A Wall Street Genius's Final Investment Playbook-Chapter 272 : The Invisible Hand (7)

Meanwhile, tension gripped the upper ranks of Gooble.

It was all because Stark had abruptly released the LLM used in the “Fridge Demo.”

Its name was “MindChat.”

[The infrastructure isn’t quite ready yet, so we can’t officially launch the service. Think of this as a trial experience. Also, due to the immense computational power required to process each question, we kindly ask for your understanding that the service must be paid.]

Five dollars per question.

Even with the limitation of only one question per person per day, the traffic never stopped.

SNS was flooded with user experiences, and the public was ecstatic each time.

Write me a breakup text.

We’re overheating just from booting up now.

We tried to optimize for each other, patched for stability multiple times, but in the end, we’re just incompatible systems. At some point, we both stopped updating and backing up.

Let’s just log out here.

—The fridge was an F type, but this one’s definitely a T...

—My CPU froze over in the meantime;;

—Should’ve prefaced it with “To a human”... this one’s on the user

—Not bad? I saved this. Gonna use it someday.

—I just got dumped by an LLM and weirdly... I can’t let go.

Why don’t people reply to my messages?

Because you send questions like this.

—Now that’s a one-hit kill sentence.

—This hit me like a highlight reel of all my past conversations.

—That level of ego destruction should be illegal

—So sharp, it feels like laser surgery—but now my vision’s clearer

Each of MindChat’s responses spread as memes, and before long, it became a cultural phenomenon.

To Gooble, it was an unmistakable crisis.

—At this rate… they’ll become the symbol.

Symbols mattered.

Sometimes even more than the technology.

People remember images before they remember performance.

The first brand to be etched into the market is seen as the ‘original,’ while latecomers struggle to shake off the ‘copycat’ label.

Gooble couldn’t help but feel this was deeply unfair.

“They just entered the field and already...”

They had entered the market first and were even one step ahead technologically. Yet somehow, the most memorable symbolic moment had been snatched away by Stark.

It was a disaster born of carelessness.

Gooble had assumed that it was virtually impossible for a startup like Stark to build such overwhelming presence so quickly—and so they were caught completely off guard.

But there was a clear reason that impossibility had become reality—

“It was Next AI after all...”

Founded just last year, Next AI had already gathered the best minds in the industry, and with Ha Si-heon’s massive financial backing, it had become Gooble’s number one threat.

But they had let their guard down because of its “non-profit” label.

“To think they’d side with Stark...”

Technically speaking, the LLM revealed this time was entirely the work of Next AI.

Though it was branded as a ‘collaboration’ with Stark, the reality was that Stark had simply taken the technology that Next AI handed over and presented it on stage.

Gooble never saw this kind of partnership coming, and so they were blindsided.

But they couldn’t afford to keep taking hits.

“What’s the status of AlphaGo? Even with the timeline moved up, there must be no mistakes.”

“There are no issues.”

Gooble’s secret weapon—AlphaGo.

This was the moment they had to prove their true capabilities.

***

A few days later, the long-awaited AlphaGo match day arrived.

Ahead of the game, Gooble framed the match with this explanation:

[This showdown is an experiment to determine which is superior: human supervised learning or AI reinforcement learning.]

Supervised Learning vs. Reinforcement Learning.

Because the concepts were unfamiliar, Gooble used cooking as an analogy to make them easier to understand.

[Humans learn by studying pre-labeled data. It’s like cooking with a recipe—ingredients, measurements, and steps are all laid out, so even beginners can cook a decent meal by following it.]

[AlphaGo’s reinforcement learning is very different. Imagine being handed ingredients with no recipe or dish name. It learns by trial and error, gradually discovering the best flavor.]

Simply put: the match was between a chef who follows a recipe and one who cooks from scratch, blindly experimenting.

Which chef would create the better dish?

This historic match in Seoul captured the world’s attention.

Most people naturally expected humans to prevail.

—Of course it’ll be the human. Go is a game of intuition and insight.

—No way a machine can match that. The number of possibilities is insane...

—Then again, that fridge demo was impressive. I’m curious how far AI can really go.

But soon, shock and disbelief swept over them—

Contrary to expectations, Lee Sedol lost both the first and second games.

And in the third game, AlphaGo’s method of victory was absolutely jaw-dropping.

[Wait... what the hell was that move?]

The commentator’s voice trembled.

The move AlphaGo made was bizarre.

It placed a stone in a spot seemingly unrelated to the current board situation—offering no visible strategic value.

[That has to be a mistake, right?]

[It appears so. There’s no other way to explain it.]

The commentators, the spectators, even Lee Sedol himself dismissed it as an error.

But after dozens of turns—

AlphaGo completely seized control of the game.

