anwhile, as the storm called AI was sweeping through the industry, I was focused on another mission.
Specifically, tracking clues about the "madness" switch of the Castleman disease.
The place I visited today was one of the biotech investnt firms.
I had previously entrusted them with decoding Milo's sample.
"Have the test results co out?"
Milo's sample contained important information about the "madness" switch, but at the mont, it was locked away in a secure vault.
It ant that we could only access that information after unlocking countless locks placed on the vault.
The only key to solving this complex problem was Spatial Transcriptomics.
It’s an advanced technology that deciphers gene expression at the cellular level in spatial context.
To that end, I had invested in a startup specializing in the technology, funded their acquisition of the industry-leading Swedish firm, and thanks to that, a prototype analyzer had already been produced.
We tested Milo’s sample using it,
"The data has been secured, but the core signal causing the pathological pathway has yet to be identified. To interpret it, we need an analytical frawork capable of systematically identifying and classifying high-resolution molecular patterns."
The conclusion was that proper analysis was difficult since what we had now was only raw primary data.
For the next step, we needed to filter out aningful data from thousands of pieces of information...
"But the required computing power hasn’t been achieved yet..."
The problem was computational power.
"The Parser architecture GPU I secured for you isn’t enough, is it?"
"Correct. Spatial transcriptomics data doesn’t just deal with expression levels, but also spatial-temporal interactions between genes based on location. This requires both image processing and graph computation simultaneously, and with the Parser, bottlenecks occur."
In short, current technology still falls short.
However, this was sowhat expected.
Why do you think I’m waging this AI war in the first place?
Because there are limitations to current technology, and I’m trying to push those boundaries forward.
Still,
'I thought we might at least be able to extract so information...'
I was rely holding on to a small hope.
Anyway.
After talking with the researcher, it beca clear what exactly we needed.
'For now, there are three things we must secure.'
First, we needed a better-performing GPU.
In vault-opening terms, this would be 'brute force' — pure strength.
Massive computing power was akin to the physical muscle needed to move the vault and turn the handle.
Next was GNN.
This acted like a stethoscope.
A tool essential for listening to the vault, identifying its complex internal structures, circuits, and locking patterns.
Lastly, Ignus was like a pair of tweezers.
A precision tool used when working delicately within that complex structure.
GPU, GNN, Ignus.
To open Milo’s vault, all three were essential.
'Well then... maybe I should start with the easiest one.'
***
I decided to first tackle the GPU problem.
The reason was simple.
'Because it’s the only one I can act on right now.'
With GNN and Ignus, I couldn’t even begin developnt — I didn’t even know where the developers were.
I had to locate them first...
Even for a regressor, tracking down the whereabouts of obscure developers was no easy task.
Of course, there were ways...
But that was sothing to worry about later.
For now, it was more important to flip over the sure cards already in my hand.
So, I imdiately convened an Envid board eting and got straight to the point.
"I’d like the next-generation series, 'Bolton,' to be moved into the production cycle as soon as possible."
Silence filled the conference room.
One director tilted their head, another blinked slowly while tapping the table with their fingertip.
After a brief pause, CEO Jackson cautiously spoke.
"The Parser architecture has only been released... not even two weeks ago."
"Yes, I’m aware. But I believe it would be best if the next product is released within Q3 of this year."
I didn’t say it outright, but their puzzled expressions said it all.
If I were to put their expressions into words, they’d probably read:
'Is this guy seriously insane?'
Honestly, it was a natural reaction.
Just like I once explained using the iPhone analogy, upgrading too frequently tends to do more harm than good.
Besides, to get it out by Q3, the tiline itself was absurdly rushed.
The new product had just been released — why rush things?
After a short pause, CEO Jackson interlaced his fingers and slowly spoke.
"Is there a specific reason we need to hurry this much?"
"What other reason would a company have to act quickly if not for revenue? The market is exploding with demand just as I predicted. If we move now, we can maximize our profits."
However, Jackson didn’t seem convinced.
"I acknowledge there’s demand, but the current situation doesn’t match our projections. This appears to be a temporary 'gold rush' triggered by a lone pioneer’s provocation."
Stark had certainly stirred the industry with his attention-based paper, but Gooble hadn’t made a move yet.
From the outside, this was seen more as one man’s show than an actual war.
"Of course, we can’t ignore the current demand. But gold rushes often cool off quickly. There’s no solid proof that this demand will continue through Q3. If this is just a temporary spark—"
"No. Demand will grow. Soon, Gooble and Stark will collide head-on, splitting the market in two and launching a competitive race for resource acquisition."
