New Employees With an Annual Salary of 1 Trillion Won - Chapter 179

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

Chapter 179. Filtering (4)

Once all external problems disappeared, the short-form service entered a completely stable orbit.

Starting with sports videos, various content like news, daily life, and music exploded onto the platform.

Now the number of content that users voluntarily uploaded had increased hundreds of times more than the videos we directly produced and uploaded.

“CEO! The short-form videos have become a complete hit.”

“Faster than expected. Since the revenue structure is solid, content providers are flocking here rapidly.”

“Even people who used to rip illegal video clips have all moved over to our side. Bloggers and community operators are also migrating to Rollbook one by one. The revenue is several times more than what they made from their previous activities.”

In the end, it was about money.

From people who uploaded videos as a hobby,

to those who professionally operated communities – everyone flocked here seeing the profitability.

Of course, they didn’t come just for the money.

Uploading videos to Rollbook meant they could receive greater interest and attention, so it might have been a natural choice.

“As content increases, management becomes important.”

“That’s exactly why the review team has entered 24-hour emergency duty. From provocative adult videos to illegal advertisements, thousands are being uploaded daily.”

Having a lot of content wasn’t entirely a good thing.

Harmful videos that caused damage to Rollbook were also increasing proportionally.

“If such videos run rampant, Rollbook will eventually become a harmful platform.”

“That’s why we’ve deployed even the development team for reviews, but we can’t keep relying on manual work forever.”

“We need a system that does primary filtering with algorithms.”

“We’ve already tried applying basic review algorithms, but the effects are minimal. People who upload such videos know too well how to slip through.”

It was a battle of how sophisticated an algorithm we could create.

We needed a system that immediately expelled users who uploaded harmful videos.

To do that, we had to create algorithms based on the data we had compiled so far.

“Is harmful video data stored separately?”

“We’ve collected it all.”

“We need to create a system based on that. First, let’s apply a delayed publication system. Keep uploads private for 10 minutes after upload, and only make them public after passing primary filtering.”

“We’re already operating that way, but it’s causing too much overload.”

“Please bear with it a little longer. Let’s work with the development team to create a primary filtering system as quickly as possible.”

I visited the development team to prevent Rollbook’s overload.

We selected only the best personnel from over 300 development staff.

At the center were P-Project members including Chad Hurley.

“We need to create a filtering algorithm.”

“What approach are you thinking of?”

“A structure that automatically filters out rule-violating content by analyzing video titles, subtitles, color tones, and voice patterns.”

“So you want to extract harmful video data and apply all algorithms to it. That will take quite a lot of time.”

Time could be shortened with the number of developers.

If dozens of developers were simultaneously deployed for data extraction and pattern analysis, we could complete the basic structure in just one day.

“We need to stop uploaders trying to ruin our short-form service as quickly as possible. I’ll provide unlimited personnel support. Let’s complete it as fast as possible.”

“Of course. Do you know how hard we worked to create this? We can’t shut down when we haven’t even started the US service yet!”

Work progressed at tremendous speed.

Chad Hurley led the developers in extracting data.

I created algorithms based on that data.

For two full days.

We devoted ourselves to developing the filtering system without resting for a single moment.

And finally, the first result was completed.

“The filtering system is complete. Apply it right now!”

“Really? After reviewing tens of thousands of videos for two days, my eyes are about to fall out!”

Manager Kim’s eyes were bloodshot.

He wiped his flowing tears with his sleeve and activated the primary filtering system.

“It’s operating without problems! It’s already started finding harmful videos!”

“The speed is definitely fast.”

“The system is handling alone what hundreds of people would have to do! Now we can finally catch our breath!”

Thanks to the algorithm, the review team’s workload was drastically reduced.

But their happiness didn’t last very long.

This time it was only two days again.

Two days after applying the review algorithm, problems began occurring again.

“Bypass videos have started appearing.”

“Already?”

“They’re bypassing by moving subtitle positions, changing screen ratios, and slightly modifying colors or audio.”

We all let out sighs simultaneously.

At the same time, I looked at Chad Hurley and said one thing.

“Let’s go work.”

“We just need to extract bypass video data and update. This work shouldn’t take too long.”

We thought so lightly and dove back into work.

But even while modifying code, new bypass methods kept appearing endlessly.

Eventually, we had to stay in the development team office for an entire week, pouring all our time into blocking newly discovered variables.

“Once we apply this update, we should be able to block most bypass videos.”

“The frequency of bypass video uploads has definitely decreased significantly. It seems uploaders have also noticed that we’re responding in real-time.”

“They won’t give up like this, right?”

“I wish they would, but… they’ll probably come back with other methods eventually.”

Should I call it creative?

No matter how much we blocked them, illegal uploaders kept coming with completely different methods.

This was also because it was profitable.

Of course, illegal videos themselves couldn’t receive advertising fees, but they could get free advertising exposure effects even by uploading illegal ads, so they didn’t stop.

There was an even more terrible category.

Those who uploaded harmful videos for simple pranks or fun, causing damage to others.

They enjoyed destroying the platform for their own satisfaction.

