see, we utilized GPUs for certain computations and drastically reduced rendering speed.”

“Was it immediately applicable?”

“It wasn’t easy. Since current GPUs have difficulty with general-purpose parallel computing, we created a separate system and modified shader operations. As a result, we were able to improve both compatibility and performance simultaneously.”

“So you mimicked parallel computing.”

Mimicking was the best we could do.

Unless we manufactured GPUs ourselves.

With current GPUs, this was the limit, but even that alone could achieve tremendous performance improvements.

“Even though we only mimicked it, we achieved this level of performance. But if general-purpose parallel computing becomes actually possible, it could be dozens of times faster than now.”

“Amazing! We’re also trying to find that method, but this is the first time I’ve seen a place achieve such results with their own system.”

“It’s also thanks to Nvidia GPUs’ inherently fast basic computational speed. Their computational capability is definitely superior to competitors’ products.”

This was both Nvidia’s strength and weakness.

A product that boldly sacrificed other functions to extremely boost computational speed.

While shunned by the gaming industry, it was actually the optimal structure for us.

“Thank you for the kind words. But honestly, this GPU is close to being a failure. The moment the gaming industry turned away, market demand dropped sharply.”

“I’d like to fill that insufficient market demand.”

“Do you have a place that would use GPUs in large quantities?”

I turned my head to look at Professor Hinton.

He was still staring intently at the rendering video on screen.

When I called his name, he slowly turned his head to look at me.

“Professor, how about trying to use GPUs for artificial neural networks as well?”

“Using GPUs… there’s certainly potential to increase data processing speed. But since I’ve never tried it, it’s hard to predict what the results would be.”

“Current artificial neural networks have CPUs process calculations one by one. So even if the learning volume increases slightly, speed drops dramatically. But GPUs can process thousands of operations simultaneously.”

GPUs could fill the limitations of CPUs.

I had already confirmed this possibility primarily through developing the rendering system,

and I could be confident because the gear structure in my head was gaining more and more reality.

“Would that be possible?”

“If we apply this structure directly to neural network learning, we should be able to distribute calculations in parallel for each neuron.”

“That would be great if possible, but GPUs are graphics-dedicated chips. Would compatibility be feasible?”

“I’ll help with that part. Instead of directly controlling the GPU, we can design an intermediate system to translate the operations processed by the GPU. That way, neural networks can use them sufficiently.”

Jenson Huang, who had been listening quietly, slowly nodded.

Then he summarized the entire discussion in one sentence.

“You’re talking about using GPUs as a second brain.”

“Exactly. And it’s not simply having one more brain. If we apply this method, we can create a structure where hundreds, no, thousands of brains operate simultaneously.”

“Then machine learning speed would increase dramatically.”

Professor Hinton was also starting to understand.

To completely capture his heart, I pulled out a calculator.

“Even conservatively, it would increase by more than 20 times, and with optimization, over 50 times is possible. And just bundling two or three cards could achieve triple-digit speed increases.”

“At minimum 10 times to maximum over 100 times.”

“Ten years’ worth of computation could be completed in just one year or even one month.”

Professor Hinton’s eyes changed.

But he was a cautious person and needed some time before making a decision.

So Tiger Fund directly stepped in to extend the flight schedule and even arranged new accommodation.

—————
This chapter was translated by Lunox Team. To support us and help keep this series going, visit our website: LunoxScans.com
—————