They didn’t believe .
Well, that’s the way the world works.
No one believes you, even when you’re telling the truth.
"As I’ve said before, we act based on facts, not predictions. And right now, we don’t see any such evidence."
I didn’t argue further with words.
Instead, I smiled and said,
"So you’re saying, if I show you proof of war, that would change things?"
My words made CEO Jackson and the other board mbers flinch slightly.
I smiled and continued.
"Then I’d like to reconvene the board in three weeks. Let’s revisit this issue then."
"Three weeks? That’s not enough ti to gather substantial evidence."
"No. Three weeks is plenty."
***
At this mont, Gooble was likely preparing for 'war.'
Naturally, there’s no way they would recklessly charge in without a plan.
They were definitely preparing to deliver a precisely calculated blow.
But since I had regressed, I already knew exactly what that “one blow” from Gooble was going to be.
'It’s the match between AlphaGo and Lee Sedol.'
If my mory serves right, that match was scheduled to take place roughly a month from now.
So technically, all I had to do was wait for that day to co…
But being on borrowed ti, I couldn’t afford to just sit around and wait.
Then the question beca: how could I make Gooble move faster?
The thod was simple.
'I just needed to light a fire under their feet.'
So I planned a new event.
“We call this structure a Transforr Architecture, and a large language model trained on this architecture is called an LLM, or Large Language Model.”
At this event, Stark took the previously introduced concept of the attention chanism a step further, presenting LLM as a new paradigm.
And this ti, he prepared a demonstration of the LLM model in a form the audience could see for themselves.
At a glance, it may have looked like a typical chatbot, but Stark firmly drew the line.
“This is fundantally different from an ordinary chatbot. A chatbot rely spits out pre-programd answers. But an LLM thinks and generates its own responses. Is there anyone brave enough to test it?”
Stark invited a randomly selected volunteer onto the stage.
Then he gave the person a special request.
“Ask the most outrageous, imaginative, and unique question you can think of. Sothing so unexpected that we could never have prepared an answer for it in advance. That way, the LLM will be forced to think and respond on its own.”
The man pondered for a mont, then finally asked the question.
“Explain what heartbreak is to a refrigerator.”
It was an absurd question.
A question so bizarre that no one would believe Stark’s team had predicted it or prepared a response ahead of ti.
Now the real question — how would the LLM answer such a difficult prompt?
And what ca out of the LLM left the audience stunned.
[Have you ever thought that when the refrigerator door closes and the light goes out, it might never open again? That’s the pain of heartbreak. Even though I perford my duties perfectly, they left and never ca back……]
The audience erupted instantly.
Laughter, applause, and cheers filled the room.
Honestly, no one had expected much of a difference compared to previous chatbots.
Up until now, chatbot technology had been more or less the sa.
So everyone assud this would be another rigid, chanical reply based on preset patterns…
But what the LLM showed surpassed all expectations.
It delivered a creative answer that seed to understand and empathize with human emotions.
That response went beyond simply conveying information — it emotionally resonated with the audience.
So even said the LLM surpassed human creativity.
While everyone was in awe of the LLM’s amazing performance,
Stark spoke again with a clear, powerful voice.
“This is the infinite potential of LLMs! LLMs don’t just mimic human language. They understand emotion and context holistically and reproduce human thought perfectly. This will bring about a revolution in AI technology!”
Then he lowered his voice and put on a slightly regretful expression.
“Until now, AI developnt was trapped in the old frawork of Reinforcent Learning, or RL. That’s why it responded chanically and could only make limited judgnts within simulations. But now, things are different!”
Stark looked out at the audience with blazing eyes.
He clenched his fist tightly.
“What we are pursuing lies on an entirely different level. LLM aims for a realm that RL could never reach — the point where AI truly thinks like a human!”
Why did he bash RL and hype up LLM so hard? The answer was obvious.
Because Gooble had long focused its efforts on Reinforcent Learning-based AI technology.
But now, through this demonstration, Stark essentially declared that RL was a relic of the past and LLM was the true future of AI.
Stark’s declaration, along with the “refrigerator explanation” mont, quickly spread across social dia and the press.
The subject was so intriguing that it didn’t just reach the tech industry or Wall Street — it penetrated deep into the general public, far beyond the AI space.
‘This should make Gooble move faster.’
If this kept up, RL would be seen as outdated and obsolete.
Gooble now had no choice but to prove RL’s superiority — and fast.
To do that, they would need a massive “event.”
And my prediction hit the mark.
Just a few days later, Gooble rushed to announce so breaking news.
User Comments
0 comments from readers