“Let’s find and add all bypass methods that haven’t appeared yet.”

“I’ll scrape all data from overseas sites too.”

Manager Kim spoke enthusiastically.

But his eyes had already lost focus, and his body was half-tilted, about to collapse.

“Mister Lee, at this rate people will really collapse first. Let’s rest a bit and organize.”

“We should. We’ve bought some time with this update, so let’s catch our breath for a moment.”

“Everyone go home!”

The developers had been waiting for Chad Hurley’s words.

They practically crawled out of the office.

Only we and the night shift personnel remained in the review team, resting while leaning back in chairs.

“Were you all still here without leaving?”

“Min-jeong, haven’t you left work yet either?”

“I came to work late because I had school assignments left.”

Yu Min-jung looked at us pitifully.

But her appearance wasn’t much different either.

As her grade level increased, her lively appearance had disappeared.

Now she went around in comfortable clothes with a pencil as a hairpin, her hair in a bun.

“School assignments seem to be increasing.”

“Arts is originally like that. As grade levels get higher, assignments become bombs.”

“Engineering school is the same. Thinking back to university days, only the engineering building and arts building had lights on every night.”

Manager Kim spoke while reminiscing about the past.

Having done such all-night work during university days enabled him to handle work like this, and I nodded while listening to his words.

“Is the harmful video filtering work all finished now? I’ve spent too much time on that work too.”

“The color scheme issue was easily resolved thanks to the design team’s help. But there are so many different workaround methods that it’s hard to say we’re finished yet.”

“Then when will it be finished?”

At her words, everyone simultaneously lowered their heads.

Because it was work with no known end date.

No matter how excellent an algorithm and system we created, illegal uploaders would find ways around it.

“We’ll have to perform emergency updates every time new workaround methods appear.”

“So you’re saying it might never end.”

“….”

“Don’t illegal uploaders learn new methods from somewhere? Then couldn’t our system also learn on its own and block videos?”

It was Yu Min-jung speaking out of frustration.

But the developers knew well that what Yu Min-jung suggested had no feasibility whatsoever.

Chad Hurley in particular shook his head and explained in detail.

“Algorithms ultimately require people to input commands one by one. Some automation is possible, but it’s impossible for them to learn on their own and completely block new methods.”

“It appears often in movies. Isn’t that level possible yet?”

“It’s only possible because it’s in movies. The concept of machine learning is being researched, but it’s still far from commercialization. Even if it were commercialized, it wouldn’t be sophisticated enough to use for video filtering.”

Machine learning or deep learning.

It referred to artificial intelligence where computers learn on their own.

It appeared frequently in science fiction movies, but it was still technology only possible in dreams.

But was it really impossible?

The moment that thought crossed my mind, a massive structure revealed itself somewhere in my head.

Countless gears began rolling simultaneously.

“Chad, what would happen if sophisticated machine learning were commercialized?”

“What would happen? It could be used beneficially in many places. Not just video filtering, but various industries, especially the financial industry. It could analyze all existing financial data and produce results.”

“Literally every industry could use it.”

“Most importantly, once completed, the review team wouldn’t have to struggle like this anymore.”

It was clearly technology impossible to implement right now.

But what if Tiger Fund’s massive financial resources and Rollbook’s development personnel were invested?

Technology that would take 20 years might become reality in 10 years, or perhaps even less time.

“I should seriously consider this.”

“Mister Lee! Are you really going to invest in artificial intelligence?”

“CEO, AI is treated as a dead field of study in academia. No one invests in it, and very few people research it.”

Chad Hurley and Manager Kim simultaneously poured out negative comments.

But I couldn’t give up.

Because I had already seen the gears that started rolling in my head.

“Let’s just look into it for now. Please find AI research teams for me.”

“I’ll check right away. All paper data is linked to Omnis, so I can search quickly.”

Manager Kim entered the keyword ‘artificial intelligence.’

Normally, hundreds or thousands of papers would appear at once, but this time was different.

AI-related papers were few enough to count on one hand.

And even those were mostly papers written by one research team or results citing those papers.

“It seems Professor Jeffrey Hinton at University of Toronto is researching artificial neural networks that mimic the human brain, and that’s about it. Other universities are conducting related research too, but there don’t seem to be any major results.”

“Professor Hinton at University of Toronto. Does anyone know him?”

Everyone shook their heads.

Their reaction suggested they were unfamiliar not only with the professor’s name but even with the term artificial neural networks.

“I should meet him directly.”

“You’re going to go all the way to Toronto yourself?”

“How many hours does it take to get to Toronto?”

“Flight time alone is 14 hours, and if you add all the travel time, you should expect a full day. And there’s a 14-hour time difference too.”

The moment I heard that, I broke out in hives all over my body.

Just the thought of wasting time on an airplane or inside the car was horrifying.

“Let’s invite Professor Hinton to Korea.”

“We could ask Professor Park. If we request it as a Korea University guest lecture, we could naturally invite him.”

The butterfly effect created by video filtering.

I couldn’t know where its end would lead.

But one thing was certain. A small wind had begun to blow.

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