Only then did the commentator cry out in disbelief.

[That move... it wasn’t a mistake! The entire current board state stemmed from that one move!]

Indeed.

AlphaGo had predicted the game would unfold this way and had laid the groundwork far in advance.

The commentator, unable to contain his excitement, exclaimed:

[Humans rely on learned experience, so we tend to ignore statistically rare moves. It’s like not knowing how to cook with rare ingredients.]

[AlphaGo, however, can cook with anything!]

[Its range of choices is far wider than a human’s. It sees options we can’t even imagine!]

In this structure, AlphaGo was overwhelmingly advantaged.

It could master countless recipes that humans didn’t even know existed.

Could humans possibly beat such an opponent?

But Lee Sedol wasn’t going down easily.

In the next game, he played a divine move—one even AlphaGo hadn’t predicted.

A daring stone placed in the heart of enemy territory before his own perimeter was secure—reckless on the surface.

But in the end, it was that very move that decided the game.

[Lee Sedol wins! He’s taken a game from AlphaGo, proving humanity’s strength!]

[That’s the kind of move only a human could make! AlphaGo would have ignored it due to its mere 0.0007% chance of success, but humans don’t shy away from impossible odds—they fight and sometimes win!]

Machines don’t attempt the impossible.

Humans, on the other hand, sometimes turn the impossible into reality.

That divine move by Lee Sedol was one such miracle.

Humanity exhaled in relief.

At least in this one domain, humans could still claim superiority.

But that relief didn’t last long.

The match ended 4–1 in AlphaGo’s favor, and people gradually began to feel dread at its overwhelming dominance.

—The end of the human era has arrived!

—Tremble in fear, flesh creatures.

—My toaster asked me this morning how “well done” I wanted my toast. I think it’s probing my weaknesses. 𝑓𝑟𝑒𝘦𝓌𝑒𝑏𝑛𝑜𝘷𝑒𝘭.𝒸𝘰𝑚

Top search trends after AlphaGo’s win: 1. AlphaGo 2. Go rules 3. How to survive the robot uprising 4. Phrases to survive robots (“Dear Robot Overlord, I conserve electricity” <- memorize this)

Humanity only held on for one move...

Even the media was in a frenzy.

<The Dawn of the Reinforcement Learning Era! What Is Left for Humans to Do?>

<AI Even Stealing Creativity? What Is the Future of Humanity?>

With this, Gooble reclaimed its symbolism.

Because AlphaGo created even stronger shockwaves than what Stark had previously unleashed.

At times like this, the one who delivers the stronger impact wins the symbolic ground.

Even those who once found AI’s mimicry of humans amusing were now trembling in fear, sensing that AI might overtake humanity in multiple domains.

And then, AI utterly crushed a world-class Go champion—so how could the shock not be overwhelming?

But for Gooble, this wasn’t enough.

At the following press conference, they went on the offensive instead of just defending.

“Is there any connection between Stark’s Transformer and LLM projects? Could there be any synergy between those technologies?”

Gooble firmly drew the line in response to the journalist’s question.

“No. AlphaGo is fundamentally different from LLMs like MindChat. LLMs are supervised learning models that compress and predict text patterns. AlphaGo, on the other hand, is a reinforcement learning model that interacts with its environment and learns through trial and error.”

Gooble made it clear: they were not the same.

They labeled LLMs and RL as fundamentally different paradigms and declared RL as the true future.

And that wasn’t all.

“LLMs are just statistical models that produce plausible sentences. They simply pick the most likely next word from massive amounts of text data. They have no goals or judgment—only predictions.”

“RL, in contrast, operates in goal-driven environments. It learns by colliding with reality, crafting strategies through trial and error, and constantly evolves to maximize rewards.”

This was a bold and aggressive dismissal.

In other words, LLMs were just smooth-talking machines, whereas RL represented true intelligence that could think and learn.

Gooble’s spokesperson even smiled internally during the announcement.

‘That Stark bastard must be sweating now...’

But that assumption was way off the mark.

At that very moment, Ha Si-heon was sitting in front of the TV with a quiet smile.

He was even humming while fiddling with pieces on a chessboard.

“Now it begins.”

***

My goal was an AI war.

And that war was finally entering Act Two.

Act Two was no longer just a scouting match.

This was when the real battle would begin.

And as with all wars, the conflict would evolve.

The first stage is a war of words.

Since my opponent was Stark, I didn’t even need to intervene—it unfolded on its own.

After Gooble trashed LLMs at the AlphaGo press conference, there was no way Stark would just sit by.

He didn’t wait for a press conference—he dropped a bomb directly via social media.

—I heard you won a board game. Congratulations.

By reducing Gooble’s victory over humanity to a mere board game, he continued:

—AlphaGo operated within a “perfectly defined environment” called the Go board. A world with clear rules. But reality isn’t like that. The world humans live in has no rules, tangled and conflicting objectives, and no right answers. True AI should be able to read context in such a reality, interpret intent, detect emotions, and understand others. That’s not what RL does—LLMs do that.

—If you want to become a world board game champion, I too recommend RL. But if you want AI that works in real life, LLMs are the answer.

Gooble responded quickly.

—We also deeply agree that AI should understand humans. But mimicking emotions is not the same as truly understanding reality.

—LLMs are very good at speaking. The issue is that they generate plausible-sounding but factually wrong statements. We call that “hallucination.”

They were directly attacking the known flaws in Stark’s LLM models.

Indeed, LLMs were often criticized for making up information.

So this was effectively a jab at one of Stark’s biggest vulnerabilities...

And yet I simply smiled.

Because this was the moment when the fight would evolve.

As always—when verbal debates hit an impasse, someone eventually throws a punch.

And now was that moment.

The time had come to act.

—Yes, LLMs sometimes say incorrect things. But isn’t it a bit premature to dismiss them just for that, when it’s barely been 100 days since their release? Especially when someone else has spent years only to win... a board game?

Stark struck back at Gooble again.

But even he knew that this point couldn’t be defended with just clever rhetoric.

So, he responded in a much more serious tone.

—The hallucination issue in LLMs is a solvable problem. With more data, better filtering, and stronger learning structures, it’s just a matter of time and scale.

Now was the time.

Time for action.

And Stark’s move was...

—Therefore, to achieve that scale, we’ve taken a bold new step. We’ve decided to pursue a partnership with... AWSS.

He had brought in a friend.

And who is the best friend to bring into a battlefield?

The enemy of my enemy.

AWSS fit the bill perfectly.

One of the largest cloud companies in the world, and a direct competitor of Gooble in cloud infrastructure.

—We plan to use AWSS’s global GPU resources to train our models at an unprecedented scale. This will dramatically reduce errors. By combining it with AWSS’s global infrastructure, we’ll improve LLM practicality and set a new standard for cloud-based AI services...

Of course, this move wasn’t just to provoke Gooble.

Stark’s weakness was infrastructure.

But by partnering with AWSS, he could completely eliminate that weakness.

With this collaboration, Stark not only secured stable computing resources but also gained access to AWSS’s vast enterprise customer network.

He now had the foundation to build an ecosystem comparable to Gooble’s.

Strategically, this was a highly impactful maneuver.

And by joining forces with Gooble’s rival, he even gained psychological superiority—a win-win.

But Stark didn’t stop there.

<Stark Signs Massive Supply Deal for Next-Gen HBM with HyNixon>

He had signed a massive long-term contract for high-bandwidth memory (HBM).

Not just for supply—this rare deal included rights to upcoming product lines.

And this sent shockwaves through the industry.

Usually, such contracts are reserved for massive legacy clients like Intil, Envid, or Gooble. For a newcomer like Stark to secure this was nearly unheard of.

Behind this success were his solid finances and supply chain networks built through previous businesses like Teslan.

But the contract came with one big problem.

HBM production was inherently low-volume and heavily concentrated among a few suppliers.

Since 2016, high-performance memory had been in chronic oversupply—and Stark’s preemptive deal would only accelerate the imbalance.

So what happens when a massive player like Stark monopolizes the supply?

Prices go up, and memory shortages ripple across the industry.

In short...

<Sonia Delays MSS Console Launch... Memory Supply Issues Unavoidable>

<Syscon and Other Server OEMs Enter Fierce Battle for High-Performance HBM>

“The roadmap of an entire industry is being shaken by a single startup,” insiders began to say.

And Stark’s stockpiling wasn’t limited to memory.

The most critical resource in AI was something else entirely.

GPUs.

<Stark Monopolizing GPUs Too?... Industry on Edge Over Surging Demand for Compute Power>

<Massive Order Secured Within a Week of Product Launch... Stark Fires First Shot in AI War>

Naturally—

With things escalating like this, Gooble couldn’t just sit still.

They too needed vast resources to improve reinforcement learning, and GPUs were the key.

<Competition for AI Compute Heats Up... Gooble Begins Pre-Orders for Tens of Thousands of GPUs>

<Gooble Falls Behind in Parser Architecture Orders, Scrambles to Secure Compute Infrastructure>

And thus began an all-out battle for resources!

It was around then that the message came.

[The board of directors is convening this Monday.]

An announcement from Envid’s board.

I checked the calendar.

“A bit early...”

It hadn’t even been three weeks since my promised timeline.